Case No In the United States Court of Appeals for the Ninth Circuit. MICHELLE FLANAGAN, et al., Plaintiffs-Appellants,

Transcription

1 Case: , 1/2/218, D: , DktEntry: 17-9, Page 1 of 292 Case No n the United States Court of Appeals for the Ninth Circuit MCHELLE FLANAGAN, et al., Plaintiffs-Appellants, v. XAVER BECERRA, et al., Defendants-Appellees. On Appeal from the United States District Court for the Central District of California (CV JAK-AS) APPELLANTS EXCERPTS OF RECORD VOLUME X OF X C. D. Michel Sean A. Brady Anna M. Barvir MCHEL & ASSOCATES, P.C. 18 East Ocean Blvd., Suite 2 Long Beach, CA 982 (562) cmichel@michellawyers.com Paul D. Clement Counsel of Record Erin E. Murphy KRKLAND & ELLS LLP 655 Fifteenth Street, NW Washington, DC 25 (22) paul.clement@kirkland.com Counsel for Plaintiffs-Appellants October 2, 218

2 Case: , 1/2/218, D: , DktEntry: 17-9, Page 2 of 292 NDEX TO APPELLANTS EXCERPTS OF RECORD VOLUME Dkt Date Document Description Page Amended Judgment Ruling Re Plaintiffs and Defendant s Motions for Summary Judgment Order Ruling on Plaintiffs Objections to the Declaration of P. Patty Li & Evidence Submitted in Support of Defendant s Motion for Summary Judgment (Dkt. 59) Order re Defendant s Motion to Dismiss Complaint for Declaratory and njunctive Relief VOLUME Plaintiffs Amended Notice of Appeal and Representation Statement Plaintiffs Notice of Appeal and Representation Statement Judgment Plaintiffs Objection to Defendant s Proposed Judgment Order Ruling on Defendant s Objections to Certain Other Evidence Filed in Opposition to Defendant s Motion for Summary Judgment (Dkt. 66) Order Ruling on Defendant s Amended Objections to Certain Evidence Filed in Support of Plaintiffs Motion for Summary Judgment (Dkt. 6) Plaintiffs Supplemental Brief Re Summary Judgment Standard for Competing Expert Evidence on Constitutional Questions Notice of Errata: Defendant s Amended Objections to Evidence Filed in Support of Plaintiffs Motion for Summary Judgment ii

3 Case: , 1/2/218, D: , DktEntry: 17-9, Page 3 of 292 Dkt Date Document Description Page Amended Judgment Defendant s Objections to Evidence Submitted by Plaintiffs in Opposition to Defendant s Motion for Summary Judgment Plaintiffs Request for Judicial Notice in Support of Reply to Opposition to Plaintiffs Motion for Summary Judgment; Declaration of Anna M. Barvir; Exhibits Declaration of John J. Donohue Re Defendant s Reply in Support of Motion for Summary Judgment Declaration of Jonathan M. Eisenberg Re Defendant s Reply in Support of Motion for Summary Judgment Exhibit 1 to the Declaration of Jonathan M. Eisenberg Exhibit 2 to the Declaration of Jonathan M. Eisenberg VOLUME Exhibit 3 to the Declaration of Jonathan M. Eisenberg Exhibit 4 to the Declaration of Jonathan M. Eisenberg Exhibit 5 to the Declaration of Jonathan M. Eisenberg Exhibit 6 to the Declaration of Jonathan M. Eisenberg Notice of Amendment to Cal. Penal Code Defendant s Amended Objections to Evidence Filed in Support of Plaintiffs Motion for Summary Judgment Plaintiffs Objections to the Declaration of P. Patty Li & Evidence Submitted in Support of Defendant s Motion for Summary Judgment iii

4 Case: , 1/2/218, D: , DktEntry: 17-9, Page 4 of 292 Dkt Date Document Description Page Exhibit 1 to Plaintiffs Objections to the Declaration of P. Patty Li & Evidence Submitted in Support of Defendant s Motion for Summary Judgment Exhibit 2 to Plaintiffs Objections to the Declaration of P. Patty Li & Evidence Submitted in Support of Defendant s Motion for Summary Judgment 5 54 VOLUME V Part 1 of Exhibit 3 to Plaintiffs Objections to the Declaration of P. Patty Li & Evidence Submitted in Support of Defendant s Motion for Summary Judgment VOLUME V Part 2 of Exhibit 3 to Plaintiffs Objections to the Declaration of P. Patty Li & Evidence Submitted in Support of Defendant s Motion for Summary Judgment Exhibit 4 to Plaintiffs Objections to the Declaration of P. Patty Li & Evidence Submitted in Support of Defendant s Motion for Summary Judgment Plaintiffs Additional Uncontroverted Facts & Conclusions of Law Excerpts of Plaintiff s Opposition to Defendant s Motion for Summary Judgment on Complaint for Declaratory and njunctive Relief VOLUME V Declaration of Sean A. Brady in Support of Plaintiffs Opposition to Defendant s Motion for Summary Judgment; Exhibits Defendant s Objections to Evidence Filed in Support of Plaintiffs Motion for Summary Judgment iv

5 Case: , 1/2/218, D: , DktEntry: 17-9, Page 5 of 292 Dkt Date Document Description Page Plaintiffs Statement of Uncontroverted Facts and Conclusions of Law in Support of Plaintiffs Motion for Summary Judgment VOLUME V Plaintiffs Request for Judicial Notice in Support of Motion for Summary Judgment; Declaration of Anna M. Barvir in Support; Exhibits Memorandum of Points and Authorities in Support of Plaintiffs Motion for Summary Judgment Declaration of Sean A. Brady in Support of Plaintiffs Motion for Summary Judgment Declaration of Rick Travis in Support of Plaintiffs Motion for Summary Judgment Declaration of Jacob Perkio in Support of Plaintiffs Motion for Summary Judgment Declaration of Dominic Nardone in Support of Plaintiffs Motion for Summary Judgment Declaration of Samuel Golden in Support of Plaintiffs Motion for Summary Judgment Declaration of Michelle Flanagan in Support of Plaintiffs Motion for Summary Judgment Defendant s Uncontroverted Facts and Conclusions of Law in Support of Defendant s Motion for Summary Judgment Defendant s Request for Judicial Notice in Support of Defendant s Motion for Summary Judgment Exhibit 1 to Defendant s Request for Judicial Notice Exhibit 2 to Defendant s Request for Judicial Notice 155 VOLUME V Exhibit 3 and Part 1 of Exhibit 4 to Defendant s Request for Judicial Notice 1624 v

6 Case: , 1/2/218, D: , DktEntry: 17-9, Page 6 of 292 Dkt Date Document Description Page Part 2 of Exhibit 4 to Defendant s Request for Judicial Notice Declaration of P. Patty Li in Support of Defendant s Motion for Summary Judgment Exhibit 1 to the Declaration of P. Patty Li Exhibit 2 to the Declaration of P. Patty Li Exhibit 3 to the Declaration of P. Patty Li Exhibit 4 to the Declaration of P. Patty Li Exhibit 5 to the Declaration of P. Patty Li Exhibit 6 to the Declaration of P. Patty Li 1862 VOLUME X Exhibit 7 to the Declaration of P. Patty Li Part 1 of Exhibit 8 to the Declaration of P. Patty Li Part 2 of Exhibit 8 to the Declaration of P. Patty Li Part 1 of Exhibit 9 to the Declaration of P. Patty Li Part 2 of Exhibit 9 to the Declaration of P. Patty Li Exhibit 1 to the Declaration of P. Patty Li Exhibit 11 to the Declaration of P. Patty Li 2128 VOLUME X Answer of Defendant (Kamala D. Harris) to Complaint Excerpts of Plaintiffs Omnibus Opposition to Defendants Motions to Dismiss Complaint Complaint for Declaratory and njunctive Relief 2195 ** District Court Docket 2216 vi

7 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 1 Page of 517 of Page 292 D #:56 EXHBT 7 Li Decl. Ex

8 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 2 Page of 518 of Page 292 D #:561 Expert Report of John J. Donohue Flanagan v. Becerra, United States District Court (C.D. Cal.), Case No. 2:16-cv-6164-JAK-AS June 1, 217 BACKGROUND AND QUALFCATONS 1., John J. Donohue, am the C. Wendell and Edith M. Carlsmith Professor of Law at Stanford Law School. After earning a law degree from Harvard and a Ph.D. in economics from Yale, have been a member of the legal academy since have previously held tenured positions as a chaired professor at both Yale Law School and Northwestern Law School. have also been a visiting professor at a number of prominent law schools, including Harvard, Yale, the University of Chicago, Cornell, the University of Virginia, Oxford, Toin University (Tokyo), St. Gallens (Switzerland), and Renmin University (Beijing). 2. For a number of years have been teaching at Stanford a course on empirical law and economics issues involving crime and criminal justice, and have previously taught similar courses at Yale Law School, Tel Aviv University Law School, the Gerzensee Study Center in Switzerland, and St. Gallen University School of Law in Switzerland. have consistently taught courses on law and statistics for two decades. 3. am a Research Associate of the National Bureau of Economic Research, and a member of the American Academy of Arts and Sciences. was a Fellow at the Center for Advanced Studies in Behavioral Sciences in 2-1, and served as the co-editor (handling empirical articles) of the American Law and Economics Review for six years. have also served as the President of the American Law and Economics Association and as Co-President of the Society of Empirical Legal Studies. 4. am also a member of the Committee on Law and Justice of the National Research Council ( NRC ), which reviews, synthesizes, and proposes research related to crime, law enforcement, and the administration of justice, and provides an intellectual resource for federal agencies and private groups. (See online for more information about the NRC.) 5. My research and writing uses empirical analysis to determine the impact of law and public policy in a wide range of areas, and have written extensively about the relationship between rates of violent crime and firearms regulation. My complete credentials and list of publications are stated in my curriculum vitae, a true and correct copy of which is attached as Exhibit A. 6. filed an expert declaration in each of two cases involving a National Rifle Association ( NRA ) challenge to city restrictions on the possession of high-capacity magazines: Fyock v. City of Sunnyvale, United States District Court (N.D. Cal.), January 214. Herrera v. San Francisco, United States District Court (N.D. Cal.), January 214. also filed an expert declaration in a case involving a challenge by NRA to Maryland s restrictions on assault weapons and high-capacity magazines: Tardy v. O Malley, United States District Court (District of Maryland), February Li Decl. Ex

9 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 3 Page of 519 of Page 292 D #:562 n all cases, the relevant gun regulations have (ultimately) been sustained in the relevant federal appellate courts. 7. am charging a total of $21,25 to the California Department of Justice for preparation of this expert report. will charge $85 per hour for deposition testimony, and $5 per hour for trial testimony, in connection with being an expert witness in the above-entitled case. SUMMARY OF CONCLUSONS 1. A considerable body of credible statistical evidence based on both panel data analysis and the use of synthetic controls finds that the adoption of right-to-carry ( RTC ) laws (sometimes called concealed-carry laws or CCW laws ), permitting individuals otherwise allowed to possess firearms to carry them concealed on their bodies in public places, leads to increases in overall violent crime. 1 Earlier panel data studies that purported to find different results are less reliable because they have not analyzed the full array of data through 214, which have analyzed, or because the earlier panel data studies are marred by specification or other econometric problems Given that the best statistical evidence suggests that the adoption of right to carry laws leads to statistically significant increases in violent crime, it is a sound, evidenced-based, and longstanding crime-fighting strategy for U.S. state and local governments to place substantial limits on the carrying of concealed weapons in public While the vast bulk of the empirical literature on the impact of gun carrying on crime has focused on laws facilitating the concealed carry of weapons, one can use this literature to draw inferences about the likely consequences of allowing the open carry of guns. n general, there is no reason to think that the social harm from gun carrying imposed by RTC laws that was just referenced would be lower under a regime allowing open carry of guns. ndeed, there are valid reasons to believe that a policy of lawful open carry could impose even greater social costs in terms of further facilitating criminal activity, burdening the police, and elevating citizen fear and anxiety. 1 Panel data analysis has been the primary tool for evaluating the impact of law and policy interventions for at least the last 3 years. Synthetic controls is a newer technique designed to better approximate the type of treatment and control analysis that would be found in a randomized study. Further details of both are discussed below. 2 To generate credible results, panel data evaluations must be conducted according to sound statistical practices. f the models used do not have the appropriate mathematical form or do not capture the appropriate explanatory variables, then they would not be deemed to have an appropriate specification. Since much of the development of the elements of modern panel data analysis came from economists who were trying to perfect these tools, any violations of the best practices in conducting such studies are often referred to as econometric problems. 3 All of our estimates of the impact of RTC laws on crime are accompanied by measures designed to gauge whether the results are likely to be caused by chance variations as opposed to a true causal effect of the RTC law. f the estimated effect is large relative to the likely chance variation in crime, then we deem the estimate to be statistically significant. 2 Li Decl. Ex

10 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 4 of 51 of Page 292D #:563 DSCUSSON Background on Panel Data Models 1. There is a very substantial literature on the issue of the impact of laws allowing citizens to carry concealed handguns. My first published article in this literature appeared 18 years ago, 4 and the latest of my 11 articles in this area was just issued as a National Bureau of Economic Research working paper this month (attached as Exhibit B) Virtually all of the published literature on this question has employed an econometric approach referred to as a panel data model with state and year fixed effects. Panel data refers to the fact that the researcher will have crime data over a period of years for many different states (or counties or cities), which can then be analyzed to test whether some legal or policy intervention (such as the adoption of an RTC law) leads to a change in crime that is not seen in states that do not experience that legal or policy intervention (i.e., do not adopt RTC laws) Panel data models can be useful to examine a change adopted by selected states (preferably at different times) so that one can compare what happens in the states that adopt the legal change to the states that do not adopt the legal change. This is an appealing empirical strategy because it allows the researcher to separate the data into the treated group, which is the set of states that adopts the law during the relevant data period, and the set of all other states, which serves as a type of control. Nonetheless, it is now well-known that panel data crime estimates of crime can be inaccurate if they are not undertaken with meticulous care and substantial econometric sophistication. The Most Up-to-Date Panel Data Estimates of the mpact of RTC Laws on Violent Crime 4. Despite some initial claims that RTC laws could actually reduce violent crime, the 24 report of a special committee the National Research Council ( NRC ; with only one dissenter out of 16 committee members) emphatically rejected this conclusion based on the committee s review of the then-current information with data through 2. 7 Noting that the estimated effects of RTC laws were highly sensitive to the particular choice of explanatory variables, the report concluded that the evidence was too uncertain to determine the impact of these permissive gun laws on crime. The 4 an Ayres and John Donohue, Nondiscretionary Concealed Weapons Law: A Case Study of Statistics, Standards of Proof, and Public Policy, American Law and Economics Review 436 (1999). 5 John Donohue, Abhay Aneja, and Kyle Weber, Right-to-Carry Laws and Violent Crime: A Comprehensive Assessment Using Panel Data and a State-Level Synthetic Controls Analysis, NBER Working Paper ( DAW ) May 217. A true and correct copy of this working paper is attached as Exhibit B. We plan to have the article published in a peer-reviewed journal. 6 The state fixed effects simply capture the fact that some states have enduringly lower or higher rates of violent crime (for reasons that may not be fully reflected in the explanatory variables that are available to the researcher). The year fixed effects are designed to capture the common movements that occur in all states each year owing to factors that operate nationally (and which again may not be fully reflected in the explanatory variables that are available to the researcher). 7 National Research Council. Firearms and Violence: A Critical Review (Washington: National Academies Press, 24). 3 Li Decl. Ex

11 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 5 of 511 of Page 292D #:564 Committee suggested that more data and new and better statistical techniques would be necessary to resolve this uncertainty. 5. Since then, 14 more years of data with 11 more states adopting RTC laws have improved the previous panel data estimates. n addition, new statistical techniques have enabled much more compelling and consistent evidence on the impact of RTC laws on crime to emerge. 6. The best evidence now shows that RTC laws substantially increase violent crime rates, so that, ten years after adoption, an RTC state is estimated to have a percent higher rate of violent crime than it would have had if no RTC law had been adopted. A violent crime increase of this magnitude is obviously a major burden on a state and its citizens, and given current estimates of the elasticity of incarceration with respect to crime, a state would need to double its prison population to offset the violent crime increase imposed by RTC laws Many of the early studies that tried to estimate the impact of RTC laws typically using panel data for all states across an extended period of time were undermined by the fact that the period from 1985 through the early 199s was anomalous. Over that span, violent crime rose sharply in certain areas, such as California, New York, and the District of Columbia, owing to the introduction of crack cocaine. Since all three of those jurisdictions and a number of other states with the worst crack problems did not adopt RTC laws, any panel data analysis that could not properly control for the criminogenic influence of crack would necessarily generate a biased estimate of the impact of RTC laws that would make them appear to be less harmful (or more beneficial) than they actually were in influencing crime. 8. This was a major problem for the original Lott and Mustard study and in fact plagues every panel data analysis of RTC laws, except for those that started after the impact of crack had been fully dissipated (in the very late 199s or early 2s) One quick but admittedly crude way to address this problem is to present a difference-indifferences comparison between the 37 states that adopted RTC laws over the period and the nine states (including the District of Columbia) that did not adopt these laws. By comparing the change in violent crime from a period before crack emerged to a year after its impact had dissipated, one can eliminate the impact of crack on crime (although of course this simple comparison does not control for other influences on crime that differed over this period for the two sets of states). Figure 1 shows that the nine non-rtc states enjoyed a 42.3 percent drop in their violent crime rate, while the 37 RTC-adopting states had a sharply smaller decline in violent crime over this period (a decline of only 8.7 percent over a 37-year period). The five states that had adopted RTC laws prior to 1977 similarly showed far smaller drops in crimes than the nine never-adopting states. This graphical display provides suggestive evidence that RTC laws tend to exacerbate violent crime (controlling for the influence of crack, albeit not for other explanatory variables). 8 Current consensus estimates suggest that doubling the incarceration rate will lead to a roughly 15 percent reduction in crime (which means the elasticity of crime with respect to incarceration is.15). Since RTC laws generate about a 15 percent increase in violent crime, one could offset this increase by doubling the prison population. See generally John J. Donohue, Assessing the Relative Benefits of ncarceration: The Overall Change Over the Previous Decades and the Benefits on the Margin, in Steven Raphael and Michael Stoll, eds., Do Prisons Make Us Safer? The Benefits and Costs of the Prison Boom (29). 9 See the discussion of Zimmerman (215) below, which supports my finding that RTC laws increase crime. 4 Li Decl. Ex

12 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 6 of 512 of Page 292D #: Obviously, Figure 1 would exaggerate the extent of the harm of RTC laws if, say, the nonadopting states had increased their per capita rates of either incarceration or police personnel more than the adopting states, thereby suppressing violent crime through those mechanisms (which could then potentially explain the relatively better experience with violent crime over the period in the non-adopting states). n fact, the opposite is true. The states adopting RTC laws in the period had considerably larger percentage increases over this span in their rates of incarceration (259% vs. 25%) and police staffing (55% vs. 16%) relative to the non-adopting states. 1 The relatively better crime performance of non-adopting states in the raw comparison of Figure 1 would be even greater if one were to control for the influence on violent crime of both police and incarceration. 11. Of course, many factors in addition to police, incarceration, and crack influence crime and the challenge for researchers who seek to find the impact of a single factor such as RTC laws is to develop an appropriately specified statistical model that accounts for those factors that may also be 1 The five states that had adopted RTC laws prior to 1977 had the largest percentage increase in their incarceration rate (262 percent), and a 38-percent growth in size of their police forces (per capita), which was more than double the growth in the police forces in the states not adopting RTC laws. 5 Li Decl. Ex

13 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 7 of 513 of Page 292D #:566 correlated with RTC adoption. DAW (see footnote 2, supra, for full citation) begin their study with a state panel data analysis that extended the NRC data by 14 years, during which time 11 additional states adopted RTC laws. Table 1, below (which reproduces Table 4 in DAW), provides state panel data estimates of the impact of RTC laws for the period using state and year fixed effects, with both a dummy variable model (estimating a shift in the level of crime after RTC adoption) and a spline model (examining whether RTC laws induce a change in the trend of crime). Table 1 suggests that violent and property crime are both higher after adoption of RTC laws. Specifically, the dummy model suggests that violent crime is 9.5 percent higher after adoption of RTC laws, while the spline model results are not statistically significant DAW then show that if one conducts a panel data analysis over this extended data period from based on another major crime study that uses different but still plausible explanatory variables, one generates almost identical results. Specifically, taking the variables from the 215 Brennan Center report (BC) on the crime decline and using the longer data time frame of 11 Both models control for the same additional factors that influence crime, such as police, incarceration, state income and unemployment, the demographics of the population, etc. (The full array of variables are set forth in DAW, Table 3.) The dummy variable model of Table 1 a dummy variable is just an indicator of whether a state has an RTC law or not estimates that on average crime is 9.49 percent higher after RTC adoption, holding other factors constant. The spline model of Table 1 is harder to interpret because it suggests that for each year over the entire 36-year data period, RTC states were experiencing a relative increase in violent crime relative to non-rtc states for reasons not well captured by the panel data model (and that this enduringly worse crime performance did not get significantly worse after adoption). t seems unlikely that an adverse linear trend could be projected to continue for so long without being buttressed by further exacerbating factors, which implies that the spline model is not informative about how much of the posttreatment increases in violent crime is caused by the consequences of RTC laws versus pre-existing but unknown attributes of the 33 states that happen to go on to adopt RTC laws. This suggests that the more precise identification of appropriate control comparisons for each treated (RTC-adopting) state that the synthetic controls approach provides is likely to yield superior pre-treatment equivalence, thereby better facilitating superior insight into the causal impact of RTC laws on crime. 6 Li Decl. Ex

14 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 8 of 514 of Page 292D #:567 the DAW data period generates almost identical results, predicting an 11. percent increase in the rate of violent crime with the dummy variable model employed by BC. 12 These results are shown in Table Another post-nrc panel data crime study that also finds an association of RTC laws with higher rates of violent crime is Zimmerman (215), which uses a state panel data set from The advantage of using this data period to explore the impact of RTC laws is that it largely avoids the problem of omitted variable bias owing to the crack phenomenon, since the crack effect had ended by The disadvantage is that one has less data and can derive estimates based only on the eight states that adopted RTC laws over that twelve-year spell. 14 Zimmerman describes his finding as follows: The shall-issue coefficient takes a positive sign in all regressions save for the rape model and is statistically significant in the murder, robbery, assault, burglary, and larceny 12 See Donohue, Aneja, and Weber, supra, Table 5.A (the same as Table 2 here). The official citation for the Brennan Center (the BC ) report is Roeder, et al., What Caused the Crime Decline?, Columbia Business School Research Paper No (Feb. 12, 215). t conducted its analysis on the abbreviated period from , while we use more complete data from the late 197s through 214. Although the BC report did not estimate a spline model, Table 2 shows spline model estimates for four crime measures and again shows the same statistically insignificant results of the DAW model that are discussed in footnote Because crack caused sharp crime increases in largely non-rtc states starting in the mid-198s, and we do not have good measures to control for the criminogenic influence of crack, this variable is omitted from the Table 1 and 2 panel data estimates and likely biases the results in a way that obscures the criminogenic effects of RTC laws. 14 The relatively short time span makes the assumption of state fixed effects more plausible, but also limits the amount of pre-adoption data for an early adopter such as Michigan (21), with only one year of data prior to adoption, and the amount of post-adoption data for the late adopters Nebraska and Kansas (both 27), leaving only three years after adoption to estimate the impact of the law. 7 Li Decl. Ex

15 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 9 of 515 of Page 292D #:568 models. These latter findings may imply that the passage of shall-issue laws increases the propensity for crime, as some recent research (e.g., Aneja, Donohue, & Zhang, 212) has suggested The next two tables (both part of Table 3) show further panel data confirmation of the Zimmerman findings using the DAW and BC specifications to estimate the impact of RTC laws on crime over the period. Both tables buttress the view that the states adopting RTC laws after 2 suffered from relatively higher violent crime in the aftermath of doing so. Synthetic Controls Estimates of the mpact of RTC Laws on Violent Crime 15. n addition to the benefit from having more years of data and more RTC adoptions to provide further evidence of the true impact of RTC laws on crime, we now have a powerful new and already widely implemented statistical tool to assess the impact of a legal change the synthetic controls analysis, first introduced in Abadie and Gardeazabal (23) and expanded in Abadie, et al. (21) and Abadie, et al. (214). 16 We have already alluded to some of the difficulties with the panel data methodology because of the inability to effectively control for the influence of crack and other 15 DAW also found that running their preferred model on post-crack-era data from generated an estimate that murder rates rose about 1.1 percentage points each year that an RTC law was in effect. 16 A. Abadie, A. Diamond, and J. Hainmueller, Synthetic Control Methods for Comparative Case Studies: Estimating the Effect of California s Tobacco Control Program, 15 Journal of the American Statistical Association 49, (21). 8 Li Decl. Ex

16 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 1 Page of of Page 292 D #:569 (known and unknown) explanatory variables that influence crime. 17 The synthetic controls approach is designed to address some of the challenges posed by panel data methods. The goal of this research is to find a set of states that do not have RTC laws that can be used to generate a good approximation of the pattern of crime prior to year X for a state that adopts a RTC law during that year. The pattern of crime for the synthetic control after year X can then be compared with the actual pattern of crime for the adopting state to generate an estimate of the causal impact of the RTC law adoption. 16. Figures 2 through 5 illustrate the synthetic controls estimates of the impact of RTC laws on violent crime for four different states, beginning with Texas. As one can see in Figure 2, a weighted average of three states California, Wisconsin, and Nebraska mimics the time-series of violent crime in Texas rather well over the period from 1977 through adoption of the Texas RTC law in Although all states were benefitting from crime reductions in the 199s, the fact that California, Wisconsin, and Nebraska did not have RTC laws led to their better performance in reducing violent crime. t is the comparison between the better performance of the synthetic control (the composite of California, Wisconsin, and Nebraska) with the actual performance of the state itself that generates the synthetic controls estimate that Texas s RTC law elevated crime by 16.6 percent above what it would have been if the state had not been burdened by an RTC law. 17. Similar results can be seen in the successive figures for Pennsylvania, North Carolina and Mississippi. t is worth noting in Figure 3 that Pennsylvania adopted its RTC laws in two phases (as indicated in the two vertical lines), with the second phase extending RTC to Philadelphia in Note that violent crime noticeably moved in an adverse direction in the wake of that legislative enactment. The attached paper by DAW (see Appendix D) also shows that when in 23 Alaska moved from an RTC permitting system to a regime of unencumbered RTC, violent crime jumped noticeably. 17 As alluded to above, the assumptions of panel data analyses needed to ensure validity can be demanding. One can question whether it is plausible to assume implicitly that states like New York can serve as good controls for treatment states like South Dakota, that state fixed effects remain fixed over a 36-year data period, and that linear trends can be projected far into the future (as in the spline models run on the full 36-year period). 9 Li Decl. Ex

17 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 11 Page of of Page 292 D #:57 Figure 2 Figure 3 1 Li Decl. Ex

18 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 12 Page of of Page 292 D #:571 Figure 4 Figure 5 11 Li Decl. Ex

19 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 13 Page of of Page 292 D #: While the individual state synthetic controls estimates are interesting, a more useful estimate comes from generating an average estimated effect of RTC laws on crime, since the average across 33 states will eliminate some of the noise (or random variation) in the state-specific estimates. 18 DAW use the synthetic controls approach over the full period and find that, on average, RTC laws increase violent crime by percent ten years after adoption. mportantly, the synthetic controls estimates do not show the sensitivity to specification changes that concerned the NRC panel. Whether one uses the specification of DAW, the BC, Lott and Mustard, or Moody and Marvel, the synthetic controls estimate that RTC laws generate large increases in violent crime is highly robust. These results strongly support the panel data estimates cited above, so a body of evidence using the most complete data and different statistical approaches has now emerged that RTC laws tend to cause harmful increases in violent crime in the first ten years after adoption. 18 As Nate Silver writes, The signal is the truth. The noise is what distracts us from the truth. The Signal and the Noise 17 (New York: Penguin Press, 212). Thus, the average across over 3 estimates will bring us closer to the truth than any single observation because averaging cancels out some noise, allowing the signal to emerge more clearly. 12 Li Decl. Ex

20 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 14 Page of 51 2 of Page 292 D #: Li Decl. Ex

21 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 15 Page of of Page 292 D #:574 The Mechanisms by Which RTC Laws ncrease Violent Crime 19. The prior discussion of the panel data evidence and particularly the consistent synthetic controls analysis illustrates that, on balance, RTC laws seem to increase violent crime. This implies that even if there are any benefits from private gun carrying, they are substantially outweighed by their costs. t is important to understand some of the mechanisms by which adoption of RTC laws lead to increased crime, which the statistical studies do not directly address. 2. First, the supporters of RTC laws frequently cite evidence that permit holders are arrested for violent crimes at relatively low rates. Since permit holders have been screened to try to remove known felons, it is not surprising that they will tend to have overall lower crime rates than the broader population that does include convicted felons. But claims about the prevalence of misconduct by permit holders need to be understood in context. For example, the fact that NRAsponsored laws have been passed to shield information about the permit status even of arrested criminals has made it hard to fully assess how often permit holders engage in criminal activity. 21. Advocates for more permissive gun-carrying even among government officials are often highly inaccurate in their claims about the behavior of permit holders. For example, in a 213 speech, Sacramento County Sheriff Scott Jones told his audience that no one has ever been shot by a holder of a concealed weapons permit issued by this office. Yet only a few months earlier, on October 31, 212, a letter signed by Jones revoked the permit his office had granted to Hun Chu Saelee, after Saelee had shot a college student in the head at a Halloween party a few days earlier. 19 ndeed, since incidents involving the unlawful use of weapons, shootings, killings, and other violent crime by permit holders in California are not systematically collected and reported, any claims about the frequency of these events in California based on a reported number is necessarily an undercount of the true incidence. 22. The evidence that we do have from some early work in Texas when the number of RTC permits was far lower than it is today suggests that the involvement of permit holders in aggravated assault and murder is many times higher than we might expect for a group with their demographic configuration. 2 RTC permit holders in the early days in Texas had very low levels of criminal disposition as shown by their low arrest rates for rape --.3 per 1, CHL holders, compared to 12 per 1, people for total Texas -- and for robbery --.7 per 1, CHL holders, compared to 35 per 1, people for total Texas. n other words, the criminal propensity in the permitholding group was about 1/36 the average Texan (if we use rape as the baseline or even lower if we 19 John Donohue, Be Skeptical About Claims Of Benefits Of Concealed Carry Permits, Sacramento Bee (Oct. 6, 216) (available online at < 2 Sturdevant, An Analysis Of The Arrest Rate Of Texas Concealed Handgun License Holders, (2) documents the arrest record for Texas concealed handgun license holders in the initial four years after Texas adopted its RTC law in See 14 Li Decl. Ex

22 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 16 Page of of Page 292 D #:575 use robbery). But now consider the involvement of these otherwise law-abiding citizens for the two crimes in which guns can play the biggest role: aggravated assault and murder. The arrest rate for aggravated assault in the Texas study was 56 per 1, permit holders, compared to 121 per 1, people for total Texas. Since 1/36 of 121 = 3.4 per 1,, it is clear that RTC permit holders commit assault at a highly disproportionate rate compared to the crime of rape. Specifically, early Texas permit holders committed aggravated assault at about 17 times the rate we would expect from their preferred demographic (as reflected by their underlying likelihood of criminality for the crime of rape). 23. Similarly, for the period of , there were 27 arrests of Texas permit holders for murder a rate of 4 arrests per 1, permit holders. n comparison, the total Texas arrest rate for murder during the same time period was 5.2 per 1, people. Since 1/36 of 5.2 =.144, we see that early Texas RTC permit holders were 27.7 times more likely to commit murder than we would expect based on their underlying level of criminality. 24. Some advocates of RTC cite the low numbers of revocations of licenses as proof that the concealed carry permit holder are exceptionally law-abiding and no threat to anyone. But revocations will clearly understate the misconduct of permit holders since many crimes are never solved. Moreover, the cases in which RTC permit holders have been killed during criminal behavior (thereby obviating the need to revoke their permits) shows that permit revocation and/or arrest is not a flawless reflection of misconduct. For example, on September 18, 213, two Michigan drivers with RTC permits -- James Pullum (43 years old), driving with his wife and mother, and Robert Taylor (56) were angered over a tailgating dispute. They pulled into a nearby car wash, stepped out of their cars and exchanged fire. Both were hit and died at the same hospital later that day. Neither of their RTC permits were revoked by the State of Michigan naccurate claims are often made that permit holders play an important role in stopping mass shootings, but a 213 FB study of 16 active shootings over a 13-year period when RTC laws were prevalent found that only one was stopped by an armed individual who was neither a police officer nor a security guard. But even this case in 28 in a Nevada bar was not any average permit holder but an active-duty Marine, who was able to kill the shooter who had stopped to reload. Note, also, that the FB study found that 21 times unarmed citizens disarmed the shooter. 26. Since 27, the website Concealed Carry Killers has documented 885 homicides, accidental deaths, and suicides attributed to permit holders, including 29 mass shootings that killed 139 individuals, including the Orlando shooter who murdered 49. How many more deaths were caused by permit holders is difficult to know because the NRA-backed secrecy laws are designed to keep the public from knowing the full extent of this mayhem. 21 Hunter Stuart, Two Concealed Carry Holders Kill Each Other n Road Rage ncident, Huffington Post, Sept. 19, 213 (available online at < 15 Li Decl. Ex

23 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 17 Page of of Page 292 D #: A second critical consideration must also be acknowledged: since the statistical evidence shows that RTC laws increase crime by an amount larger than that committed by RTC permit holders, then RTC laws must be increasing crime in many ways even if the permit holders are not committing it. The ability to carry a gun may embolden some permit holders to engage in provocative behavior inciting criminal responses in others, as some have alleged happened in the well-known George Zimmerman case leading to the death of Trayvon Martin. Criminals may also be more likely to carry weapons in response to RTC adoption and more likely to be aggressive towards their victims if they fear armed opposition. Guns carried outside the home because of RTC laws are potentially more likely to be lost or stolen, which is a major pathway to arming criminals. 28. Finally, the presence of more guns can complicate the job of police and simply take up more police time as they process applications and check for permit validity when they confront armed citizens The primary cause of death to police officers from intentional assaults is from guns (which is the plurality cause of death for police officers), and the nature of the threat is reflected in the fact that states with high rates of civilian gun ownership are more dangerous for the police. A study published in the American Journal of Public Health examined data on the number of homicidal deaths of police in two groups of states with roughly equal number of police officers the eight states with the lowest levels of gun ownership and the 23 states with the highest rate of gun ownership. 23 The study found that, over the period from 1996 to 21, the rate of police homicide in the high-gun-prevalence states was three times as high as the rate of police homicide in the low-gunprevalence states. Specifically, in the states with low-gun-ownership rates, there were.31 officer fatalities for every 1, employed officers over the study period, and contrasted with.95 fatalities per 1, officers in the high-gun-ownership states. Not surprisingly, only one of the lowgun-prevalence states had an RTC law during the study period (Connecticut), while 21 of the 23 states with the high level of murders of police had RTC laws during all or part of the study period (and the other two owa and Wisconsin both adopted RTC laws in 211) Anything that impairs police productivity or that serves as an effective tax on police serves to elevate criminal behavior, and RTC laws do both as police have to spend valuable time processing permits and contending with armed citizens. The more fear that police have, the greater the threat 22 Christopher ngraham, More Police Officers Die On The Job n States With More Guns, Wash. Post, Jul. 8, 216, (available online at < 23 Note that, since the low-gun-ownership states had higher populations, the effort to equalize the numbers of police in the two groups required looking at more high-gun-ownership states, which tended to have lower populations. 24 As the authors note: Because the low-prevalence states were typically more highly populated and had many more officers than the high-prevalence states, the final 2-by-2 analysis had 8 low-prevalence states and 23 high-prevalence states, covering approximately 2.75 million LEO-years per group. David. Swedler, et al., Firearm Prevalence and Homicides of Law Enforcement Officers in the United States, 15 American Journal of Public Health 242 (Oct. 215). 16 Li Decl. Ex

24 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 18 Page of of Page 292 D #:577 they will pose to the community, which can strain the bonds that are needed for effective law enforcement. Open Carry Versus Concealed Carry 32. While the empirical literature discussed above has largely focused on the impact of laws allowing citizens to carry concealed guns, this literature can be used to make informed predictions about the likely impact of allowing citizens to carry arms openly. 33. The concealed carry of guns has a potential advantage of deterring criminal conduct if it leads criminals to fear that someone they are considering to mistreat may be armed. But the prior discussion shows that whatever deterrence is generated by RTC laws, it is apparently outweighed by the factors just discussed that tend to encourage violent crime. These facts suggest that open carry of guns would be less socially desirable than concealed carry since the latter at least has the prospect of a deterrence since the criminals cannot know who is carrying weapons. 34. Open carry might conceivably confer a benefit if it could dissuade potential criminals from targeting certain individuals if they or someone nearby has a weapon, but in general the greatest effect of open carry would likely only be to move crime away from the armed target to an unarmed target. n general, spending resources that shift burdens of crime from one group to another without reducing the overall burden is a net waste of resources. ndeed, the billions of dollars that are spent each year buying guns for self-protection without any statistical support for the claim that they diminish crime could easily confer substantial crime-reducing benefits if the money were directed to known crime-reducing expenditures Moreover, as we saw in the Boston Marathon bombing case, criminals can easily target open carriers of guns, either to eliminate the threat from gun carriers or to help the criminals secure a weapon (as the Boston bombers hoped to do when they killed an MT policemen to obtain his gun). 36. n addition, open carry of guns can spread fear and alarm in the community. An openly displayed gun in public also gives a muddy signal about the gun toter and could draw undue attention from police officers, directing law-enforcement resources inefficiently, which again makes law enforcement less effective, thereby further promoting crime. Respectfully submitted, 25 John Donohue, Fighting Crime: An Economist s View, 7 The Milken nstitute Review 46 (25). 17 Li Decl. Ex

25 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 19 Page of of Page 292 D #:578 Exhibit A JOHN J. DONOHUE Stanford Law School Stanford, CA 9435 Phone: donohue@law.stanford.edu Web pages: EMPLOYMENT Full-time Positions Stanford Law School, C. Wendell and Edith M. Carlsmith Professor of Law, September 21 to the present. Yale Law School, Leighton Homer Surbeck Professor of Law, July 24 to August 21. Stanford Law School, Professor of Law, September 1995 to June William H. Neukom Professor of Law, February 22 June John A. Wilson Distinguished Faculty Scholar, March 1997 January Academic Associate Dean for Research, since July 21 July Stanford University Fellow, September 21 May 23. Northwestern University School of Law: - Class of 1967 James B. Haddad Professor of Law, September 1994-August Harry B. Reese Teaching Professor, Professor of Law, May 1991-September Associate Professor, May 1989-May Assistant Professor, September 1986-May Research Fellow, American Bar Foundation, September 1986-August Associate Attorney, Covington & Burling, Washington, D.C., October 1978-July 1981 (including last six months as Attorney, Neighborhood Legal Services) Law Clerk to Chief Justice T. Emmet Clarie, U.S. District Court, Hartford, Connecticut, September 1977-August Temporary Appointments Fellow of the Society for Empirical Legal Studies, Visiting Professor, Bocconi University, Milan, taly, October- November 212, April 214, and June Li Decl. Ex

26 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 2 Page of of Page 292 D #: Faculty Scholar in Residence, University of Denver Sturm College of Law, April 21-22, 211. Visiting Fellow, The Milton Friedman nstitute for Research in Economics, University of Chicago, October 29 Schmidheiny Visiting Professor of Law and Economics, St. Gallen University, November December, 27. Visiting Lecturer in Law and Economics, Gerzensee Study Center, Switzerland, June 27. Visiting Professor, Tel Aviv University School of Law, May 27. Herbert Smith Visitor to the Law Faculty, University of Cambridge, England, February 26. Visiting Professor, Harvard Law School, January 23. Fellow, Center for Advanced Studies in the Behavioral Sciences, Stanford, California, Academic year 2-1. Visiting Professor, Yale Law School, Fall, Professor, Center for the Study of American Law in China, Renmin University Law School, Beijing, July Visiting Professor of Law and Economics, University of Virginia, January Lecturer, Toin University School of Law, Yokohama, Japan, May-June Cornell Law School, Distinguished Visiting Fellow in Law and Economics, April 8-12, 1996 and September 25-29, 2 Visiting Professor, University of Chicago Law School, January 1992-June Visiting Professor of Law and Economics, University of Virginia Law School, January 199-May 199. Fellow, Yale Law School Program in Civil Liability, July 1985-August Private Practice (part-time), New Haven, Connecticut, September 1981-August nstructor in Economics, Yale College, September 1983-August Summer Associate, Donovan Leisure Newton & rvine, New York, Summer Summer Associate, Perkins, Coie, Stone, Olsen & Williams, Seattle, Washington, Summer Research Assistant, Prof. Laurence Lynn, Kennedy School of Government, Harvard University, Summer LSAT Tutor, Stanley Kaplan Education Center, Boston, Massachusetts; Research Assistant, Prof. Philip Heymann, Harvard Law School; Research Assistant, Prof. Gordon Chase, Harvard School of Public Health. (During Law School). 19 Li Decl. Ex

27 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 21 Page of of Page 292 D #:58 EDUCATON Yale University, University Fellow in Economics; M.A. 1982, M. Phil. 1984, Ph.D Dissertation: A Continuous-Time Stochastic Model of Job Mobility: A Comparison of Male-Female Hazard Rates of Young Workers. Awarded with Distinction by Yale. - Winner of the Michael E. Borus Award for best social science dissertation in the last three years making substantial use of the National Longitudinal Surveys--awarded by the Center for Human Research at Ohio State University on October 24, National Research Service Award, National nstitute of Health. Member, Graduate Executive Committee; Graduate Affiliate, Jonathan Edwards College. Harvard Law School, (J.D.) Graduated Cum Laude. Activities: Law Clerk (Volunteer) for Judge John Forte, Appellate Division of the District Court of Central Middlesex; Civil Rights, Civil Liberties Law Review; ntramural Athletics; Clinical Placement (Third Year): (a) First Semester: Massachusetts Advocacy Center; (b) Second Semester: Massachusetts Attorney General s Office--Civil Rights and Consumer Protection Divisions. Drafted comments for the Massachusetts Attorney General on the proposed U.S. Department of Justice settlement of its case against Bechtel Corporation s adherence to the Arab Boycott of sraeli companies. Hamilton College, (B.A.) Departmental Honors in both Economics and Mathematics - Phi Beta Kappa (Junior Year) Graduated fourth in class with the following academic awards: o Brockway Prize o Edwin Huntington Memorial Mathematical Scholarship o Fayerweather Prize Scholarship o Oren Root Prize Scholarship in Mathematics President, Root-Jessup Public Affairs Council. 2 Li Decl. Ex

28 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 22 Page of of Page 292 D #:581 PUBLCATONS Books and Edited Volumes: Law and Economics of Discrimination, Edward Elgar Publishing, 213. Employment Discrimination: Law and Theory, Foundation Press, 25, 29 (2d edition) (with George Rutherglen). Economics of Labor and Employment Law: Volumes and, Edward Elgar Publishing, Foundations of Employment Discrimination Law, Foundation Press, 23 (2d edition). Foundations of Employment Discrimination Law, Oxford University Press, 1997 (nitial edition). Book Chapters: Drug Prohibitions and ts Alternatives. Chapter 2 in Cook, Philip J., Stephen Machin, Olivier Marie, and Giovanni Mastrobuoni, eds, Lessons from the Economics of Crime: What Reduces Offending? MT Press (213). The Death Penalty, Chapter in Encyclopedia of Law and Economics, Spring (213). Rethinking America s llegal Drug Policy, in Philip J. Cook, Jens Ludwig, and Justin McCrary, eds, Controlling Crime: Strategies and Tradeoffs (211), pp (with Benjamin Ewing and David Peloquin). Assessing the Relative Benefits of ncarceration: The Overall Change Over the Previous Decades and the Benefits on the Margin, in Steven Raphael and Michael Stoll, eds., Do Prisons Make Us Safer? The Benefits and Costs of the Prison Boom, pp (29). Does Greater Managerial Freedom to Sacrifice Profits Lead to Higher Social Welfare? n Bruce Hay, Robert Stavins, and Richard Vietor, eds., Environmental Protection and the Social Responsibility of Firms: Perspectives from Law, Economics, and Business (25). The Evolution of Employment Discrimination Law in the 199s: A Preliminary Empirical Evaluation (with Peter Siegelman), in Laura Beth Nielsen and Robert L. Nelson, eds., Handbook of Employment Discrimination Research (25). Divining the mpact of Concealed Carry Laws, in Jens Ludwig and Philip Cook, Evaluating Gun Policy: Effects on Crime and Violence (Washington D.C.: Brookings, 23). 21 Li Decl. Ex

29 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 23 Page of of Page 292 D #:582 Articles: Right-to-Carry Laws and Violent Crime: A Comprehensive Assessment Using Panel Data and a State-Level Synthetic Controls Analysis NBER Working Paper. May 217 (with Abhay Aneja, and Kyle Weber). Comey, Trump, and the Puzzling Pattern of Crime in 215 and Beyond, Columbia Law Review (217, forthcoming). Did Jeff Sessions forget wanting to execute pot dealers? The Conversation, January 23, 217 (with Max Schoening), o Reprinted in Huffington Post, o Reprinted in Salon, Jeff Sessions, The Grim Reaper of Alabama, The New York Times, January 9, 217 (with Max Schoening), Testing the mmunity of the Firearm ndustry to Tort Litigation, JAMA ntern Med. Published online November 14, (with David Studdert and Michelle Mello). Empirical Analysis and the Fate of Capital Punishment, 11 Duke Journal of Constitutional Law and Public Policy (216). Available at: Firearms on College Campuses: Research Evidence and Policy mplications, Johns Hopkins Bloomberg School of Public Health, (October 15, 216)(with Daniel Webster et al). Be skeptical about claims of benefits of concealed carry permits. Sacramento Bee, (October 6, 216), 22 Li Decl. Ex

30 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 24 Page of 51 3 of Page 292 D #:583 The Death Penalty Does Not Add Up to Smart Justice, California State Treasurer ntersections (September 216), Reducing civilian firepower would boost police and community safety, Stanford expert says, Stanford News (July 216), Domestic Violence and Effectively Terminating the Gun Rights of the Dangerous, Legal Aggregate Stanford Law School (June 216), 4 Gun Control Steps U.S. Needs Now, CNN.com (June 216), The Demise of the Death Penalty in Connecticut, Legal Aggregate - Stanford Law School (June 216), Empirical Evaluation of Law: The Dream and the Nightmare, 17 American Law and Economics Review Capital Punishment Does not Deter Homicides, Casetext, August 3, 215, There s no evidence that death penalty is a deterrent against crime, The Conversation, August 8, Glossip v. Gross: Examining Death Penalty Data for Clarity, Stanford Lawyer, June 29, How US Gun Control Compares to the Rest of the World, The Conversation, June 24, o Reprinted in slightly modified form under the title Ban guns, end shootings? How evidence stacks up around the world, in CNN.com on August 27, Li Decl. Ex

31 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 25 Page of of Page 292 D #:584 The 1 day period is reasonable, San Francisco Daily Journal, September 3, 214. An Empirical Evaluation of the Connecticut Death Penalty System Since 1973: Are There Unlawful Racial, Gender, and Geographic Disparities? 11 Journal of Empirical Legal Studies 637 (214). The mpact of Right to Carry Laws and the NRC Report: The Latest Lessons for the Empirical Evaluation of Law and Policy, NBER Working Paper Revised November 214 (with Abhay Aneja and Alexandria Zhang), Do Police Reduce Crime? A Reexamination of a Natural Experiment, in Yun-Chien Chang, ed., Empirical Legal Analysis: Assessing the Performance of Legal nstitutions, London: Routledge, Chapt. 5, pp , 214 (with Daniel E. Ho & Patrick Leahy) Reflections on the Newtown Shooting One Year Later, Stanford Lawyer, December 5, Outlier Nation: Homicides, ncarceration, Guns and Gun Culture, TAR 9 (Verona, taly: 213). Gun lunacy rides high in America, Special to CNN, September 13, Why the NRA fights background checks, Special to CNN, Wed April 1, Substance vs. Sideshows in the More Guns, Less Crime Debate: A Comment on Moody, Lott, and Marvell (with Abhay Aneja, and Alexandria Zhang) ECON JOURNAL WATCH 1(1) January 213: More Guns, Less Crime Thesis, Guns in American Society: An Encyclopedia of History, Politics, Culture, and the Law (volume 2:G-Q, at page 585) (212). Jury Nullification in Modified Comparative Negligence Regimes, 79 The University of Chicago Law Review 945 (212) (with Eli K. Best). What Can Be Done to Stem Gun Violence? San Francisco Chronicle, December 21, php#ixzz2G4qkJJ2 24 Li Decl. Ex

32 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 26 Page of of Page 292 D #:585 When Will America Wake Up to Gun Violence? CNN opinion, July 21, 212. Posted to: Time To Kill The Death Penalty? The California Progress Report, June 28, 212. Assessing Post-ADA Employment: Some Econometric Evidence and Policy Considerations. Journal of Empirical Legal Studies Vol. 8: No. 3, September 211, pp (with Michael Ashley Stein, Christopher L. Griffin, Jr. and Sascha Becker). The mpact of Right-to-Carry Laws and the NRC Report: Lessons for the Empirical Evaluation of Law and Policy, Am Law Econ Rev (Fall 211) 13 (2): (with Abhay Aneja and Alex Zhang). See January 214 Revision released as an NBER working paper above. Punishment is a Cost, Not a Benefit, Review of Mark A. R. Kleiman s When Brute Force Fails: How to Have Less Crime and Less Punishment, XLV Journal of Economic Literature (March 21), The Politics of Judicial Opposition: Comment, Journal of nstitutional and Theoretical Economics, 166(1), (21). ntroduction to the Death Penalty Symposium, 11 American Law and Economics Review. v (Fall 29) (with Steve Shavell). Estimating the mpact of the Death Penalty on Murder, 11 American Law and Economics Review 249 (Fall 29) (with Justin Wolfers). The mpact of the Death Penalty on Murder, Criminology & Public Policy (November 29, Volume 8, ssue 4) at pp The mpact of Legalized Abortion on Teen Childbearing, 11 American Law and Economics Review 24 (29) (with Jeff Grogger and Steven Levitt). More Guns, Less Crime Fails Again: The Latest Evidence from , 6 Econ Journal Watch (May 29)(with an Ayres). Yet Another Refutation of the More Guns, Less Crime Hypothesis With Some Help From Moody and Marvell, 6 Econ Journal Watch (January 29)(with an Ayres). Measurement Error, Legalized Abortion, and the Decline in Crime: A Response to Foote and Goetz, The Quarterly Journal of Economics (28) 123 (1): (with Steven Levitt). AntiDiscrimination Law, in Steven Durlauf and Lawrence Bloom, eds., The New Palgrave Dictionary of Economics, 2d Edition, Li Decl. Ex

33 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 27 Page of of Page 292 D #:586 Murder in Decline in the 199s: Why the U.S. and N.Y.C. Were Not That Special, Punishment and Society 1: 333 (28) at Understanding the 199s Crime Drops in the U.S. and Canada, Canadian Journal of Criminology and Criminal Justice, Vol 49, No. 4, p. 552 (October 27). The Law and Economics of Antidiscrimination Law, A. M. Polinsky and Steven Shavell, eds., Handbook of Law and Economics, Volume 2 (27), Pages Economic Models of Crime and Punishment, Social Research, Vol. 74: No. 2, Summer 27, pp Rethink the War on Drugs, Yale Law Reports, Summer 27, pp More Cops, Brookings Policy Brief #158, March 27 (with Jens Ludwig), Studying Labor Market nstitutions in the Lab: Minimum Wages, Employment Protection, and Workfare: Comment, Journal of Theoretical and nstitutional Economics, 163(1), (March 27). The mpact of Damage Caps on Malpractice Claims: Randomization nference with Difference-in-Differences, (with Daniel Ho), 4 Journal of Empirical Legal Studies 69 (27). The Discretion of Judges and Corporate Executives: An nsider s View of the Disney Case, The Economists Voice: Vol. 3: No. 8, Article 4. Available at: The Knicks Boldly Go Where Companies Have Not, The New York Times, July 2, 26 Sunday (with an Ayres). The Death Penalty: No Evidence of Deterrence, The Economists Voice, (with Justin Wolfers) (April 26), o Reprinted in Stiglitz, Edlin, and DeLong (eds), The Economists Voice: Top Economists Take on Today s Problems (28). The Costs of Wrongful-Discharge Laws, 88 Review of Economics and Statistics (with David Autor and Stewart Schwab)(26), pp Security, Democracy, and Restraint, 1 Opening Argument 4 (February 26). o Reprinted in Loch Johnson and James Wirtz, ntelligence and National Security: An Anthology (2d. ed. 28). Uses and Abuses of Empirical Evidence in the Death Penalty Debate, 58 Stanford Law Review 791 (25) (with Justin Wolfers). 26 Li Decl. Ex

34 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 28 Page of of Page 292 D #:587 o Reprinted in Steven Levitt and Thomas Miles, eds., The Economics of Criminal Law, Edward Elgar Publishing (28). o Reprinted in Robert Cooter and Francesco Parisi, eds., Foundations of Law and Economics, Edward Elgar Publishing (21) Does Terrorism ncrease Crime? A Cautionary Tale, (with Daniel Ho), 25. Fighting Crime: An Economist s View, 7 The Milken nstitute Review 46 (25). o Reprinted in Kurt Finsterbusch, ed., Social Problems (McGraw-Hill, 26). Guns, Crime, and the mpact of State Right-to-Carry Laws, 73 Fordham Law Review 623 (24). Clinton and Bush s Report Cards on Crime Reduction: The Data Show Bush Policies Are Undermining Clinton Gains, The Economists Voice: Vol. 1: No. 1, Article 4. 24, The Employment Consequences of Wrongful-Discharge Laws: Large, Small, or None at All? American Economic Review: Papers and Proceedings May, 24 (with David Autor and Stewart Schwab). Further Evidence that Legalized Abortion Lowered Crime: A Reply To Joyce, 39 Journal of Human Resources 29 (Winter 24)(with Steven Levitt). The Final Bullet in the Body of the More Guns, Less Crime Hypothesis, Criminology & Public Policy (July 23, Volume 2, ssue 3) at pp Shooting Down the More Guns, Less Crime Hypothesis, 55 Stanford Law Review 1193 (23) (with an Ayres). The Latest Misfires in Support of the More Guns, Less Crime Hypothesis, 55 Stanford Law Review 1371 (23) (with an Ayres). Can Guns, Or Gun Violence, Be Controlled? (Reviewing James Jacobs, Can Gun Control Work?), The American Prospect (December 16, 22), p. 35. The Search for Truth: n Appreciation of James J. Heckman, 27 Law and Social nquiry 23 (22). The Schooling of Southern Blacks: The Roles of Social Activism and Private Philanthropy, , Quarterly Journal of Economics (Feb. 22), (with James Heckman and Petra Todd), pp o Reprinted in Legal Decisionmaking section of the American Bar Foundation Anthology, ABF Press (27). o Reprinted in American Bar Foundation, Anaylyzing Law s Reach: Empirical Research on Law and Society (28) 27 Li Decl. Ex

35 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 29 Page of of Page 292 D #:588 The mpact of Race on Policing and Arrests, Journal of Law and Economics, vol. XLV October 21)(with Steven Levitt), pp The mpact of Legalized Abortion on Crime, Quarterly Journal of Economics (Vol. CXV, ssue 2, May 21)(with Steven Levitt) pp o Reprinted in Steven Levitt and Thomas Miles, eds., The Economics of Criminal Law, Edward Elgar Publishing (28). o Reprinted in Robert Cooter and Francesco Parisi, eds., Recent Developments n Law And Economics, Edward Elgar Publishing (21). Understanding the Reasons for and mpact of Legislatively Mandated Benefits for Selected Workers, 53 Stanford Law Review 897 (21). o Reprinted in Michael Zimmer, Charles Sullivan et al, Cases and Materials on Employment Discrimination (6 th edition)(23). Nondiscretionary Concealed Weapons Law: A Case Study of Statistics, Standards of Proof, and Public Policy, American Law and Economics Review 436 (1999)(with an Ayres). o Reprinted in Steven Levitt and Thomas Miles, eds., The Economics of Criminal Law, Edward Elgar Publishing (28). Why We Should Discount the Views of Those Who Discount Discounting, 18 Yale Law Journal 191 (1999). Understanding The Time Path of Crime, 88 Journal of Criminal Law and Criminology 1423 (1998). Discrimination in Employment, The New Palgrave Dictionary of Law and Economics (1998). o Excerpted in Lynne Dallas, Law and Public Policy: A Socio-Economic Approach (23). The Legal Response to Discrimination: Does Law Matter? in Bryant Garth, Austin Sarat, eds., How Does Law Matter? Pp (Northwestern University Press, 1998). Some Thoughts on Law and Economics and the Theory of the Second Best, 73 Chicago-Kent Law Review 257 (1998). Allocating Resources Among Prisons and Social Programs n the Battle Against Crime, 27 Journal of Legal Studies 1 (1998) (with Peter Siegelman). o Excerpted in Sanford Kadish & Stephen Schulhofer, Criminal Law and ts Processes (8 th ed. 27) Guns, Violence, and the Efficiency of llegal Markets, 88 American Economic Review 463 (May 1998)(with Steve Levitt). Did Miranda Diminish Police Effectiveness? 5 Stanford Law Review 1147 (1998). 28 Li Decl. Ex

36 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 3 Page of of Page 292 D #:589 Some Thoughts on Affirmative Action, 75 Washington University Law Quarterly 159 (1997). Executive Compensation, 3 Stanford Journal of Law, Business & Finance 1 (1997). Some Perspective on Crime and Criminal Justice Policy, Lawrence Friedman and George Fisher, eds., The Crime Conundrum: Essays on Criminal Justice 45 (1997). The Selection of Employment Discrimination Disputes for Litigation: Using Business Cycle Effects to Test the Priest/Klein Hypothesis, 24 Journal of Legal Studies 427 (1995) (with Peter Siegelman). Employment Discrimination Law in Perspective: Three Concepts of Equality, 92 Michigan Law Review 2583 (1994). Reprinted in Frank Ravitch, Janis McDonald, and Pamela Sumners, Employment Discrimination Law (24). o Translated into Chinese and published in Peking University Law Review (27). The Effects of Joint and Several Liability on Settlement Rates: Mathematical Symmetries and Meta-ssues in the Analysis of Rational Litigant Behavior, 23 Journal of Legal Studies 543 (1994). Liberal Law and Economics, (reviewing Rethinking the Progressive Agenda by Susan Rose-Ackerman), 13 Journal of Policy Analysis and Management 192 (1994). Review of Richard Epstein s Forbidden Grounds: The Case Against Employment Discrimination Laws, 31 Journal of Economic Literature 1477 (1994). Law and Macroeconomics: Employment Discrimination Over the Business Cycle, 66 University of S. Calif. L. Rev. 79 (1993) (with Peter Siegelman). Advocacy Versus Analysis n Assessing Employment Discrimination Law, 44 Stanford Law Review 1583 (1992). o Reprinted in Christopher McCrudden, Anti-Discrimination Law (23). Excerpted in Professors Michael J. Zimmer, Charles A. Sullivan, & Rebecca Hanner White, Cases and Materials on Employment Discrimination (Seventh Edition 28). The Changing Nature of Employment Discrimination Litigation, 43 Stanford Law Review 983 (1991) (with Peter Siegelman). The Effects of Fee Shifting on the Settlement Rate: Theoretical Observations on Costs, Conflicts, and Contingency Fees, 54 Law and Contemporary Problems 195 (1991). Re-Evaluating Federal Civil Rights Policy, 79 Georgetown Law Journal 1713 (1991) (with James Heckman). 29 Li Decl. Ex

37 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 31 Page of of Page 292 D #:59 Opting for the British Rule; Or, f Posner and Shavell Can t Remember the Coase Theorem, Who Will? 14 Harvard Law Review 193 (1991). o Reprinted in Saul Levmore, Foundations of Tort Law 16 (1994). Continuous versus Episodic Change: The mpact of Civil Rights Policy on the Economic Status of Blacks, 29 Journal of Economic Literature 163 (December 1991) (with James Heckman). o Reprinted in Paul Burstein, ed., Equal Employment Opportunity, Aldine De Gruyter, New York (1994). The mpact of Federal Civil Rights Policy on the Economic Status of Blacks, 14 Harvard Journal of Law and Public Policy 41 (1991). Studying the ceberg From ts Tip: A Comparison of Published and Unpublished Employment Discrimination Cases, 24 Law and Society Review 1133 (199) (with Peter Siegelman). Prohibiting Sex Discrimination in the Workplace: An Economic Perspective, 56 University of Chicago Law Review 1337 (1989). The Law & Economics of Tort Law: The Profound Revolution, 12 Harvard Law Review 147 (1989). Using Market ncentives to Promote Auto Occupant Safety, 7 Yale Law and Policy Review 449 (1989). Diverting the Coasean River: ncentive Schemes to Reduce Unemployment Spells, 99 Yale Law Journal 549 (1989). o Winner of the 1989 Scholarly Paper Competition, Association of American Law Schools. Reply to Professors Ellickson and Stigler, 99 Yale Law Journal 635 (1989). Law and Economics: The Road Not Taken, 22 Law and Society Review 93 (1988). Further Thoughts on Employment Discrimination Legislation: A Reply to Judge Posner, 136 U. Pa. L. Rev. 523 (1987). Judge Bork, Anti-Trust Law, and the Bending of Original ntent, Chicago Tribune, sec.1, pg. 15, July 22, Posner s Third Symphony: Thinking about the Unthinkable, 39 Stanford Law Review 791 (1987)(with an Ayres). Determinants of Job Turnover of Young Men and Women in the U.S.--A Hazard Rate Analysis, in Schultz, T.P., ed., Research in Population Economics, vol.6, Greenwich, Conn.: JA Press (1987). 3 Li Decl. Ex

38 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 32 Page of of Page 292 D #:591 A Comparison of Male-Female Hazard Rates of Young Workers, , Working Paper #48, Center for Studies in Law, Economics and Public Policy; Yale Law School (1986). Hazard Rates of Young Male and Female Workers--Recent Developments, Working Paper #51, Center for Studies in Law, Economics and Public Policy; Yale Law School (1986). s Title V Efficient? 134 U. Pa. L. Rev (1986). o Reprinted in Paul Burstein, ed., Equal Employment Opportunity, Aldine De Gruyter, New York (1994). Section Cases, Sherman s Summations, Vol.3, No.2, Sherman Act Committee of the A.B.A. Antitrust Section, Fall, 1982, at 49. An Evaluation of the Constitutionality of S. 114, The Proposed Federal Death Penalty Statute, Hearings before the U.S. Senate Judiciary Committee, April 27, 1981, at 151. Godfrey v. Georgia: Creative Federalism, the Eighth Amendment, and the Evolving Law of Death, 3 Catholic University Law Review 13 (198). Criminal Code Revision--Contempt of Court and Related Offenses, Hearings before the Subcommittee on Criminal Justice of the House Judiciary Committee, July 18, 1979, at 187. Blog Posts: Moore v. Texas and the Pathologies that Still Mar Capital Punishment in the U.S., March 29, 217, Trump and Gun Policy, Stanford Law School Legal Aggregate Blog, November 12, 216, Facts Do Not Support Claim That Guns Make Us Safer Stanford Law School Legal Aggregate Blog, October 12, 215, When will America wake up to gun violence?, CNN.com, July 2, 212, t Takes Laws to Control the Bad Guys, The New York Times -- Room For Debate: (January 11, 211). Have Woman-Protective Studies Resolved the Abortion Debate? Don t Bet on t, (September 28). 31 Li Decl. Ex

39 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 33 Page of of Page 292 D #:592 Dodging the Death Penalty Bullet On Child Rape, (July 28). Why d Stick With Yale Clerks-- Some Econometric Ruminations, (April 28). WORKSHOPS AND ADDRESSES Comey, Trump, and the Puzzling Pattern of Crime in 215 and Beyond, University of Texas School of Law and Economics Seminar, April 24, 217; Faculty Workshop, UC Davis School of Law, April 1, 217; Law and Social Science Seminar, Texas A&M University School of Law, March 6, 217; Quantlaw, University of Arizona Law School, February 17, 217. Debate with Kent Scheidegger on Capital Punishment, Philosophy of Punishment Seminar, JFK University School of Law, March 18, 217. The Evidence on Guns and Gun Laws, Federal Bar Council Program on Guns and Gun Laws -- Rancho Mirage, California, February 23, 217. Guns, Crime and Race in America, Stanford s Center for Population Health Sciences, Stanford Medical School, October 17, 216. Evaluating the Death Penalty, Forum on California Propositions 62 and 66, Stanford Law School, September 14, 216. Empirical Analysis and the Fate of Capital Punishment, Colloquium, Presley Center for Crime and Justice Studies; University of California, Riverside, October 24, 216. Gun Violence and Mental llness, Department of Psychiatry, Stanford University, August 25, 216. The Battle Over Gun Policy n America, Physicians and Social Responsibility seminar; Stanford Medical School, October 3, 216; Bioethics Committee of the San Mateo County Medical Association, April 27, 216; The League of Women Voters of Palo Alto, April 19, 216; Human Rights and Health Seminar, Stanford University, April 12, 216; Bechtel nternational Center, Stanford University, February 23, 216; Stanford in Government Seminar, Haas Center, Stanford University, February 2, Li Decl. Ex

40 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 34 Page of 51 4 of Page 292 D #:593 American Economic Association Continuing Education Course The Economics of Crime (with Jens Ludwig), AEA Annual Meeting, San Francisco, January 5-7, 216. Race and Arbitrariness in the Connecticut Death Penalty, University of Connecticut School of Law, Nov. 2, 215. Connecticut v. Santiago and the Demise of the Connecticut Death Penalty, Faculty Workshop, Stanford Law School, August 19, 215. Do Handguns Make Us Safer? A State-Level Synthetic Controls Analysis of Right-to- Carry Laws, Second Amendment Conference, Covington and Burling, New York, May 14, 215; NBER Summer nstitute, Cambridge, MA, July 23, 215; Faculty Workshop, Stanford Law School, November 11, 215. U.S. Criminal Justice Under Siege: Will Becker or Beccaria Prevail? Faculty Seminar, Bocconi University School of Law, Milan, taly, June 18, 215. Can You Believe Econometric Evaluations of Law, Policy, and Medicine?, Stanford Law School, Legal Theory Workshop, March 1, 27; Faculty Workshop, Tel Aviv University School of Law, May 14, 27; Faculty Workshop, University of Haifa Law School, May 16, 27; Law and Economics Workshop, Georgetown Law School, September 19, 27; Law and Economics Workshop, St. Gallen Law School, Switzerland, November 29, 27; and Yale Law School, February 25, 28; Law and Economics Workshop, Swiss nstitute of Technology, Zurich, Switzerland, May 21, 28; Faculty Workshop, University of Virginia Law School, October 24, 28; Plenary Session, Latin American and Caribbean Law and Economics Association, Universitat Pompeu Fabra (Barcelona), June 15, 29; Google, Milan, taly, June 8, 215. Commentator: Throw Away the Jail or Throw Away The Key? The Effect of Punishment on Recidivism and Social Cost, by Miguel F. P. de Figueiredo, American Law and Economics Association Meetings, Columbia Law School, May 15, 215. Broken Windows, Stop and Frisk, and Ferguson, 215 Justice Collaboratory Conference: Policing Post-Ferguson, Yale Law School, April 17, 215. Assessing the Development and Future of Empirical Legal Studies, Stanford Law School course on Modern American Legal Thought, February 25, 215. Commentator: Payday Lending Restrictions and Crimes in the Neighborhood, by Yilan Xu, 9 th Annual Conference on Empirical Legal Studies, Boalt Hall, Berkeley, CA, November 7, Li Decl. Ex

41 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 35 Page of of Page 292 D #:594 An Empirical Evaluation of the Connecticut Death Penalty Since 1973: Are There Unconstitutional Race, Gender and Geographic Disparities?, Faculty Workshop, Economics Department, Rice University, Houston, TX, Feb. 18, 214; Law and Economics Workshop, University of Virginia Law School, September 11, 214; Faculty Colloquium, University of San Diego School of Law, October 3, 214. What s Happening to the Death Penalty? A Look at the Battle in Connecticut, Hamilton College, Clinton, New York, June 6, 214. Panel Member, Research Methods Workshop, Conference for Junior Researchers on Law and Society, Stanford Law School, May 15, 214. Logit v. OLS: A Matter of Life and Death, Annual Meeting of the American Law and Economics Association, University of Chicago, May 9, 214. Guns: Law, Policy, Econometrics, Second Amendment Litigation and Jurisprudence Conference, Jenner & Block, Chicago, May 8, 214. The mpact of Antidiscrimination Law: The View 5 Years after the Civil Rights Act of 1964, Renaissance Weekend, Liguna Niguel, CA, Feb. 15, 214. Concealed Carry and Stand Your Ground Law, Renaissance Weekend, Liguna Niguel, CA, Feb. 15, 214. Reducing Gun Violence, Forum on Gun Violence Reduction, Mountain View City Hall, Mountain View, CA, Feb. 8, 214. Gun Policy Debate, C-SPAN. National Cable Satellite Corporation, Jan. 16, 214. < Trial and Decision in the Connecticut Death Penalty Litigation, Faculty Workshop, Stanford Law School, November 2, 213. Rethinking America s llegal Drug Policy, Law and Economics Workshop, Harvard Law School, April 2, 21; NBER Conference, Economical Crime Control, Boalt Hall, Berkeley, CA, January 16, 21; NBER Summer nstitute Pre-Conference Economical Crime Control, July 23, 29; Whitney Center Lecture Series, Hamden, CT, October 5, 29; Law and Economics Workshop, University of Chicago Law School, October 13, 29; Seminar for Spanish Law Professors, Harvard Law School, October 23, 29; The Criminal Law Society, Stanford Law School, March 31, 211, University of Denver Sturm College of Law, April 21, 211; Law and Economics 34 Li Decl. Ex

42 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 36 Page of of Page 292 D #:595 Workshop, Boalt Hall, Berkeley, CA, October 17, 211; Shaking the Foundations Conference, Stanford Law School, November 2, 213. The Challenge to the Connecticut Death Penalty, Yale Law School, Death Penalty Clinic, November 5, 27; Graduate Student Seminar, November 11, 29; Stanford Program in nternational Legal Studies Seminar, Stanford Law School, Nov. 11, 21; Faculty Workshop, Stanford Law School, June 8, 211; Faculty workshop, Duke Law School, April 13, 212; Program on Public Policy, Stanford University, May 2, 212; Annual Meeting of the American Law and Economics Association, Vanderbilt Law School, Nashville, TN, May 18, 213; Faculty Workshop, University of Arizona Law School, October 17, 213; 8th Annual Conference on Empirical Legal Studies, University of Pennsylvania Law School, October 26, 213. Commentator: How to Lie with Rape Statistics by Corey Rayburn Yung, 8th Annual Conference on Empirical Legal Studies, University of Pennsylvania Law School, October 213. An Empirical Look at Gun Violence in the U.S., University of Arizona Law School, October 17, 213 Discussant, Sex Offender Registration and Plea Bargaining, NBER Labor Summer nstitute, Cambridge, MA, July 25, 213. What Works in the War Against Crime? Renaissance Weekend, Jackson Hole, Wyoming, July 5, 213. Seminar Presentation, Statistics and the Streets Curbing Crime, Realities of the Death Penalty, and Successes in Public Safety, Renaissance Weekend, Jackson Hole, Wyoming, July 5, 213. Flashes of Genius (Glimpses of Extra-ordinarily Novel Thinking) -- Stemming Gun Violence, Renaissance Weekend, Jackson Hole, Wyoming, July 5, 213. Can Laws Reduce Crime?, Safe Oakland Speakers Series, Holy Names University, Oakland, CA, May 1, 213, Presentation on The Death Penalty in America on a panel on human rights and criminal justice systems in the world, Science for Peace conference at Bocconi University in Milan, taly, November 15, Li Decl. Ex

43 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 37 Page of of Page 292 D #:596 Seminar Presentation, America s Criminal Justice System, Renaissance Weekend, Santa Monica, CA., Feb. 19, 212. Statistical nference, Regression Analysis and Common Mistakes in Empirical Research, SPLLS Fellow s Workshop, Stanford Law School, February 2, 212. New Evidence in the More Guns, Less Crime Debate: A Synthetic Controls Approach, Conference on Empirical Legal Studies, Northwestern Law School, November 4, 211. Drug Legalization and its Alternatives, Lessons from the Economics of Crime: What Works in Reducing Offending? CESifo Venice Summer nstitute Workshop, July 22, 211. ncapacitating Addictions: Drug Policy and American Criminal Justice, in Rethinking the War on Drugs through the US-Mexico Prism, Yale Center for the Study of Globalization, May 12, 211. Plenary Session: Flashes of Genius (Glimpses of Extra-ordinarily Novel Thinking) -- Has Legalized Abortion Reduced Crime,? Renaissance Weekend, Liguna Niguel, CA., Feb. 18, 211. An Evidence-Based Look at the More Guns, Less Crime Theory (after Tucson) The American Constitution Society for Law and Policy (ACS), Stanford Law School, January 25, 211; Renaissance Weekend, Liguna Niguel, CA., Feb. 19, 211; Faculty Forum at the External Relations Office, Stanford Law School, April 5, 211. Empirical Evaluation of Law: The Dream and the Nightmare, SPLS Fellows Lecture, Stanford Law School, January 15, 215; Legal Studies Workshop, Stanford Law School, Feb. 7, 211; Renaissance Weekend, Liguna Niguel, CA., Feb. 2, 211; University of Denver Sturm College of Law, April 22, 211; Presidential Address, Annual Meeting of the American Law and Economics Association, Columbia University, May 2, 211. Death Sentencing in Connecticut, American Society of Criminology Annual Meeting, San Francisco, Nov. 17, 21. The mpact of Right to Carry Laws and the NRC Report: Lessons for the Empirical Evaluation of Law and Policy, Conference on Empirical Legal Studies, Yale Law School, Nov. 6, 21. Comment on Bushway and Gelbach, Testing for Racial Discrimination in Bail Setting Using Nonparametric Estimation of a Parametric Model, Conference on Empirical Legal Studies, Yale Law School, Nov. 6, Li Decl. Ex

44 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 38 Page of of Page 292 D #:597 Commentator, A Test of Racial Bias in Capital Sentencing, NBER Political Economy Program Meeting, April 23, 21. The (Lack of a) Deterrent Effect of Capital Punishment, Faculty Workshop, University of Chicago Economics Department, October 21, 29. Keynote Address, The Evolution of Econometric Evaluation of Crime and Deterrence, 1st Paris& Bonn Workshop on Law and Economics: The Empirics of Crime and Deterrence, University of Paris Ouest Nanterre, September 24, 29. Comment on Cook, Ludwig, and Samaha, Gun Control after Heller: Litigating Against Regulation, NBER Regulation and Litigation Conference, The Boulders, Carefree, Arizona, September 11, 29. mpact of the Death Penalty on Murder in the US, Faculty Workshop, Law School, Universitat Pompeu Fabra (Barcelona), June 18, 29. Comment on Joanna Shepherd s The Politics of Judicial Opposition, Journal of nstitutional and Theoretical Economics Conference, Kloster Eberbach, Germany, June 12, 29. The Great American Crime Drop of the 9s: Some Thoughts on Abortion Legalization, Guns, Prisons, and the Death Penalty, Hamilton College, Clinton, NY, June 5, 29. The mpact of the ADA on the Employment and Earnings of the Disabled, American Law and Economics Association Meetings, University of San Diego, May 15, 29. Crime and Punishment in the United States, Eastern State Penitentiary, Yale Alumni Event, Philadelphia, PA, April 26, 29. Measuring Culpability in Death Penalty Cases, Conference on Applications of Economic Analysis in Law, Fuqua School of Business, Duke University, April 18, 29. Autopsy of a Financial Crisis, Workshop on New nternational Rules and Bodies for Regulating Financial Markets, State University of Milan, March 23, 29. Yet Another Refutation of the More Guns, Less Crime Hypothesis With Some Help From Moody and Marvell, Law and Economics Workshop, NYU Law School, March 1, 29. ntelligence-squared Debate: Guns Reduce Crime, Rockefeller University, New York, October 28, 28. The D.C. Handgun Controls: Did the Supreme Court s Decision Make the City Safer? Debate, The Contemporary Club of Albemarle, Charlottesville, VA, October 23, Li Decl. Ex

45 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 39 Page of of Page 292 D #:598 Evaluating the Empirical Claims of the Woman-Protective Anti-Abortion Movement, Panel on The Facts of the Matter: Science, Public Health, and Counseling, Yale Conference on the Future of Sexual and Reproductive Rights, Yale Law School, October 11, 28. Empirical Evaluation of Gun Policy, Harvard Law School, October 9, 28. Assessing the Relative Benefits of ncarceration: The Overall Change Over the Previous Decades and the Benefits on the Margin, Russell Sage Foundation, New York, May 3, 27; Law and Economics Workshop, Tel Aviv University School of Law, May 28, 28. Death Penalty Debate with Orin Kerr, Bloggingheads, April 11, 28. Evaluating Connecticut s Death Penalty Regime, Faculty Public nterest Conversation, Yale Law School, April 9, 28. The Death Penalty in Connecticut and the United States, The Whitney Center, Hamden, CT, November 5, 27; Seminar on Advanced Criminal Law: Criminal Sentencing and the Death Penalty, Fordham Law School, April 8, 28; Law and Economics Workshop, Swiss nstitute of Technology, Zurich, Switzerland, May 2, 28. Radio nterview, The Death of Capital Punishment? Morning Edition: Where We Live. WNPR. Connecticut, March 1, 28. Comment on Thomas Dee s Born to Be Mild: Motorcycle Helmets and Traffic Safety, American Economics Association Meetings, New Orleans, Louisiana, January 4, 28. The Empirical Revolution in Law and Policy: Jubilation and Tribulation, Keynote Address, Conference on Empirical Legal Studies, NYU Law School, November 9, 27. The Optimal Rate of ncarceration, Harvard Law School, October 26, 27. Empirical Evaluation of Law: The mpact on U.S Crime Rates of ncarceration, the Death Penalty, Guns, and Abortion, Law and Economics Workshop, St. Gallen Law School, Switzerland, June 25, 27. Comment on Eric Baumer s A Comprehensive Assessment of the Contemporary Crime Trends Puzzle, Committee on Law and Justice Workshop on Understanding Crime Trends, National Academy of Sciences, Washington, D.C., April 25, 27. Comment on Bernard Harcourt, Third Annual Criminal Justice Roundtable Conferemce, Yale Law School, Rethinking the ncarceration Revolution Part : State Level Analysis, April 14, Li Decl. Ex

46 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 4 Page of of Page 292 D #:599 Corporate Governance in America: The Disney Case, Catholic University Law School, Milan, taly, March 19, 27. The U.S Tort System, (Latin American) Linkages Program, Yale Law School, February 13, 27. Panel Member, Guns and Violence in the U.S., Yale University, nternational Center, January 24, 27. Economic Models of Crime and Punishment, Punishment: The U.S. Record: A Social Research Conference at The New School, New York City, Nov. 3, 26 Comment on Baldus et al., Equal Justice and the Death Penalty: The Experience of the United States Armed Forces, Conference on Empirical Legal Studies, University of Texas Law, School, Austin, Texas, October 27, 26. Empirical Evaluation of Law: The Promise and the Peril, Harvard Law School, October 26, 26. Estimating the mpact of the Death Penalty on Murder, Law and Economics Workshop, Harvard Law School, September 12, 26; Conference on Empirical Legal Studies, University of Texas Law School, October 28, 26; Joint Workshop, Maryland Population Research Center and School of Public Policy, University of Maryland, March 9, 27. Why Are Auto Fatalities Dropping so Sharply?, Faculty Workshop, Wharton, Philadelphia, PA, April 19, 26. The Law of Racial Profiling, Law and Economic Perspectives on Profiling Workshop, Northwestern University Department of Economics, April 7, 26. Landmines and Goldmines: Why t s Hard to Find Truth and Easy To Peddle Falsehood in Empirical Evaluation of Law and Policy, Rosenthal Lectures, Northwestern University School of Law, April 4-6, 26. The mpact of Legalized Abortion on Crime, American Enterprise nstitute, March 28, 26. The mpact of Damage Caps on Malpractice Claims: Randomization nference with Difference-in-Differences, Conference on Medical Malpractice, The Rand Corporation, March 11, 26. Powerful Evidence the Death Penalty Deters?, Leighton Homer Surbeck Chair Lecture, Yale Law School, March 7, 26. Uses and Abuses of Empirical Evidence in the Death Penalty Debate, Faculty Workshop, University of Connecticut Law School, October 18, 25; Faculty Workshop, UCLA Law School, February 3, 26; Law and Economics Workshop, 39 Li Decl. Ex

47 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 41 Page of of Page 292 D #:6 Stanford Law School, February 16, 26; ; Law Faculty, University of Cambridge, Cambridge, England, February 28, 26; University of llinois College of Law, Law and Economics Workshop, March 2, 26; Faculty Workshop, Florida State University Law School, March 3, 26; ALEA, Berkeley, CA May 6, 26; University of Chicago Law School, Law and Economics Workshop, May 9, 26. s Gun Control lliberal?, Federalist Society Debate with Dan Kahan at Yale Law School, January 31, 26. Witness to Deception: An nsider s Look at the Disney Trial, Distinguished Lecture, Boston University School of Law, November 1, 25; Center for the Study of Corporate Law, Yale Law School, November 3, 25; Law Offices of Herbert Smith, London, England, February 23, 26; Law Faculty, University of Cambridge, Cambridge, England, February 27, 26. Understanding the Surprising Fall in Crime in the 199s, Rotary Club, Orange, CT, August 5, 25; Faculty Workshop, Yale School of Management, September 21, 25. Panel Member, The Board s Role in Corporate Strategy, The Yale Global Governance Forum, Yale School of Management, September 8, 25. Crime and Abortion, Museo de la Cuidad de Mexico, Mexico City, October 2, 23. Allocating Resources towards Social Problems and Away From ncarceration as a Means of Reducing Crime, MacArthur Foundation Research Network on Adolescent Development and Juvenile Justice, San Francisco, CA, February 28, 23. Shooting Down the More Guns, Less Crime Hypothesis, Stanford Law School, Law and Economics Seminar, January 28, 23; Faculty Workshop, Center for the Study of Law and Society, Boalt Hall, University of California, Berkeley, Feb. 24, 23; Development Workshop, Stanford Law School, April 25, 23; Faculty Workshop, Stanford Law School, July 2, 23; Law and Public Affairs Program Workshop, Princeton University, September 29, 23; Stanford Alumni Weekend, Stanford University, October 17, 23; Faculty Workshop, CDE, Mexico City, October 2, 23. The mpact of Legalized Abortion on Teen Childbearing, NBER Labor Summer nstitute, Cambridge, MA, July 3, 22. Do Concealed Handgun Laws Reduce Crime?, Faculty Workshop, Stanford Law School, October 4, 2; First-Year Orientation, Stanford Law School, September 5, 21; Faculty Workshop, Harvard Law School, April 26, 22; Faculty Workshop, Columbia Law School, April 29, 22. The Evolution of Employment Discrimination Law in the 199s: An Empirical nvestigation, Fellows Workshop, American Bar Foundation, February 11, Li Decl. Ex

48 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 42 Page of of Page 292 D #:61 The Role of Discounting in Evaluating Social Programs mpacting on Future Generations: Comment on Arrow and Revesz, Colloquium on Distributive Justice, Stanford Law School, Oct. 18, 21. The mpact of Wrongful Discharge Laws, NBER Labor Summer nstitute, Cambridge, MA, July 3, 21; Labor and Employment Seminar, NYU Law School, October 16, 21; Faculty Workshop, Stanford Law School, September 18, 22; Yale Law School, January, 24. Racial Profiling: Defining the Problem, Understanding the Cause, Finding the Solution, American Society of Criminology Conference, San Francisco, CA, November 15, 2. nstitutional Architecture for Building Private Markets, Conference on Latin America and The New Economy at Diego Portales University in Santiago, Chile, October 26, 2. The History and Current Status of Employment Discrimination Law in the United States, Unicapital School of Law, (Centro Universitario Capital), Sao Paulo, Brazil, March 1, 2. Corporate Governance in Developing Countries: Opportunities and Dangers, Conference on Neoliberal Policies for Development: Analysis and Criticism, University of Sao Paulo Law School, March 13, 2 Legalized Abortion and Crime, Law and Economics Workshop, University of Pennsylvania Law School, September 21, 1999; Faculty Workshop, Yale Law School, September 27, 1999; John Jay College of Criminal Justice, October 7, 1999; Faculty Workshop, Quinnipiac Law School, October 13, 1999; Faculty Workshop, University of Connecticut Law School, October 19, 1999; University of Virginia Law School, October 25, 1999; Faculty Workshop, Baruch College, November 9, 1999; MacArthur Foundation Social nteractions and Economic nequality Network Meeting, Brookings nstitution, December 4, 1999; Faculty Workshop, NYU Law School, January 21, 2; Faculty Workshop, University of San Diego Law School, February 18, 2; Public Economics Workshop, Department of Economics, Stanford University, April 28, 2; Law and Economics Workshop, University of California at Berkeley Law School, September 18, 2; Faculty Workshop, Cornell Law School, September 26, 2; OB- GYN Grand Rounds, Stanford Medical School, October 2, 2; Center for Advanced Studies in the Behavioral Sciences, October 11, 2; Faculty Workshop, Graduate School of Business, February 5, 22. Panel member, Session on Executive Compensation, Director s College, Stanford Law School, March 23, Exploring the Link Between Legalization of Abortion in the 197s and Falling Crime in the 199s, Law and Economics Workshop, Harvard Law School, March 16, 1999; Law 41 Li Decl. Ex

49 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 43 Page of of Page 292 D #:62 and Economics Workshop, University of Chicago Law School, April 27, 1999; Faculty Workshop, Stanford Law School, June 3, s the ncreasing Reliance on ncarceration a Cost-Effective Strategy of Fighting Crime? Faculty Workshop, University of Wisconsin School of Social Science, February 19, What Do We Know About Options Compensation? nstitutional nvestors Forum, Stanford Law School, May 29, Commentator on Orlando Patterson s presentation on The Ordeal of ntegration, Stanford Economics Department, May 2, Understanding The Time Path of Crime, Presentation at Conference on Why is Crime Decreasing?, Northwestern University School of Law, March 28, 1998; Faculty Workshop, Stanford Law School, September 16, 1998; Faculty Workshop, University of Michigan Law School, February 18, Commentator, Conference on Public and Private Penalties, the University of Chicago Law School, Dec , Some Thoughts on Affirmative Action, Presentation at a conference on Rethinking Equality in the Global Society, Washington University School of Law, November 1, Commentator on Chris Jencks Presentation on Welfare Policy, Stanford Economics Department, October 8, The mpact of Race on Policing, Arrest Patterns, and Crime, Faculty Workshop, Stanford Law School, September 1, 1997; Law and Economics Workshop, University of Southern California Law School, October 23, 1997; Law and Economics Workshop, Columbia University Law School, November 24, 1997; Law and Economics Workshop, Haas School of Business, University of California at Berkeley, February 19, 1998; Annual Meeting of the American Law and Economics Association, University of California at Berkeley, May 8, 1998; Conference on the Economics of Law Enforcement, Harvard Law School, October 17, Crime in America: Understanding Trends, Evaluating Policy, Stanford Sierra Camp, August Executive Compensation: What Do We Know? TAA-CREF Committees on Corporate Governance and Social Responsibility, Center for Economic Policy Research, Stanford University, June 27, 1997; NASDAQ Director s Day, Stanford University, June 3, Li Decl. Ex

50 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 44 Page of 51 5 of Page 292 D #:63 Panel Chair, Criminal Law (Theory), Criminal Law (Empirical), and Labor/Discrimination/Family Law, American Law and Economics Association, University of Toronto Law School, May 9-1, Commentator, Diversity in Law School Hiring, Stanford Law School, February 25, Keynote Speaker, The Optimal Rate of Crime, 11th Annual Conference, The Oklahoma Academy for State Goals, Tulsa, Oklahoma, May 7, Panel member, Session on Executive Compensation, Director s College, Stanford Law School, March 28-29, The Power of Law: Can Law Make a Difference in mproving the Position of Women and Minorities in the Labor Market? The Fellows of the American Bar Foundation, Baltimore, Maryland, February 3, Public Action, Private Choice and Philanthropy: Understanding the Sources of mprovement in Black Schooling Quality in Georgia, , Stanford Faculty Workshop, January 24, 1996; Faculty Workshop, University of Virginia Law School, January 22, 1997; National Bureau of Economic Research, Cambridge, Massachusetts, Labor Studies Conference, April 3, Commentator, The Effect of ncreased ncarceration on Crime, Meetings of the American Economics Association, San Francisco, January 6, Commentator, Symposium on Labor Law, University of Texas Law School, November 1-11, Panel Member, Symposium on Criminal Justice, Stanford Law School, October 6-7, Commentator, The Litigious Plaintiff Hypothesis, ndustrial and Labor Relations Conference, Cornell University, May 19, Commentator on Keith Hylton s, Fee Shifting and Predictability of Law, Faculty Workshop, Northwestern University School of Law, February 27, The Selection of Employment Discrimination Disputes for Litigation: Using Business Cycle Effects to Test the Priest/Klein Hypothesis, Stanford University, Law and Economics Seminars, October 31, s the United States at the Optimal Rate of Crime? Faculty Workshop, ndiana University School of Law, ndianapolis, November 18, 1993; Faculty Workshop, Northwestern University School of Law, April 18, 1994; Law and Economics Workshop, Stanford Law School, April 28, 1994; Meetings of the American Law and Economics Association, Stanford Law School, May 13, 1994; American Bar 43 Li Decl. Ex

51 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 45 Page of 51 of Page 292 D #:64 Foundation, September 7, 1994; Faculty Workshop, DePaul Law School, September 21, 1994; Law and Economics Workshop, University of Chicago Law School, October 11, 1994; Faculty Seminar, Stanford Law School, October 31, 1994; Law and Economics Luncheon, Stanford Law School, November 1, 1994; Faculty Seminar Workshop, University of llinois College of Law, Champaign, November 22, 1994; Law and Economics Workshop, Harvard Law School, November 29, 1994; School Alumni Luncheon, Chicago Club, December 13, 1994; Northwestern Law School; Law and Economics Workshop, Yale Law School, February 1, 1996; Faculty Workshop, Cornell Law School, April 1, 1996; Faculty Workshop, Tokyo University Law School, June 4, 1996; Panel on The Economics of Crime, Western Economics Association Meeting, San Francisco, July 1, The Broad Path of Law and Economics, Chair Ceremony, Northwestern University School of Law, September 3, Commentator on Paul Robinson s A Failure of Moral Conviction, Northwestern University School of Law, September 2, The Do s of Diversity, The Don ts of Discrimination, Kellogg School of Business, Northwestern University, May 17, Does Law Matter in the Realm of Discrimination? Law and Society Summer nstitute, Pala Mesa Lodge, Fallbrook, California, June 25, Commentator, The Double Minority: Race and Sex nteractions in the Job Market, Society for the Advancement of Socio-Economics, New School for Social Research, March 28, The Effects of Joint and Several Liability on Settlement Rates: Mathematical Symmetries and Meta-ssues in the Analysis of Rational Litigant Behavior, Economic Analysis of Civil Procedure, University of Virginia School of Law, March 26, Debate with Richard Epstein on Employment Discrimination Law, Chicago Federalist Society, February 23, Panel Chair, Optimal Sanctions and Legal Rules in Tort and Criminal Law, Meetings of Annual Association of Law and Economics, Yale Law School, May 15, Panel Member, The Law and Economics of Employment at Will, The nstitute For Humane Studies, Fairfax, Virginia, March 27, The Efficacy of Title V, Debate with Professor Richard Epstein, University of Chicago Law School, February 26, Moderator, Using Testers to Demonstrate Racial Discrimination, University of Chicago Law School, February 13, Li Decl. Ex

52 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 46 Page of of Page 292 D #:65 Law & Macroeconomics: The Effect of the Business Cycle on Employment Discrimination Litigation, Law and Society Workshop, ndiana University, November 6, 1991; Faculty Workshop, University of North Carolina Law School, Chapel Hill, November 8, 1991; Faculty Workshop, Northwestern University School of Law, December 11, 1991; Law and Economics Conference, Duquesne Law School, March 14, 1992; University of Chicago Law School, April 2, Panel Chair and Commentator, New Perspectives on Law and Economics, Society for the Advancement of Socioeconomics, Stockholm, June 17, 1991; Law and Society Meetings, Amsterdam, June 29, Panel Chair, Regulation of nternational Capital Markets, Law and Society Meetings, Amsterdam, June 27, Panel Chair, The Law and Economics of Discrimination, American Association of Law and Economics, University of llinois Law School, May 24, The Economics of Employment Discrimination Law, ndustrial Relations Research Association, Chicago, llinois, March 4, Does Current Employment Discrimination Law Help or Hinder Minority Economic Empowerment? Debate with Professor Richard Epstein, The Federalist Society, Northwestern Law School, February 26, Panel Member, The Law and Economics of Employment Discrimination, AALS Annual Meeting, Washington, D.C., January 6, Re-Evaluating Federal Civil Rights Policy, Conference on the Law and Economics of Racial Discrimination in Employment, Georgetown University Law Center, November 3, 199. Opting for the British Rule, Faculty Seminar, Northwestern Law School, September 11, 199; Faculty Seminar, University of Virginia Law School, September 14, 199; Law and Economics Seminar, University of Michigan Law School, October 18, 199; Faculty Workshop, NYU Law School, November 14, 199; Faculty Workshop, University of Florida Law School, March 18, The Effects of Fee Shifting on the Settlement Rate: Theoretical Observations on Costs, Conflicts, and Contingency Fees, at the Yale Law School Conference Modern Civil Procedure: ssues in Controversy, June 16, 199. Studying the ceberg From ts Tip?: An Analysis of the Differences Between Published and Unpublished Employment Discrimination Cases, Law and Society Meetings, Berkeley, California, May 31, 199. Panel Discussion on Tort Reform, University of Pennsylvania Law School, April 27, Li Decl. Ex

53 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 47 Page of of Page 292 D #:66 Panel Discussion of The Role of Government in Closing the Socio-Economic Gap for Minorities, at the Federalist Society National Symposium on The Future of Civil Rights Law, Stanford Law School, March 16, 199. Continuous versus Episodic Change: The mpact of Affirmative Action and Civil Rights Policy on the Economic Status of Blacks, University of Virginia Economics Department, February 15, 199; Princeton University Department of Economics, February 21, 199 (with James Heckman); Law & Economics Workshop, University of Toronto Law School, October 8, Sex Discrimination in the Workplace: An Economic Perspective, Fellows Seminar, American Bar Foundation, October 16, The Changing Nature of Employment Discrimination Litigation, Law and Economics Workshop, Columbia Law School, March 23, 1989; Faculty Seminar, University of Virginia Law School, March 24, 1989; Law and Economics Workshop, University of Chicago, April 25, 1989; Law & Society Meeting; Madison, Wisconsin, June 8, 1989; Labor Economics Workshop, University of llinois, Chicago, November 1, 1989; Law & Economics Workshop, University of Pennsylvania Law School, November 9, 1989; Law and Economics Seminar, University of California at Berkeley, October 4, 199; Law and Social Science Workshop, Northwestern University, February 3, 1991; Law and Economics Seminar, Stanford Law School, March 21, 1991; Faculty Workshop, Cornell Law School, April 3, 1991; Visiting Committee, Northwestern Law School, April 5, Law & Economics: The Third Phase, The Association of General Counsel, Northwestern University School of Law, October 14, Employment Discrimination Litigation, Northwestern Law School Alumni Monthly Loop Luncheon. Chicago Bar Association, May 31, The Morality of the Death Penalty. A debate with Ernest Van Den Haag. Northwestern University School of Law, April 19, Models of Deregulation of nternational Capital Markets. A presentation with David Van Zandt, Faculty Seminar, Northwestern University School of Law, April 1, 1988; Visiting Committee, May 5, s Title V Efficient? A debate with Judge Richard Posner, Faculty Seminar, Northwestern University School of Law, November 2, The Senate s Role in Confirming Supreme Court Nominees: The Historical Record, Northwestern University School of Law, September 22, Diverting the Coasean River: ncentive Schemes to Reduce Unemployment Spells, Yale Law School Civil Liability Workshop, March 3, 1987; Faculty Seminar, 46 Li Decl. Ex

54 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 48 Page of of Page 292 D #:67 Northwestern University School of Law, March 18, 1987; University of Southern California Law Center, May 1, 1987; and Seminar in Law and Politics, Department of Political Science, Northwestern University, May 8, 1987; Labor Workshop, Department of Economics, Northwestern University, October 27, 1987; AALS Annual Meeting, New Orleans, January 7, Women in the Labor Market--Are Things Getting Better or Worse?, Hamilton College, February 23, The Changing Relative Quit Rates of Young Male and Female Workers, Hamilton- Colgate Joint Faculty Economics Seminar, February 23, Living on Borrowed Money and Time -- U.S. Fiscal Policy and the Prospect of Explosive Public Debt, Orange Rotary Club, February 22, Capital Punishment in the Eighties, Hamilton College, April 6, Terms and Conditions of Sale Under the Uniform Commercial Code, Executive Sales Conference, National Machine Tool Builders Association, May 12, 198. PROFESSONAL ACTVTES Member, Committee on Law and Justice, National Research Council, October 211 present. Co-Editor (with Steven Shavell), American Law and Economics Review, May 26 August 212. President, American Law and Economics Association, May 211 May 212. Co-President, Society for Empirical Legal Studies, November 211 August 212. Member, Board of Directors from November 211 November 214. Testified before the Connecticut Legislature in Support of Senate Bill 135 and House Bill 6425 (A Bill to Eliminate the Death Penalty), March 7, 211; Testified again before the Connecticut Judiciary Committee on March 14, 212. Member of the Special Committee on AL Young Scholars Medal, October 29 February 211. Vice-President/President Elect, American Law and Economics Association, June 21 May 211. Secretary-Treasurer, American Law and Economics Association, June 29 May 21. Board of Advisors, Yale Law School Center for the Study of Corporate Law, July 24 August Li Decl. Ex

55 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 49 Page of of Page 292 D #:68 Evaluated the Connecticut death penalty system: Capital Punishment in Connecticut, : A Comprehensive Evaluation from 46 murders to One Execution, Member, Panel on Methods for Assessing Discrimination, National Academy of Sciences, September 21 June 24. Resulting Publication: National Research Council, Measuring Racial Discrimination (24), Member, National Science Foundation Review Panel, Law and Social Sciences, September, 1999 April 21. Editorial Board, Journal of Empirical Legal Studies, July 23 present. Editorial Board, nternational Review of Law and Economics, October 1999 present. Editorial Board, Law and Social nquiry, February 2 present. Board of Editors, American Law and Economics Review, August 1998 April 213. Consultant, Planning Meeting on Measuring the Crime Control Effectiveness of Criminal Justice Sanctions, National Academy of Sciences, Washington, D.C., June 11,1998 Member, Board of Directors, American Law and Economics Association, June 1994 May Member, ALEA Nominating Committee, July 1995-May Member, Program Committee, July 1996-May 1998 and July 2 May 22. Statistical Consultant, 7th Circuit Court of Appeals Settlement Conference Project (December, 1994). Testified before U.S. Senate Labor Committee on evaluating the Job Corps, October 4, Assisted the American Bar Association Standing Committee on the Federal Judiciary in evaluating the qualifications of Ruth Bader Ginsburg (June 1993) and David Souter (June, 199). Chair, AALS Section on Law and Economics, January 199 January Economic Consultant to Federal Courts Study Committee. Analyzing the role of the federal courts and projected caseload for Judge Richard Posner s subcommittee. February 1989 March 199. Member, 199 AALS Scholarly Papers Committee. Member, Advisory Board, Corporate Counsel Center, Northwestern University School of Law. Since December Associate Editor, Law and Social nquiry. Summer 1987 December Li Decl. Ex

56 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 5 Page of of Page 292 D #:69 nterviewed Administrative Law Judge candidates for U.S. Office of Personnel Management. Chicago, llinois. May 23, Member, Congressman Bruce Morrison s Military Academy Selection Committee. Fall Candidate for Democratic Nomination, Connecticut State Senate, 14th District (Milford, Orange, West Haven). PRO BONO LEGAL WORK Death Penalty case: Heath v. Alabama. Fall 1986-Fall Wrote brief opposing death sentence in Navy spy case. Court ruled in favor of defendant on September 13, Staff Attorney, Neighborhood Legal Services, January-July Appealed sentence of death for Georgia defendant to the United States Supreme Court. Sentence vacated on May 27, 198. Baker v. Georgia. Court-appointed representation of indigent criminal defendant in District of Columbia Superior Court, February-July 198. RESEARCH GRANTS Stanford University Research Fund, January 1997 and January The National Science Foundation (project with James Heckman), December 1992; (project with Steve Levitt), July Fund for Labor Relations Studies, University of Michigan Law School, March BAR ADMSSONS Connecticut - October 1977; District of Columbia - March 1978 (Currently nactive Status); United States Supreme Court - November 198; U.S. District Court for the District of Connecticut February 14, Li Decl. Ex

57 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 51 Page of of Page 292 D #:61 PROFESSONAL AND HONORARY ASSOCATONS American Academy of Arts and Sciences (since April 29). Research Associate, National Bureau of Economic Research (since October 1996) in Law and Economics and Labor Studies. American Law nstitute (since September 29, 21). American Bar Association American Economic Association American Law and Economics Association PERSONAL Born: January 3, Li Decl. Ex

58 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 1 of of Page 292D #:611 EXHBT 8 Li Deel

59 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 2 of of Page 292D #:612 Highl-to-Ca.rry La\v:-; and \ /iolent Crin1e: A Cornprchcusive i\ sse~~n1ent Using PanPl Data. and a St.ate-Level Synt hct..ic Confrob John J. Donohue.\ Abbay Aneja, and h.yle D. \.Veber J o hu.. Donohm! (cnm!>pnnd iug n.nt.hnr): 8un1ford Law Sd1,jol, ;i.:ib ::.Jat.h;u1 Abbot t. \Va y, St.m1t'ord, Ci\ f);l3(,. Brnnil: 1louob,1ci1Jlft w.:c;uw1forcl.e,l11. Ahl1ay Ancjn : HfUL'C: Sr.hnr1J nf B1tsiue.s~, 2~2U P ic:clmoul. Av()llll('. 13,~rkc!ley, CA U4 7'.W. J<ylo D. Wr.lwr: Cnlumbia lj Hiver~it,_v, '1 :W \V l l h tb St.n i~t., i\n', Y(irk, \, luuu. Al,st1 ar::t 'l'h;; 2ll,J n,porl. o f t i,,,,\j;, t.i,., 11 ;:1. H,. :;,. ~.d, Cl!um:il (NftC ) on F.i : ~HJ!h wt,i Viu.i..-nn- l'l)<.'1.,gt d,,1.,,j l hijl viu l~11l c 1 in11! \\ 1.J.~ hie;j~( 1 l ~ put,;{-~)i.l~:;n g:c p,_~rio d (rf: ]i.1.l:i\'e t ri 11at:i(1n n.l 1--:r il)1d pa1.l.1 r11sl fc1.- :.ti.11c.s ud'-j5.1 Lj1Lg d gh t c;ur_v (HTC) conceule c.l l1andg un la w.~, btit h cn rnr~ o f mc11fol "''l'''"'k1kc tl11~ ~,v.. w~ wu.:; uua bk to idimr.ify tit~ f,n>i' r.i,1~,,l dfocr: of tl1es l' lawr from t.!ie th,\n-,~xi!\l. iil e; J"'.n, l iutu evidc 111. T lii~ &1,ndy 11,A'fl l,l 1uldit:i1,n:d,V":\rs,,, J>lH,d <l,..1..lf1 (t. h1 rn o; li '21.H ( (:a p l.. lll'tu t: a.11 ad tjjliunal l l HTC ad<tp~il)r,:, irn,i 11~w!\l.;)1 i; t.ic,11 tediniqucs t<., ~cc ij 111ure rnnvi1wi11 e; :u1d rol. 1lsl t:uncb1:;iou:-. cnn <! JC!r~11. O,ir pri>fc'.1tnl r,r,11..t,fo t :, ' "!:/~~:~i.>11 ~h11:djil.:uti,.ln (tlie "DA'.V mr:,ch-l''} n,1d 1.h, H,.,11 11:111 C1: 11J.1: r ( 8 C} 1w..,dcl, a~ wr; ll,,~ otlwr ~Lu t ii l,icuj rnodel:; hy Lui.. tmd tdn,-;t nrd (.d) :u1,j.\,k,,.,jy,'>-iii. ),forvdl ('.'<11'. J) t l111t lmd previr,:18ly h,,,,,, <Jlfon.,tJ a.~ c \. ic.lmkt: of crirnc-red uciup: BT(' l:w.r,1, no w,,, 11 ~~i8 k,11ll,1 gc1u~rnl<.' c:;uumtc:, Hhowin;:!; HT C' Jaw~ i,;, :rn rne c1\ e r.1jl. inknt crinw l'rnd / or rn 11n l1.,f' w l1t 11 n 111 1:,11 Ui~ UHJ~l t ornpkt,: dnt,1. \,Vi t,h, P., r.h,; ~j ilfh1: l.i1 r:11lru l Hppro uch o f Alberto Ah: Hli r, anti J :wi,-,. <_;:i nk;i:dtlml 1..iou;1) t,~ ge u,irn tp ~,_.:l l :-~p,:nli,;,::; Ll 111.1Lc:; of t l1e impact. t, ' RTC lno.vr,n1 N i 11i,. 11r 111.ij,:,c li11d i11i; i,; Lhu l, under a : J f<inr sped:knl,io 11~ [D:\1.V,!)( \ LM, a nd /,M,-l ), lfl'(' l,j\\'fl nr u (1~;.()Ciiclt..'11 wilh h igh,::r fl)!;;'.).tl)p:>11.e 1 io lf:m n im,, r.i t,;,.a;, a 11d t h : :;il<.' <>f t.h,,,t,,1,,1.,,r i<>i>~ 1:Jrr.~:t.:; t h,il <'>r : a:;sodi.llecl with t he pn:-;!-m i:;:<! nf HT C' J~ws d i,nl,s '"'l'j' lim ;. \ '.',.,~slimal-e 1.l,,lf: t.h c: adoptirnt o f ntc l:~wh ~11h,,tani i:, ll y f'l t)','d1c.~ v:o le nl crime roles,.mt ~<-'lllll~ t-h lm\'!! nn irnr:h t. nn p rof11jrl,y c ri:111. 1md m111 r1,,,. rnt,,s. T<:11 St,: in, :,[Ll:r l,ht nc.lopliou of TlT C tmv,i, vic,hnt., rim,-. \'-,-,n irnal.1:d i,., l,,j 13-l.'.>';r;, p rci:11. l, ig l;er Limt1 it. WQt.tld havec bc<!rt wit.l1on t t lw l{'.j'c.: l,hv. l),)lil«, L],., p,h11: d u.t u..,;.c Ltiug. l.]u:,i;c n :s11lt,, nr,:, n,,t!'!1 1~i, iv1 t.11.ii!! c(wn, i:lt.p.~ ind1.1d1?, l ii~ J>l'1. dk l.dr:;, Tltc m i.l.{.\ttilud ~ of \.he C;'l l.t111hl,f'd i11crf'm,f! in v io l,;m crl 111e fr,.. m RT L' [..,,,.~ i,; :; ub~l,mlial iu l ba.l. u.~ing a cu 11scns11s cstim1tt.,1 for t lw Hi.1~tfr i1.y,, n i 1111; wi L, rr:,;y,:t:l lu rnrnrccrntilll 1,1'.1 ~.. tl1 f' ;1vc r:1p:c1 n'l"c!"till,p. wm1kl h.,.v~ t n dou l ol, it."l priso n J)(Jpulat.imi l.o t:1111lcn1('t. t l,e flt C-in.-i11cP.d inn,~.~p.e i11 viol.im l (' t'i i ll t~ ~ '\Ve 1,h;u,t, n,1,1 Hn, S!di11 iu l.k li.,vigtiu, rl.ub Tibshira.ni, 1\ cv,,r last.k, S t,,fan \V~f'.uf, ;, 11 d 1.:1,nluru n<:<: parli ~1panLs ;ii t.b 2Ll C uufor~ui:c n f Ernpirirnl Lewd St11di1'.s {C!E.S), '2fll2 A,n, 1 i,_:;1ri J..iw a m. Ecu11<JrnK;; rl.cvic.'.' (.-\LEn.:1,<\1i1111al.\-1..,,1i n '.~. :J(Jl: Cmmdian La,.,;,111d E rnnn rnic;;. ;\, :,(,d:iti,,r. (Cl,J-;,V: 1\111Lu<1. J\ill'dmg, a 11d 2() Ul i\"[3.n S11mrm,r h1~t.i1.l1l.f': (C, i,;i..,') l'ur t.hd, "o m 11ient. :wd lll'lpt'11 ~,_igge1;tic,t\f,. ~'il1i~tir:la l &Up (H. rl Wu!, pro...tck<l by St.,rnford La w Schr,oi. \VP. :ci r<~ indd,t., cl i,o,\jb,: rlu Abad i~,.~ 11,Yl ;cj Di:i.111,:,1,d, :,n <l.jct1s H:.\huuLu;lli.,r for Ll wir wor;_ tkv1.: lo 1>in;'.!, t.l,i~ ~y 11 i-.l,;;tk n m trnl a lg,,ri1.j1 t11 :-,ttd JH''-'~l'u 111mil,g. Lhc ~L11tu m<.u l11l1: ns1;rl in t.!i,s JJ,~per 1rnd fur tltr:ir l1dpl'ul enn11m,nt.,. T l, e ;,111f;l,11 r~: wr,11 Jd :~i~., i~:, 1.,, U1;,11k Ak x A lhn)lltt. Aod s.iw Bl'lk<"r, Ohiirguv C upal C rysi.al ll11 111)\, [~ml<: [h1hh11n i, Aksh:ty l:l:1", ~,.nd Vikfa1u R,11.\ v.-lw pruvic.l,:d cx,~1:llo:!ut. rc:,ca rdt /L<t~i~.r:1.111'1', :cs wdl a~ Arl<ii~ O'C.111mm. rnd A l;,x C h,-kh., Jl,11 a1. t-b" Rc~<;,lrdt Ct,mµuL1u g c.livisil1 JJ uf :-t.;rnf,)ro:l's lnlm m.o_t:inn Tcc,ltnn lof,y S, 1 vi,~,~ Jo r 1.h~:ir 1.r,ch,ii1 ai :-snppc11t. Exhibit B Li Deel

60 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 3 of 34 6 of Page 292D #:613 Part ntroduction: Far nearly two dor.adee, them has been a spirited academic debate over whothcir "ehfill irau()" conoooled carry laws (also known OB right,..1,o...carry 1 RT'C lnws) hove an lr'nportant; impact on crime. The "More Guns, Leas Criruu" hyp1iun-!kit> origimj.uy articulated by John Lott and David Mm;lanl (1.9117) clrimed that R'C lawr dccr=cd violent crime (p88lbly shifting criminals in the direction of rommlttiug moro property crime to a.void armed citizens). This resea.rch =y well J~ve en<.:ou!"3ged at.ate legilllatures to ndopt R'l'C c.ws, arguably umkiog t ho 1-mir'ij 1997 paper in the Jo-urnal of Legal 8tuclic;.o; ono of the mob-t conseq_uent.inl criuu11ologir.al article!> publi'lhcd in the lo.<lt twenty.five years. The original Lott nnd Mu.E.trud pap~r as woll M subsequent work by.john Lott iu bis 1998 book Afore Gu 1w, i,ess Crime used a panel data an.a.lysis to support their cheory that R'C.1'lWS rednco violent crime. A lf\rgc numbo' of papen; examined the Lott thcri.'l, with decidedly mixed results. A number of sludies, primarily Ulling the limited data initially employed by Loit. an<l.mustard for the period , supported the Lott and M~t.ard thesis 1 while \ hui;l of other papers were skepticil.l of the LoLi lindiags. 1 t was hoped that the 24 National Rcscn.rr.h Council (NRC) report Firearms and Violence: A Crinrol Review would resolve the controversy over the impact of Rl'C le.wi;, but this was not to be. Whiln one member of the committee -.T o.mc.q Q. Wilaon - did partially endmsc Uic,<,t t t.hcais by eaying there WllS cv!<lcnco tliat murders fell when RTC laws "'ore s.dopted, the other 15 members of the panel pointedly criticized Wilson's claim, anying th.nt "the scientific evidcuci~ doca not llupport hls position. 9 Tue 111.jorlty cmplmaizcd th.ot the eetlmat,ed ~ffecre of RTC laws were highly sonsitive to t.he particul11.r choice of explanatory variables and thus concluded that tbe panel data evidence through 2 wa..,; too fragile to support uny conclualon a.bout t.he true effects of!.h~c Jnwr,. Thil:t paper begins by reviaitlug ibtt p~nel data evidence to sec if oxt.ilndlng!.he data for an additional H ~. thereby providing additiorui.l crime do.ta. for!)tlor TC states 118 well ns oo ll newly adopting BTC statr~~, o[t'.1'9 auy dp.&1:e'.r picture of the camm.l impll.ct of n.!lowing d t izens to carry concealed won.p1>u K. Across seven dilforcnt permutations!tom four major sets of cxplalmtory w.rfables-including our preferred model (DAW) plus models u.qcd by ~he l:kcu.oan Center {BC) 1 Lou and bt'ltt.'3lnrd (LM), and Moody 11.nd.MAtvf!Jl (:MM)-RC laws c.re assocint.ed wllh lii9her rates of overall violont crinip. ma<l/or wurder. To an.bwer the call of the NRC roporl f.o.( new approaches to estimate lb.o lmpa.cl. o( RTC lavro, we use a now AU!.tlHticu.l technique de:ngo.ed to nddrcsa some of tlw w~nc66eb of panel dato. models that 41:u; gained prominence in 1.hn period sinco the 24 NRC report. Using the AynthoUc controls methodology, we hope to preaent the type of convincing l\fid l'llhnat resulta that can reliably guida policy in H1Jt; nrea. 2 This synthetic controls methodol( ' - first introduced ia Abadio a..ntl Garden.zabal (23) o.nd ex.ponded in Abadie et o.l (21) a.nd Abadie et al (214) - uses o. m(u.ching methodology to create a credible "syntjict,ic conlrol" baaed on a w1:jighted average of other states tlmt la*!l mr-1.chllli the pre-passage pattern or crime for 1Juch "tre.ated" 1 1n irupport of the orl11u11lj l997 Lott nml Mm;tnrd pa.par, see Lor:t' s book Mure Gun.,, U43 Grime (EJ.Dd the 2 and 213 eciltione of thi,~ book). A3'rl!6 o.lld Donohue (23) Md.he 21 :-iational Rel!elU'cb C<>uncil report Fi= 11.lld Violence: A CrJtkal Revlaw dlllmlelll.-d ibo LoW./Munlard b.ypot.km,ls llill lacking credible BLaililLlt:1'1 Sl>JJJX>r~, M did Anoja, Donohue, Jld Zbang'a 21 l Amerlc,m!Aw cmd BwrwmiC3 paper (and the 211 Nl::!ER. pap~ forthr.r cxpn.'lding the ALER paper). Moody &nd M~veu (28) &nc.l Moody, MMV :U, Zlmmonmm, D.J1d Alemante (214) conunuetl l.o 11.l'itUC n favor of a crl:me-red.uclng eltecl of RTC laws, ~lllough Zlmmerm:u, (214} concludce th~ RTC!&"' inare1me vlulenl r.rlrn~, as da.cwllcd ia Soctioo l.b.o. 3 Abadie ct el (214) ld<jn~ify u. num~'t of pullaiblc probkma wi~b pmol rvg'n..!u.1ir,1r1 t chniquc:i, i-ncludiag tbo danger of extrap,olaciou when U1~ ob~et\'llblc ehru-ocwriritlai of.ho t n.'61ed area are outside tlte fluls!c of Lh<: oorregpondlng cbaracterlstlra for Ute otlicr obsctvll~lons in lic ~mplc. 2 Li Deel

61 Case 2:16-cv-6164-JAK-AS Document 45-9 Filed 9/11/17 Page 4 of 34 Page D #:614 Case: , 1/2/218, D: , DktEntry: 17-9, Page 61 of 292 atate, whldi can then h1c1 ul\.ed to ertiru/\te the lilmly path of crime if RTG-nilupl.ing Rtn.tes had not ndupl:ed u. RTC law. By compnring the n.ctu1ll crime pattern for RC-o.doptillg states with the estimated synthotic coutrohi in the poot-pa.ssage pcrio<l, we derive year-by-yrjar estimates for the impact of RTC lnws in the ten yc.n.rs following o.doption.3 1b preview our ll<\jur findings, the synthetic 11trols esti.u.iate of the o.ver3ge impact of RTO l.1.m a cross the 33 sta.tcs that adopt between 1981 and 27' 1 iudica.te thnt violent crime is &-ubstnntia.lly higher after ton yeare tha.u would have be<>,n l,he case h.nd the RTC law not hr.en ndopted. Essentially, for vtulent r.rime, tho synt.hetic contrultt t1,pproach provides a siruih1r portrnylll. of RTC lo.w:-1 w, that prt1..-ided by tl,e DAW o.ml BC panel data models and undermines the results of the LM and \,M imnel data. models. According to tho J\ggregate synthatic control nmdelll - whether one URP.H.he DAW, BC, LM, or MM covaria.te! - R'.'C lawti led Lo Ulctealles in violent crim1} of percent after ten years, with pobitive b ut not statistically significant effer.tr ou property crime and mnrder. Tho tnedian elfod of RTC adoption after 1 year!! ii; 14.1 pr..rcent whether ono considers all 31 statcr with ten )'CJ'tl of dnla. or limits the o.o.alysis t.o t.he 26.stntce with the m6t. oompelliug pre-pa.saage fit between the adopt.ng state.c; AJtC their S}'lltliHtic contmlh. Comparing our DAW-llpecificnt!CJn findings with the rooul~ geuernt.ed Ufillf?; pll\.cebo treatment~, we!us able to reject the null hypothesis thnt RTC Jaws hfwe no impact on aggregate violent crime. Tho slrncture of the paper prt.m;.eocls na follnwll, Pari l disc."1.181.hc!h the pane] 1fot1.t resul LH for the fmtr different modol.s, showing ~hat the DAW and BC models indicate that RTC laws hnvq increased ylolent and proparty crime, while the LM fl..cd?,,.11\,1 models provid{) evidence that RTC AWll have incr!wled murder. We u.rguc thol lhe DAW llol of e..xplriuolory variables are thr. ruotlt plnuhibla and show LhaL mockst and ndvisuble correctiollil to the LM a.ud ~Thi ~iflcatiollb also generate estimates tba.t RTC lawr increaso violent crimti. T he reml.lluder of the pt1per Bhows l:l11:1t the B}'llt,hetic contmlll app'ollch 1wder all four 11ets af explanatory variables uniformly supports the conclusion that R'rC laws lead t.-o substanllal increa.'jes in violent crime. Part describes U1tJ ir...atistica.l uuderpinn!ng11 of the synthetic c.ontrols approach Wld specific details of Olll' lmplemen(jl(lon of this l-e(;hnique. Part V providt!s our synthelic cant.rob! estimnu:s or lbe impo.c;l of R'rC lmvi;, and P1\rl. V co:ncludett with sowb thoughts on the m<-1d1su:tibms by which RTC lawb incref!jie violent crime. 3 Tbe BCl;UtDcy of thl& uto.tcliiog can be qualitatively 11,~oouood by ~mining the ro1.1~ Clll'..llJ1 exiuiue prodici.!on errw (RMS PE) of the synthetic C(llrol in the pro-treatmec.l ~tiod {or a, armuon on Lbt.9 RMSFE impltlmcotod in Ulhl pupor}, and tbe signmcan~ of tho est!mat~ ttcatmea.t offcct can be appr<ncimaied l,y nmning a e1;rri(:tl of plaoelx, estima tes auj e"'ii.miniog ~1' ~zc of tlte m,irnnated tt:eatmo:ot effect n (lc)mpllrison to thu <&trlbutlon t>f pmcebo tl'lll,t.ment effec!a, "Note that we do not generate \ 15Yllthetlc control aa.tinu,te tor lndiana; ev,;n though t Pll.'l'!lod. its RTC 11.., icl 198, owing to the fnci, tbnt wo do t\t'' lui.vo cnoug1, pro-troatmonl ycoj' to e.ci.:ur:\tcly me.tdt the ste.to with n.r. ~ppzoprlale Rynt,bolic ronl.rol. We cuullidcr the cffec;t of making lndil:\11.l, El trcs.tmc,ut i,t4;c as a i:oti,.rntnaia check a.ud find tbe.t Lhw chndgc do.-.., uot mca.n.inl!.full.y cruulj(c our rcaulte. Sbutlarly, we do no~ generete ~yuthot:ic contri;,l crtimate! fo.r owa and Wl1<CUnsin (wh6e R!'C laws weut le.to err... :1 la 211) tu1d for lllitola {214 RTC law), bf!ca;uj!o of the l l nihod poat-pl'.,,,,agc date.. 3 Li Deel

62 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 5 of of Page 292 D #:615 Part Panel Data Estimates of the mpact of RTC Laws A. The No-Controls Model We follow tho N ltc report by beginningwltb t he basic facts a.bout how crime has w1foldcd 1 elalive to national trends for st.o.t!!h wltipt.ing RTC laws. F igurn 1 dep icts percentage d1uugel:l in the violout c1rime rate ovor our e ulir~ dat.a period for throe groupll of states.: those t hal never adopted R1'C l.t1.w11, those that adopt.ed ltc lo.ws i;om.atlmti between 1977 nnd 21'1, M d th8e thn.t adopted RTC laws prior tc> t is noltfwtjrthy Lhat Lbe nine eln\.ca tbnl never adopted R'TC lawg experienced declines (in percentage terms) n violent crime t,hat. ar:e greater than four tuuoo the reduction oxpmienced by states th1it, t«:lopted RTC oit.lrnr prior to 1977 Or U.Urlug Our period of o.nn.jysir. & Figure 1 Tho Ol!cllne n Vlolent Crime Rates ha been Far Greater l11 Stat.o11 with No Rl'C Ls.wa., 19n-: ')4,..ll! ac...,..., JCR let a.mo mat; Ca_,~ i1>t!1> p.,rrulmilu / / / / / ' ', --~szjj &O;, ~/.., /.;,/ R1u, ='.\U5.e / / llsmn ~ -:!NPoo;,'.11-8,7% -9.S'Jf, Sta~ 1/'iQ hll'yg never ~dopl&rj RTC Lawu., / /' ~ / - -- $1etl)'J Uiel J,ave adoplsd R'TC ~ b~w-n 19n and 214 / / / - Slstas D'8!!ldop1ad RTC li!\lm priotto 19n The ~"RC rnpntt presented a "no-crmt.tnls" estimate, wbid, 1H j1i.llt. t.he coefficient. P.>11.ir1Ja.te on the vnrij:\.bl!-! indicating the de.tc of adoption of a RTC law in a crime ra.to panol data model with state and yca.r fixed novor tb.o same 11177?. 14 ix:riod, tho rrtatm ~ l~ c.voidi:::d ooopt:ing RTC lo.ws hu.d :subt!to.ntially low<:t iacrw.'i<.~ in tb.ei.r r~to!l of incaza!a.lfoc B.Dd JJolk.o employment. The nine n~~l.1.dopting states lnerel.\u<::d their iccazcere.tioc tilt~ by l!5 percent, whll"' th.ci l11e&rcara,lon ra1.41!< n tho >ldopllng statee ' " hy ~6'J =.cl 25 poroo,it, for tmho ntloptlng RTC law~ hnfl)ro o.nd after UY/'T rwpcctiwly. Si:milM)y, tl1e mtc of police crnploy1uwlt rol'll: by 16 pcrce,;it lu the ocvn-il<lopting ital.es 1\!1r.l by 38 ond 65 pcrrcent, for tb6e ado]ltlng befon and after 1977, re11v~ctivcly. 4 Li Deel. 8 ~

63 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 6 of of Page 292D #:616 clfocts. According to the NRC niport, "Eotimn.tlug the model L'>lng rlatr t.o 2 showa.hat.states adopting right.to-carry laws i;aw 12.9 pcrcont int.-reases in violent crime - a.nd 21.21>orcent increases in property crime - relative to nationa.l crime patterns." We now estimn.to this same model wi[ng 14 additionnl yonrs of dat n (through 214) and 11 nddltional t\cloptlug t1ta.tes (listed at the bot;tom of Toble 8). Row 1 of 'l'oble. iibows the result.~ of this "no-contwl~" pnncl data approach using a dummy model, wbich jub't estimntcs how wuch on avcrngo crime changed o.ftor RTC lnws were passed (relative to u3lional trends). According to th.ls ruudel, the average i.,ostrparsngo jncrease in viulent crime was 2.2 p~n;w1t, while the OOHJptuable incrense iu property crime WUH 19.2 percent. now 1 also rep-rte the!lllpact of R'C laws on the murder rate (Column 1) and the murder count using a negative binomi.nl model (Column 2), which provide statistically irl!iigni.ficant ostin111tp.8 Uiat RTC law~ incrcosc murder by 4-5 percent.. i The NRG Report tj.loo pre::;ented n. 11pllne model to estimate haw RTC a.dopuon might. alt.er the trend in r.rimo for n.dopting sl.nlcfl, which suggested violent crime o.nd property declined relative to trend in tho do.ta through 2, while the trend io murder was rmc.ha.nged. Row 2 of Tuhle 1 recompntaa t.h.i.s "no-controlr" spline model ou <lat-a through 211, which eliminaloo the earlier suggertion thlit RTC lnws were associa.ted with any drop (rolat.ive to trend} in violent or propntly crime, and rt>affir.mb ~he null finding for murder. 7 Lu other words, more B.lltl better data llij.vtt strengthened t.1111 duljlmy varinbl.8 m11del finding that R:'C law1> inctcllso violent crimc 1 and eliminated the carlfor spline model showhtg of possible declines in violeu~ and property crimf.l. 'table 1: Panel Data Estimates Showing Greater mcreases in Violent and Property Crime Following RTC Adoption: State and Year Fixed Effect.is, and No Other Regressors, 197'7-214 Murder Rate Murder Count Violent Crimll Rate Property Crime Ra.te (l)_ (2) (31 (4) Dummy VD.r!ablo "Mociel 3.8.J (8.19) 1.49 (.53) 2.21 (6.83) (6.6) Spline Maciel -.2K (.61) 1.4 (.fl4).22 {.79), 14 (.5) OLS eetlmationb inch.ade year and stn.ui fixed effects a.uc.l 9.rll weighted h) stato populat.lo:o. Robust Ht-!lllda.rd crrur11 (dntitc.rcd a.t lhu 11te.ta level) arc provitlcd next to J)Oint cist.imatefl in parcnthcl!q, ncidence Rate fl..otios (M) ei!lim.i.ted using Negative Binorni.a.l R.egrceelon, where state popululion is ind11ded ~ a control varlnble, are presented in Column 2. T he null hypothmls i6 that tho mr equals 1. The l!iource of all the crlmo rater is t.hc Uniform Cr.imo Reporui (UCR). p <.l, '«* p <.5, *** p <.1. All figure.'l rcporled in percent.age Lemm,. While the Table 1 dummy 1nodcl inwcates that RTC states e."'q)filiencc e. wori,e post-passage crime pattern, Utls does not prow that HTC lnws incroo<le crime. For example, it mjght be tho case that some sta tes decided to fight crime by a.llowing citizen.,; to carry concealed handg11ru1 while othera decided to hire more p ouce and lncarcerntc n greater number of convicted criminals. f police and prisons wcro more effective in st-opplng crime, the uno control:!n model might Hlmw tha.t the cri.ma exp~riem::e in R'C RtB.LP.6 waa worse tlrn.n in other JTbe dummy,ruiabli! mo.dej report,, ihc coefficient assoclntc<l. wlth &l HTC vnrbtble bo\ is given a value of rero i an RTC law 1B nol hi otcct in th.at year, a vn.luc of one if B.D RTO!Jlw ~ in clicct th.at. entlro y<>ar, o.nd a,'llus oqiw LO lhc portion of th!! ~ lilt n 'l'c law ill in effect oth~ii,c. Tho date of adoption for each R:nJ state s ~hc,\ n in Appendix 'nr.ble A 1. fj'he rjpluw model reports ~ult.it for a variable whjch 1..~ t!&li.gdod a value o~ w.ro beforo the HTC law l.11 ln offoct a11ci a,'alue eqtl'l.l t.o the J)Orlioo of the "f('.:u thfl Rl'C laws was in effp.r.t. tl,e fim yan.c a.ft.er adt1ptjl'ln. 1\ft.cr this ye~r. the v;i.luc oft.he thlj l'oo!ible 111 incrnmented by oue anauu.lly fo:: states t.bat ~uplud right-to-cal'l'y luw~ but~'oon Dtl 214. Tho i,plide model aho locludc:i a. &OCood trend v,u~blc rcprwmitlng the numbe~ u yoo.m that have p-d sinco 1977 for the staie11 adopting RTC ltlwtl ovor tho sa...'llpla period. 5 Li Deel

64 Case 2:16-cv-6164-JAK-AS Document 45-9 Filed 9/11/17 Page 7 of 34 Page D #:617 Case: , 1/2/218, D: , DktEntry: 17-9, Page 64 of 292 Rtnte; even if thir were not a true. cnumi.j result of t.h~ adoption of Tl'l'C luws. AR n turns out, though, R'C states not ouly cxpcrioncoo higher rnios of violent crime but they also had larger incre.8.lles in incsucere.tion and police than other Atiltea. \Vhile the roughly 7 percent greater incrc!jje n the incnrccrstion rate in RTC at.ates is not stntistlcally significant, the mcrcnrcs are large.11d sta.tl~tically signi.lleant for police. Accordingly, 'li_:i.ble 2 confums Ulai ltc statee di<l not have declining rates of iuca.rce.ration or l.ol1.1,l police employees a.ftcr adopting t.heir RTC lt\ws that mil!:ht explain their relatively ood criuio p&formancc. Tuble 2: Panel Data Estin1ate1:1 Showing Greater lncrea~es in lncarcerntion and Police Following R'TC Adoption: State nnd Year Fixed E(l'ecta, and No Other Regressors, loc.arceratlon (1) DWlllllj Variable Modd 6.18 (6.22) Police l :mploynurni Per 1k (2} (3.l!i) -====- Policu Olllcen, Per 1k (3) 'f.8" (2.76) EBtline.linns include ~ar and state lliced eftecui 11.lld are welght8cl by state popultl.llon_ Robu.s.l. lltnndl\.rd error:, (cloot.cr:cd at tnc atat.c luvcl) a.re pro\'idod next to point eetimal,\,~ in p!l{cnth~. The source of the police 1!.mploymenl. tale nnd the sworn. police ollicar rate la the Uniform Cr!m. Reportij (UCR). The source of t.ha Llcu.n:enition ratti itj the Dure.au of Jlllltir.e Statlsti1 (DJS) * p <.!, p <.5,..., p <.1. All figuroo reported in percent.age tc-,rul.6, B. Adding Explanatory Variables We know from U1e e.n!l.lysie of the dummy model in tho NRO report o.ntl io 'l'able 1 f,hat Rl'C lnw adoption ls followed by hig/1er rates of crime {relntivc to national trcmle) and fron-1 Tobie 2 that the poorer crime performance after R:.l.'C law adoption occut:1 de.spite tlte fact thot n:rc states continued to invest at ll".a.st as heavily in priaons and nclusdly invested 111oro heavily!ji police thn.1t 11on~RTC atn(a'.(l. While the thcoreticnl pre<l.ictioru;.j.bout the effect of TTC laws on t~rime a.re indeterminate, the~ two empirical fu.ct1j bmed on the n.ctual patternh of crime!l.lld crime-fighting measures in RTC and non-rtc &-ta.tes Rnggc.st that the moot pla.u~iblc working hyj>othe.sis is Umt RTO laws -inc~-e crime. The next slop Jn a panel d.at11 a.nalysi.s of Rl'C law? would be to te11t this hypolhemli by introducing an appropriate sot of explb.1atory Vl:lriables thtit plnusibly influence criu1. The d1oke of these Vl).ri!~bles is important beca.uso any ~1.ufablc tha.t both influcnce8 crime ruid itj 1:1imult11rucoualy corrc!atotl with RTC lf~wa must be included if we am to generate 1111bi!U!Cd esiiu1at~ of the impa.ct of R'l'C laws. At the same time, including irrelevant. (Uld/or highly collinear va.ri.o.bltl8 ca.n al1() undermine efforts o.t vnli1l estimation of the impact. of RTC laws. At.ho very least, i~ seems advlsnhlo to control for th.e lcvcla of polico and incarceration because these a.re the 1.wo most import(l.jlt ctimillll.l jtl8tice policy instruments in the battle against crime. l. The DAW Panel Data Model n addition to the state and year fixed effects of the no controls modol a.n<l the identifier for the presence of n HTC law, our preferred "DA\V model" indudefl an array of other f.actor11 t.ha.t might he expected to influence crime, sudl Q; the lovohi of police and in<:nrctjrationt vnriou.s incomo, po~rty!.d<l untiniployment m,~ urcs, and slx demographic conltols designed to capture tho proocncc of males in three racial categories (1:llack, 6 Li Deel

65 Case 2:16-cv-6164-JAK-AS Document 45-9 Filed 9/11/17 Page 8 of 34 Page D #:618 Case: , 1/2/218, D: , DktEntry: 17-9, Page 65 of 292 \\'bite, other) in two high-crime nge grn11pingli (15-19 nnd 2-39). The full 1ml. or t!xplanato:ry vnrial1let1 is liated in Thblc 3, along with the regrcs.siou models nsed in throe other i;tudies that havo estiwatoo the impact of rrrc laws on crimo. 8 'fh,1 ljaw panel data model ill Thble 4 (run on data from 197!l-2[)]4) is consistent with the ao.tne basic pattern obaervej lu Tuble 1: 9 RTC law11 on average incre!l.<je<l violent crime by 9.5 percent and property cruuc by 6.8 percent in the yell.ta fotiowin!!',; o.doption according to the durruny model, but a.gain showed no statistically significant effect in the spline model. lo As we saw in the no--conttols model, the c11timate<l 1:,,ffect of RTC laws in llhle 4 on the murdm n1i.e is alsu not stn.li11tically significant. 2. The BC Panel Data Modol Table 3 liilt.s the vn.rfablr.s Ullod iu the Brennnn Couter (BO) crime rcgretltf:ion model, which differ in n fow re.'!pects from the DAW model (Altb.ough to a lesror degree th.an the LM Md MM modclr) (Roeder et n.l, 215). The BC modol controls for both lnca.rccmtion nod police rates (ns iu DAW), bot the BC model td...1.ccs the log of both these rates. The J:C model alone <:ontrols for lhe number of cxcc11uon11, and unliko DAW docs not control for either the stnto povert.y rate or tho percentage of tho atat:e population livuig in a. ~etropohtan SJ.\Watlca.J. Arca. t.'loroover, while DAW.luclud<!a aix d<'.mogn\.phic i,'ll.rin.hh1, tlc uses three o.gc grnup.ings over tho nghll , and simply controls fur tho Linck percentngu uf the state popufation. Tho rcsull.s of running the BC model over the period from arc presented in Tobie 5, :Panel A. With the cxcnptlon that the BC dnmmy vnriable model estimate of the increase in violant crime L.<1 eomawhat higher than Lhat for DAW (1.98 pt.ltoont incren.sr. vcn!uh 9.49 pcrc.(:nt- i.ncreu.sc), t.hc 1JAW and BC model e!ltlma.tea are almost ldentica.! in suggt'\llting higher rates of violent and property cdme (the dummy modp.ls) but. nu irupact in the!!jlline models. f WH replace the four DC demographi<: variables wi.l.b the 6 DAW demographic va.r!a.blea (Table 6, Panel B}, the sizo ur the est,iu\t.-od increases in vlol<mt crime and property crlrne (in the dumm.y model.a) arc only modestly lower thfill the DAW reaults in Thblo '1. 3. The LM Panel Data Model Table 3's rccjto.uon of the explanatory vnriabke couta.incd in the Lott and Mustard (LM) po.ncl dn.ln model reveals two obvious omission.<j: there are no cont;rol<j foe the lev<'ls of police and i.!j.c.j.l.fceratioa in <'.nch stale, <n-oo though a l!ulibtantial litcrah11:c ha.a found thnt: these factors bnvc a large impact on crime. mlcc<l, as we saw above in Tobie 2 both of these factors grew ofter RrC law oooptloo., and the incre1.111e in polico employment after Rl'C acloption i6 aubstnnlivcly and BtntiAliCS1.lly significmit. A Bayesian n.nalysis of the impact of ltl'c laws found that "t;.be incarceration rat.e is a J}owcrful predictor of future crime rates," aod apecifically fnull.od this omission from the Lott n.nd Mustard model (Strnad, 27: 2[)} fu 8). Withont more, thou, we hn.-ve reason to believe that tho Ll,1 model ii; mis-specified, but in addition to the obyious omiti.cd K\iVbilo we 1\\ t,cmp~ ea include 111:1 m1>,ny atatoa in tlte.~.; n,grtj!l!lione as p,~iblc, Di.lS~ct of C<,hnnl,iu incarceration rjatu i~ mi.!uring a.ft.el' the Y'JO.r 21. 1n ad(!j\[on, a handful of obfier,"utioll6 are a.lao dr 'ppcd from the LM a~d MM regrl!llllio!ls owing to sto.t:es that dl<l not roport &ny u~ab],: urrwt data in VllllOU ya,.rs. Our regresslowi a.re performed wll.h robwt stan<laru error! th.at me clull~ ~ the state lo;. el,!l11d we lag the STteflt rl\lcij uirod il both ~he U,f Md MM rel{reblliou modele. The r&tivua.lc!! underlyia~ bot11 of theee ch.anger,,,..., do1k.-ribed la more <fottl.il in Aneja et nl. (214}. All of toe rngro!rio pr119entl!d in this paper a.re wi!lp;heod by sts.te popu!jl.~111. The oompln~o ~ot or entbr>j\tce lot ~11 o.'qllanar.ory varfa.hlet, (axoopt.he dort1l'>tf11pbic vn.rl.ablu!) ror t,bo DAW, BC, LM, and :lvfm dummy iu d 11pliao models l~ ~hu,,;11 i11 appendix 'nl,l.,lc A2. 1 Defens!Vt: Ulla! of guns are onon: llkcly for violent crlruw because the victim will clearly be pte:mut. llor property etiroca, the victl.m lr t) pic'j\lly absent, thus ptovidiug Jcee oppql'tunity \v dofurui with a gnn, l is UJtclear whether \.ho ma.n,y wayit in which lu'c Laws c11uld leod to more crln:rn, which we dlqr:ujill ln l'11rt V, would be more Ukely to far:jlitutu violent or pro tilt.tr crime, but Ofil lnh:ltlon la that violent r.rlma would be r.iore ~troo.gly in1uenced, wblcb le n fact whp.t T:i.blo '11mgge!lts. 7 Li Deel

66 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: 17-9, Page Page 9 of of Page 292D #:619 Table 3: 'l'nble of Explanatory Variables Tor li'ou.r Panel Data Studies Explanntory Vo.r1ablea lught to Cru-ry Law Ltigp;ed Per Capita nc.uceratlon Ra.to l,aggod Log of Per Capita ncarceration Rat- Lngged Police: Staffing P~r 1, Rr.eirlm1tR Li1ggeri Log of Swurn Polke OffiCC'! Per.Ralidmt Population Lllgged Number of Bxec1,1tlon9 Povudy Rate Unemployment Rate Por Capita Eiluwol Consumption from Deer PcrccnlA~~ of the Stole Population living lu Mot ropolitn.n SLntiRticnl Areas (MSAe) R.ea1 Per Capi~a Personal lncoma :.'.llnmi:nal Per C'n,pita ncome (Mcdl.a.n n~me in BC) fuml Per CaplLa ncome.mainte,urwco Real Per Capito Rlitlrement Pa,ymonts Real Pur Capita Ullfilllployment nauranru Pli.yment.6 Popula.tiou Density Lugged Vloleut or Property A.nest RatB $Lo.to Population Crllclc lridex Lagged Dependent Variable DAW X X X X X lc X JC HC X X X X X X X LM MM X X X X X X X X X X. X X X X X X X X )C X {j Ago-8c.x-Rncc Demographic Vu.riablc~ -all 6 comhi.uatioll8 of bl~, white, o.ud other maloe in 2 age groups (15-19, 2-39) indi.c-a.ting the pcrccn~ of t.he population in r.l\ch group X :-t Agc--Group PercP.ntagcs (15--1!), 2-24, 25-29), =d mack PcrccnLage of Popull.l.tiun 36 Ago,Se;x-R.a.ce Demographic Variable~ -nll pollslblc oomhluatiolls of blac'k and white malcr in O ogc group.,; (1-19, 2-29; 3-39, 4- i9, 5-64 aud over 65) EJ.nd rr.peating ti.ii~ wl for fojjlj.\cll, indicating the perr.entae;e or tho population l.n co.ch group l{ X :'.llotq: 'the DAW mooel ie lld:vo.uc:cd in this po.per, while thr. other tbrcc modcla were prcv[our!y publl&bc<l by the Brenmm Centw- (BC), Lott and ~lub-tard (LM), and Marvell and Moody (.MM). See footnote 18 in Apponrlix D for nn explanation oft.he d.iffcrc.nccs in the retirement pa.ym1>.nts variable dp.finition bctwt:.m:. t.be l,m a.nd MM Hpecifica.ti.ons. The cni.ck index vnrid.blo n t he )..1M StJLci.llc:at lon ill available only for 19ro.2DOO. Li Deel

67 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 1 Page of of Page 292 D #:62 Tuble 4: Panel Data Estimates Suggestlog t.hat RTC Laws incronso Violent and Property Crlllle: St.ate and Year Fixed Effects, DAW Regreasors, Murder Rat.o Murder Count Violent Crime lnto Pmpr.rty Crime Rate 1) (2) {3) (1} Dummy Vl\l'inblo Modol.3 (5.35} l.u5 (.52) 9.4'.. (2.96) (2.73) Spliue Model -.31 (.53) 1.2 {.,1) o.m; ~o.64).1'1 (.38) OLS estimatio1111 incl.ude ye!\1' and etatt. lixod effectu a,nd oo-e weighted t,y ~to population. RobttBt suw.dard errors (cl11&tered 11t t.he atn.t.e level) e.re provided next to point eatime.tea ln pareuthosas. lucid1mce Ro.te Re.tlos (llr) esuma.t~ wsing Nt:gatiw Dinomial Regret1sioll, when:!statu population ie includt:tl i.'! 11 i.:.antrol ~,uiahle, 111' pmmnted in Column 2. 'l'bc null hypotho!ub in th11t. the RR equals 1. Tbo source or all the crime ratr.s i~ lhe Uniform Crime Reports (OCR). SLx demographic variables (based Ofl dille.ront ngo-ile."t-race categories) Bro ioc:luded os couttulii in the rngt'{'-~'flion 11hove. Oehm: crm&rol~ indnde the. Jagged incacccmtjon mtc, the lagged police employee rate, coal pm capii1l peroont1.l income, the unor.nploymen t ra.te, poverty rate, b(.oer, and percentage of the population living in Ml:lA.8. * p <.1, p <.5, ** p <.1. All.6gura1 reported. in porcen.tll,6-a tenru1. l'. Table S: Panel Data Estimates Suggesting that RTC Laws increfl.9e Violent and Property Crime: Stato and YMr Fixed Effects, BC RogrAA..'lOl."S 1 197S-214 Dummy VH.ria.ble Model Ponel A: BC R eynmior., ncluding 4 Demogropluc Varwbles Murder Re.tc Murder Cot111t Violent Crime Rate Propert.y Crime la.to 1) 3.45 (5.67) (2) 1.5 (.51) (3) (3.65} (4) G.86 (3.26} Spline Model -.49 (.51) 1.3 (.,l).19 (.).12 (.35) P_a_r_id_ B_ :_B_C_ RCQ'1'f.llfJ!Ors with (J DAW Dcm.ogrophic_V._ana _._hl_u ~1 u.rdcr!late Murt\P-.r Count Violent Crime Rate l ' ropcrty Crllllc Rate () (2) (J) (4} Dummy Variable Model 1.88 (6."17) 1.57 (.51) (3.29) 5.57' (2.85) SpHne Model -o_. 3,1 (~.4,8} l.3 (.4) '.24 (.,59).16 (.34) OLS estjmationh include yffil' and state fixed etecte a.nd are weighted by et.ate population. Robut,t 6t..'londard crrora (clwrt.erad a.t the stati: le-yb!) nre provided nrud. to point B11ti.mate 1 in parenthc&'s. lncldance late llatioo (RR) Cilt,i.nuJ.t<..'ll UJ1ing Negative 3innmial Re~ion, where lll.u.tu po{)t.!ll\tion is indutlcd a:i ll conlrol va:ria.hle, nre prescni-od l.n Column 2. Tha null hypothesis Jl! t.ho.t th BR CJ.~ 1. The source or o.11 tho crime ratei; lll the Uniform Crime fu..1>orf.8 (UCll). Four demogr11phic variables (perc:cnt black, percent aged , percent aged 2-2 1, ru:id percent o.gcd 25-29) arc lncludcd in the PMel A regrcsbiorui. The: {i DAW demogrnphic vnrir.ibl~ lij:c u.scd in thl! Pl;lllel B reg;rreslons. Othe.r co:cr...rols im:lnde Jog of the lagged. lncarcora.ti11 rn.t.e, 18,{l;ged police employment par re'jlir!p.nt population, the unemployment rn.to, nominal per cap[fa income, lnru,,cd m1mber of executioll5, gallon.e or bp.er r.onmi.med per cap[tn. " p <.1, p <.Ofi, H p <.1. All rlgw cs reporl.od in perr.entol\ge terms. 9 Li Deel

68 Case 2:16-cv-6164-JAK-AS Document 45-9 Filed 9/11/17 Page 11 of 34 Page D #:621 Case: , 1/2/218, D: , DktEntry: 17-9, Page 68 of 292 varinblo birui, we have di11ct!b8ed nn iu ca.y of otlillr infumitim1 with the LM model iu Anejn, Doouhnc, ru.111 Zhe.ug (214), including their re!ie111cc on flawed arr(m rntcs, and h1ghly collinear demographic variables. As noted in Anejo, Donoh11c, ttnd Zhang (214), "The Lou. o.nd Mustard ntre!!t ro.tc' n.rc a ra.tio of nr~ta to crimc-..s, which nw:ooo that whou one person J.:illfl many, for example, the arrest rate falls, bnt when many people kill one person, the arrest rote rises since only one co.n be arrested in the first i.usta.nee n.nd runny CSl.!1 in the second. The bottom line is lha.l Lhla 'aucsl rate i8 uol a probability and is frequently greater than one because of tlrn multiple nneats por crime. For a.n ex"ten.ded discussion 11 the abundant ptoblems with ~his pseudo arrest rate, sec Donohuo and WoUors (29).» ThH Ll,{ arroet. ratee are ahm economutrically prohleruatic!rincfl the dmmruina.tor c)f the arroill. rate ill lbe.numermor of the dopendent variable crimo rate, improperly lc,wing the dcpondent variable on bolb. sides of (.he regrcfll'lion equation, We la.g the /l..test mt.aa by ono YP.Jl.r to rcdur~ LhiR prol1l1jm of rntlo bidb. LotL nod ).fu!lta.rd's uso of 36 demographic vorlables is nlso a. potcnlial concern. With so many cnotmuusly collinear variable6, the high llkellbood of introducing noise itlto the estimation process is revealed hy th.e wild Onctuations iu the coofficleut estimal.e! on these vnrinblc!s. For oxau1r1je, comll1ler the LM explmultory variable:. ((nllither black nor white male. aged 3-39" and the identical corr~n<liug fem,i.l e category. The LM dwnmy varin.blp. model for violent crime 1mggests that tbe mo.le group will vastly!ticrease crime (the c~fficient is 211!), but their fornale counlorparts have a.n enormously da.mpening clfect on crime (wilh n. coefficient of -2551). Both uf those highly imph1.usible estima.te.s (nol i;huwn in '!'able A2) ttre statl~lically signi.fk.ant nt t.bti 1% levt!l, and they rue almost cerla.irlly picking up noise mther tlian Teve.i.ling true rolationahil)8.!jizane rcrults a.re common in the L1'.'1 estimates among these :16 deruographk varinbl,111. ll Table 6, Pru1cl A shows the results of the L:M pa.nel dnt.n model cstlljl&ted ovor the pcrio<l As i:;een above, the DAW model generated estimates tha.t R"C lawi; raised vtole.nt and property crime (in the rlummy ruorlcl of Tobie 4), while having no ohvious imvnct on municrs. The J.M model 6ip11 theec prediction!! by showing atrong estimot.e:i of increased murder (in t-he spliue model) u.nd no evidence of lr1creased violent or property crime. We crui almost perfectly restoro the DAW 1 1'able 4 findings, however, by!l:imply following t,he typical pattern of crime regressions by lirniting the inclusion o( 36 higl1 ly collin~a.r clemograph!c variables and including m ea.'urell for polh.:\l and in~rceration. These res\1lt-s appenr in Panel S uf Tobie 6, and thili modified LM dw:nm.y variable model suggests that RTC laws increase crime. This finding is similar but sornewhnt stronger tho.n the DAW dummy vnrlable model e:iumato of higher violent nnd property crime. 1n sununa.ry, the L"M model thnt had origi ually b<!cn cw ployed using data through 1992 to nrgno that R.'l'C laws reduce crime, no longer showi; any stall:itically significant ovlden.ce of crime reduction. ndeed, using more complete dn.to. 1 the L.M spli11c model (Plluol A of Tf\.ble 6) sug,gests that RTC laws increase tho murder ra.le and couul by about 6 or 7 porcent.after 1 years, which a.re the only statistically sigrtifica.nt results in Panel A-no other crilfle category ill affected. Thooe who are skeptical of thcsa ehults bocauoe the LM specification is plagued by omitted variable bias, flawed pseudo-arrest rntce, too 111.ny highly collinear demographic wriabl~, and other problems> might prefer.he estilrult~ in Panel B, which!'ilrnply limit l;he LM demograpwc variables from 36 to 6, and add the incmcctation and police cont rols. These changre once a.r,n.in U Anoja., Donol.iuo, a.n.d Zhit.ug (214) te11t for the severl~ of the mulllcollinearlty problem u~iug the :!6 LM demogr11pluc -vatl11ble&, Uld ~ problem la lmlood &Ol'loUR. Tho Varlant:e nflation F;ictor (VF) ll tibown to l)e in the,a.age of 6 to 7 for tl:e R.TC variable in both the l,m dumm;,u,d Hplino models when the 36 demographic controla ore 11BOd. l.16ing tho 6 DAW varl11blo~ reducoo tho multkolllnet\rit.y fo1 the RTC dummy lo a tole?"abli! level (with \fll. 'ti alwa}'11 htlow the da1[r;\blc thr!!!!bold of 5). [ndoed, lb!! cjcgroo of multloolll.nearil)' for tbo i.ddlv!dul\l ili!mograph!at of the bl.ticl<-ll1ub catogt1nai arc SALonMungly hlr,h with 3B de1ru1g'h1p:wc controltl- iij ~he ncighl.m,,-jmad of 14,1 1 1"his 1m,.Jy,,it1 mam 11~ WMY of esti~1mtar of th~ lrcopnct of RT:C 1~-..-. th.at e1npluy the Lott. MuHlard &et of 313 domographk controls (1\11 dooo the MM model). 1 Li Deel. 8 ~

69 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 12 Page of of Page 292 D #:622 Tnblc 6: Panel Data Estimates of the mpact of RTC Laws: State nnd Year Fixed Elfeda, Using Actual and l\fodifled Ll\11 Regressorst 11'7~214 Ptmel Ai LM.R grnssors ticluding,'j6 Denwgraphic Variabll!}J Murdar Rat e Murder Count Viulc:it Crime Rate Property Crime Rnto ~~~~~~~ ~~(1..!...>~~~~(:!-2l. ~~ ~ (3) Dummy Variable }l;fodel (3.44) l.3 (O.D32) (3.15) ( l) -.33 (1.'ll) =Sp=l=in=e =:\ =fo=dll=l=====.=66= = =(=.33= =) ==1=.=(l= = =(U=-==3)=-=--;.;...4_1 (.47) o.~ ~.o.2a)_-= Panel B: LM Regressot s with O DAW Demogrophw VarialJlc~ and Addinq CcmtrolA fnr ncn.1r.eratinn and PnlirR.. Murder Rat<. Jr.'lurdcr CouuL Violent Crime lata 1) 2 (~) Dummy Vru'illble Model 3.6 (5.67) l.58 (.54} (4.54} Prni)crty Crimfl Rat~ (4) 8.9'" (3.63) Splin11?-.foclru.3 (.43) l.3 (O.Q4) ~-~U _(~57).5 (.34) E:tti.w.nJ.:iulltl lndutl~ ye1:1r wttl ~tutc lix1::<l elicct:,i u.u<l nru weighted by 1Jfott, populu.tiuu. Robtlllt 11taudE1rd Brrorn (clustered a..t the strte level) a.re provided next to point estimates in parentheses. fo Pa.ncl A, 3 dctnogrnphic variables (bmed on different ag&-sex-rnce categories) are included as cont.rol.6 in the reeressiona above. u Pal.tel B, only 6 d=ographic variables am included a.nd contmla ll'!! ndc!oo for inci:m;:eration 1mc! police. Tur both Panels, othllt controls includu tho ptovloub ~rja :tnurdor, violent or- propmty crime arre~t ni.tfi {depending on the crime catcgocy of the dependent variable}, smtc population, pop~tlon dedbity, real per roplta income, tell! pc,r capit11 unemployment lilb"l1r!wc(;l ~_yment e, r~l per capita income maintenancl.l payments, and real ret irement payments per pc!l!lon <:M:r 65. 1' 1J <.1, -.-. p <.5, ~"'* :p <.1. All ligun:a reported in parc:entngo ic:rma. 11 Li Deel

70 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 13 Page of 34 7 of Page 292 D #:623 rt!t:il,ore the Tohle,1 DAW dummy vn.r-lhble model r~ull that RTC laws increase hoth violent and pro11~rty crime. 4. The MM Panel Data Model 'fable 3 reveals tho.t tle Moody a.o.d MELr.-ell (~Thi) model improvea on the Ll-.{ model lu that it includeti the key lt1ce.rteration vnriablc, but Ml'\'1 nlrlo omit the critic.al police measure fow1d in the DA\V Specificntlou. The MM model ttlso contains the problcmn.tic pseudo-!lll'c!lt re.tea and over-saturated and highly collinear demographic vn.rlnble11 that LM omplt,y. 12 Panel A of Table 7 eslimate11 the 1[},f modtll for tho ptiriod While M?v's use of a potontinlly problcmotic lagged dependent. variable control risks purging some of the i>jfoct of the RTC h,w, again we e~ evidence that n:rc laws fru:rf.a,rn violent crilll~. The only n~her st.atiatically slgnifkant estimate ib for tho murder rn.tc in the spline model, whicli suggr.flts that the murder mte would bo roughly 4 percent Wgher ten yearr after RTC adoption. This finding is roughly Rlmilar to thp. Table i, PiLn.cl A finding of incrcw!cd murder in the LM model. Pll.l'.lcl D of Ta.bl<; 7 mlmics our previous critique of the L.M model by including a measure of police 1.ttd using more apprnpriato cfowographlc coukols. Thetle modilkauorn, once ago.in revive a dummy variable mo<lel estimate of increased violeut crime. 5. The Lessons from the Panel Data Studies Estimated Over the Full Data Range All four mod('jb shown in Tobie 4 through Table 7 showed evidence tb.al TC lnws increaced rnurdo.r B.lld/or overall violent eritt1e. DA\V aud DC showed almost identicnl ncreases in violent crinrn of 9-11 percent and prop11rty crime oi G-7 percent. Tb<.: LM model ('.l'able 6 1 Panol A)-tbc liciltt of the origiuru More Guw, Less Crime hypothesis ()9timates a si7.ee.ble and st.o.ll.stically signlflcmt incn'..a,qc in murder will follow RJ'C adoption. A airnimr finding 1:1merges for the MM model ('l\wle 7, Pam![ A), which nlw predicts a.u increase in violent crime. f we look at the modified versions of Lhe LM and MM models in the[r respective Panel B's, the LM modol (Thble 6) almoa1. perfectly roplicates the iucrew;ed violent RUd property cr.im.e cstimo.te.s of DAW and BC, whilo the MM mo{lcl (Tuble 7) conllr1uei to show a 11ta.Ust\C/\lly s1gnificant increase i11.he violent erlrne rate. The stroi1gcst result t.o crncrge from the sovun pn11cls across the four sets of pa11ol dat.a. spcciclcalions in Tables 4-7 L!1 that 6 of these 7 pa.neb show si.ati8tically slgntficant evirle.twe that Rl'C law!l incrca.se violent r.rimtl. The Dilly mc.ception (LM Panel A) sl1ows statistic.ally significant. cvidenc.e of i11.c:n:me-1j in murder. n other v... ords, all 7 panels support tho conclusion tlui.t RTC laws facreaae overcll violent crime OJ1d/or murdor. ACTOSfl..he 56 estimated effects in the AC~n po.ncl.:;, not one abowetl 11.ny cv3dcncc of a decrooac in crime at the.5 level of significance. 1 2,\'hll,; our Thblo G MM p&\<'j dat.a apoc:uk,;t,ltm fouowa ~foody f l'.ld Mon-ell {2~) in includlag J;,,p;S"'J vnluoa o{ thi, d.,penjolli varje.ble M TCgral!lor, no ru1alol(<ju11 varlo.ble A 1'.xplicitly include<( boow ia our synthetic control W\.lyj!i~ fcnturing the Moody~ Me.n'lill pre<l!ctor varlab-les. Slnet1 n.11 Jagged values or the dependent varfa.ble a.re 11.ltt~y ncluded a.a predlctora n the R}'Tlthot!c coatrolb ~1tlll,Y11i~, iacludlng tha lafly,(:d DV would ho rodundanl. l~m~1 UK<J tbo crack index or Ftym- at al {213}, but thia c.om,<!il nt th.~ prioe or luo.ilu.41" the avalll\llu dntn yea.tb for Om.Mlti1 pon.el data 11.fuJ.Y!U'. to the yeal't, 19S-2CJOO. We e!ith.ootod ilia MM rt.ujdcl on the data period &om 198-ZOOO with and wl~hout ~ho crnck ooc:ai,ic vurlablo, whkh yfoldcd vi.rt ua.lly ldanucal tooulla. Thctcloro, in 'Thb].:, 7, we oxclude.be c.r~ COCUDC va.rlriblo, which a.llow!l uu to ure 15 yo!u'! uf u.dditioou.l data. 11.i c~timntc the efre~t of n:rc laws (from 1979, as well 1111 :.luol t hrough Wl4). 12 Li Deel

71 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 14 Page of of Page 292 D #:624 Table 7: Panel Data EBtimatcs of the mpact of RTC Laws: State and Year Fixed Efrccts 1 Uslug Actual and Modified l\1m Regressors without Crack Cocaine, , Panel A : MM Reyre.,8ors With-Out CracJo Cocaitic a.nd lncludtfl9 36 Demograph!e Variabl('..11 Murder Rate Mtudar Count Violr.nt Crime Rntc Property Crime Rat.c Dummy Varlablu.Modol (1,_) _ (2) - l.81 (l.85) l.2'.j (.27) (:l).69 (. 77) (4).48 (.69) ~pllno Mod.ii.38 ' (.16} 1.3 (.2).17" (.8) -===O=-l=- {.7..:. )-== = Panel B : MM Rcgrc.uvrs WithQUt Ct'Dck Cac.ame, WW, 6 DAW Dem<>{lrupllic Variables, and Adding a Control /or PnUce Murder lwt- Murder Count Vlolent Crime llnie Property L'rlme Rate (1) (2) {3) (4). Dutnmy Variable :Modol 1.22 {l.75} 1.:11. (.35) 1.5 (.53).52 (.53) Spliuo Modcl.24 (.17) 1.1 {D.:J) O.H (.9).6 (.5) OLll estimatiolu include yoar and 5ta.te furnd effectr and oo-o weighted oy etaw populnt1on, H..obu~t Btn.nrfard enors (clustered a.t the state loy11l) are provlded next to point- esljjnatl!e n parejjthcscs. ncidence Rat-c lntios (RR) estilll.ll.lccl llllin,r Neg11.t.tvc Bluuminl Regression, w.horo etatp. populnlloa is inc;ludcd wi 11 control vnrlabk, o.ni pra'll!nted in Columu 2. The null h,ypolhl!flie ie that the mn eqllals 1. Jo Pa.tie! A, 36 demographlc Vl\riable11 (b:iscd on diffe:rcnt agc,scx:-rnco cntegor[cs) arc includi:d as controb in the regrwious abov1c?. Jn Panel D, unly 6 demographlc varlables are J..nclutloo. and 11. control ls a<lded for pohco. For both panels, olhar c:nntrols Lnclude the ls,gged dependunt variable, the pruviowi yea.r's murdur, violent or propwty crime arrest rafo (dep.lndi:ng on thu crime r.at.egory of tho dapendr.nt vn:rio.blc), 11tnte populo.tlon, the lagged incarccrat[on rate, the poverty rn..t.e 1 the unamployment iatc, rcul per capita income, n:al per capita. unemployment Wl!uumce p1.1ymcnts 1 real p~ capito. ncome maintenanco pnymenlr, nnd renl pur cupil.a rntiremem. payments. p <.1,.. p <.6, '*** p <.1. All fi.gurei roported in percentage t.m-m.,,. 13 Li Deel

72 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 15 Page of of Page 292 D #: The Zimrncnnan Model and Our 4 Panel Data Models Estimated for the Post-Crack Period Our previour cilljcut1flion ll/ul focm1ed on panel data e5timn.tee of the impact of RTC laws on crime over the full period from the late ln7a through 214. Zimmcrii:Ul.l {214) examines tho impact. of various crime prevention measures on crime 11..'illlg a state panel dflt.a t1et from Be fincl,; that RTC ln-y,."m i.nr.n;rued murder by 15.5 percent for the eight states thll.t adopted n'c laws over the period he ann.ly1md, The advanlafle of using thi5 <ll\ta period to explore the impact of RTC lawr; R th.at lt largely avoids the problem of omitted varia.ble bias owing to the crock vheuomonon, since the crack effect had eubsii.lr.<l hy The di.5advantnge ls LLAt one ce1n only gain. estimates baaed on the eight stalei, lhat adopted RTC laws over that twelve-year spell. 14 ZimmornuL1 dorcribes hit1 finding a.a follows: "The ahn.l.1-lssu~ COt!ffi.(,'ient takes a positive sign in all regr-ions sa.ve for the rape model and is stotietlcally ~ignincant in the murdcr, robbery, M!lautL, burglary, and larceny model!:!. Theae latter finding.~ D1lY imply 1,hat the pb..lll:lage of shrul-is.<tue b,wli inc;rnaheh the properu1ity for crime, as some recent resea.rch (e.g., Aneja, Oonohue, & Zhang, 212} has suggcstcd."ls n 'nlbl~ 8, we ihow the results of nll four oo.'j]c mmjeh! that we di'>curs~d nbove DAW, BC, LM, and fv[m- wl1ca nm o,,-cr the per!ocl w 'l'lir. DA\V model mimics the Zi:rm.r1crmnu fimfuig of a large jump in the murder r ate. '!'ho BC model weakly support., tho lncretwe in murder, and more sc.rongly shows ru, 8 porccnt increase in the overn.11 violent crime mtc. T'nc results for Lliis shortened period using lhc LM and MM models flte uever statistically significant at ~be.5 level. 7. Summary of Panel Data Analysis T.he uncertainty a.bout the impa.ct or RTC la.w! on crime expressed u1 the NRC report WM based on an analysis of dato. only through 2. The preceding cvuluation of nn nrmy of dlfforont spccificatioru, over the full da:ta. period Crum the late 197s through 21'1 bat1 eliminated any suggestion of beuign effects on crime from the adoption of RTC l8;wll a.nd conaiatcntly ebmn1 lnoi<lence that RTC laws incrcaso murder u:nd/ or orerall violent crime. Three of five models estimated on post-cro.ck-era data {Zimmermn.n., DAW, a.nd BC) provi1lc further support for this conchislon. Durlauf et o.. {216) &LLempts to rort out the dill'ere11l specification ch.oic.es in ev~ua.ling RTC laws by using n Bayesian model uvern.g:ing approach using county da.ta from Applying this teclmique, the a.utl= find that in their prcfcrreti s_pllne (tr~nd) model, RTC lnws olovate violent crime in the Lhrco yeare itfter TTC a.<loption: "All a result of the law l11~ing introduced, violent r..rime incr~~es in the first yea r 1iod continues to iucroa.so n.ftorwarcb? By the third year, their preferred model suggests o 6.5% increase in violent cr1mc. Siru-.c their J)fl.!if!r only provides estimntc.<j for threa p11f\t-paa:,;age years, we CAnnot clrnw concluaiona beyond this but note thnt. thoir finding ihat violent crime increases by over 2 percent per yeo.r owing to RTC laws is a. substantial crime increruie. Moreover, t;he authors note that "For our estimates, the effect on crime of introducing guns continues to grow over time." Despite the substantial panal <la.le. evidence in the po:st-nnc Ut.eralure tjmt supports the finding of Lhe pernicious influence of RTC!awe on crime, the NRC suggestion ilw.t new techniques should be employed to HTb.e rplatlvely ahott. time spo.a mo.kas tb.e assumption of atat11 Jtw<l. clructe more plausible buc lt uo llm1t6 tho amount of pre-adoptlnu <lala fllt M ~~ly adaptor such as MlchJg.~n ('21) Mil.be. runount of poi;.lra.doptlon dala fo r Lh<) la.lo adopterr c'lebro,.~4 ~ru.l l<a,,sojj (both in 7.7). 15 AnllJtt. e, lot. (211) nloo ran the ADZ model O\'l!l the aaro,: period tha': Zimmerma.n empl<i.)'k 1 whicli gonerated &11 e.~timate that mutj,:r tlltq r&e 11bout 1.6 percentage palnb ea.ch y~r tbal!. R:rC 18.w was n effect. 16 \W r<tlltt..: twij time period in WOO because the llluu'p cdn:.c: c:kc.,;-.~ of the l OllCJa ended by l:heu 11, l quuc atnr~ing in 2 WM mor., 11tllblo for tbo romz.inder of our data p,erlo<l thl'!.11 t bad provioumy bee Li Deel

73 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 16 Page of of Page 292 D #:626 Tnblc 8: Panel Data Estinuile1:1 of the mpnct (lf RTC Lnws Ui;ing DAW, BC, LM, and MM specifl.catlollll Pc,.nel A : P(lnel Da.to. EstirnaWt SU9{JeBting th(lt RTC Law,; iricret1<1e Murder: State and Year Fized.Eflecta, DA 'W ll.egrehol's, 2-~1, Murder RaLfc! Murder Count ViolcnL Crhtm Rn.Le Property Crime RaLo (l) (2) (3) (4) Dummy Variable Model 5.7 (3.69) 1.'21 (.43) (3.49} (2.25) Splinu ~lode! (.56} 1.13"~.53 (1.11}.4 (OA2) ====- ;;._=== Panel B : Panel Data E11Umates SuggeAting that.to Laws increcue Violf!nt Crime: Slate and Yoor ~ Effecu, Brent1.n Center Regrca,ors, 1?-214 Dummy Variable Mo<lcl );fw:det lillto Murder Count VioleuL Crime Rate Property Crime l./j.t.o ~~~) _ 1.29 {4.1) 1.31 (,42) 7.9r (3.56) {4) 1.9 ('.MA) Spline.Model ::::=:..:::_ (.6~) 1.12 (U.6).5~ (1.29}.35 (.45) - -===----- ==---- Panal C: Panel D(J.ta E11timate.a With,'l(J Oollinro, DP.mOfJfYJpltfo Variabfet Show No Effe.cl of RTC L-.wa: State and Year Fiud. Effecta, LM RegresaorB, 2-2n14 MW'dcr Ra.to Murder Count ViolcnL Crime Rate Property Crime Rate P2 (2) (3) ( 4.) Dummy V11rlablu Mndel 2,8';? (3.63) 1.26 (.'.18) - O.B6 (3.34) -a.m; (J.95) Spline Model.92 ~.78) 1.1 {,8) -.3 (. 72) -.31 (O.al) - - Panel D: Panel D<1t1J Etitim.atea With 96 Collinear Dcmugraphic Val'iablt!.'f Show No Eifcr;l of R.TC Laura: State and Ywr 1''~ed El]e.c-t,, MM ~/faora {Vith.otit Crack Coca(nc, 2-21.J, Murder Rate Murder Count Violent Crime 11.aLe Property Crime Rate (1) (2) (3) (4) Dummy V11riable Mu<lcl 2.49 (3.3) l.23 (.38).13 (1.5) (11.86) Spline Model.7 (.8) LOO!) (.9).3 (.33).17 (.18) R.,;timatlons htdude y,iat and state fixed effect s n.n<l n~ weighted by st11,t,a population. lwbu!'lt standard errors (clustered at tho at.ate levcl) arc provided next to point ootirnateb in patunthefle.~. Po.ncls A, 13, C, and D replicate the Bto.ndard spocillcatlons on dllt3. To allow for estlmntlori n Lhis period for lha 'MM m.odcl, the 1--rar.k index variabw ls dropped. 'l'he following 11 sta.t!lls adopte<l RTC law~ during the period of coru,iderauou: C{23), TA{:211), L(2M), KS(27), M(21), JvfN(23), MO('..l4}, NB(27}, NM('2D4), O(W(Jti), and W1(2ll) "'p <.1, *$ p <.5, ~ p <.1. All figureji repork'ci in percent.age henna. 16 Li Deel

74 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 17 Page of of Page 292 D #:627 estimate the impact of these ltt.ws ir fitting, 'l'he important pa.per by Stmrul (27) rnw.d a B11.yesian B.Ppronch to argue that none of the published models used i.u the RrC ew.luntfon literature rated highly in hia model selection protocol when applied to data from , Moreover, one member of the NRC pnnel (Joel Horowitz) doubted whether n rmnc:l dnt.a model could ever convincingly <:stabush the causal imp11ct of R:!'C 16.wv: ''the problems posed by.b.igh-din1ensiull esu,ua.tion, miebpecified models, and l,',lc}t of lmowledge of the correct set of explanatory variable8!:leem iilbunnountablc with observational data..n (NRC, 24: :lo!:l.} lut owing to the substantial chnllcngcs of est.imathig effects from ohscrvotiona.l datu., it will be useful to sco f it differtlilt Bta.tiBtical approach tha.t brui dlffonmt {.UributeB from the pnnel <lalu. methodology c.art be brought. to bear on the issue of the impact of RTC laws. The root of this pa.pcr will pr~ont this new approac.h. Part Estimating the rr1pact of RTC Laws Using Synthetic Controls Tho synhuit.ic controls methodology, which s becoming incre!l.!lingly prominent in economics and other social 6denccs, ib n promiarug lll'w 11tatiatical approach for t>.ddreshi.ng the impnct. of R.TC laws.17 A nlllllhc:r of papers have used the eyntb11tic control technique to Avruuate varioub influences on c:rlrne. Rudolph et al. (2lii) construct a synthetic control for the 5tat.c of Conncctit~11t yfold.ing evidence that the i.;1.i1t.c\; fuenrm homicide ra, L-e (but not ita non-firearm homicld~1 tale) fo]j appreciably after ~he lrnplerneutation of a p erml1,-t.opurcliase handgun Jaw. MruiaHib ajj.d Guettabi (~13) usa tlri::1 methodology to cxrunine thfj effect of Florida's ''Stand Y<Jur Ground" law, concluding ~ha!;. this law was as.5cialed wllh a.r1 increase in 11emll gun deaths. Si:milnrly, Cunn.ingha.m and Shah (217) Hb1dy the effect of l\hode Tulancl'H wiexpected dec.rimmjtlhmlion of indoot pr~utlttion on 1..he state's rape rate (fl,ffiong other out.come variables); LoD;~rom a..t1d Raphael (213) t>..stima.te the effect of CRHfomia'~ public safety ronlign.mant un crime rates; and Pinottl (213) examines the coru;cquenccs oi ru1 inilux of Oiltanized crime into two talian provin.ces in the lnte 197s. vvh.ile these papers focus on a!lingle treatment in a single geographic region, we look 11.t 33 HTC adoptions t.hroughout the country. For end1 F).{.iopting (treated) atntc we will find a. weighted average of other states d ooigned lo serve!b a good counter-factual for Lhe impact of RTC laws, be<!l.\u&i this "synthetic control;' had a sim.iln.r pattern of crime to the adopting statii prior to RTC adoption. By corupu..ring what actunll, happened for t he ndop~lng siate post-paasa~e to Lhe crime performance of the synthelic control over the same period, we genernto esti.mat~ or t,he raubal. impact of RTC lawll on crime. l & A. The Basics of the Synthetic Control Methodology The Bynthetic control methorl at.tampt~ to gene.rate representative 1mmterfactua.l unitb by comparing il.1,rtrn,tmcnt unit ( i.e., a state adopting n tl'c law) to a. set of control units ocroas a set of explanatory w,,riablcs over a pre-intervention period. Tb. algorlt.hm l:1(-',ll,rches for similn.riti~ hetwp.p-t1 the tree.tment state of inter!!l{1. ltthe synthetic ctmtrol rui:tho<lology has boon deployed in e. wide variety of liddn, lnclucling health economlc<1 (NM11cmak~r ot al., 2D11), immigraili,u ~onom.il'.>i (Bohn cl a.l., 214), polilk!!!.l. oconori1y (Koo,l(:, 2), urban c,co:oomlcs (A,do, 215}, tho econo!l((:ll of Mtl.Vlll h:ttourcc:~ (MJdokua, 213), and the dynamics of economic growth (Csvnllo et al, 2()l 3). l8far a mo:re detalli.d Wdiuical desc;riptiou of this met.hod, we dlrecl lh1: n:1.\dcr to Abadie a.ud G~Tdea:,;aba.J (23), Abadie et al. (21}, aod Ab11,dic ct. l\l. ('.lij14). 16 Li Deel

75 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 18 Page of of Page 292 D #:628 nml.he cont.rol ~t.atei; during this period nnd thl'!jj gv.uemteh a synthetic countorf.act11h.l 11n it, for the treatment state that is a weighted combll.llition of the comp-oont contr ol states. JO Two conditions n.ro placed on these weights: they must be nou-nega.uve a.nd they must sum to one. n g1meml, the matching process underlying the synthetic control technique uses pro-lrcatmeut values of both the outcome vnrinble of interest and ot.hnr predlctol's believed to influence thla vnriablc.~ NJ ju1:1ulled in Appendix F we use ovory lo.g of the dependent "\'Ul'inblc as predictor-a in the DAW, DC, LM, and ~f spccificatiod 1. 2 l Once the synthetic cow1terfactual is generated Bnd t he weights associated with each cont.rol unit lll'o oss!gne<l,.he synfli program thr.n ctucul6tes vuluetl for the outcome variable o.saodti1.p.x with thih counterf.nctua] tmd t.l1e ruot. mean Hquared predidiun error (nmspe) bruied on differences bctwcon tho trefltment and synthetic control units in the pre-treatm ent period. The offoct of thti 1,reat.ment, can then be ootin:ulr.d hy c;omparing the actual vn.lucrn of Liu~ dependent -.,mfoble for the treatment unit to the curre11ponding,,atucs of tho synthetic control. B. Generating Synthetic Controls for 33 States Adopting RTC Laws During our Data Period To il111t1tn1..ta the procedure outlined nbovll, co5ider the Cfille of To..=, wb~.a RTC law went into effe.d ou Ja.nua.ry 1, 19l)6. 'fhc potent.la. control group for ~h trcatmant state consists of all nine states mth no HTC legislation e.a of the year 211, M,w.U M ot.atel!> that pn..<n RTC LH.wli 11.t le&"!t ten ye.nm nil:ar tha J'lll.'<.'!ff.&e of the trca.tmcnt state (e.g., in this cllsc, those atotes {)ll88ing RTC lnws alter 26, such ll.l:l NebrllBk.n. o.nd <ansru;, whose R'C am1 wont llll,o effect at the beginning of 27). Since we estimate rr.attlta for up to ten years po.st-po.aaage?~ this restrict.ion helps llil avoid including H1.u.ta1:1 with their own parmissive m uceale<l ca.rry laws 1n t.he synthetically consin1ciod unit After eutering the neccbb!l.r'j' sper.lfic:al.inu information into the synth prngn1.m (e.g., treo.tmcnt un.t., ll'lt of control slales, explanatory variable.<!, clc,), tho e.lgorit.hm proceeds lo conslruct the b-ynthet.ic unit from the liat of control 91.ut~ 1:1pecific to Texas and gcnera.t.~ value5 of the dependcnt \l'fil'iahle for the counterfa.ctual fur both the pre-tre11tmont. e.n.d post-treatment pcrl<lfl. The rationale behind t his mr.1.hodology is thn.t n d6e lit in these time series of crl.lnh htt(,ween the treatment state a nd RYU!.hetic control in tho prn-pl\fl6age 1 Our mi.o.lyllis i9 done lo SLa.l.a Wllng Lbe sy11lb Hofl.w11ro pc\d(.a.ge de,-elope<l by Al'bt;rl.o A bu.di", ~ Dlan.lll!ld, arid Jans Ht>inmucll~'T- 2tl~URbly Mpooldng, the algorithm that we 1~~ 11udo W (the W11[ghts of tbe componp..n.ts of the ~yntnetic control) that mlnlmw.'1s, /{X1 - XoW)'V(X1 - XoW), wharo V la a. diagonal mo.~rix incurporating information e.lmut the; rcltltivc wrights pl11ced on cilfferem predlc.torll, W!ti a. wctm of n.oo-oegntlve w,;lghte th.ll, lfllfli to one, X s a vector contll.lnlng J)r f>oo lrco.tment lnfonnat!on about the predlcuirb 11.M<)eL"\Le,d wlt,h tho treatment unit, ar.d X 11 A a n,atrlx conta.inlng pre-lrealmeni lnforni.;.t.jon about the predic:ors for &.! or tb~ cont.n>l uniui. For our mnin tmb.!yeie, we ~ tho ne,,ted option in Stato to generate tlc relownt weights. 'l'hie option wiee standa.:t<l Ot>t.lmlwt,ton tccliniquu to find the welghtil MtWcio.tcd with each prooictor t hat ml:i1lmraa tba pr& trootmont RMSPE of the ref!u.ltltlft. syulhol.tc control, 'th.e StatB. module ~ w1: tillc lililo can g~nerate the relevant wcl.tt.bwi using a lc&j1 comput.ntiodally exp~nrlw r~~to11-btuw.d todmiquo. Owing lo co,uput.,illo.11~ wn5l.raiut6 1 we use tbja npp.ro,u:l, lu our pla.oebo &lllllysis. n w e comdnroo using ono lag, t.bl'(l( ~!:, nud )'1llll'lY Jags as predictor11 and wn ovc\jlually choso to uso yeraly pro.lrmlllloot crime mtta l'>inoo that oplluu minimi>:od 1hr] on,rago c.oafficient of wri~1tiun of thu RMSPE during t he validation pwiod. 1t is worth noting that the eetiru,ale<l tre,i\n:wul d foct nkl!ocfoted with the pwlist\f(e of o Hlot1>-lc.."V<Jl n'c lnw reme.lde slmllilt fur violunt. crime regardiee& whether oue, ~bree, or cp,-cry poo9lble lag s ncluded along wlt.h the DAW, DC, LM, and Ml\.f predle,'toru (or whethlll' thes<; ~ are excludc<l from Lhe li.u\ of prod[ctora entire!}'), 22 ur cboioe of ten y,ilitl! ln tljja coutuxt is informed by the trru:leoff~ Wl!Ociatod with. umng a different thr.e fmrrlc, Uei.ut:: a longer po5t-j)8811o.ge period would cai!oblo ue to eetima.te the impact of R'T'C!awe for 8t~t<111 in which tb.ru:e we1e more tba.n ten ycru' of po11t.pa1,1111ga dfltl\, but ~ would liko}y roduco tho aa:w B.Cy of o,tt r.i\llmnlce oe tho cfl'oct. of tho \N!ll.tml'..ut ii\ earlier pcriod.y. This dogrl\ja,.ion wuuld occur owing to t bo cxdus[on o f aclcllliunul c;urdrol u~otc:1 irom coouidc:ation [n thu compociition of our s:ynthetlc C:Ollt::Ol, whlch would lend to reduce the quallty of our 5YlLLhct.ic control ~stimatee for the eru:ller ponlvn of tho p 3t-trentment 1)8rWd. Usu,a ll Mhortor pollt pan&ge period :JRkll faillug to <:apturo cfl'oct& of rrrc la.'11111 Uult truta \ dccil.do to unfold. 17 Li Deel

76 Case 2:16-cv-6164-JAK-AS Document 45-9 Filed 9/11/17 Page 19 of 34 Page D #:629 Case: , 1/2/218, D: , DktEntry: 17-9, Page 76 of 292 period generntrui greater confid~uctl in the nccurncy of tho collhlructed counier&u:tual. Computing tho pot-11,.. trcatmeut difference betwoon the dopendeut vnrinblcs of the treatment b1:atc nnd the syn1hctic control unit provides the synthetic controls est.imate of the trc~tmenl. effect attributable to RTC adoption in that tita.te. 1. Synthetic Controls Estimates of Violent Crime in '.Pour States Figure 2 shows the!,'yllthetic coutrol!l gtaph for violent crime iu Texas over tho period frotll l 977 through :W6 (ten years n."tcr the adoptiou of Texas's Rl'C faw). The solid black line showll the nc:hio.l pattern of violent c.riu1e for Texas, u.ricl the vertical liue indicates when t he Rl'C lt1w went into ~ift&t. hnplom1c!11ung tho synthetic control protocol identifies throe states that generate a good fit for the pattern of crime cxporionced by '\,..xl\8 in the pre-1996 period. Thetie states!u"l:l Canforn.in, which gets o. WPJght of 57.!:l perr.p.nt owing to its simjlar attributes t.o Texas, Nchrnska with a woight of 8.6 pcrcont,!l.ld WiRoollilir.1 with a weight of percent. One of the advanta~cs of iho synthetic controll> methodology is thal one can ru;scss how well the synthetic coulrol (call it "aynl,h.etic TCJtM," which is id!!nw1ed in Figu,~ 2 by the da..'uled line} rnald1es the pre-rtcpnsflltge pattern of violent, crimo to see whether the m ethodology ill likely to geuernte a good fit in the ten )'6lUB of post.-pn.<j.<jagc data. Rote t.hc fit looks rather good ic1 mimicking the ri5cs o.nd frllll in Tcxn.4, yfoleni crirno frum 'hla pattern increases ou:r conlidenc.o Umt eynthetic TH:l(ti.:; will provi,ie a good prp.diction of who.~ would have happened in TolCU had it no~ adopted a R'TC law. AnoLher advanb.b'l-l o( t.hc synlhetic control,; prot.ocol is thnt one can con.e.ldar the attribl.ltes of the three atates that ln6ke up syntuetic TexnR to see if they pln.usibly makh tho foaturcs.hilt generate crirue rates in states actos8 Lhe country. Looking at Figure 2, we ~..ee that, whilo bot.h Texrui and synthetic TP.:xaa (the wr.ighted average violent ctirm1 perfonnanc of the throe rnt1ntfoned stnl.h!j) 11how decllniug crime tntl'.j iu the post-pu... ~ge dec;ule aftm 1996, U1e crime drop f11 aubstaniially greater in synthetic Toxa8, which had 11 R'.'C law over that period, than in actual TuCllS, which did. AB Figure 2 no1~, t en yen.rs after adopting lth RC lnw, violent crime in 'Texas wa.s 16.6 percent... MrAer Lhan we would have cxpeded ha<i il not t1,dopted n. ll't'c law.'}_,3 Figure 2 ahiti illu11trates pt~rlm~ the most. important ]R.'-'SQn of camia.l inferenco: 11e ca.nnot sinaply look before and after nn ovcnt to dct.-ortnine the consequence of ~ho event. fut...hor, one ne<'-d.'> to estimate ~he diffc'.rnn,~e hetween whai did unfold nnd the counterfuctual of wha.l would hava unfolded without t he evm1l. The value oi the synthr.t.ic controls methodology is t.hat it provl41cs a highly tmubparent c.stiiuate of thnl. counterfuctual. Thus, whon Lot.t (213} quoteti a Texas Duitrict Attorney suggesting that he h..-i.d reversed his earlier opposition to the at.ate's RC lnw in li.~t oi.he perceived fnvora.blc c.-qicricncc with the la.w, we i;ee why it can ba quite easy to draw the inaccura.l:.e c.awm.l 1nrerence tha.t 'l'exaa' crime decline was facilllated hy its RTC lnw. 'l'he public ma.y perceive the fall.ing crime rate potjt..lrr!){l (the solid blaclt line) but our analysis suggests thr~t Tex11.S would have experienced a more si.ln.blo violent crime decline if it hn.d not po.aaoo a RTC law (the dotted liue). More spedfic!l.l.ly, Texas experienced n 19.7% decrea.<1e h1 it.s a.ggreg11.le violent crime rntc in the ten )'C!l.1'8 folluwing its RTC law (betv: and 2tX>6), while the sta.t e'a sym.hetic control experienced a larger 3.8% decline. This counterfuctnnl would not he apparent to residents of the st<\t.e or to law ct1foccomeut offi.ciafa, hut our result.q suggest that. Toxas'a Rl'C lnw hnpooed a large social cort 11 the st.ate. ~atux11&' violcn, 1,"rimc rato tan ye{l.rlj po,rt,..e.doption \i.'<c~ tba.t of 11 1J)'Tlthctic Taxas~ by i;i~;:t~ x 1 = 2.2%, Flgi1n: 2 shcjwll ~e B!ltlrru,t~ violent crime lncroase D 'ex.'!11 ol 16.6%, which Come& from subtf:14;1,lng from 2.2%, tho amount by which 'l'exm' violent etllno rate exr.eed""l that of synthetic. Tc;xas tn 199G = ~;t 1 X lcj() = :l,'7%. (See footmie :ll for furthet dill(,"1.jl!won of thi.,i u,?~ri!at ion.) 1R Li Deel

77 Case 2:16-cv-6164-JAK-AS Document 45-9 Filed 9/11/17 Page 2 of 34 Page D #:63 Case: , 1/2/218, D: , DktEntry: 17-9, Page 77 of 292 Figure 2!1 C Q 'O 1 ~ - :: ~ CJ..., OD i Cl. C( ) E B (D g C] l CJ ~ ~ ~,- Texas: Violent Clime Rate Effatl of 1996 RTC Law 1 Yearn After Adoption: 16.6 /o D 8 N --- treated unit synthetic control unit COmpoalllon of SC: CA (57.8%), NE (8.8%), W (33.6%) Slall!s Never Paulng RlC Laws ncluded h Synt:hefic Control: CA RTv-AdOf)ting Ste.CO ll)cud&<:' n Syntn~o Control: NE (27), W (212) The grroter transpa.rt'.ncy of the synthetic conlrols approach is one lliiv.wtage of thl~ methodology ov1.1r the panel data models that WH CQm1idered abovp., ligure 2 ml.w«ltl clear what Tm-1\8 ll being r.:111pe.red to, and v."'el can. rcuoct on whether tbi8 match is plausible and whether anything othor than RTC laws changed in theoo three Bta.tes rlurin..g the po.<rt- pa.-.."bge decade t.lu,~ might compron1.i.,;,e the vnl.ldlty of the synll1etic controls osdll.late of tho impact of RTC la.w!. Spr.ci.fiea.lly, if one ag?ood wi1,h sooe of Jolu1 Lott's writt.p.u work that the clt>.atb penalty ih E\ powf'..rfn1 deter.rent one might be concerned that Texas's far greater use o( the death pcno.lty during the poot-passagc period t.rum in the Atal.P,.~ r.ompriring ~yntl1(lt.ic Texas lllg11i. undermine the pr~iction tha.t RTC Jaws inr.rp.mp.!l crime by l6.6 percent n T6Xtu:t. 24 Dut the death. permit), O(A;Onllni to Lott, depr~ crime, llo to the cxtcut the dt:mth penall-y played a greater roio in 'f'e.xas than in Aynthetic Texas durlng the pog1,.pn.9s&gc pr.rio<l (relativo to the pre-possaga period), then our 1;:;tiruute of the in<:re!l.8e in violent c.dme generated hy the RTC lsw would ncl uo.ily Wldcrstnlo t.he Lrue increase. Figure 3 show~ our aynthetic co11tru.w e:rl.imn.te for P1mnsylvania, which adopted n U'C law in l~~j tlmt did not. cxloud Lo Phlfadc)phla until a eubscque.nt lt.1w vrcnl into effect on October ll., n thir CiJ.~e, aynlhcuc Penn.qyJvmi.la is comprisod o( eight states and thp. pre-))8.'!sngo flt. ib nearly pcrf~t. Following ndopuon of the RTC Llv. s, tiy11thetic Pennsylvania shows substantially b<?v...er perfurman.ce thrul actut1l Pennsylvanlu after tho 2 1\:=,j l!xl!cuted 27Ci cum'.ict& during the VQ1t.--pa1:mnge decntle whll" Cclifornia W1.~ct,t1;.<d 11, Neb~~ 2, nnd WiaC11Jjln!'-'\... '111.ed 11 oa (Death Pimt1lty 1Dfomu1tiou Ceut,or, 2U 16). Othet kt1}' uxpl.a.:natory 'ltlf[n blr.l1 might be.hf. lnra.rc.eratidn :uid poil l!c'l tt\l<a, BocaUAO 'TulM hlld &n eno: mo1;.q juuip l.n ll.ll lncarcara\loo rmc from 11Hl2-199G, 1,ho growth la tbo TO;lt(l.!i lncnrccration ru~c &om l91j6..2dti WM c:mly 'f.fl%, while fc.,.r "uyntbctic 'buut» the gruwtb rate v;1i,, 22.()<Y.,. ndividu~ly, the growths fot lic: Uw:c lly:ithet.lc control litatas were 6J!% (CA}, 6'J.O% (Wl), 26.6% (NE). The growtb ra!.o in the ThxM poliw employmet1~ rate 1)~ 11r t ho doc-ads wall...6%, while for "syn.l.h~tlc ''axas" the growth rala WM 1.11%. Jntlivtdi;aJ1y, the grow~h15 for the t.hrea 11ynthctio control stat~ were 9.()% (CA),.,.!)% (WT), 8.2% (NE). Cf>!n(' plo.wiibl.o crime elll6\,ititit:! for po11oo tltld prip1a R11g-g~WJ t~t 1w;,:ounting far thes1: ~wo factors could (:()nw.lvabiy shrink tl.ie e&fon.utod lmpa.ct. on vlofout crime by 3,,1:rtcDt,.Evun tli.ts wouhl l'lu&kcl!t that the Tollll RTO Law l:11:reaaed violent crime in ~ ~euth yeo.r by ronih.1.y 1 percent, AR we fio\ud in Ta.bl~ 2, l,hn <m1r1lu lncrnb,1 l.t1 l=,rc.omtion and pol.t;;, ni.tt1 were grnl\lcr ll' ntc irl.ates t..hiui iu 11-:kTC!t.at..,;, wh.ieh -v;ould tend tt, mokc our EStimatES ur tho JncreeEed vloumc.: io R:'C states OOf\/lef''1!,l;ivo. 19 Li Deel

78 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 21 Page of of Page 292 D #:631 RTC ln.w ls extended to Philiidelphia ill l!!t.e 1995, as illu~c.rn.tccl hy lht! tjecond verticil lint.l >~: ~l'i The synthetic contro!.s motho<l clitimntcs tlui.t RTC lnw13 in Pcnnsylvn.nin incrco.sod ita violent crime rato l>y 26.5 percent. 1l C: ii) -u ~ ~.O ~.., <D V a..l! tl ) '"] ~.: 8 - t.) '? li';,, 1: ii) ~ ~- N Figure 3 Pennsylvania: Violent Crime Rate Effect of 1989 RTC Law 1 Years After Adoption: 26 5% \ \,..,... \ 'a treated 1Jnit syn1hetic conttol "liiia"j =~:,i!lrri~~tjl~d:~~~~c~~:'8' ::.'~8%),l'.1(SU%1 PC-"""1>'~ 81,,,,.. lrt,mod n S\-Corlm: Ne (2CWL '4 (21»1), W (2:11') \ \ ' ' FiF;UCS 4 s\mi 5 1:ihow the comporn.blc aynthctic controls matches for l'iorth Cn:olina nnd MissiooippL Again both states allow good pre-pi,1..~ge fit be~ ween the vfolent, crime rat~ of tho Lreatment state.iad the synthetic control. The 1J1ctho<lotogy estimates tle.t RTC lnws led to 1111 increase in violent crime in NortJi C.a.rolin8, of 18.3 percent and in Ml!isillsi_ppi of 34.2 percent n D\1< 6yu~hcUc co:it.rols a µprq\\(h, wu ~root tlto year of p(l."1ligo t.o bo the fusl :,«ix it1 which n. RTC L'lw wm in effect for the majority of Umt year. Accor<l!Ja,\y, wu umrlc Phile,de)phfa'ij u,w J><"11!611g8 in 1996, aa dt>c\unonted in Appenclix B.J. :i.irn Ap]l<!Jldll( D l\'o kclude all 33 g?a.ph~ Khuv.iag tho pa.t.h of vttll!!ltl crinw for the treatm~11~ ~t!ltca lllld the synlj.letlc r,1mtrols, ~long with lnfarmawc,1:.1!lboul tho compoolt1oa or \l:c!kl i!ymlurt.ic conu'olll, Lha do.tea of RTC a<lopuon (ir B!\Y) rar 51.atOB lnc)u.<!od in thc:,c ayn~hatic co11hvlo, 1md t.hc estimated ttr,ut mcnt effect (axpm.~.etl in 1'\."Tillll of the percent. d111ngc in a pw-licu!jlr 1,;rlmc,,.tt,) t-on :yoora after &doplio u {or 7 yearn aft.er adoptlot1 for 2 ~tatoo th.at adopt HTC lllws in 27, sinr.p. 11.r data ur.ds in 214) Li Deel

79 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 22 Page of of Page 292 D #:632 F ig1lro 4 D C ~ "U Cl.;; '- ~ :,c... D... < 8 -. ~ : ~... C 8 C: G '<t J ~ a, North Carolina: Violent Crime Rate Effect of 1996 RTC Law 1 Years After Adoption: 18.3% ;. : '~-! '\ ' \ \ \ g N --- treated unit synthetic control unit Compo,lllon or SC: DE (9.2%), L (39.6%), NE (51.2%) S:atos N11vor Paoal ng RiC Luws nduood in SyntheUc Con1rol: DE RTC-1\dopllng stalaa lnd11dod n S~nthotic Control : L (214). NE (27) Figure 5 Mississippi: Violent Crime Rate Effect of 199 RTC Law 1 Yeal'1J After Adoption: 34.2%,,,,,, '\ ' '\ '... - treated unit synthetic control unit Composl!lon of SC: HJ (72.1'>' ), A {1.8.'l'J. N~ (1.1 'lli), OH (~.2"/o) &.a.n NeVO! Pustf-Q R'TC Lswa lncluc!ila n Syr.lhetic CO!V,rol: H RTC-Adoptlng stbls!i ncluded n Synthetic Cofllrol: A (211 ), NE (27), OH (24) 21 Li Deel

80 Case 2:16-cv-6164-JAK-AS Document 45-9 Filed 9/11/17 Page 23 of 34 Page D #:633 Case: , 1/2/218, D: , DktEntry: 17-9, Page 8 of Stat e-specific Estimates Across all RTC St.fltcs Bocalllle we a.re prnjectlng the vfolont crime experience of t he syutb.etic control over n tcn-ytlat period, there will undoubtedly be a deviation from the. ~tme" counter u.ctual a.nd our estidj1tr.<l cu1uiterfact1ui.l. One of tht! a.d-vn.ntngcs of our twtlc ib tluit we have a large number of st.at.es ndopt ing RTC Jaws $ that the avcr-est.imn tcs and und er-estinu1tw wi.ll tend to wush out in our Ul{'.an treatment estirnnt!!!l. Figure 6 Hbmvs the synthetic control esti.mat.cs on Yiolent crime for all 31 stutci, for which we have ten yoora of post-passuge do.ul. For 23 of th1;1 3l states adopting HT C lnwr, the incrf'.a.<je in violent crime is noteworthy. \Vhllc three stat es were e!>-timatcd Lu have crimo re<lu~tfom gt!lfiter thnn t he -1.6 percent estimate of Sou1.h Dakota, if oue averngos a.crolis u.u 31 R1,nlEltl, Lhc (pop11latio11-wcigbtcd) me.an trcn~ment eliccl aft.er Len years ir n 15.1 percent, tm:reatje in violent crime, f mu, instead ue~ 11n ( unwc.lght.ed) medio.n matlldure of cantt»l t.endoncy, 1-rl.'C laws nrt-i 1reen to incm:i8e crime by J.4.1 pe.rc.ent. 3. Less ETective Pr~Passage Matches S ection 1 above provided four examples in which tb~ synthetic ooritrolij approod1 generated ijy11thetic cont.rnls that mntcbcd lbe crime of the treatment states wcll in the pre-po.ssnge period, but thia docs uot nlwnyr happen. Again, one a.dvantago of the syntheuc controh1 lj.pprc.ach i'l tha.~ one can &1!6es:; the ~ture of thi!j fit n the pro-pruiimgc p eriod in ol'clor to dctcrn1ino how mur.h confidence ouo can have in the p-offt-pnssnge prediction. 'h't'o etata for whlc:h we would have consl<lerably less eonndence w Ule quality or t.he 5ynthetic control,; estimate 1m1 South Dakota. and MniJJe, both of which happen to show dcclfoe.<; in crime after RTC adoption. l 11deed, these are L-wo of the eight ata.~ell showing improvements lu crime fol1owing R:rC adoption as indiretc<l n Figure 6. Figure 6 ~ 2. -V,,.,,,af RrC?Nn (91 a 11'(8) i5-1l9 (1~) -~(61 1 Th Effect llf RTC L&wa on Violent Crime ~r 1 a Y ynlll llc Colllrat Elitlm11ln to, 31 ~i. ( ) An examinn.tion of Figl.lrC!l 7 u.nd 8 showing the ayuthetic controkl e::ltimates for theso t.wcj i.-tates provicl.es dnunntic visual confirmntiou thu.t t he mothodology l.t&i failed to vrovidc o. good pro-possagc fit between the 22 Li Deel

81 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 24 Page of of Page 292 D #:634 1:.Time performance o( t.ht> t re.atment statcr and 11, llltitable synthetic: co111.rol. Figure 7 South Dakota: Violent Crime Rate Effoc:1 or 1985 RTC Law 1 Years After Adoption: -1.6% --- troatod l111il eynthello control unit] CO'T111()9/11or> of sc: \ (62.6%}. W (37.~~) St11ta, N<l"1!f f'aa&ll'o RTC ur"" f~jd ltj 6yr,ll\8Hc CGl!li,:;i: R.TO-Adoptlf1\J Stal.in lr.dudm::l n Synil>t!1l~ Control: A (211). \\1 (2t12} Far South D~co1,1.1, 1 ona sees th.at tho synthetic control and tho RAte violent crime porformance diverged lung before RTC nrlopt.lon in 1985, and thllt 1 by tho da.te of adop1:ior1, ~yn~betic South Do.ko1.l\. lu1d a far higher violent crime rat.e that wae r!aing while actual South Dakota had a violent crime rate that was falling ln A similar pattern can bu f!~u for Maine, which ng11ln Uldermines confitl1m<.:e in the synthetic contml1, est.ima.tes for these two sta.tes. The difficulty in generating good pre-1,aasage maichcs for South Dakota and Maine Atoms fnim their tmll!!ually low violp.11t r.time in the pre-pn.'l.'lllg~1 period. Figure 8 Maine: Violent Crime Rate Effect of 1986 RTC Law 1 Years Aftor Adopllon: -16.5% ' r===- 1/&aled v l'iit synthetk control u~ Cornpc,sticnd6C: ft (18.C71'), A (84.~l & HV9r Pl!68fno R'rC w..,. ~,dldod n Synltleil~ C"'1trol: H RTC>-A.dapfng S1B1Bi1 nell.dud n SynlhatJij CQol~: A (2C11] Li Deel

82 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 25 Page of of Page 292 D #:635 Figme 9 Tho Effect of RTC L..awa on Violent CM~;;i afll,t 1 Yc.\... Sl(~lh lk: COr,trul l!l!llrn'ltm for 21 6talaa (1977-2D1~l r.m11frtc~ o U~;a) 46(~) 96-(1~) CC-o4(B) TW t.18 Fl Mwdl.an:: 14.Ul a... n ~ UT Figure g rcproducw Figure 6 while lenving oot th e four statea for which the qunllty of pro-pmsage Jit is clearly lower tha.u in tho remainiug 27 statcs. 27 Thia knoclul out ND, SD, MT, and WV, lr..nving n slightly lower estimated mr,rui but the same wedian effect of RTC lnws. As F 1 igur1: 9!!hows, the (weigh~(!d) mco.n increase in crime across the listed 27 TTC-t'\<lopting states is 11.3 pcrcc'ut: while tl.te (unweighted) median increnso la l4. l percent.. ncreases in violent crime of thh magnitude o.re trou hli.ng. Coneerisut1 estlma~ of the clast!clty of crime with rei;pect to incarceration hover fu:olwtl.15 today, which suggcsl11 tl:at; to olfscl the increase in. crime caused by R'l'C adopt.ion, the median RTC stato would need to approxi.ma.tely double lt8 prioon population. Part V Aggregation Analysis Using Synthetic Controls A smnll but growing litorature applie:; ~ynthetic cont!'ol techniques to the analy11is of multiple trca.tmoni.s. 28 We ~-timate tho percent.agll dufonm.ce in violent crime botwoon i!acl1 ~niatunmt (HT C-adopting) state rrnj 17 n pa1tic11lar, for tl.ese four stc..t.os, t l,o pro-~o.go CJVRlYSPE-tJiat t&, t l:o fu,.,spe t r;\o!fonnod into a. coo!ucie11~ of vnrlatian by dlvid.lng by avl!!-nge pt~~ijnau ~Tim~wna algnijicamly gre-atar tj~1 for \ hu other 27. See J:ibotnote 33 for fmth~e d isclb!do.ll of this statistic, ~~Tbo clonest p,ll)et t.o t he prt!!lruli MLudy R f>nb!, ft.od Zlppenir ('213 ), who lnt.ro<lur.e Lhdr,,wn tnot b odoloq- fur aggiegat\ng multiple evcnt,g mto e. llipgle l, tjt.lml\tt<l t,reutmco t offoct a:id calc:ulating ltu e~l18.1,.'t\llcc. T'ncir study contcr~ on the enet.:1 of increases in!.ha mlnlmwn w~e on 1:Jll))loyu1cnl oulcomce, 11,od, 11B = do, the autbor~ eiitlma tc tho paroontnge dlffernnce between t.l.!ll tre11t.mp.nl and tbe RynHrnt.lr. trhl l.ro n.he po.ht.-trae.tmenl period.,vhtl., R<'lrvc, p;lplltll analyui multiple traat..mentl: b:t aggregaujjll &lie l\l'wd c,[~cd by thc,o t reatments iato 11. single unlt., tlli& opprooch i!l nm wall-equipped to dsal wilh a Ct\6 sugh as RC law a.c.\optton where trw.ru~-:1l6 affect!be majority of panel t!j'.lt s e.rtd tuorc thun two decides sei)l!late t he dates of the tlrn ~11tl!Mt \rcll\tl'c 11t 11adcr cur11<idcrll.!ion, DB bigh.lig)itcd b Figure O. 24 Li Deel

83 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 26 Page of of Page 292 D #:636 the corresponding synthetic cont rol in both the year of tho tz-e11.tment n.nd lu 1.h~ itm yp.arn following it (we obvfously 118e dnta from n..>wer puat~treatment ye.ars for the two treatment statcs 2 1l that hnd ltc lo.ws that took effect less tho.u ten years before t.he end of our sample).:m We could 11sc each of thaic ten percent.age diffcrcnec! &S 1tr estimated eff<..octs of R'C lawa on violent crime for the ten poot-paaaage yea.rs, but we make one adju6tment to these :figure.a by subt.ractiug from each the pcrconta.go <lifforonce in violent (,Time in the adopt.ion ~ between the treatment and synthetic control stuf..cs. hl other words, if ten years aftor a.dopt.l.ng a RTC law, the violent cr.im.e ra.te for the state was 44 and the \ iolont crime mte for the ll)'llthouc coutrol w~ 4, oue e~tima.t.ti of the effect of the RTC law could be 1% (._44-4/4). Rather th.an use tl:tis cstimo.tc, howovor, we would.subtract. from this figure the percentage difference between the synthetic and treatment states in the year of RTC u.doptlou. f, tiay, that. value ware 2%, we would subtract 2 frotn 1 to obtain an estimated effect of RTC lnws in ycm 1 o[ 8 pc:rccmt. 91 We thm1 nggregntc all tho Rbl.tP.-Apedf:k ~t.lniate~ oft.he the impact of RTC laws on violent crimo and test whether they arc significantly diffcrc.ot from?.cm. 32 As we saw in Figures 2-5 e.nd 7-8, the validity of using the po!j.1.-1,rc'..&tmenl <Hfr.re11ce belw1i crime rates hi the trna.tmeut utate (the particular b"tate adopting a n'c Jaw that vro a.re annlyzing) and its corretlponding synthetic c-.nt.1:ol o..,q a rocns1uc of t.he e[ecl of the treatment dcpcn.cfu on the strength of the match hctwcc.n these two t.ime serjes in the pro-tre.atment perlotl. To g,m~rnte an eatimate of pre-treatment tit that takes into account cllircrenccs in prc-tr~atmcnt crime levels, we cst.imo.t.e the coefficle.ut. vru:lalion for the root mean :,qu.ared prediction error (RMSPE), whi.ch is the ratio of the synthcuc controps pr-trmtment fuv!spe to the p1 t,-trl:.ntmc~nl o.vorn.go k:vol of ~he ouicoruo variable tor Lhe LreaLment atalc.jj After genera.ting the 1ggnigaW synthetic controls estimates of the crime impact of HTC Lwn, rle.serlbed t~ the pa.rngraph above UBing the full ~1Lmplti 1 we comlider t.wo aubsamples of t reatment states: states whoeo coefficient:;i of Vi1ri1itlcm axe le!:!ll than t;wo t.imes Lhc o.vorl\ge coefficient or vtj.rlalkin ror 1;1,ll thirty-three treatments and states whose coefficient~ of variation me less tbll.d thi'l avurngo, We then re-tun our 1,-ynthetit: controlb protocol using each of these two :l8t1tese two ~ta.t.l!! are l(an.'1811 anct Nahr!Vllgj., whlcli Moptoo. ntc laws n 2U7. See footnote 4 ctiacu!l!ll11g the siatp_s for whleh we cannot emimate the impact ofrtc Ami u11!.ng Byntl!etk r..ol\l.rol!. DJ)ube and Zipperer (213) 11.varago ha cst.lmll.t<.'ti p<,ul-lro:tl,tnr..nt pcrt.cn~e,gc tlifl'orc111:cs nod convert th.is averngo Jn.t.c, iill claaticity. Thua, our work?eports eoperato BV(Jrlll!,e 1.rtl1;1lu1eul,;Jfoctl! for ten yrorly intcn'n!l following the time of the treatment, while Duba and Zipperer (213) emphalli2e an aver:age tre~tmer:it i:j,~, (.:xp[aisod 1118 nu elosticity) ootimatoo = the entire p~t-trl!stm<int porio<l, ~iuoth approaclm, should geaa:rate slmllm ll!ltlme.tee on average, and h:1 r'll~t nnr metljj.,uitim;:,.tlld orni1:ti, {m morn OO13en-aUw with our profomid approocli. T he intuitive rationale fur our choice of outcome,,,ui.nbla WM that pr&.tr~.t,ttnent diffm:11oes JJ,1.wcon the trei:,tmcnt ~tn~ tmd ibl! synu,ct.ic WnLrol a~ tho time of R'C a.doption likely roflecleu imj)qrfec;wqrl1j in ~he proc.t'ffl of genecl.!.4iug ijyuth,..1,i<; wulrr,,1 uud 1:1bould not wutributo t,;, our eutimnted treatment emict if pooaible. n othltr,,.,,rd~, if the treatment state had a crln!e rate that. wa.' li% K~tcr U1an Lhot of tho synthetic co11trol in both the pxe-trea.tment.,,ud pogt;..treatm~t pe.r!od, lt would.argue.bly be mleli?tuilng t.o!gnoro tl\,c pro-treatment difforonai acid ded&re that t.he treatm.mi.l incron.!loo crime rates by 5%, }'or violent crime, t.he 11San {median) percentage diffe-renee between th<:; t=t.mcn~ ~tatc ruid tho syu~luitic control in the year of lhc trootmcnt was 1.9% {1.3%), with. 18 crnatment Rta.tM flhawlng g.tt-11t.>.t r.:titb!i r,\t;f,il ~l1.n1 their aynti1c~ic controre and 15 treatment states showing lc~s crime than thoir syr.thetic c-.ontrol lu th11t yutr. A~ ;:- mi ult, our cmimo.to of th.c OJDoun~ by which H.TG w.ws incnaaed violent crim.e WS lmuer than would b.e.ve heet1 tbe CMe 11::de.r tht: altc:mativc appraoch. a:.ttbib ta.st i~ p mformcd by rogr881!1.ng tb.aoo dlfferences in & model 1Lsin8 only t. <:OOl!Vl.tll wnn imd oxamirung wbecher that conirt=t ill sta.tistically sig))ifirnnt. 'l'h1!6(1 regreooione are '\Wighted b) t.)u, popujtl.tic;,h 1,1.i' t br: l;.r~tmco.t ~tufo itl th.c p~-t-trcatment year under con.sldoratfon. ltobust sts.ndard enors corrected fo~ heterookedmtlr.lt,y 11,re ul'!l!d lu th.w 1mnlyniu. 3 3Vv"hllo tho ll'mspe i& oft.11 used to BBSES11 th.lg fit, ;,.,e b ell6ve tlmt iha llsl'e or 1,hl~ ml'.mllte 18 not ldcal in ~he prce(jnt conl.b.l<l uwing to the wide vu.fotion tlw.t exist. in the avw.ng~ pr,rtrea.t.mi:-ne ur,,m., rat.<:~ among t,hc 3.:1 tr<:atm1.,'tlt lft1d.u; tha t we conaidor. l<ur example, the pm-treatment H.MP.SE B&!Ocla.ted wlt)l our aymllc:tk 1.-ontr,,l anru~i~ u.:,ing tho DAW predic:tor variables a.nd a.ggrognte v:lolent c1ime 1111 the outcome variable la sunllar :/or Colmado (:lt,. t} and Malhc (36.'}, but the pro-troo.tmant le.vela of Colorado'& aggrogato violent crime rate Mil far greater tj11u1 Maiuo'R. 'f'o \,,,~ rnom ApOC.iRr:, Cr'llorndo'g Bvorngo violent crime mte prior to the implementation of ite rrc law (from 1977 t immgti 22) WM 467 vlofout crimeg per 1, resideuti., wbi!e the corresponding figure fol Maine vras 186 v-.olent crimes per 1, resl~1~11. Fr.:>r t bll! rcv.iion, M: b.avc groo.ta con!:ideuce. in our cmi.jnatcg ~At in the teath y,.:iar, Coloca.do's RTC luw llil.<l det;r~ vlok>ot critnc by 1.23% th.au we do in ru1 cswmate tlllit Maine's law bed decrell.!18d violent crime by UL5%, &.nee t.be per<:et1~g1: lmpr~ion in our tjynihotlc pm trea.tment. me.tclt for Main.a la so mu.ch greater than Coe Colorado. 25 Li Deel

84 Case 2:16-cv-6164-JAK-AS Document 45-9 Filed 9/11/17 Page 27 of 34 Page D #:637 Case: , 1/2/218, D: , DktEntry: 17-9, Page 84 of 292 fh1b1;amplea to exruniu~ whether re8tricting our etitinbtion of (.he average i rell.tment offed t.o stnt cri for which a rolath'\lly 'bettor syntht--tic control could be idontllied would meaningfully change our fin~. A. RTC Laws ncrease Violent Crimo We now turn mu attention to the ai;gregatcd results of our synthetic corucol analysis using predictors derived from tha DAW epccifk.jlj:ion. Tablo 9 ahows our l'tlthll~ on tho full aa.mple OXAt:nining violont crime. a~ Our estim.nt~ Huggest th.at siu.tao thu.t p8l!a-ld RTC laws llxperitmcoo mum deleterious 1:hang~ il1 vl1mt crimlw1l ai..-1.ivity than Lheir 1,-ynthetic controls in the Len yem"s aft.er adoption. On average, Lreatment st.ates had nggregate violent c:rime rates thnt were o.row1d 7% higher thun. their eiynthetic controltl live ycll.nl after passage and almost 15% higher ten yea.rs a.rt.er p!l.~sa.ge. Thblc 9 sugr,este thnt Lhe longer tho l~tc law is in affect (up to the tenth yoo.r thf.1t we e.nalyza), the grmter the cost ln terlll.ll of increased violent crime. Tnble J: The mpocl of RTC Laws on tho Violent Crime Rate, DA'W covariate~, L4 l"litwlo " l'-\'aju1. Proi,,rnlml d. C..-...,.t:,1 l~..:<>bo - R~ &t.d i;m,1 l'nl,,..'tilan.t.c:~ ~i...,. Rod..,...,s~., lll>jl'l'll l'n,()ot11oo d. ClftUl'OMl.ff PM:<bo l!arl...,,,, 8wJO-,t aj JJl l.md () ('J) /Ji l ~,---{~!O} /;;; i!)_--id) (lrir- O.ll9 2.14'\ -:r;~,- ~b- f.t1u~1r io.iii~~m ' 1uw""""ti.~ rr- (1.1:.1~1J (1M2}~ _ (2.!41J ~(2.ro) (J.'m) (l.6'1"1)~) 1J.l JJ ~ 3l l<l J.J ~ ~ 8l J O.i& 11.lle,l (UJ2 D. l~i D (1111 n.1,13!1 U Ulll Q.Ud um, ().(U:l 11.l!lll ""n ll.l"i 11.l ll'l O. R U.1.2l 2 U.JJ4 n.1:12 o.~14 O.OO ~.~'!.l(j 11.M.113 n.1:1q 1~13!1 n.,.. U.l~B n.,q Q,l)a;j ~.mo Q,1114 Wllll.11,'> DX.i n.r.ti lltg8 1() UJ)(il "B~ At Ul)..,..ii.. - Cll(uu.a. u;umwl" ia:b:t.l ~~,-:r-uul. r~;.im; N a-c:ml"' oj t"klw!a u.ap.\. "~ n1:ltlu! ja ~ ~ tirtw~ \bt p:ua..p iltlkttuo$ ia t.m -,td.a:c. i:,uo,t r.1,k la L.rou.mu.i:.,.,.-,.U:,tlic ~ uiee t i. it\-ta ~L-1nlo; ~ am. l dtt..a,:,t c.t.rt \.it&lucd. Poi:iw~h-tb,,r.r.cr+W~~~,w.~ $( ft"cl.-s, AR" An A J pt, QA ru '(! xv t.a t.,tk, MJ Ml"! WO Mi ltl' NO NTl p;z..":m ~ Oil QK oq P,\.so SD 1'N.x CT Vft,, y.", W\/ Tl:.u,-,::t.:utic ~,v,e.1 ~,o.ttata UM ;.4a.:'&, ~\l&b.t.. ~ 1h,a bu. 4b:M- ""=n sc::-r t"'1 Giat, 1a ~ ~ &ot)' oca,::rtb«,cs ta tl,t ~ td.1. <MO..., <a.a.,,... p<ncu 'Tu.blc lll r~peatb tho Tablt1 9 annlysiti while dropplog the four tit.ates with n UV of the F{.l\,SPE that i~ above twice tho average or t.ho sample. Toble 11 u.<ics a more stringent measure of assessing how well the synthetic control fit~ the pre-pashage data. by dropping the sbt states with an nbave average CV for tho Rl.1SPR. t i~ striking how n.ll three; tables show roughly idr.nt.ical conclusions: RTC la.~ ru:e consistently tthown to increase violent crlme i;tarting three yea.r11 A.fter paasail'o. Tobie 1: The mpact of RTC Laws on the Violent Crime Rate, DAW covadate:1, < 2x Average Coefficient of Varlatlon of tho RMSPE (J 11)~ {~) (51 (&) (7) (~) -----r,;r \t~ D.c67 2~ t.91" J.2J~"' t:wt - -1=.~111G"" = &t.1S"'iifil"' j,}. oll!~)~l-:. 1!4) (~~ ('J.~l8i <,.~ ~) (1;!!'1) {2,'lr':~ - <~u,.,r,_., 4} (MlOl (~.l>jij ~ (1...,i"' ~ ,, V...,w,l'.\ m, n11<1n!m.1111< 1\,l'lll 11.lll2.!l! U.Cl>l ll 1~1, llllu 11.m.: l'ropcrt\ul ti C:<rnapond!a~ PlMicbo ;:.,1,,,... Bu,,lfla,,l it. D f,.,d <1.1~1.1< l'Ni.:11:fl O ~U.2Ml O.lllt U~ l'n,pml!< d. Coat,p-U.1 l'\jqbo!ztwab!lipl&a.nl,u:, U'O<,ll!H O.llD O.M O.llK Oil& U,Jll ~i O, Llff n.rn~..11.( ~d Co,r'"p~ PMr.~u bttm.1"9~1.t,ill L<,nJ O.W 11.Q:.,i U.11'-'l l!,;1 o.im n,!<li _ o'"'o'=lll====n~.o..:. '..:. ' <1.:i.1 _ s~i::nqnt.~ (!filwm mw.l,u, lall:.&ba ~~ ~ ~ rauf.,l,ir~ lf - aa.mha of~ lu aj.qj,a Dt~ ""(\t\.w la Lu t!:jj...:i.a h ntttu \l.a pm:u&q, ~ tu U. ~ a~ 1111,i.al bat.~ U1(,y.,Q,U a,utd rt..l ~ lf"'9.1 """'"...~ ~Lw...t.JJl ' Ha.9 ~ c.b,. 4"""\uaw. ~"" ~ b: L~ c-ui:flj.1 hrm rwulusa t1w1.,tu~ - L et<,\( AA A7. W Ft. U./l D X9 ;.-y ~A r.m Ml l,uf :WO lo!!l SQ )<r; l'>< m OB ox OJ< VA ~o Tll TX UT v~ WY 6wla-J..dol/t,rp<>1_._,!: l.(t"o!owv ~ flyllhdk a:unrla 'MC\ t.o ~le,nc ~ ~,Jrq.k,t '",ha ta\j.t. \nrr. ~ ~ird 'Uin{ u.. rt1i"'tfi1na c:ar:thr::uwtm 4~fbc.:J.b. \bt l"mb {,,!'i>\, P<:.hJ,..,,.::'1':K., u 4 p( ru;n 34 We dl!l<:um the,ynthetic control~ astillllllffl fur murder wtl property crlmu i:n sectio! G bulow. [n au ~"1.W4."ll, the t~1~b-yooj effect is allll'd.yli 1J11itiv&-Ulffi<lt!lin.g -.hat RTO ln.wb increase ttlme-but not ~t.atlstical!y sll(llillcaot. The poinl cstima.tai llcr69 lhe four!w)(u'jh 1<UggoRt RTC AW lnaewm Lhc t1111rdor rb.te by 5 to 6%, Uld the property r.ruur. rate by 1-3% llller!en y~m~. 26 Li Deel

85 Case 2:16-cv-6164-JAK-AS Document 45-9 Filed 9/11/17 Page 28 of 34 Page D #:638 Case: , 1/2/218, D: , DktEntry: 17-9, Page 85 of 292 Table 11: The mpact of RTC Laws on the Violent Crime Rate, DAW cov11date.s, < l x Avcrag<~ Coefficient of Variation of the R l'\1spe _ ::-{l) (1) (~~ l«(d) {1) _ (B} : ~ jloj-, 1\11 UY.l. ~kl G 19r" ui.- 7,O.. inn. u..un 111.'l!llr'"' UM- ~ - (1.tl'<) J l,14) (l~) (\Ull:ll.._~,.ltl) ~~(!UO} - ~ (l,llfj 1' Tl lt 77 :rt ----W- r. ~1 ~ 2 ~ - 4 Pww.ln f VMoo MU o.j1s n.1p;,.118 n.ooo o= l'rcl>crllo<. ol C:n,"Ptndln1 f'llall,o lr.<!mftl<tl Slr,tl O>nl ol.1,..,.1.1 l}.1~.1n ~.,a, o 1n cuss _..., o/ Comw1,otdiu31'1>«,',., Tl«lo"W 81grlll "" M,«; U,,. r:mporl.bs ol c-1""'.t!"«,~... F.tlm.w StotlltaM u ~ 1..-'ll (lo&:).16.11ft.::, Q.{132 ~ ~ O.GO.11$ O Ol! ~cltff!lla pvctl.ara (".alum. :uunb.n r""""" J:"'U.-,~.,-. ~ ~, N' - a..wn al,!aial m ~.1\44 o.ou o.roo a.can b.joo (l.1'll.24 {l~.13-.1:m -l ~ Cll~ o.= oo-.i (U)r-) a~ _ ~dad.,...,w:>. r. ua- it11tt1:,.,.. ~ tu 1W--ORM.p dl!.aa... t. ~ ~ 1:rm::a t&ja UJ lr~ a.d. ~ tmkd.ata:t.i a.\ sfl,1" f Jlll"...ln.knm ~ and.i w::. a1 '-\a ""'\m:!i Jt..dt,i,-,..-vud hr,i.. ~ 1:.-m..,...J,q;.mm. dm ~ :R,tllrill la 1,TVU7, A.K. Alt Ali (; YL Q.A W!';! l<y L-A M ~tl MO M!l NC NM N\ 11 Ot< C) H PA SlC 'J'til,x t: r VA Vt"Y!1.-1 ~J.4 ina ~1-a~1 K&MTNiJ..Nt'C :.tl> WV TW ijo-.lnlcc cm:i..""d.l 1.,.J kl,-,a&w lk.. pl,.,-41)) t,tttw.j,. la Lt.a q,i:vt ~.. _. ~td ~ \b. ~ ~cw.uv ~ la lb» Ulla t.ut, p < CJ.111, <:O.iU,,._.., p < UJJl B. The P lacebo Analysis Our abillly to make val.id inferences from our synthetic:; control oeliroates depends on the 8CCU'acy o{ our 11tru1dard error estimation. To test tha! robustnes!i of the eto.ud!'rd errors th.l:\t we prctumt wider t.he first row of Tables 9-11 and to gei a sense of whether the coefficien t.a that we measure are qualitn.livoly lu.rge compured to those tlml would be produced by c:hance, vm incorporath an ana.lyrlr1 using plarnho treatment effects Hlncils.r to Ando (215). 36 For tllis analysis, wo generate 6 ects of randomly generated Tl'TC dates tho.. a.re designed to re:iemble tho fl illtribution of n.ctual RTC pa..,@.ge elates that we mie 1n our analysifl. 3 tl For each nf lbe 5 RHU! of randomly gonerat.ed RTC dal.ee, wo then usc the synthetic control methodology nnd.he DAW predict.ors to estlmate thirty-three synthetic controls for ea.ch state whose randomly generated adoption year 13 between 1981 a.ra.l 21. \Ve tl'ic th.is <lala to estiruat.c the percentage dilfcrcnce betwcc:n eacii placol>o treatu1cmt and its corresponding synthetlc control during both t.he year of the treatment and each of the ten post-trea~ment years (for wh!ch we bnvc <l11l.a) thnt follow it. We then test wlict.hcr the e1:1timntcd tn.~.1.tmcnt cfjcct for mcli posti-treatment year is statistically significant (using t he methodology described in footnot.es 23 a.ncl 31). We afao repeat our esti:mation of the average treatment effect associated with (>.lj.d1 of the hm poat trcntment years aft.er excluding stat.cs whose ooefficicut of variation is either ouo or two &hues the average obscr-vcd fur all (placebo) treatment 11tate.s, leaving u.s with 3 coefficient~ and p-volues corresponding to &i.ch of the 5 sets of randomly generated pla.ccbo treatments that we cousi<ler. At the bottom of 'l).ble 9, we list t,he proportion of each post-tren.tmeot year's plocebo rcgrcsl!ions th.at wore ~1>1gnificru1t" at the.1 h.m!l,./'i lovel, and.1 luvel. We: provide tl.ir.so proportions to give.ho reader J 6 Ando (215),:x1>mincs th<l lmpoct of CO UJ-tructing nud~ plants o n local real p~r ropite tbjmhlc in<;omo iu,jnpan by generl\liug a eynt.het~ ooutrol for ev,:ry coostal municipality that irwtalled a nuclcac plant. Wbi!o the ave~ll)r,\l ~roatmenl cllcd me,a.qlj.t'«j. in our paper di/tcni from Lne one used by Aodo, l\'8 folk..,.. Ando ll, ~pc.olcdly eattuu:iiing averagfl placebo ell"t:elft by rancloialy uelecting d(lforunt aceaa to lli.'rs'll as plaeltbo trnatmects; ('l'ho sheer number of trejllmocj\:8 that we torn conalderull!" in tb.lr ajl'u)-m5 p'6vbilt.a Utl from Um.tt[ng our placebo treatment 11n11.l:niill to statei1 that never 111fopt ltc law!\, but this ~lmj)]y n,mnr ~b~t our place.ho ostimatui will likely be uiiw.d ag~rul!lnd.iog a qurhlf\livol,y sigr,jfirjl.jj~ o!fect or t't'c laws Oil crimo, lll.n~-.: oom.e of our pluccbo t re&:me.r.i.tii will be ca-1,'turing the e!f<.'l.t of the p8!1&15 of B'.rC luw~ on crima r11tu~.) 'l'he 11<,1 un.l aveft.&u trootment effect cnn theii be compared co tho dfatrlbu1lon of aver8e p4ocobo trea~t cffeda. Hee.llfiL.. o.nd P el<:roon (214) also perfonn A ilimillll rluldo.dtlzatton procedure to estlmalc ~ho signlllc11.t1co of theil eiul/llltod avru-i-,(!i! Lroatmer.L cfu:ct. ~vnro c.t e.!. (213) perform a Hlmflllr test to CP<Urninc bow tlui n~rago of <liffaront p lai;el,o effects comp~~ to ilia llv'l...,.agt: tre«.tment effect that they cneasnre Wtlng synthetic control techn iques, al:t hougb!jura randomitation proceriurc di.ffar.' from olll'll by tcztrlctid~ the tlrlllng of placebo ~ren.t-mont.b ui Lbc axacf. dat P.R when a.ctu11l treo.tment..~ took pl1m:e. Mt>,forc spocilic{l.lly, we 1andornly chooso olgt,~ "'o.w& to oe ~"(...- pwsa ln'c li;.w,s, sec sta~c,i l,o pg."js TC luw11 bdorn 1981, 33 ~\.ut re to paai ltrc lnws between 1981 nod 21, t1ud three sta~ to pa.ga theit R1'C laws bel.w~w ld 214. (Wll!lhlugton, D.C. is not inclurlc,<l io tl:e placebo Jl4l} &is Slll.C() i~ ill excluded Crom our me.ji; ll!ullybie.) Th<:C ligure?b... ~re cbooan t o J'n.irror tho number Qr 11tlltC8 in each of tlic:1e categuric.:ti i:j1 our adu;:il datu set, 27 Li Deel

86 Case 2:16-cv-6164-JAK-AS Document 45-9 Filed 9/11/17 Page 29 of 34 Page D #:639 Case: , 1/2/218, D: , DktEntry: 17-9, Page 86 of 292 fm i1.1tuitive 8Cruie uf tbe poesiblo hin11!u:!bocin.tcd w[th our stand1ml!!rtor e.'ltimn.tiou, 8.lth.ough (for LhH reason noted in footnote 35) it is likely thn.t these placebo estimates n.rc cnpturh1g 1,ome of the effect of HTC laws. Table 9 show11 that the pl.n.cebo results tippear to be significlllt at the.1 level 2.2 percent of the timo for our first ycb.l' n..ftor pasaa.gc to ii.8 percent in Lhc t.e11th yeru. fo other words, Lho stan.dn.rd errors~ report nt. the Lop of Table O o.n, potentially underestimated, as our placebo a.verages are stotii;tically signifiolnt more often than would be expected by cha.nee. 37 As nnother check on the i,tatistical sign!ftcuncc o! our results, we compare l'tlcl1 of the ton codllcieut etttiruates in Tuhlt1 9 with th11 J ii;tribution of U,e 5 averugs placebo treatment effect!! tlni uae the samh crime ra.1e, posurcntment year, ond ea.rnplc o.a tho given cstimote. To 11BBist in this comparison process, we report o.!jhtmdo p-value which is equal l.o the proportion of our placebo treatment ttteects whose ahaolute value is grco.tt:ir than the n~olute value c,f the given estimated treatment clfet.-t. This pseudo [rvaluc provides ~\llother intnitiye measure of whether our mrtinul.ted averago treatment f'..ffoct11 are qualitativrly ln.rge compated tn tho distribut ion of placebo effects. Our confidence that t he treatment cftcet that we a.ro menauring for RTC law11 is real incr(-1..a.,-;cs if our C!Slimatoo treatment effect. i'< siguilico..ntly gtea,l,er than tho Vl\.St majority of our cstinw.ted average placebo troat:tmmt effects. 38 Examining ou.r ]Jl:ielldo p-valus':l in Thbles 9-11, we sea t.hat our violent crime result.\ Me always slnt.lstically!rlgnlficant in compntmon to the <llstribution of placebo coeffident.::i a.t tho.oo level after R~n years or rnure have p!uq:joo :;ince the treatment dii.ltt. C. Synthetic Control Estimates Using the BC Explanatory Variables Table 12 provides syntheljc cont.rol esu.tml..es of the impact of RTC laws on. violonl crime using lhe BC mo<lal'n tret of µredkt.orn.aa This model est imn.ltjh t hat R'C 111.wa increase violent crimo cum;istently ii.ler adopl.ion, rli;ing lo ~3.3% after ten yca..cs (algnilica.ni M.he.1 lcvcl). 'T'his tenth-year effecl is n.lw quite clru!o to the corr<'.aj)ouding DAV.l model's synthetic control ostitnate (Table 9), aa well as tha DAW o.nd BC panel data models' dwruny vn.rinblc cooflicicnts (Thi>lca 4-5). 3t 1n gooeral, the ulil'mmce between ~iio propo rtion of plooooo results li.l&nificaa.t. at a i;lve11 lovt1l and the ~l~lfica.nce level lklolf ~1uics across crlnu, rnwu und t:rentm,"'t 11clcction crite.i!l. We do not nbhl;l"v': ajjy con.ql!it.t;,1t ~cmkm:y inr the Migni6cance [.,.. d~ i,nd proportion of pl1icubo result1, slgnl.licn.nt 11t tho&e levcli, to converge whun rootridng th~ ~runplt1 to atlll.lj:l 'Wi.tb a relauvely lt>w R.\f8PE. 38JlctnUJ9C oft.he oompu~mion.al den,f.lndl! required to J,>Orfonn t!u! aualy6is uni~ U-,a mo...~i11111rri li.lrolihoo<. C6~imstion t.ecj, 11lqu,..--t!tt1 nut.ed optlr,iu Umt wtj employ~d in our main aa,11.}irio (mentione\l in footnote 2}-wc iwrform twa plooobo analysis using the ~yut~ module 'a!le Co.ult regresalon-t..t.:tcd technique fur aitimatlng the M--ighhl!SSigned tu en.ch predktot wbon constru.ctj.ng the &ynthetic cont.rot. 'llis change Rhoulcl bw our eetll'nnk.11 o.gaimt ftndhig n signilica.'lt effi:<:t of kl'c lawfl on crimo. Sinro the n('j\,od option tcnc!s to imprnvo the r.r,i,tc;hiog process bl)twccn atutas, we W\luld expect Jargttr deviations betwwro our placebo trea.tmeot& and thelr synthutic controls v. il,;n this option s 111 utilized. Th!.11 would auggeat n\nro dispersion n olu e&tima~od pl11cehn t.rcatment elie~ut ru1d.,.,greater ljlcel.lbnod of mtlmatu1g tha.t our a.ct11ru tt~lmant affects ~c, not slgnifklilltly dhn.,rcnt from the d1j1tribution of p!ocebo dl'ccta. ln<locd, when pcriorn,lng c-11 corlicr version or thi!i plaoobo {m~lyisij;, wo found that our e&tin1ate<l pooudo p-valueg wi:ro comll'vlltlvely wtinw.ted when comparing our non-11~-d pseudo J)-~ "'1= with il::ofk: produced ualng t he ucsiod' functloc. 3 For certain tnmtment ststee wjt)j O cx.ecutiorui prlo, eo n:rc adoptlou, the synthetic: c:untrol program UJ ttoable to gener,.lli; a countaruu:tua unit. To l:'esolve t h!jj problem, and t.q 1..J.ntillll conslllt6tlcy in the procbrr of generating a. r.ount8lmctu.e.l unll for the 33 LreaLmenL.ste.tcs, the E1Jtei: 114lom vari11ble is drnppcd from Lhe BC lljodol in the synlhc\ec control!; aoal~[!. 28 Li Deel

87 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 3 Page of of Page 292 D #:64 Table 12: Tho mpact r,,f RTC Laws on tho Violi:mt Crime Rate, BC covari(les, '1 (1} J?)~i) {4 {~) {O) {'} {~) ('ij. (LO) -.~.---,Ncror.lllllJ<d --,=-,-::;,l's' =-- _-;;-D.3~:47.111,ur.f.li1-t.ma 6.21,!, f..1! t" n.~&r-- l : : n.urr' 13.JJT... {1,17) (U.!11) {l.9'1) (WU) (:1.-9,8} _ 13.o!o) (a~) (4..B<t!) ('1-8l8l ( 3,~ N ~.JJ ~ Rl ~ J:i~ - =:13~==31=====- Sl ~ 31 --=--- --'===----:: Skoda,J.~la~ G:.hJrn.a!l'l'lh.n l Uc.414!)'1M,-~,-aula a::nld...~ r< - mn1nf 1Jf ~ ta-.:zi;:it. ~ ~!a ~ddsa,e,, b«waq ~ptt~twia.:u..la llll ~rr.maj~ lo trtlliu.j.lm~kc,ni\ta.l '31.lu U.fJwn pnt1fnw)ka ltlw~aul \um a!\ftai-..trn.at R-lh.1 r~ 111 u.- wulad. '-rm nnh:::a.& 1nm llm ~ _... _,... ~,\lu,lc() '!. 1M W 119KY LA M!tl.ll U>U{Ol.l9 ill' )1~'» rm ~-..1!\' OOK ORPA 9C!D'r~,-x O\' "' WV WY p <1'1, - ; <.S,... ;t <llill D. Synthetic Control Estimates Using the LM Explanatory Variables.n our Pnrt prurnl d.ut.n ann)ytlil;, we snw tbo.t RTC la.w11 were a..' cintod with 11lgnificantly hlger ratbh of violent crlrne in the DAvV model ('ahle d), the BC model (Thule 5, Panel A), and the 11M model (Tuble 7, Panel A), but not in the L~ model (Table 6, Panel A), nlthough both the LM and l\im modo~ did show RTC laws increased murder. Table 13 cetimalics the impnct of R'l'C lnws on Yiolent crime using Lhe LM :ipecilication. 4 T'he cieldniental effects of R;TC la.wt! on violent crime rntm1 ar1.111t11"ti1:1tically :.ignifica.nt at ihe.5 lovel starting five ycnra after t.he p8888ge of n RTC lnw, a.nd appear to incrcnsc ovor time. The trcntrucnt effects associated with Yiolent crime n Table 13 range from 1 'l.% in the sevent,h postrtret~tment ye11r to 12.8% in lhc lcnth poet-treatment ycnr. Rr:markably, the DA\\\ BC, and LM llynthetic control C'i;lt.imates of the impact of R1'C lam1 on violent crime are nearly dentical (compare Tables 9, l2, and 13), and this la true even when we limit the eruuplc of eta.tea in the mnnner ilp.acribed in TablCll Table 13: The mpact of RTC Laws on t he Violent Cd.me Rate, LM covariates, (l) P> 1s1 < l,~r ci5f m:::: (A) M 1 A,,,op;-1...,J,t,,,cl'!'tl> -DOJl Bl uw 7 &jj'* tiii'- 1.!ll.4'";- J~.mm - ~o--;.1;,:,!m"" 1="~' --- (1,:m) (1-~) (3,'?!!J ('J.2M) ~ (J.211) (3,1~ f.).!ll4} (3i!H) (~.,-l)) N 3:1 3l :13 J(J,13 ll 3:1 Jl 3l ~ 9itrnJ,JJt1N:1lu~ l.1i;lnnm nu"dl;wtj lr4jlu.t,,, Jo)fi.it.-.P )'lat wdcr ~bft; N = "lw.tw" of,..kit lti l!im~ Oc;uamlt ~ ta t.'if_ dlt\f'ln,a; J,,rtft,cill "-c ).dc:ct,hcc ~ to 'Lli Tlr.,l.rz.t ~ ry.1,1! ' " tr~ Md Q1li!utic cnttral,-.1;at.&t ~ lk ft tjttl~tm lblcn l -.d.. dsu= ef,ht 1~'.ftl NU111~~fi,-ll-atttt.Y.t.l kr.111.,-ult~,., 1h\i~ 8!.aiN!n v,;:o- Al( A1\,.l'. co.f(,.o~ m K! KY LA YT., JdT MS ~iu M rs i.ft ~ o lo) rifp: t,'j,,1 NV OB OC. OR. "'PA. f!,!) T h' rx vr \ 'A W\' WY.. P..: o..ro,.. P : o OJ,..., < Q.c;n E. Synthetic Control Estimates Using the MM Explanatory Variables Tobie 14 provides synthetic control estimates of the impact ofrtc laws on 'Violent. crime using the Moody und Mf'.lrYeU (28) pred.ictors. 4 ~ The tn.hlo revoolil that R'C statc5 c:xporionc.cd ov~rnll violent crime rate.<i that 4 Tho modified panel data Mal}'S<:! of L:M and MM, ehcrwn ln Panel 8 or Thblc11 6 ruid 'r, djd 6nd RTC law!\ ner= violent criino. n conductiug \lie l.m vaucl llo\.a tm.,,lym, we 11M?d ~ho vlolan\ am. propcr~y ru-r~t mtce r&thor ti$, \l,o crimc-~pcci.fic nrrest ratb!l d~bed by r.oti autl Mw.Laro (1997) ow:idg to the fad.but ~W,. " 'OulU - mticj.ly (a.nd impl'qperly) pl.lu;,i tho imm.e vnrlabw 11 both al dee of tlu! regreaalon model. Thill obj ectlon s Jesa import.lint tnitler ~ho ayn~hetlc control framew(ltk. fobr thle rcmoc, 1Yll uso the!? oonj.empcuancoua e.rlmo.!i,ccillc n.rroet ra.te6 Jn our synlhallr. ".nnt,rtj roodol UJ1lng the Lou 1U1d M 11~Wd (lllg'r) control VBJinbleR. ~ 1 The tentl!-yt!ar effect n tl.te 8}'U~UQUc controls an.alyiiie UBing the L~{ vu.l'l..~bl~j N 12.5% when we elimlna:e t he ~Le. Le:., with oore thlul twlce tbe averare CV of Lhe RMSPE. Knocldng out the flix stl\tefl with 1'-bovu-n.n~ngo m.luee of tb!s CV gr~u}rllle:i = almost idoatice.l Lll.G% effect. We ~1~ C!ltimotod the i mpact of Rte lav."!l on violent cnmc ulliog tbc &ynthetic controls npprooch a.nd the LM model rnoclu!ed to \111(1111.x DAW domogmphic V1W.ah!ES. This r.hfll-.ko lncteu.eod tho eetims.ted teoth-yfllu' ln<:ri:ooo i:i violent crlme1 from 12.8% to 15.3'Y- i~ For U1e ~me n :,,..uolf c.l..,..;riljcd fa ootcotc 1., we USB the lagg1-'<! violoht or property crime arrei;i; re.te ltt otv "'&rulll1ion. tabl,:,1 out 1, t.ho contempomnooue violent or property crl.tue a.crwt m\c 1JB 11, prodi.ctor in our syntbetlc contruw codo for tba 29 Li Deel

88 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 31 Page of of Page 292 D #:641 were roughly Hi% grcntcr t.ru>..n thclfle of Uwir: Hynthetic controla ten yearn nfter pnsiingo, which wa.q Al.1LLh1ti1:Rlly i;igni.licruit at the.1 levcl. The similarity of tho DA\.V nc, LM, u.tal MM synthetic cont rols estimates of tho impact of ftc la.ws on crime is striking..moreovcr 1 thew four sets or estlmateii a.re remarkably consistent with the DAW n.od BC pa.ncl data estimates of the impact o[ HTC laws, whirjt bolsters the CABe that the DAW and BC panel dat.o. specifica.t!ons provide more reliable estimat es of the impact of a:rc laws on violenl crime than. either the LM or 1,[\1 modcls." 3 Table 14: '.l.'bo mpact of RTC La.W!t on the Violent Crime Rate, MM covnrintos, lljij M () ii) M -.- ~ "' ~ i;~o,-m~,u-,,..., ~'.n>=~o.~1-~1,-,!.!34 :,.uo - '-~'Gi,.,n.,.. &.196" - ~"' N ~llol) ~.'1G) ~; &lj) ~~-W} gru) ~.171} g} 1'1) {3.91!9) (,.~~) ('3. 1~ $ ;~-,.en.n~ tw'at!tll ltlm~ ~ poa~-c:,-.i a:a1a cmald1n lka; )( a.a.'tw til. ~ ta MY.t{ Otpadfflt..,,.,t..h.J. L,:J,.,tJtbnllrA ~ \h... P""~,ll_t'.of,,,pl!d b:, U., 1.kxtl ctme n,.:k,!ti hmts::rw::c t.:1..:1..,...~ ~ -...&aa 4 f""- JWW.. t:r-1-«!m(f1,'llol.,.., ~ ffira,,r4 lln,n.rm::u JJ-.f'#SU- ~f, 1h'.t ~M 'kn1 N9fJ.L'b1 fmn t tl&,.,...ra.!lta.'ullln~o: A.k A1\A'1il"U ttr,qlt. 1n ~ R'l" r.ji, Mv..P.ff MN' \.!OW.9 1' NC ND l'i'1r-l,'u!<voa (>XOf\ l'.a W UL T K" ' l'jc 1Jl' VA Y'iV \\"Y ~ : O.O,... p...:...5,... p ; O.Ul 'Turning our 11.tt.eu~im1 to property crime5, we find littlfl syelerru1tlt.: avid B1H:H that- R.TU lawe influence property crime in the iiynthctic control approach. Our aggregate property crime a.re neyer significant. F. Does Gun Prevalence nfluence the mpact of RTC Laws? The wide vn.rijltioo in th e sto.le-speeifi.c synthetic control csti.mn.lcs thnl wl\.'l sc.cn w Figures 6 and 9 suaacsts that greater confidence!ilionlcl be rt~pooed!n ~he aggregated estimates than in nny indivldor/.l 1:1tate elltimate, as iweraging a.crob! a. e11bstantia.l number or states will tend to eliminate the noise in the cstit1111tcs. Another wo.y to distill Lhe signal from tbe noise in the E1tate-spccific estlmt.tes l!:i t.o cum,ider whether there is a plnusiblo l!xp hu11.1..tury foct11r thtj.t could explain underlying difforenc:es [n how RTC uduptiou iuflueacea violent crint11. O ne pooiible mcch.i.nism could be lha.t RTC Jawo will influence crime differently d cpendln1 on t he level of gun provakmco in the Rta.ta at t,ha 1.imil uf udoption.. MM lijleclficatlon_ ~ Ab we ha.w Reen prevlomuy, leavlug ou t Jla,le,, wi~h lnrgor CVR,1PSEe barely ch.anga;i tho re!mlt.q: F;llmluall11g Htn~c, \\1~h twice the nverago CVRMSPE J.c~K lo..,. "~U1.41Wr.l t,mtb-y-oor uffcct u1ring MM vwiabl.ee oi l!i.%, wul "l,hu.il:uliu::s tbok with 11bm'O'nvm.age CVH.i.VHP~ values leads to an est.jo..,.w,;1 i:jlect u( 14.7"'A,, Wo ru!o ~timated the mpact of!uc ltlw11 11 vjoleot crlrttr, uelog tho syn,jwttc control!; approach and.he, MM lruldel mod1fletl t n 1~ llix DAW domograpbic,'llliables. Tb!s change inw'-"llloo tho oatimatod tont b-~br incroose n violent cr!mea Frum 15.3% t o 15.4%. 8 Li Deel. 8 ~

89 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 32 Page of of Page 292 D #:642 Figure 1 nia lmp,act or GVh Prvvalonc4 on lhb lne1:,q Vlolont Cr1m11 Duo to RTC L~"'11 (6,,.ntlurllc Control Eellma!Jt, )..... _..... "' "... ~.. W r:t- ;tf/ ,--- o.ro o.n o.oo o.m o. 1.1G o.eo l'mlr&;a F111dle<i n1!11,lelcloe C<!nlfflllod \\1ll, "1'! n N 3 '\'Mrs Pllol D RTC l\c'ai,1jnn t_, The S)n:!'lali~ c:art.ni 'tl-aolmottt Elllt:tdllfllo')-.d lo t,r lf,_. m"l!(7lt,, ~. 9Ul, 1th),_, allcr R't'e l><iopl'o'l ll'ootjnen. E!lod 14,.7 +,t.l, R7 Q.J~ P-nca. t 1,M ; R-'2 2,7, Rf,g'ffllioo ~,iightad P/ pcpl!lltt!wl Figure 1 shows the 11CJ1.ttcr diagram for 33 RTC-udopting; stat-cs, nnd relates the estimated impact on violent crimq to a measure of gun prevalence. (Gun prevalence is proxied by tbo commonly used measure 11bowing the fmctlon of tltticides inn Btn.te th1j.t are committro ~ih guntt.) The last llne of the note below F!gurc 1 pro\idcs the regression equation, which shows thn.t the gun pre,,-al.encc proxy is poslt lvcly t cln.tcd to the etltima.ted incrca.~e n ctlme, but the coefficient i5 not stnt1m;icn1ly fdgnilicant (t = 1.54) and the R 2 value ib vory low.o\4 The population-m:lglj.ted me~n gun proxy lev-cl across our 33 states i.s.64 (roughly the level of Montunn)r which would bo llssociated with a 14% higher rate of violout crime 1 years nftcr El"C adoption. G. The Murder and Property Crime Assessrnents with Synthetic Controls Ueca1.15e the syuthotic controm estimates of the impact of RTC.lawtl ou violent crime 1u~formly generate st:11tistically significad.t cst.irnatc.s, we have hcrctofot"e focused on th.at!ldjllysii;. Our synthetic cont:tol cstiain.too of the impact of R:TC law:i on murd~i, an<l property crimo oppeo,r in. Tubles A3-A in the appendix. ':l,'hlle in all cases tha lcutb-year effect for these crimes is p61tivc, lu uo cwie is it statistically lliguificant at even the.1 level. 1'\lt murder, the point C8timH.tl'S suggest Bil incroo.se of 4-5 percent, and for property crime, the point estima.too l'lw1ge from 1-4 percent lnc:.r~lfflltt. The rclntivoly smaller impact of RTC lavni ou property crimo is not BUiprising. Much property crime occum when no l1ne i-1 around to notlco, so gur.i. use is rnu.r.b less p11t~atially relevr.. nt n property crime scenarios than in the C8.ti of violent crime, where victi.ms a.re ucccs&l\rily present.. Most o( the pernicious ~A bivn'[a.le rcgrt:l!sion that wght11 by tbe lnveniu of tbc CV of ~he "R.MSPE, rather the.n hy Btll,~ popufo.tion yieldi! ieoult~ sub!ltantively ld~n~lcal to tlwee (n 1''1glll'1l 1, Wo OH;O repeat tills iu1al}'813 wbcm dropping the 5!ltJ\l.il8 wlth the womt pre-p~o U\. (NEi, \'{V, M'l', SD, 1U1d ND), s.n<l 1-hlft modtflcauou o.gn.in dom nol R11b!t~11111lvdy ch11.11go ha li'lgtlte 1 rcgrcl!!llion rcftul\ft. 31 Li Deel. 8 ~

90 Case 2:16-cv-6164-JAK-AS Document 45-9 Filed 9/11/17 Page 33 of 34 Page D #:643 Case: , 1/2/218, D: , DktEntry: 17-9, Page 9 of 292 effects of R'l'C la.wll - with the axception of gun thefts - flre likely t.o opern.te fn.r more power(1tuy to im:rease violent c.rimo ni.thcr than property crime. T he.s.nuiller impact of RTC laws on murder 88 opposed to violent crime ma.y simply reflect the great.er difficulty in gc11crnling prccis<: estimates for a far Jess numerous, and hence much more "olatile, crime category. AltornaHvely, tho greater ability of police Lo atop murders tha.n overall violent crime llilly explain why tho synthetic controls estimates for murder arc wco.ker thlln those for violout crime, since we know from 'Tul.,le 2 th&t R:TC.statlll! inc.re:uied poli(,"e employment by 8.39 percm1t tnore in the wake of trc a.dopt:ol1 than did non-rtu b"tates. Thi1:1 fact al11 would have been.jc~cted to suppress mur<ltml in RTC 1:1ta.tes (relative to non RTC states) by about 5.6 perc(mt.~ 5 Since the syntheuc controls approach does not control for tho higher police employm~nt in the pcltlt-adoptiou phase for R'l'C stnteh, it may be appropriate io elevnte UtP. synthetic controls estimat.cs on murder t- reflect t.be ttrurdcr-datnpernh1g effect of 1:hoir jticrcac<l police presence. Such an ruljmitment won.d yield stntiatically significant increiuies in mtu-der attributable t,o RTC laws. Part V Conclusion The ext1m1:1ive array nf panel dat.a 8.nd ayntb.etic controh1 esti.matrui of the impiu:t o RTC l~wo that we presenl uniformly undermine tho 11 Morc Gmw, Less Crime" hypothet1is. There ib not cvon the slightt\st hint lu the dn1,11, that RTC laws reduce violent crime. lndccd, the weight of the cvlrlence from t.he panrl rljj.ta estin:mtes W well as the gyuthetic contl'ol.ti!w1llyeit1 be!lt supportu the villw that the a<loption of lttc laws t1uotjtonifojjy ra.i.s~ O\'erall violent crime in the ~en years after adopuou. n our htltial pfl!lel data u.na.lytlhi, our profor:red DAW specificatlou rui well l.u the BO spe<:if:ication predicted that HTC laws have led LO alat;.istico.lly signilicll.o.l a.ad suhs~autial incrco.\ies in violo.nl crime. When tho LM a.nd 1fM models Wate a.pproprfo.t.ely a.djllllted, they generated the la.me findlngii, but evon without adjustment, Wiese modols 11howcd!TC laws increuaed murder aign.iiir.o.ut.ly. We.hen supplcmcnled our p8.llel dnl.e). re11uhs 1111ing our RynLhe1.ic oonhol methodology, a.ga.i.u llsing tho DAW, BC, LM, and MM epecifir.jdions. Now the nn:rultll wore uniform: for a.11 four epac:ificatiorui, states t~t passed m'c laws D(perienced 1~15% higher aggregate violent. c..time rat.cs than t.hcir synthetic controls o.fter 1 yca.n, (results ihal were elguilicant nl citl1hr the.5 or.1 level aft.hr five yoo.r~). The syou1etic contrljlll estimate,j for the n11pact of R'l'C lawb 11 murder aud property crime were also uniformly posit.ive after ten yeare (but nol st.atisticolly signincant). Thfl Hyrrthetic con~rola effocth that we lfle5ijurc ropr~ent mea11ingful incr~.u;es in vio)p.nt crimf! rate.;. following th~ adoption of H.TC laws, and this conclusion remained Uilchangcd ojler restricting the sot of stn.tes con.'lidored b&!ed on model fit and after c.1ltlidering a la.rge numher of roburt.netia chc<'.k.'. 'While our placcl,o llilnlysia suggests that the stnrnlard errora nssocin.tctl wilh roritc of these estimates mn.y have bec11 bia.<red downward, the size of our average e-1:1t.imated treatment effect in comparisoo to the di'\trlbution of placebo effects indicates that tho deletcrioub e1 ecta associat.cd witk RTC laws that we estimate for o.ggregnt.c violent crime arc qualitatively large compared to tiiose that we would expect to observo by chance. Tho consistency acroe!8 different- specificatione and methodologies of the finding thnt RTC clevntr.a violent crime enables fur strouger conclualons than were pos.sible back when the NRC report was limited to analyzing 4.'STbo importaul rcc-<1nt pap~r by Profeeaor,1 Ail.COD Chalfili a.nd Ju11tlo :McCrary oonditdffl t.hat hl15her polke employmanl ha11 a dampening ejfoct on crlme, e.nd, most strlklngly, on muf,;l,:r, Specifically, Ch.e.l:in and McCra.ry (213) find e138ticitie3 of,67 for murder but ('mly -.l'l for v1olunt crimt!fl l;l(ld -.17 for property crime,. 32 Li Deel

91 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed 9/11/17 DktEntry: Page 17-9, 34 Page of of Page 292 D #:644 data only through 2 with t.he single tool of pol\nel data e,'lllunt.ion. Nonetheless, cstimrtion u!lins obrcrvauonal data nlwnrs rest~ on numerous ll8sumptions, so one must always bo alert to potcntie.l shortcomings. Fbr exu.mp le, if states that wcro exp~ted to experience future increases in crlme were more likely to adopt RTC laws, then we might exnggcrn.t.n the detrimental cifcct of RTC laws on cri.rnc. Given the very linutcd ability of politichmti, pun<.liti;, Md even ocf.ldamic ex:pert.s to corre<:tly predict crime trend~ OYe r this pel'iod, though, this problem of cndogoncity is unfilcly to mar our results. Po.ncl data. analysis cnn be susceptible! to problems of owitted variable bit;i.9, bul ihe synthetic controll:l approach WM dcslgued to better nddrel:ltt Lhat concern. The rc.o;ults prosented n this pa.per o.lso help to eicplnin iho longstanding discrq,ancy that has cx:istod beiwp.tm the ecouomot.dc res11lh suggesting tho.t RTC biws increase erhllh fljld the percoptlon "on t.lie ground" that R1'C nv.'?! are not associn!.c<l with a contemporaneous increase in crimo rnt.ei;. The conflkt. between these finding.q Lq rcaolv(~ when one renllier tllat. flince the crime aplkes of the late l!j8,(jr n.nd early 199R, lll6t states l})q)ericnced lnrgc o.nd!mportunt crime dccrcnscs, including thos( adopting RTC laws. Howt:.>ver, our n.nnlysls euggerl'l th1l; had states a.voided nrlnpt.ion of RTC.aw~,.hey would hav11 experienced greater clrnps in violnnt crime. ndeed, ati Figure l illustrated, while HTC stntet1.l.ia.ve now fallen below their violcnr cr1nm rntea of 1977 by o.boul 9 Ot' l%, the states ihn,l dlcl not adopl H'.l'C laws enjoyed violent crlme drops from the la te 197Qij of over 4%, Fin.ally, while! this pu.pcr has focused on the suu-istical estimation lhe impact o( lltc laws, it is ueof.ul to con.<.idar the mec.hanisms by which RTC laws wc1uld lead t o net i1cteru;es in violent r..rime; that is, the i;tai i.blical evidence shows us Lhnt whalever bencliciflj eliec~tt HTC lnwa h.avti in reducing violence, tj:iey rue outweighed by graa.ter harmful effects. The most obviour m.a<:ba.nif!ll is that th!l RTC permit holder may cumruit a crime that l11c1!jr ttha would not hm<ti committed witlumt; the perm.it. A uuml,iir of high profilll 1:rlmes by lu'c pennit holders would seem to follow thie pat tern: George Zimmennan, the popcorn killer at a Florid~ movie theater who W8!! angry nt a futher teicting n. bu.ll)1litter, and the angry gill! station k.!llar (shooting a black Leen for playing loud rap music) nre all lr1dividua.js who would likely never have killed anyone hl\d they not. bad an RTO permit (''tou,a, 212; Robln'!, 211; Luscombe, 21'1). Of cou'11e, aggmvnted rui!!aults arn far more corntnon thn.n murder (albeit far less visible to i hc public), so the so.me impulses the.t generate killingr nl8o work l.ci tttimulate aggrnvntp.rl a.~ults (a.nd hencf! nverali violent crim H), Somo hav-e questioned whet.her permit holders coutmit enouiµi crime io substantially clevnlc violent crimi~ nality, citing a.pparently low rates of official withdrawals from permit holder :;1 convicted of crimes. 'l\vo points need to be mnde!n rcapouae to this clninl. First., officio.l witlith n.w11.ls clearly undetslat.o cri.miruility by permit hold~nj- For example, convictlonti for violent crime are fa.r smaller than l\.c\.ij of violent crime, s,, rne.uy permit holrlcrs would never face offieial withdrawal of th.cir p-ormitb even if they committed o. violent r.riminal u.ct that would ~rrant s uch termination. Moreover, official withdra:wals will be unnecessary when the offending permit holder ih killed. n the nightma.i'h =a for R'C, two Mid1ig1W pen:rnt holdiug ru-iyen, pulled OVllr l.o battle! over a -tailgatlng dibpute in Sept ember of 213 and cn.cl1 shot and killed the other. Again., without pe.rmits th.is would likely have not hcx:n a double bomicide 1 but note that no offido.l a.ct.[11 to terminnte permits would ever be r ecorded in n cru;c Hko this (Stuart, 213). The second critic!m point w that RTC laws also increase crime by individuals other tha.n permit holders lo n variety of wn.ys. Fust, the culture of g111j carrying cn.n promol.o confro:ntntionb. P rr.s11u1a.bly, George 'limmerllu.l.!1 vroul.d no~ have b.as.sled Trayvon ~ fa.rtin if Zimmerman had not had a gun. 1f Martin. W 11BBaulted Zimmcnna..n, the gun permit then could have been viewed o.~ a stimulant to cnruc (even if the permit holder v,ns not the ultimate perpetrator). The mc.ssagc$ of the gun culture Cillt promote fear nod 33 Li Deel

92 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 1 of 51 Page D #:645 Case: , 1/2/218, D: , DktEntry: 17-9, Page 92 of 292 anger, which are e1miuons thnt cru, invite n:or11 hostile confruutationa lciullng to more vtohmce. ThiR 1-11,t.itude may be reinforced by the a<loption of RTC lnws. 'When Philndclphia permit holder Louis l\fockcwicb shot a.nd killed n popula.r youth football coo.ch (another ponnit holder carrying his gun) over a dispute concern.ng saow shoveling in Ja11uB.ry 2, the bumper sticker on Mock:owich's car hnd an NRA bumper sticker reading "Arruoo with Pride" (Gibbons and Morat1i 2). 1f you a.re lljl angry yuung man, wilh 1mmewhnt of a paranoid streak, and you JJ.1,-en 't yet f.i.ca)u convicted of a crime or adjudicn.tcc.l to be n mental defective, it is likely that. tho o.bhity to carry {L gun will both be more attractive ond more likoly ln e. HTC stote. That sucl1 h1jividuals will 1 lh~refore, be more likely t.u be nggroosi~ once armtid ll,ld hem:11 more likoly 1.o stimulalt1 violence by others should not bo surprising. Second, imlividllill.s who cnrry gtlllll around o.re a constant i;ource of ni-ming crimlnrlt1.. \Vhcn Saau Penn obtained n p ermit to carry o. gun, his r.ar wa.s stolen with two g,tos in the trunk. The cnr was soon rcx;ovured, bnt t.he guns w1m! gone (Donohite, 23). n July 215 n Se.n Francisco, tho theft of a gun from a cnr in Son Franc-lsco led to killmr; of a tourist on a. city pier tbslt ahnast certainly would not have t~currcd if 1.hc lawful gun owaer had not l1~_fl; it in tho (.'ll.r (Ho, 21&). Just a few months Aw.r, a gun AtC>len from a.u 11nlocked cu.r was used ln two separate killingb in San Ftanclt;co in Octob1.:r 215 (Ho w1d Willinnlll, 21.5). According Lo Lhe National Crime Victi_mization Stu:vcy, in 213,here were over 6G,DO a.uto tlrnhs from households. Tho tnore guns being carried iu vtihicies by permit holt.hmi, the more t-timinalr will be walking around wit,h t.hc gujjs.a.ken from the car of some permit holder. Of coutse, the Snn Francisco killer did not have a ltrc permit; although the OWD.!!f of the gun 11sed in tho killing did (Uo, 215). Lost, forgotten, and mirpla.ced gun! arc i1uo~her daugorow, by-producl of RTC la.we, os lho growing 'l'sa l:ieizum.; in carry-on lugy,ar;c 11.LLcsL. 4 6 Third, BH more citl2t>.n..-i carry gun!, more c.rlminals will find it incroo.c;lngly bent>.fidal to carry guns and t1lle t.l1em moro quickly and more violently \,tj thwart any potentilll urmed resi11tance. Fourth, the past1nge of RTC laws normaj.facs ihe practice or carrying guns in a way t.bat may etl!l.ble crimit1ala to carry ellilb more readily without prompting u. challenge, wltile making it hnrdor for the police to know who ib n.nd who ib not al.lowed Lo p~ss Kuna tn p1ibhc. Having a,cdesignaled permi~ holder" along Lo take pos.sesston or (.he guns when confrontecl by police R~lll to bo an attractive benefit for c;rim.inn.l Hl~,menta ncl.ing in con~art. (Fernru1dez ct al., 215; Luthcm, 215)- Fifth; il o.lmost ccrtaluly add& to the burden of a p<:ilicc force to have to dcol w fl,h armed dl.izen.,;, A policemen trying to gi, a a traffic t ir.kel has far more to fcv.r if (.he drivor ill armed. \Vhnn a gun is found in a car in such a situation, a gr<'.at.er amou11t of time ib needed to f.18gartain the c.lriver'e status M a permit holde.r. Polfoo may be less enth.m-ia.stic about investigating certain susplclouu a.ctivitlei given the gcca.ter risks tl1n.t widc.sp1 ciu.l gun co.rrying poses to them. Police resources used to process guu permits could ln111,ead be more efficiently utted to d\rectly fight crime. All of Lhese facto1'll a.re a ta.~ ou police, and therefore 11 would mqicct law enforcement to be lees cltcctive on the margin, t hereby contributing to crime. ndeed, thl.s may in p&d explain why ffi'c states tend to increa.<te t.he size of t heir police forces {ml.ali ve to non-adopting stateh) after RTC l11,w11 n.ro ph.,med. The fact that two different types of 6tati8tical dll.tn. - panel data reg:re;;.sioll s..nd syntlaotic controls - with varyillg strengths nnd tiliortcoming1:1 and with different model specifie8.tiomi both yield consist.ent and strongly atn.tirlltcally sig:nilicnut evidence 1.ho.t RTC laws increase violent crime COBtitutcs persuasive evidence U.1at a.ny beneficial effects frotn gun C.'\rrYlng a.re likely i,ubsti\dtially outwclghed by the lncrea.scs in violent crime that thcac laws stimulnto. 47,j(]Soo WUUi1n1~ 11.Jld WALr!fl (2 1). ' 7 lt sbo1jlc:l be noted thll.t, l.'\-"l-'ll v, itb the 1.-nonnous st.ock of guns in chc U.S., the V",'3t roa.jorl~y of tho t ime \hat oomoone ;~ tbrestimed with violent crhne.no gun will he wlelded def~m1lvoly. A fivo-yoor study or such violent victim!zatklll/1 n tbs Unltt1d Ste.tca fo1.~1j ~ruv. vid.[ru/1 fa.ilod to defe,u.1 or to threat.eu ~ho crimlua.,.,.i~h a gun 99.'..i porceui or ~ho time - this in a coun1ry with 3 million gun~ u.1 dvilian hand!! (Planty and Trumn.n, 213). 34 Li Deel

93 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 2 of of Page 292 D #:646 References Aha.die, A., A. Diamond, a.ntl J. E1inmueller (21). Srntluitk control methods for complu'n.tive case studies: Ei, t.imating the effect of cnlifornia.'s t.obacco control program. Journal of the American Sfutistical A.3.9ociatfon 15(49), Abadie, A.., A. Dlanioml, o.utl.j.. Uainmucllcr (2D4). Comparative politics o.nd tl1c f:ynt.hct;ic control method. American.Journal of PoUt-icol Science 59(2), 4!)[,--51. Aha.die, A. and J. GardcaZl:lbal {23). The economic costa of confl!ct.: A case study of the basqu\! country. A111mioo11 Economic Review 9.9 (), 1 Ul- 1 :12. Ando, M. (215). Droflmfl of urbani.zauoo: Qut\n, 11.a.Live case studies on t.he loc;ul hnpacltl of nuclear power facilities UBing tho synthetic control method. Journal of Urban Economi~ 85, Anoja, A.,,J. J. Donohue, Md A.. :lh.ang (211). The impn.ct. of right to carry laws and the nrc t<...1)ort: The la\.el:lt, lemions for the empirka,j OVl;lhmlion oflaw a.nd policy.. Ame.ri<:nn /,aw and Er.onomio; f{,wiew l3(2), Ancjn, A., J. J. Donohue, end A. Zhang (211, November). The imp~t. of right to carry laws and the nrc report: The l.at1.1st lossorjh for t.he empirical evalua.tlo11 or law and policy. Working Paper 18294, Natiollll.l Bureau of Economic Research. Ayres,. and J. J. Donohue (2a). 'l.'ho lt\test miafues in support of the 'rnote guns, less crime' hypothesis. Stanford Law Jte11fow 5S, lohn, S., M. LofRhom, and S. Raphn.cl (214, May). Did the 27 Lc.gnl Arizollll. Workers Act Reduce t.hc State's Unauthori.7.ed mm!gr!l.l.1t Population'? The Review vf l!:conomics and Statistics 96{2), Ci.wallo, E., S. Galiani,. Noy, and J. Pantano (213). Cntruitrophi<: ll6tural clli!a.sters nnd economic growth. RetJiew of r.'con&mica and Stati.itics 95(5), 1S4!H56t. Center, D. P.. (215). Executions by state and ycai. Acr.e.'JBed: Chalfin, A. lljld J. McCt:n.ty (213, February). Tho offec:t of police on crjme: Now evid1'!.uce from u.s. cities, Working Paper 18815, National Bu.C1m of Economic Research. Cunningham, S. and M. Sb.ah (217).. Decriminalizing ndoor Pru11tiLution: mplications for Sexual Violence a.nd Public lenlth. Review of Ecmwm ic Studies. Revise and Resubmit (third round}. Donohue, J. J. (23}. The final hnllct h1 the body of the more g1ms, lcss crime hypothesis. Criminolgy e1nd Public PoHcy (3), Donohue, J. J. and J. Wolfcrs (29). &ti.ma.ting the impact of the dea.tb pens.lty on murder. AmeriC'ltU Law and Bronomics Reuiew 1/(2), 24 39, Dube, A. and B. Zipperer (213). Pnoled i;ynthetic control estinmtes ror rp.currlug treatments: An. applir..».uun to minimum wage cn.5c studies. Durlauf, S. N., S, Nowuro, and D. A. Rivers (216)..\'odcl uncetla.inly and the effc.cl of i;hn.11-l"l.suo tigbt,-tocarry lawr mi <:rime. EuroJJ(:lln Eco11omic Ret:iew 81, Li Deel

94 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 3 of of Page 292 D #:647 Ferr1H.1tcll".J., M.,,, St.ack, H.11<! A. Blliuler (215, May). 9!\.fP, killed in hiker gang shootout in WRC."O. New York Times. Ft)'\1r, l. G., P. S. Heaton, RD. Levitt, and KM. Murphy (213). Measuring crack coca.inc and its lmpa.c.1:. Er.onomic nquiry 51 (3), GibbonR, T. o.nd R.. Moran (2, Ja.nua.ry}. \ fa.n. shot, killed in snow d1!!pnte. Philadelphia lnqt.jirer. Heersink, B. and B. Petenion (214). Strategic choices in election campaigns: 1!.{f!ne,uring tho vico-ptefjldettlinl home state advrurte.ge w.ith synthetic controls, Availal>le at SSRN Ho, V. {215, July). Gun linked to pier killing stolen from federal ranger. San Francisco Chronicle. Ho, V.!llld K. William.a (215, Oc'"11ber}. GWl in 2 killings stolen from unlocked ca.r in fibhcrmau>s wlluf, COJ>fl 11n.y. San Fhmciitr.o CMrmit:k Kaul 1 A., S. {lobnor, G. Pfeifer, (Uld M. Schlelet (216). Synthetic control methods: Never use a.11 preintervention outcomes os economic preuict.m,l Keele, L. (29). An obscrvo.tlounl atu<ly of t,allot in1tinl.i~s 11.11cl stfit.o out~omes. Tuchnical report, \Vorking paper. Lofutrom, M. a.nd S. Raphael (213, December). nca..rcera.tion and Crime: EVidence!'rum California's Public Safety Rea.l.igrun6llt Reform. ZA Du;cru,,'Sion Papers 7838, nstitute fur the Study of Labor (ZA). Lott, J. R. (213). More guns, less crime: Unde1'standing crime and {/Un conhvl law;r. 1Jnivart1ity of Chicago }'res.s. T..ott, J, R. and D. D. Mustard (1997). Crime, deterrence, and right-to-carry concealed ha.ndgumi. The Jom'n4l nf Lr11al Sludif:.~ 26(1), LJL<icornl,c, R. (2l4, February}. F loricla. uliln fl,(,~c11scd of killing una.rrned teen 'lobt it' over loud rap music. The G1.1ardian. Luthom, A. (215, Jtmo). C-Onceoled carry draws oppoaiw viev.'11 - and a murky middle. ltfilwa-ukee Wiscon.iin J1J.mal Sentinel. Mi,lHkSB, T. K. (21.3). TLe P.conomk imp11.d. of nat.,u~l reoourc-~1:t. Journal of Env~ronm,mtal Economics and Matia.gement 65(2), Mood)', C. E. and T. B. Marvell (28). The debate on shall-issue laws. Econ Journal Watch,5(3}, Moody, C. E., T. B. lv!arvcllt P. R. Zimmcnnant and F. Alcmante {214). The impact of right-to-earry lawa on crime: An e.,;:ercise in replirat ion. Review of Economic.~ (1 Financ.e _., aj-,ta. M=ib, A. and M. Cnottahi (2Ula). Florida stand your gt<>u1d law and crlma; Did lt ttj.ake floridia.ns more 1.riggcr happy? Avail4blr. at.'jsrn M1,5fd9{i. Nonnemaker, J., M. E.ngelen, and D. Shive (211). Arc mcthamphctaminc precursor control laws effective tool.'! to tight the methamphetamine epidemic'/ Health economic.s 2(5), Pinotti, P. {213}. Organizp,d Crime, Violence 1 and the Quality of Politir:ian'!: E-vitfome from Sout.hll'N to.ly, Chapter 8, pp UT Press. 36 Li Deel

95 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 4 of 51 Page D #:648 Case: , 1/2/218, D: , DktEntry: 17-9, Page 95 of 292 Plru1tyi T\,f. and J, Truman (2Hl, May). Firearm vio11mce BJS Special RepJrt 24173, U.S. Dcpurtmont of Justico Durenu of Jlll!ticc Stnti:itic.s. Robles, F. (2L4, Ja.nnory). Man killed during argument over texting al movie theater. New York TirM11. Rml(ler, (). K. 1 L.-l. EU',Cll, J. Dowling, J. E. Stig:lit-i., and. E. ChettiAr (215, Fel1rnary). Wh!il ca11.<lcrl the c,ritnc decline? Columbia Business School Rcscn.rch Paper No Rudolph, K. E., E. A. Stuart, J. S. Vcmick, nnd D. W. Webster. (215). A!l.'lOClation br.tm.->cn Connecticut's pfillit-to-pun:hase handgun.aw 11.ml..homici<l81i. Amerimn Journal of Public floolth 1,5(8), e49;l5 1. Strund, J. (27). Should lcgo.l empiricists go boye&ian? American Law and Eccmomics RevieuJ 9(1), Stuart, H. (213 1 September). 2 concealoo carry holders kill each othor in rood roge incidr.nt.. Huffingfon Post. Trott.a., D. (212, April). 'frayvon Murtin: Before the world hcnrd the cries. Reuters. Wellford, C. l '., J. Pepper, C. P etrie, et al. (24). Firearms atid vfolencti: A. critical review. Nat-ioDlll Ar.ru.lllmieH ProRH W&ihin&rto11, DC. Wi.Jli.nnul, C. and S, \Vt\ltrip (24). Aircrew Security: A P.roctirol Guide. New York, NY: Ashgnw Publishing. ZinunermaD.i P. R. (214). The deterrence of crime through priva.te security effortii: T heory o.n.d evidence. lr1tcmational Ueview of fow and Ewnomics 87, Li Deel

96 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 5 of of Page 292 D #:649 Appendix A: Tables Table Al: RC Adoption Dates f.\ate Ellt\.;Jw nu. cl ltn:!.<,~- ~ndy'~l'- l\"l' i.. s,a1lo11 Coo.trolo A.cu.1;-,i,T Alt.l.\o.rn lojli J9?11 Alaoka J/1/l~.:11,, l.~ /ufx... 7/17/l'ifH o.~ l!jt)~ A,lmw.. 7nT/l'i#l o.~ JG C.JllnmL,, ~-A D Colm.i.Ju l/lj/~:.m '2~ c.:.:.tntcll<>jo 11ml 1!)'71) o..i;..,.""' "1/A l)httm nf t)ib=l:d "1/,\. fbl,!a O/l/lli !~ ~ R/U/19 l).jff~ 19'.lO f{,\ n,nll!d,,ho T/1/1 1).5(11 ltl'jrj ll1lru:la 1/5/:OM 2DU lr.ii&ia 1/1&/PftO o.ooi 1n.q) Jenn 1/l/:1 ll 1. ~11 K..., 1/1/:!7 l. ~1 Kntulcy 1/ l/loog o.m 1W, - Lou.ltla.u.a 4/ lyjj<gj U.'iU'l lffl g/u/1.955 n l~.\la'7iau! ~ /A ~1,t:h,1~ ~,. M;,i.!p,, 7/1/'r.i)l o.rot DJl Mr~ d/28/7.fll O.W'T!l;W M:ioi.lllllppl 7/ 1/199 Q,lj) lll!jo Mt.can 2/26/?i:.4 -M7 14 },ltc.liua 1/l/ l Nelic,,ili l/lf21ju7 um 'l.7!,'<>,ml,. 1/l/l!l'JG ll.:lu2 JDJO -~"" lwo1~41ro 19.ls l'l-<rtrt Jw.-:y N/A New L!.c::aLn 1/l/: r.on ~llu Ncw-Yor); N/A Nortb O,,,,U.,,.. U / 1,'1119S,85 UOG N"l'lb~!/1/lffl,41~ iJo,;a/:14,732 2~ Okl.,,liom.\ 1/1/lWJ l,joo P.5..1,'<!D 1/1/19'. 1,i)(l') lilll) l'mui,yh""'-'> 6/17/lf!W O.bCJ Ul8:, l'hllarl,ll)cln l~/ll/lg9.l.12.\ lffl!\lu:d, ]cbuul ~-/A l!ou.&lj~...uu. 6/l3/l o o.a.ia l{!j'i' South Othil&,/lfli111D Q,W, 19.l.1,,.,... 1/l/lDllil.m l!l'jd 'l'l,r... 1/1/lllOO Q.'2,jl t!l!yr t;~ 6/l/l9115 ~1 16 v- wm mm V"l7llbl& &/6,11m.&6.~ l!hl~ W...tJngtcm 1ga1 11 w... Vtr,;l<lU f/7/ O.Al\! CO W1m,o,i!n 11/1/rielll w,,,olln& )ON1W4 ~.~li'l l85 Note: An RTC ocloption year of O ind.ica.tes that a state? d id not adoj)t a right-to-carry law between 1977 and the early months of 214. f the fractio11 of year in effect is less than.5, the ll1'c date u~d hi Lhe Bynthetlt: c:ont.rol a.nalyais i." thti follmving yoo.r. RTC dates before the year 1977 may o.ot bo exact, since dhicre:ucee between these dates would uoither affect our regretl!lion results nor our ::iynthetic control tables. For <'.xample, we only rcfid Vermont'!l ~tut~ up to the ycnr 197 to confirm thc:.ro were no refcrcnr.()tl to blanket prolilbjtions on carryiug c.ouceti.led weapo ns up to the ycnr 1!)7, although it- oppea.ra given widespread public commentary on this point that Ve.."'llont n e-...-cr had o. co,uprchenhivc prohibition of the carryjng of concealed wcnpona. We follow earlier convention 111 the academic literature on the RTC i&1ue in as~gning R'f'C adoption <lat~ for Alabama a.nd Connecticut. 38 Li Deel

97 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 6 of of Page 292 D #:65 Tuble A2: Panel Da~ Violent Crime Co(;lllidenta using DA"\V, BC, [,M, and M1'v1 models, State and Year Fixed Effects 11.Jgbl,.t.G-('...rry f,aw Lugs,,d!J1carc=Liou Al~ Lagged Log of Pe!..plla lnc.nn;(r&uoo Raui L!iu.ed Polloe Employoo futfo Lagged r.~ or 811,wn PollC<! O!Hc<lt11 Per rt.c.ldimt Popw1111oo u.~cd,\nu,~ fu.tc 6,r Vt<>!= Crlfoo! Lawd Dependent \'iu-table R~!'",;,- (;',ipll11 Pnm<>HA\ TutOU>ll Ref. Pe:r C.apJle. UMmploymanl. lr41lflllic(l Jlt,,u J>gr ~pjtl\ lncojbe M&e!ClDUCO n...l Par C,,;ml\ Rctl:romont Pll;(l!!C!!bt Llld Othc~ (Lott wrlllo.u} R.{,sJ Pcv Co 111 flclircrmcnt Pf\YOOfllH tu1d Olho,- (MM,'JTllh:11t) Nnmlll&l Pet' Co.plr.a l.naune llnmcploymnnr. Rat e, Pavcrt7&!.r, Lo.ggod N umh<r ol r... ecauong li,o(lr Po1>ulaHu11 P\lrcor,t of t be pof7ulatoo11 livmj! in MB,\s r opllldl<m D<ln lly Obt!t!rvatimui ('T\lblu ~) J.Wf :Modol ( 1) 9.,w (2.96)..1 (.2) -.llri (.1) o.oo (.).1 {.77) ~9 (.-19) 65.:..ir- (17.8),9-t'~ (.2!1; (Table 5.A) BC Mudd (l) 1.S... (3.65) U.OD.. (~-~G) 3.:1.7 (1:!.59) - ooo (.) - 1. (ll.67).11 (.11\) 7L74 (!.ll.ll} 1~74 (Ti,blo ~.A) f,111 Model (J} (3.1~) -.16'" (.8) o.oo (.1. (ll.1} O.U4 (O.D.1}. {ll.1). (,C(J) -. (.2) l.!! { 't'11blc 7.1 } MM hl<idel (4).69 (.77) -(l. {.) -(M,\1.. {.2) 111.or (1.17) o.oo (.).1 (.1}.2" (.Jl) -o.oo (o.oo) -.3G (.2$).13 (.lll) -.D (1).) 1'/S Panel B : Sptlfl<! Modal Re1.tulu ('lablo -t) DAW Modol ('l'nhlo li.a) U(J ll.tocfol (l) (2) ru~jit..lo-cun., L.o.w..S (.!14) U.19 (.66) TroJ1d Cot CL.a,,.g,.,r Sratc,!l~' {.4D) O.UG' (11..53) J.~Jltl)Ci ca.rcet11tj1ru!kt~ o.oo (.2) l..oggc<.or. of P11, CaJ,ila UJClt/teniLioo \AJ.e :11.19 (8.1) L&gg,:,cl l'olicc t:mpleycc RAle -o.o~ (.4) L&gg.;tl Lag u( S11-cra l'ql1ro OifJoonl Pa:r fu...,idc:it Pa;11untl.oo 1,34 (13.57) ~ Arroil~ RAte m V!clant Grim<~ L,,g~ Dcpcrnm:.nt VwloLlo Real rer Capita Pm11nl"1 locow,,. (O.CO) R...t p.., O.pot~ Un.ampl,;,ymollll~ """'"""'" RM Pe, Cap:1e. lncomn Mun\m,o,:ice R"'1l P1 Caplln R...>U:Nurwllt 1'a.ymo,:,t.., JU1d Olhct (Loll vor:ijan) n...l f\v CQpl\~ tmirollloot Pa,ymecln 11:1d Other (MM \'C111loll) Nurwwi.l r.,.. cnp;1,. 1nromo U. (.) Un«oplaymte,,~ RAl.7 (..!17) -.29 (.Bl} Po~crty Rate -,41 (O.~O} L gp;cd Number cc fu.ecuuom t!.18 (. 17) llccr '- (1.Ja) 67.lB'" (lfi.41} Popul,,tion P,m;,mt c,j. i b p,,pul,,.t;;on llvinn ln MSAn, T ~ (.23} Popul.atiou DeusiLy OJOt.u-v,ilium, 1a-2a 1~?4 ('Tu.ulo 6.A) LM l>fodlil (3).41 (,17) O.J, (O.!ltl) - o. rn (.8) U." ' (.i>} - o.oo (.'2) a.3 (.3}. (1).1) o.oo {.). (-2) ROO {'Thblc 7.A) Ml,f Mod.el {4).11 {O.C8) -.7 (.~) -. (.} --.M (D.D2) F,6.!,J'.. (1.-1.7) o.oo (.}.1 (.1}.1 {D.Dl) -o.oo.. (.) -.27 (.2.3) -.15 (.11)) -.1)1) (.) l!atb=llr.>:in Jccludo yoo.: ll!ld f'.alo ~ i:jf<:<:tj Mld a.. '<l wtlgbk;;! by 1lnt.q pop~on. CoeGk:""'b OQ d<:mo(':l&pble vatl.a.blll8 and the c:o.tm.ul omi:tq'.l ~t atl\lldard otrct11 (cll:ljtcr1xl Pl tho stato lavcl) fl.ie p=-i<le<l next to point ootl=l"" ic pnn::~~~::e. Thorolln.,,, al,j lb<> crim n,!cft.to tjw Unl.f<Jm\ C:dme ~Otl (UC!t). w p <.1,..,. p <,o;;, p <.1, All tlguc1lil tiipdrtt...!. m peg~ C.QCJO!, Tlw DAW modtd.11! ru.u uu d:.h from ()1 4, tbe BC cuudd &um 1()7S.2(J l4, ti,~ LM.tndol fn;,m l()'j''f-'14. t.lid \bi, MM 1()dc:l (wlt11ou~ \w, cru.d< c;<jl'(jll() lw.ltx} frilln 197!l-214. t7rl 39 Li Deel

98 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 7 of of Page 292 D #:651 Synthetic Control Estimates of the mpact of RTC Laws on M:urder and Property Crime for 4 Different Modols 'Thble A3: The mpact. of RTC Lnws on the Murder Rate, DAW covariates, (1) 12) (J (') (6) (Gl (7) A-S'"""'""'J 'E!' Ull1 -:rl<il -l..' l.m-1-4.m -D.m.fi.tn ("'1_!.~) (~.9) ( ) ("uu) ~_!:ll) (M'1t) (o.alloj ~ ~ ~ ~ M ~ ~ ~ tu..,.j_..j rw i.u P~ ($] (8) ('11 1.~ -..1an un (<.T.R) (~.3>) (~P.n) 31 al " a1 Ol)lµl:11 ng11bf!' tojkt,b1 ~~ ~ ;n<# c,~ iw. N e sut:&.b 9f ff:tlfll tt..m;,:. C)w;,-!ml -..rkhl. W lm Q~ 'b.twaa. th.~ d1amaai l u tb munl"'" nda n l~ruia:d p,:! 1)':tbritn ~ '41- ~ ~- :;",11.-N&J:tUSL :4"'1..id 1.L ltw. al t.l.a b'm..rnt.e, AM.1.llt tq,:t1al tar lb ~,...i. t.effi1 N!lllliq &um~ ~ ~lolo, A ~""" All AR AZ co?, Cl\ OHl! KY LA )JR 1.1] l,lj; ),(o M:J»t' l'tc N1l h E :!nl NV OU Ult: CR l'a gc 11D r.. "X 11"1' ~"A Vl'V W'l 1' <.1. - J <.'5, - p cjl1 Table A1: The mpact of RTC La'WS on the Property Crime Rate, DAW covariatca 1 197'7-214 ==--- jt, (2\- (3) - (4J (J) (~) fl) (.1) (OL..._ ([o) 1t>=Nt""'1bal TP.' -o:il6 LL!} 2.1!1~ O.l'l>J U.l'4 l.34.l ~'13 Ull 1.DL~ 11,lil :.odusl...,. 1,.,,.,..._:.'. l.'l : 1 n ~ :,!! ;:i;j ai 51..;; J.:: ;(o.d,,!) (1,:tlV) (~!~:J (~.?#) _(lm:) {2Jil.B) (1,;i.il) _ (l.;m) (2.!!')) 1 _-=== = "' === - c-'nhmn aj&lbfft lujlen.c pt1tt-puulfjll J'U 11'-4,r ((Al'\JetMW; N' :r~ ~ ~ r,i Nit.;,~.. A'.4 ~'"7 r.tud nb ln UOM.1~, illu1 ~.u. aatml cu. W 1tn:i ~U: K.nll.l '6d M ~ d ~ 1:-.tulM D,\:~J.tot.._,.i.w. ll U,. d..ia.lwlnli ~ Qe paic.:tbllll dldwnl.t 1 Rau.i.- nr;crial trr... (W".tCMlW k.:~ :"W"'..lllA& ran -ua. ~ 3'-!o(,'<W AK 1.R.UC!1L Cl,\ ll> icjkyw. loal>.!wnlto ~ ~Mr ~OlfDlfiln!W (ll ()tlal\pa SC BD Tlf':l UT VA WYWY,..:O.H, 111 <llj,..., <run Table Al>: The mpact of R'l'C Lows on the Murder Rate, BC covariates, i1> c~> <~i {4l <'J (al (o> c1n> """"' ~'..i:iiliiiod - ~..,.,,.t P=..., 2.n, ,us,.1.w, -'J.!ml ~s - ~ u 1a 1.:n;,~74 (Mtll (HT,J (4.m1:, ~i.u,i {M:Jli} (4.ril&) C i.66/jj (d.lli:ll (3-'m ) ().JW) N ,:, :fl aj ~.fl ilj_ $1 31 n ~-=-=_=_=_=~-=-- n,~ crron la pa.tma.l~ ~ u«euuc's ba.:!.ca.t. pna-~j"ltl WX.. aa::ald,in.~; t~ -...wl W.ffllt..f a. amp3it ~ \U'lii,W).. th. aj.n,mii ~1' du~--~ th 1hf. mun.it&" Ne la. NAtn'J!Uri -."4 )'lll,,;u..: atutnj.iat..,rt,p,-.j->f\..lton.ia..:d b:.t t.nd u. ltnb ttf lla DWalzunt R-Ah -r-11f:"d tnr lib Ulcu1ad. Lem t"ajll:ja4 frgd ~ ~ Hlat- la pr.tr,,: J.. x A.\ A.'1. CO?.L CA m kfl Kl' 1.A )11~ ~U ~ YO Mil.t,,t 1' NC Nil N& ;i')( NV OR ()'k. Or-J l'il JC &.o '1 TX U'r VA WV W Y ~.:.1~ J:i<Oat..... o,am Tobie AG: The l mpact ofrtc l.f:lws on tho P rop erty Crime R.ate, B C covarir.tes, (l) () l3) (4) (~1{6) ('1) [B) (U) (fo).i.~'" N""1tillWJ 1 Jil> -L!T~ ~.f!h uw o..s.i~ o~.--o.t'lll 1.,li unn _Cl.fflGJ (1.:lMJ <~> ci.ruij can.'!) e5112; 1~.WllJ c;a1151 run> c:i~>' N :l:i 33 al )3 ~ 3.! 33 ii 3!!lllllditriui:natn~ Co~ ambin ~ PJ11Jl-~,..., tdlw ~tlmu 1i -~cl,la.la lll MU.lls ~ eir.&:.ls it U. ~ lm.-..-;!i 1la ~l.aii, ~ :a \At P)'«,1.ul7 en- r..i.. ld tn..b:l,,,ffl.tau,a.we. a:atnj d;:wa 1 r,.v.,,._bm1.a.ait ~ td'.fl. rj.o.it lb, L:nw'.mai J\moltl,..a-cJ.d b th, OOM&'11 lcc:t,-ulttr..( ~ t1l.4 Jf~l'.Mo eut.,... 'ft 1Hb7,: AX Al\ AZ r.n P'J, (;A. ftj ~11 'KY L \ MC l,tj MN \ro! V.\ MT NC NO ME NM YV tm flk OR PA &C!D 1'.H rx l!' VA. '"""" WY ' -<; U W,.,. ~ <o.<r't - J,r;QQL 4 Li Deel

99 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 8 of of Page 292 D #:652 ThbJti A7: The mpact of RTC Laws on the Murder Rate, LM covariate!!, 1977~214 _ - -ci) (2) (A) {~) (~:C::(a} (T) [~) (1) Ji,_ Noriii.i.li.<d ~' Ul;6-2.TJ -.1.:j,lj2-6.2<ll ~i3 (1.nJ) (WD) {d.5!11) (U-411 (&.au) (5.155) (,MM) - p,: 3J J..1 J.1 a.3 D 111 3l ~ 19 (Mill!ii ( 9) l.uj:) (4.wt) :11 Bt411dt.tdMOtttr.""'~ =-'==== == C.du"" TMm"...,.. lr.-1~ p('lt._.~ }"U wi.t. wo.ul-.,.uw1 ~ - uud:.r rl, Li.La.. -.apl. ~ TaJ11t~:.,11t w«ro;., ~ t.m ~ dtdatci.<o<a 1~ li,wj,(,t ~ tk.troa1tocd ta.j 11Ydl1ed: o::at.nl...-., RJ- pmt..tru,.\~ ~.l.fn. M ~ «\M i:(~m Rar.al:, rwpcrta: t.m Ui, tm.rius. lm-. r.mur., (r;wn lht. ~ 81:alc» bl l(tcro>. AX,\R.u. co fl't. OA m CB KY L4 ~R J(.,K \to \A!) f.a r.'(o '.'io'll h,c SM 'Of\' 11 OK OJ\ PA :!O en 'T"N ff UT W1 W" 'WV.v<Ol(\ - J <.tli..., ~.1 ~ (UH) :J (~) (J) (1) 1.~) 2.M ~ ().m) (:z.m) (.J.1 1,1 ~ n,ttt&n,,..-,ll'l"fml..(ilil~~ C".oli;;mc. uml:c1 b.d'.ari pat:.p--.r )"lio 112:WS rnuu.u1;;q1 N - ~ 9' J\fo'iil fd ~.,-~ ~'t4+1f tt... ~ 1"'1..,... ll. p c..u~ ~!:a. U. ~ czttz. nn la matuimt azj 11')-s.l.Ut!a cral,nl pa;la 1 ~ Jlt""1~llil:.'M l.c:l.«-.,,j tu.d..i. lfu:.,, uf lb. J,nm.UJai Jiadl ~ '1c- lbc. Q:C&&M l.c:rm,,..~e innl data ~,Otr.t.M t, r-11 AK "'' A'l PT~ OJ\ lj1 KS KY.A ),fe Yl MX XO W MT NO lid NP. N.\t'. NV Off OC Oll 1~, < l.ll,,, < O.M,..., p < b.m... KO atl TN TX JT VA "ltt"v v,y Table A9: Tho mpnct of RTC Lnws ou the Murder Rate, l\-1.m covariates, 11'T~2l4 <1> m (31 c~ i4j cei -m-m-m tju> ~ N...n-.t n :.l' frn -l-mg -1.1~ -1.n~~ -3.T.l>.~J'l6----:-rli'A~ ~:1-..mn--~1-=nu=---;J;c,_9"'ro, ~ ~(l" ' :n'-'-,"- l - ~-ml {-1.t~L (~-~)_ (~l (U5<t) C-<.l:lll (,Ult!) (4,U!C) 14.~n) N 3-1,13 :J;J 33!S :13 :13 ~l 31 ~l!ludatd. ao,nri ta;&-~ Cahm&~ h:d.la.capr... ~~";lnt..~-va,,,.,. tuunlwr.t~fn ~ll ~ -..Mlui! lbr: dj..tt.ru,..:.c \.cl:wn:n. km~ dltl.:n:u:m hi L}- TTTlrba' J'W' n 1rftlmat t.rrl -.yn11ur;k: cow\'j l'.a'k.-1.,.. W:6 ~.. LAAb,.m1 lm<t...i Gd.., i.mt ol l.k 1,,,-cila:Wll Rfwll& ~ 6. lbc C'U,.Ulad. ttta. rmul".l.q f.:xaa. ~ nw-1cu.. "'""''11111,;4 A.'. All,\Z co Pt. C.l m KB XY U. 1'1B ~ti :\N:l.lO M.!l l,('j' SC~ D MU~ nn 1( ()fl PA ::,:t> r.~ ~ i' ' VA WY WY.. p < O.l<\ "J <O-O,C;~... ~ -:.1 '!'able AlO: The lmpe.ct ot RTC Law6 on th e Property Crime Rato, lvm covariates, 1977-ZOH (;dm:1n~ bj.!r.r. pmi-~'fmor.:t.'t'ow.dlra.t\.ni H l"ujnl'f\11u.lmb &u u r\\ ~ "'ill?,iu.s U. djlnna b.t-. L~ ~ J1!m.x. ~ U. pnrpwtj at.ma n.ta la. ~':a.t.kj. ~ a::aral a.a. t it--~~t m,.._, ua.t ~ fl..,.. o/ ~ cr-e,lqmt.t -.i1,.,.._,_......,...,,...,...;u,,_tlol,,_ 81.oka b. ""''' Alt.Jl.,UCO "~ CA m!k! C\' L \ ME 1'1[ 1-/ l.to 14!! llt!io NCJ lf6: til4 ~ OOK Oft l'a ti<jil) 1".' 'l?; UT VA VV WY,<Q,J 1 Jl <.. (G,... P< DO\ 41 Li Deel

100 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 9 of 1 51 of Page 292D #:653 Appendix B: Data and Methodological Appendix. Data ssues The btate-levcl data set used in t.b[s pnpcr updated through 214 earlier Jnt.o. sct::i UBed in A11r.ja. et al. (2H) a.nd Aneja et al. (211). We further updote thii; data set to incorporate chn.nges to th.e varioub primary 5urccs that. have occurred sinr.<: fin;t released, nnd to include t.hc additional prooict.or vi.t.ria.blcs thnl. ~re featured in the DAW a.nd BC models. All vnrinblcs are collected for the )~B unless otherwise noted.4.li Annual stnt<>-lcvcl crime rates arc tnkcn from c.he r'bl's Uniform Crime Reporting progro.m. 49 RtatL"'l!lvel income vnrifiblp.s (peraonal incou1p. 1 income m,'.linl uce paymp,ntj1, rhl:irement paynmnl.i, and uncmploymont iru,urancc po.ymonts) are taken from the DEA's ficg3onal Economic AccountH. The pcrsonu.l lncome, income mo.lntena.nce, and unemployment llll!ul'ance pe.ymen.t vnrin.blea a.re estimated iu real per capitl.\ terms (defined using the CP). The the LM and vrm: specificalious use u.lterno.tivo voraionb of the rotiroment vn.riablc WU\\. are described ill foolnotc 48. St.ate-level population ib generated usiug lhe Census Bureau's intercensn.j. population. estimn.tos, while the proportiotml tiize oflm's a6 i~ge-race-5e."< domograpbic groups tmi esti.tr.:.a.ted using sl.e.te-level population by a.ge, sex, and m<:8 gftthcred by 1.hc Cellilus. (n CllllCS where the most recent form of these data. were not easily ~t-'ebsible at the Ata.te level, state-liwel figures wore generated by nggrcgntlng the Census Burcuu's county-jovel population estimates by 11gc 1 sc.x, n.ndrace.} PopulnUoo deru;ity is estimated by dividing a givm1 oll!:lervation's population by tho nraa of that state l'llfa)rted in tho prnvious decenninl cctlbtta. Sta.te-kYel unemployment ro.tc d11to.. Js ta.ken from tho Bureau of Labor Statist.ics, whilo the poverty rate is Loken from two Censm1 serla. (~he 1979 R\.n.~e-level poverty rat.e lll derived from.he Decenninl Ceosu~ i,nd the puverty rat6.' rua gene:rnt.ed usil'tg t.he Current Popt1lation Survey). A wea= of incarccrnuon {ncarcerated individuals per 1, st.a.tc rcsidenl1s) is calculated from t.able.s published by t.hc Dureau of.tm1i.ice Sta.ti.sties r.ount.lug the num.bor of prhtoners undor thp. jurisdiction of different state peual syslenu;. Our primary etitimates for crim(t-apecific stn..io-level arrest rates a.ta generated by adding together e~t.l.roa.te1:1 of a:rrr.rts by age, sex, a.nd rnco 1mbmit1ed by different police agcnciei-.. We then divided thih variable by the estimated number of incidents ocr.11.rrin,g in the so.me el.a.le (according c.o the UCR) in the relevll.llt crime cnte(loi:y.~ We also U/Je the index of ere.de cocaine u.saga t;oostructed by l 'ryer et al. (213) for our analysis, which is only nvailn.ble b-ot;ween the yc!l.!fl 198 and 2, aml therefore we drop t hla variable from the MM model when v,,e ~t.imate thla model on data through 214. Sinco we already include controls thf\t incmporate information on the racial com1>t1ition of indiviju1tl t1tatea in our mml)'?lhl, we u.sc t.he 1m1:1djusted '8 Mo.ny of the dar.a ~OUl'.--W tha~ we used in ow cn.rli<.!r E.lBJylii.a are rt: 'llk.1 contin'.lously, ~ud V."8 use a newer vi.:n1ioa of the&e data llefias io thlli paper than we did in our sarl!er ADZ Grul.lygis. We eo,;,,c~tmo! m.e.de data cli!!,cj!al during the clllta clc1ujing procc!la. For il~t;a..nce, a detallt.'<l rnvicw of the raw di,l~ w1dorlying ur~l l!~ru.illuc!! UitCOVen'<l 1\ ~11111 nttmber of 11gcnciee which re1)ried their police stalilttg lc~l~ \-<\ :cc, a.nd we 11U.cmptro to delete tha,c duplicatoa whenr.= poshiblc. Mme=.-, we aomellmm u~ variables that are d6fuloc! slighuy dl!erenely from the oom!l!lpoodlng variable UBOO tn. Lott and Musl.ar-d (199'7) or Moody i>nd M=U (~8). "FUT cxumplo, nftor W<1Untning tho mcton.!ll1.1n of Lott'a county,jat11 ~ot to tho yo!ir 2, wu found thtll uur aftimntes more cl~ upproximated l..o~ ~ pur capita retlretnl!n.t puyment vnriahll! wl11.;n we (11) used U1c total population M lb.o denomioatoj t11tl11:r tbo.n population over 65 and (b) us.lld M our numerato:r a nlt:murermmt that lricludai retlremenl 1)1\ymcn\a aloog with HO!l\O olhor fora,.11 of R,O,w:uncn.t Wlllistance. A~ raiull, wo 1111e a m(l(]_lficd rotlroment,11rl.l1.blo that lnrorp<>mtl!ls thcl!c chttn{{t'h (u tho ~{M &pocifkatkm. Ow: rctirament wrl~bk in the LM. lipl!cl(\~ion, io cootrast, \~'!Q the populatloo = a de.oomfl'j.m,or and usoo a tight,~ definition of retlreme:nt po.ymcnte. -'llfot 1)1lr 11w.i1L &11a.lyRl', ~ fortnu!al.o our crlmc ra.t..)l by dividiog FB rnportcd crlmo oounl..i! hy Fill ropor~od Al,:..t,o.l.ovcl iwi,ul11tion~. ii; a robua~m;~~ chcclc wo uocd tlto rounded 11tE1to-lovol crlmu n:.k:, reported by ~i,u F8T while u!llug thu DAW regre.<mmi nnd aggregate vlo!lmt crimo an oatoo1uo:,u.cioble. We Jind tbal thb alterna&lve erlmc rah> defi.oitlon d oca no~ q11:,.llt~1.1,.,.jy a.ffoct. our li.otltn~. ~1;','11 cbol!e thia vnrm.b}q : tho primnry one thm W1l would use in r.m~ tum]~ after oonflrrulr4! that thle vnr!a\,]u "'= mo1e c!o&ely CQrrolated with Loti'a ~u,.t;o.lowl 8T6!1t varll\bke in the most recent dl'l.ta eot published on hi& 'll'ubsite (a d11.t11 6'1. whldl rum, through tllo year 25) than so~ ara.l &ltcrm1.ly1l!! that wo conslru,:t;:<l, Four 12 Li Deel

101 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 1 of 51 Page D Case: , 1/2/218, D: , #:654 DktEntry: 17-9, Page 11 of 292 vo:n;lon of tho era.ck index itl!ltead of t11e version t.h11t is adj 11sted to n.c~ount for cl ilferenccs in i;tato racial domogre.pbics. No ts for the crnck cocn.l.ne index that we U.'Q WM avii.iloble for the Dk.trict of ColumbiJJ., and our matching u1cthodology docs not allow the Di.slrlct of CotumbJa. to be included in our anclysis in spccifica.tions tlia.l includo thi1:1 varia.blo ~ a predictur. After considering tltweral different ways Lo confronl this issue, we ultimately decided to e.."'<cludo the District of Oolmnbia from the synthetic controls nna.lr-us owing to its status 88 n cloa.r out.lier whru;e character3stics aro less likely to be me{\d.ingfuuy prc<liclive for other geogrnphic arcos. Abadie et. al. (2l!l) 1iwph.asi2 Lliat researcher~ m.ay wunt to '"lr~~trictj the cumpa.riflon group to LUrltll that. are si.mila!' to the c.'xp8od units [iu terms of tho predictors which o.ro included in the model].,, GiYcn that the Dh!Lrict of Columbia. hn<l tha highest pp-r capita phraonnl 311,;r1me, murder rate, Uill-l!llJ)loymeul. rate, pow.rty r11lc, n.ud population density nt,-atioub pointll in our a~mplc,!o Abadic's ndmonition would seem to support omitting thp. DiBtrict ru; one of om potential control unlttj. 111 Wo Rhould note that even if we include DO in tho tiynthctic controls estimates, it et.ill shows RTC lnws incrcosc violent crime by 13.2% in the te nth yenr (ab opposed to th,~ 14.7% figurn shown i11 Tobie 9). We comiider two llepn.ratc pulice measures for the purposes of our anal)'bi11. Our reported results n.rc bli8e<l on the sn.mc police var.la.ble t.hnl we used in Ancja cl n.l. {214). To con&rucl this variable, we t.nkc the most ractint agency-level data provided by the FB Rnd usd ihit1 infordjilioo to e.iitunate the number of full-tirn.i pollco employce5 present in each state per 1, residenl,.s. We fill in ntl.."gi.ng observations with slrl.ffil.l!~ data from pravious yr.atij in cases where tho FB cho.so tn append thls information to t.heir agoncy entrioa, and wo divide Lho resulting calima.lc o.c the totj:1l number of police employees by Lhe populalion rcptehenlcd by tha.q~ agencies. Thill varial1le, which wa..~ origiruilly conatrud.ed for our regre;sion an.alysir, has the advantago of nol having n.ny mibsing outriea o.nd ls clooely cm.relatad ( r =.96) with an n.lte-jrnativc lll~l:18urb of polico i;taffing gentffa.ted by ex:t.rapola,ling missing police agency dn.ta bmed on the average HLaffing levels reported by agendes in tho sruue year lllltl type of area. aorved {represented by a viuinblo im:orporo.twg nineteen co.legoriee separating diterent types or suburban, rural, and urban developments.) A s an allerno.llve, we usl:l dnia publiahod by the Burr.1.m of Just.ice Statistie!! on tbh number of full~time i,quivalenl amployl!e!h worlting for pohcc ar,cncics (llgurcs Lhat were nlso included in the doll\ set featured in Loti and Musto.rd, 19U7). (\Ve rlo not rely on thia 'Hr{able i:n <iur ma.in nnalysis owiug to the liirge numhr.r of mibking yenrr p re'.sent iu this data wt nwl owing to,llilcrepnnciclj in the ra.w dn.ta provided by the BJS, which eometintcs needed to be corrected usillg publlilhoo tables.) 1No find that our estimated average treatment effects for aggregate violent crime a.ud the conclu!lions tha.t we draw from the11 a.vcrngc~ are qun.litativcly ime.ffectcd by eub.,,-t ii1tting onn police empluyment rnewmre for anotheri w h ich sugg~t.1; that nw.a.sutemc11l error associated with our estlruat.es of police activity is not driving our results. L The Dates of Adoption of RTC Laws Wa Ulie tho ffl\me effective RTC <lat~ used Jn Aneja et ul. (21<1) wit.h one b"lnali modification. Owing t- the fuct that v,c nre using annual panel data, the mechanics of the synthetic control methodology require UB to 51 AnotlLor ~Jvo.o;ngo of o:rc!uding Llo.c District of Columbll.l from our somplo ie tl,.-t lho Bu,C'{J.u of J\lll~lco StDtil!~i<:8 l!ftope e~wmnti.dg the incrucerawtl populatio11 of the D~trict of Colu.wbin after ~lie year 21 owi.ng to the transfor of the,.hutrict's lncot<:orated population to Cha Coder al prlllqn system u.nd tbs DC J BlL lh'h! lo Ml ha.ve tried to recol)a\.ruct ln=oratlojj, d11.uj. for DC for theae ycll.nl using nlbi:r data ~OUti;c-11,.he erur;m1.i;cs res1illiug from th!jo, awj.iyllu =re not, w our villw, pl,ubible &UbKLitute! for the fi./8 et1timatcs wo 11.l!C for a.11 o&hcr i;t&too. '!'he rav, d/:\\a Ct that ~ uso to gnthcr infqt1n!lotion about u.tato.lew;l arreet rahm ia nlso t»ll!lllng a.!atta nwnher of obeer, Dllortt1 from tb11 Dilltrid. of Colu:mbla'8 malo pollro deparlm.,nt, whkh further ijtrcngthnns ~lto coso for o,rcluding DO from our tle.tc. sct. 43 Li Deel

102 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:655 specify "t1pecifie y~r for each atal~'s RTC ruile.. To take ll<lvantagc of lhe informalion we hn.vo t ouected on the exact dates when t'l'c laws went into effect in each state, each state's effective year of passage is defined tuj the first year in which a RTC Jaw '\l/'88 in effect for the mp.jorit1 of that yea.r, i; 2 This causes some of the vnl.uea of our R'l'C variable to shl~ by one year (for i.nrlance, W isconsin's RTC date shifts froll\ 211 to 212, ainco the 11!.a.te's Rl'C lsw took effect on November 1, 211).~ While eh.ere have boon numerous disagrecmcuts about the exact lnws th.at should be used to determine when states lllfide the transit.run from n ~niay is.sue" Lo a "shall is8ue'' state, we believe thal lhe dl).tcs Ul!ed ill this paper accurately refl1id, the year wlum differeot trt.at~ adopw.tl their Rl'C lt~w. \,Vo eupplementetl olj' analysis of the etn.tutory history of RTC lawis in different ste.wj with an extensive search of newspaper archives to ensurn that our ch~+tn dates rop~nted com:rete chn.ngh.':l i:n concenloo carry policy. \Vo ~l;ensivcly document!.ho c.ha.ngcs th.at were llildc to our earlier selection of righl.-t.o-carry d!!.lcs and the rationales,mderlying thorp. changes in Appendix C of Aneja oi. a.l. (214). t i11 import.a.n\. to note thu.t. the coding of these dates may not re.fleet ndrninistra.tivc or logistical dclnyii thot may have prcveul,cd the full lm:plemeutauon of a R.TC b~w after aut.horitiea we.re legally dcui~d a.ny dircnition in roj!lding tho lrr;uing of RTC permiti;. deally, n. rai.t!a.rcher would be able to control for the ac.tual l<rvd of R.TC µerruits in existence each year for cnch state. Alt.hough thir dma would bo prcfcrablo t,o a mere i)1dlcator vnrio.blc for tho presence or an RTC lnw, i;uch comprh) 1ell!li ve inforrnl\l;ion unfort. u.nately is not avni!ablh. l>2 A enble llhowll111 uoch sta.te;s orikinru adoptl.uo dnte a.nd 8.(ljuBtod adoptlc;,11 d».te la shown io Table A 1 t>f Appendbt A. 6Jgy di:coul~, we a!aq t4wl U1le BdjuR\.mcnt into ar.t:ount when d11clding which 8L~ta1 o.dopt RTC WW wllld.11 ten year~ of Lho lroo.tmeoni e\m,o'e adoption of.he given!(!lw. AB a. rohw1,nc8s cbock, we ro-um our aggregate vlolimt crime coclcs 11ad(!J' tho naw &poc:ilicatlon withou:: cojl~kforing the modlfiod n:rc dati,, in our!1leci.lon of conirol 11.uit!, finding that this chan11 did not ~,:ct hur qualltntlve findings mean1nrfully. 44 Li Deel

103 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page 1213 of 51 of 292 Page D #:656 Appendix C: Replicating Our Analysis Ono i.ssuo which s mroly ndd:rossed dir1.1ctly ill tha rod.i,tl!jg lli.en~tun~ H\TTOlllidixig the Et.pplic:atfon of the synthetic control technique is the sensit ivity of t he sdcct.ion of the syntbotic c.ootrol to S(lClningly inconsequential details when llfil.lg maximum likelihood to select the weights Msociated with different predictors in our analysis. More SJ}l.'Cifically, whon using the excellent "synth n pnckago for Stata created by Abadie, Ha.in" muc:jlrn:, nm! Dia.monrl nlnng with tho 11ooti:ic1 opt.ion (whlch. implalllauf;,:i the npl,irnizatimi i.p.tliniqn.e dt>.m:ribed in footnote 2), both the vernion of St atn. (o.g. 1 SE vs. 'MP), the spocifico.tiotl8 of the oom1,1uter rwmir.i,g thti command, and the order in which predictors arc listed can affect the composition of the synthetic cont,rol uml liy HXl.enHion the tthm of t\1e eht.itm~te<l treatment effect. The root. caubc of the ditcrc11ccs between S-at.a veo;lou.s is a.x.p)l;llned by a 28 Stati~Corp memo, which noted that: 'Vlbc.n mmc. 1.hflll m1c ptoeeaaor is 1u;ed in StM.a/MP, the computl!.tions for the lilrnlihood arc split into pieces (one piece for t>.. ac..h processor) and thej1 aro added a.t t.he en<l of the csk'tlla.lion on each. iteration. Beca.uae of round-off error, nddition is not mmaeia.tiw in eo1putet Actenc:e i!:! it ls ill mathematics. Thia may cause a 6lip.;ht difforcncc in results. Fbr example, nl+a2+a.1+ n4 can ptodnr.a different. reimltr lnnn (a l+il.2)+(a3+a4) in numerical comput11.tion. '\Vhen cljinging the number of processors w.cd in Sti1ta, tho order in whlch the resulta from each processor are combined in calculations may not be the S<1D1C depending on whlch proc~r completes its calcula.tioru; fl.nit. 5 J Moreover, t.1.tls document goes ou to uo~c that the differences associated wilh ubing different vcrsions of Stata = be minimized by ~etti~ a higher threshold for nrtolcnm~(). Thi5 optimi:i,a,tion condition is actually rela."ced by the Hyrrth routine in situations where setting th.is tlrre~holtl at. itl' clcfo11l1. level causes the optlrr1!za.tio11 routine Lu cnwh, and we would therefore expect the results of Sta.ta SE and lv[p to div()rge significantly whenever this occurs. n OU' nnnlysis, we u.'jc tho UNDC vorsion of Stata/\olP owing to the well-documented performance gains 11.o;socint.ed with this ve111ion of tho soft;ware pe1ckage. Another discrepancy that we en.countered is that memory limitations eomet.lmes r.&tlfled 1tr riynthetic control ana.ly8 B to era.sh when using the nested option. '\>Vhcn this occurred, m would generate our syn~ th~tlc conttul u~1ng t.he regre&ilm1-ba.h&i technique for detennining the rela.tive weights assigned t- differ~.nt prnc.lictota. Wo ouoouulcrcd this sltmtiou sovcral t.irnoo wheu ruuui.ng our Stat.a code rm standard desktop computers) and these errors occurred le.o,s often when using more powerful computers with g;rea.t.er amounts of memory. For th.is reason, to rnplicate our results with the f,'1'c.dtci;t runount of prcclaiou,,..,.e would rcc.om,ncud that other resesrchern run our c:ode on the aame machine11 that we ran our own analysis: a 24-core UNX machine with 96GB of RAM running Stat.a/MP. One final discrepancy thot we arc still n t:he process of invebtl.gating ia the effect of changing the variable order n the 1,ynthet.it. l:t)nl.rol ccum~ud iin the compooitio11 of the Hynthetic cunt:rol whfln ru;ing the nested option. Unfortunately, the large number of predictors included in the LM and Z...11\o apecifications ma.kc t dl.ffienlt to lllw. A. fixed r..rlterla (e.g., ml.nlrnlz!ng the!t.vi>.ta.gp.,:;-.r,fficierit or varia.tion of the JU\o1SPE) for dot.crmining the order in wwch variables should be listed. WW!c we have not modified the order in which predictors were listed in our modem iutcr oboorving the results that we de.rived from that variable order, il is uscrlll Lo b e aware that ditcrcnl:. va.r inblc or<l<:rn tinri n.lt<:r C"St.imat~ slightly. Huwovcr, the observation that our syulhetic controls cst.imatcs for violo:nt crilllo (eaults n.re ~utia.lly w1chauged a.ler t, ylttg multiple ~ ' 1 '!hib mmno can bo found ut the followin,g link; / 45 Li Deel. 8 B

104 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page 1314 of 51 of 292 Page D #:657 Bp Cificntiom1 fcntnring different sets of predictors gives U.'l grcntcr confidence that onr canclllilion.s about these spt."cifications sre robuat to changes in vnriable order u.s well. 4(1 Li Deel

105 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page 1415 of 51 of 292 Page D #:658 Appendix D: Synthetic Control Graphs Estimating mpact of RTC Laws On Violent Crime Using the DAW Model~ 5 J9 C o - ll co "O 'i.13 ~ :: ~ f',, """ L. ~ Cl)... co (\1 cc: ) E 't: LO (.) -C Q.l 8 Alaska: Violent Crime Rate Effect of 1995 RTC Law 1 Years After Adoption: 14.8% ' - 9\ \ ~--- 5 "ii' treated unit synthetic control unit J 65 Rec11ll t l.jtlt ee.cll state ~ effec.tt~-e yeo:r o( po,!j~i,le ia tle!inoo u,i t.bo firlrt. yout n which a TC h1w " 'Bu ln 1;1ffet.t fur the majority ai that year. 47 Li Deel

106 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:659 Arkansas: Violent Crime Rate Effect of 1996 RTC Law 1 O Years After.Adoption: 23.7% c.o C\ -- treated untt synthetic conlrol unit j Ccmpo6itioo of SC: OE (1 S.6%), L (23.1%), A (58.~%) Slales Never Passing RTC Laws ncluded ln SynU,e\ic Control: DE RTC-Adopting Slates ncluded n Synthetic Comrot L (2141 A (211} 48 Li Deel

107 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page 1617 of 51 of 292 Page D #:66.!! : CD }l VJ OJ r,.. Ct'. :: Q a l!l w '"". OJ. (! «LO :: r) ~ 8 Arizona: Violent Crime Rate Effect of 1995 RTC Law 1 Years After Adoption: 8.8%. ; c:: LO, '\... \ C l!l ~- ~ a ~ ~ > -- treated unit synthetic control unit Corn~ositioo orsc: CA (32.7%), H (4(),9%), MD (22.9%), NE (3.. 5%} Slnl.Gs Never Passing RfC Lam lnduded rn S,,nlheUc Control: CA, Hl, MD RTC-AdopDng Stales locluded tn Synthe~c Ccnircl: NE (27) LO N 49 Li Deel

108 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page 1718 of 51 of 292 Page D #:661 ~ ill 'O 'iii <O!}_ ~ Colorado: Violent Crime Rate Effect of 23 RTC Law 1 O Years After Adoption: -1.2% Co Oo,... ll L. Cl) a. CD.. ('J a::,r;t Q) E.:: ~ - ---~--~ ~M~ ~----~ > ~ treated unit ---~- synthetic control unit Ccm~ition of SC: Hl {49.9%), NY {29.3%), R (2.8%) Slaioo NQ,\18r Passing RTC Laws lnduded n synlheuc Conlrol: H, NY, R RTC-Adopling States r.duded n Synthetic Corrtrol: (") ~ (),- N,'SO Li Deel

109 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page 1819 of 51 of 292 Page D #:662.! C Q) ~o ~ Q) N :: ~ ~ O... L. Ql ~ a. ~ (ll a:: Ql co E ;;: (.)... C: ~ (JJ r > t,.. t-.. O>... Florida: Violent Crime Rate Effect of 1988 RTC Law 1 Years After Adoption: 34.8% / _ /...,,,... \,,,,,,.. / \ co (X) CD,- \ \ \ \ \ \ \ ~ - trealed untt - ~--- synthetic control unit Com~ sijfon of SC: CA (22.3%}, Ml (11.%), NY (66.7%) Sta~ NeY-Or Possir.g RTC Laws lnduded n SynlheUc Conlrol: CA, NY RTC- Adoptjng States ncluded n Symhelic C-On1fd: Ml (21)... ' ' ' co ) en Li Deel. 8 ~

110 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page 1911 of 51 of 292 Page D #:663 ~ ~ g 'iii co ~ ":i a o r,.. r... Ql a. 8 Ql to... 'll " dl E ~ c (.)... C O. Georgia: Violent Crime Rate Effect of 199 RTC Law 1 O Years After Adoption: 6.6% /.,. / / - ),... ' \ \ \ \... '\ ' ' \ ~~~ ~ , 5,... 8 ~ ~ ~ ~ N -- treated unit synthetic control unit J Coff1Xl&mon of SC: CA (l.1%}, A (46.2%), MO [9,?1~1. NY {15.%} S1at8S r-lmr Passing Ric Laws lr.d~ed in S'Jnlhalic Conltol: CA, NY RTC-Adopling Slilllls ncluded n Synl11e1ic Control: A {211 ), MO (24} 52 Li Deel

111 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page 2111 of 51 of 292 Page D #:664 ti]... C: ll :2 N ~ M G'. ~ (t) ''"".. NCO ll. ll.! (. ~ N dl. ~ daho: Violent Crime Rate Effect of 199 RTC Law 1 Years After Adoption: 5.3% ' \...,1 ',,,-) " \ \ l) C: N ~N ~ w~ ~ 5 ~ ),- Ol O'l T"' 8 N -- treated unit synthetic control unit j Cof'r4)Clsilion of SC: H (96.1%), A {3.9%) Slates Naver Passing R'f C Lawo lr,dudoo n SynlheUc Coolrol; HJ RTC-AdopUng States [ncluded n Synthetic Cantrol: A (211) 53 Li Deel

112 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:665 ~ Cl) Bo ~ g cc ~ [) S:,:i-,. Q) a. Q)... st ~ (J) E ltl - (f) "-... C: 11 i"') >,.._,.._ ) T"' Kansas: Violent Crime Rate Effect of 27 RTC Law 7 Years After Adoption: -6.3% - - treated unit synthetic c-ntrol unit.,.._ st T"' N C'l ColrpositiCfl of SC: DE (18.3%), H (58.3%), MA (23.5%} Slates Nev&r Passing Rf C Laws n duded in Synlhelic Conlrol: DE. H, MA RTCQAdapling Slates ncluded n Synthetic Ca~~: Li Deel

113 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page of 51 of 292 Page D #:666!.? C,) "CJ O " 'cii ({) QJ er.: ~o O lo... L. ~ ~ & (ll g. E (') c (.)... C: Kentucky: Violent Crime Rate Effect of 1997 RTC Law 1 Years After Adoption: 3.9% ~N~ ~------,- > ~ ),- -- treated unh synthettccontrol unit Co1Tl{)Sition of SC: L (19.8%), Wl (8.2%) States Never Passing RTC Lawa lnduded n Synli1etic Conltol: RTC-AdopU~ Stites [ncluded n SymheticC-Ontrct L {2i4), W (212) j.. N 55 Li Deel

114 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 23 of 51 Page D Case: , 1/2/218, D: , #:667 DktEntry: 17-9, Page 114 of 292 Ul... C G) 3;l N ~ r :: ~o ~ ~... ii a. cu a:)... (l,l :: Cl) E g ;:..., t:..!~ 5.,... OJ " Louisiana: Violent Crime Rate Effect of 1996 RTC Law 1 Years After Adoption: 15.4% ( ) ) r -- treated unit synthetic control unit l ( N Coo,positm of SC: CA (3.8%), DE (2.7%), L (75.6%) stales Nova- Pa&si11g RTC Laws tnduded in Synlhetic Coolrol: CA, DE RTC-AdopUng States ncluded n Synthetic Coo1J'of: L {214) 56 Li Deel

115 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 24 of 51 Page D Case: , 1/2/218, D: , #:668 DktEntry: 17-9, Page 115 of 292 ~ C ~g 'ii; () Q) C O ~ a M... \.. u, 4) (' CL $ ~~ d,l E u, c r u..., C O - Maine: Violent CMme Rate Effect of 1986 RTC Law 1 Years After Adoption: -16.5%,, \,..., \ r-- \ / ' \ / ' ~~ ~ ~ ~ 5 -- treated unit synthetic control unit \ Ccmpositior, of SC: H (16.%), A (84.%) Slates Never Passing R'rC Laws ndud ed fl SyntheUc Control: H RTC-Adopti~ States!ncludod n Synlhellc C<mlrol: A (211) 57 Li Deel

116 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 25 of 51 Page D Case: , 1/2/218, D: , #:669 DktEntry: 17-9, Page 116 of 292 j9 C a) tj a.iii ~ co : ~o Oo,...,-.. Michjgan: Violent Crime Rate. Effect of 21 RTC Law 1 Years After Adoption: 8.8% Q) Cl. - :.l!! ( \ : (ll ' -, a: ~... CDE O : \ U),.._..,-, ~ ' (). ' Q ' c o - ',, Q)~ ~ ~ ~ ~ >,- C\,,- N - - treated unit synthetic control unlt \ Compositiai of SC: MO {36.2%), NY (33.4%), W {3.3'%) Slates Never Passirg RTC Laws lnduded rn SynUietic Control: MD, NY RTC-Adop6ng States ncluded n Synthetic Centro!: Wt (212) 58 Li De el

117 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page of 51 of 292 Page D #:67.l!l C ell 32 ~ LO ~ ct) ~,..... (") ell n. o t ~ G E 't:: Oo... C N () 5 Minnesota: Violent Crime Rate Effect of 23 RTC Law 1 Years After Adoption: -.7%... \ / \ \ \ /\ / \ \ \ (\') N... ~... " ' \ \ \ /./ -- treated unit syntheuc control unit C~&i1irn of SC: DE (8.4%), H (91.6%) State., Never Passing RTC Laws ncluded if\ SynihoUc Conlrct DE, H RTC-AdopUng Slates nclude<! n Synlhe11c Co.~ircl: 69 Li Deel

118 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:671 Mississippi: Violent Crime Rate!l C Effect of 199 RTC Law 1 Years After Adoption: 34.2% G) 1J - ui () ~ : Q ~ ~,--... Cl) '<j' a.! J C'.: l() (') Q) E c l1> u... C: l!l ~ N,... 5,... (j) T'" '! :', / /\' : - --, \ ' ',..., (\J -- treated unil synthetic control unit j Ccrni;osition of SC: H {72.. 1%), A {1.6%), r-le (1.1%), OH {25.2%) Slaloo Never Passing RTC util$ lndllded in Synlhetic Conlrol: H RTC-Adc~Ung Stales [ncluded n Symlletic CorMll: A. {211), NE (27), OH (24) 6 Li Deel

119 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 28 of 51 Page D Case: , 1/2/218, D: , #:672 DktEntry: 17-9, Page 119 of 292 J} C ~g 'iii ( Ql ~ O ~ - r,..,-... (!J Q. ( 1! «l a::. ~ O E c c ~ Ql 5 Missouri: Violent Crime Rate Effect of 24 RTC Law 1 Years After Adoption: 14.1% , ~ i N -- treated unit synthetic control unit \ Cofl1P()sition of SC: CA (4.1 %}, DE (26.5%), H (33.3'/,) Slates Navar Passing Rf C Lawa lnduded in Synthetic Cootrol: CA. DE, H RTC-Adop~ng Stales lnduded rn Synthetic Centre{: 61 Li Deel

120 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:673 E Ql 1J O - "{ij (") () :: ~ a (') t"'.. C) () N a.,2:l O:l O. :: N Q) E u, ~ c:,-... Montana: Violent Crime Rate Effect of 1992 RTC Law 1 Years After Adoption: 9.9% ()..., C:. ~t"'~ ~ a > ~,..- N ) ),- N N -- treated unit syntl1etic control unit j Ccmi;osi1ion of SC: H {12.1%), W (87.9%) Slates Nev8r Passing RTC Laws lnduded in Synlhelic Conlrol: H RTC-Adopling States lrdlded n Synthetic Corrlrol: \Y {212) 62 Li Deel

121 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page 3121 of 51 of 292 Page D #:674 Nebraska: Violent Crime Rate!'3 C Effect of 27 RTC Law 7 Years After Adoption: 9.7% Ql!! O Ul ' Gl a: ~ 'f r lfl aj M / O.o /\ ro M y \ 1 - CJ: \ / Q1 / '-/ E ~ c ()... - C N ~ 5,.. (1) ~ / \/,,' -.. " '',, \- : ' \ t,.. g N -- treated unit synthetic control unit COT4)smon of SC: DE (17.3%), H (82.7%) States Never Passing Rte Laws lnduded in SynthelicConltol; DE, H RTC-AdopUng States lnduded n Synthetic Con1rol: \ ~- \ q,,.. N fl3 Li Deel

122 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 31 of 51 Page D Case: , 1/2/218, D: , #:675 DktEntry: 17-9, Page 122 of 292!!l C ~ - ~ '!"" ::: ~o O') t" '- - 4) co Q, Cl>... ~ " Q) E o c <D (.) +' Nevada: Violent Crime Rate Effect of 1996 RTC law 1 Years After Adoption: 23.7% C: - ~~~ ~ ~ 5 -- treated unit synthetic control unit Ccq;osition of SC: H (16.7%), MD (63.3%) Slaloo Never Passing R'fC Laws ncluded i'l Synlhe~Conlrot H, MD RTC-Ad~ting States locluded ln Synthetic Comra: lo N 61 Li Deel

123 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:676 J!! C -8 - ~,- :: Y'.,- a'.)... Cl) a.! ti! :: - Q) (!) E 'i:... a C O New Mexico: Violent Crime Rate Effect of 24 RTC Law 1 Years After Adoption: 14.7% ~~ ~ ~--- -~ 5 i N -- treated uni synthetic control unit J Ccmposition of SC: CA (48.5%}. DE {S1.5%) Slrlles N8ver PaGSlng Ric Law& lndu<loo in Synthetic Conlrol: CA, OE RTC-AdopUng Slates lnckjde<i n Synthetic Coolrd: 65 Li Deel

124 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page of 51 of 292 Page D #:677!l C: l1l 'O 'iii '-, t1) Cl'. ~ ~ O,. QJ a. D -o ~g Q} E c Oo... ~ ~ North Carolina: Violent Crime Rate Effect of 1996 RTC Law 1 O Years After Adoption: 18.3 % t f ' "! r- ' " \ \ '\ \ '-r ~ 5 (:: {'1 ~ ~ ~ r ~ N -- treated unit synthetic control unit J Ccrnpositioo or SC: DE (9.2%), L (39.6%~ NE (51.2%) Slates Never Pasi;lng RTC La'ilS lndud ed in Synlhetic Conltol: DE RTC-Ac!Ci)lir.g Sli!lecs r.eluded n Synttletic Control: L (214~ NE (27) 66 Li Deel

125 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:678 ~ G> ~ g Q) 'i " l() o N.- a '- () N a. ()... ) ~,- () E o 'C: r (.)..., C lo Q). 5 ~ ),... North Dakota: Violent Crime Rate Effect of 1986 RTC Law 1 Years After Adoption: 13.4% (D O' O'l,- -- treated unit synthetic control unit j Composilia, of SC: W {1.%) Sfal&s Never Pa&Sing Rte laws lnauded kl Syn1ha1lc Control; RTC-Adopting Stales ncluded n SynlheticControl: Wt (212) 67 Li Deel

126 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:679!! C Q) 1J 'ui ( Q) o:'. ~ O L(') r ) ~ ~ -('J ~ Q). E M 'C ()... C O.! (\j,..._ 5,..._ Cl) \"" Ohio: Violent Crime Rate Effect of 24 RTC Law 1 Years After Adoption:...8% / :\ -- treated unit synthetic con1rol unit Com~ositio11 of SC: CA (19.5%), H (2.7%), R (59.8%) Slates Nev!lrPassJng RTC Laws lndudoo in SynlheticConll'ol: CA, H, R RTC-AdcpUng Slales fnc/uded n Synlh~c Co(llrd: tl8 Li Deel. 8 ~

127 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:68 Oklahoma: Violent Crime Rate j'3 C Effect of 1996 RTC Law 1 Years After Adoption: 9.7% Cl) 'O Q 'iii " (1) a:: 'Y <O ~. QJ.. QJ LO... (!J ~ QJ E ~.::... C! (".) 5 ~ ) ~ :.,... \ \ \ \ \.,,,--. _,,,..._ '---,;... _.,,. ( N -- treattid unn synthetic control unit \ Cooiposition of SC: DE (27.%), L (24.6%}, NE (48.4%) Slates N8"er Passir,g RTC Laws nduded il Synthetic C<lnlrol: DE RTC-Adopling Stales ncluded n SyntheticComrol: L (214~ NE {27) 69 Li Deel

128 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 37 of 51 Page D Case: , 1/2/218, D: , #:681 DktEntry: 17-9, Page 128 of 292 ~ G) U O - 'iii (!) Q) er: ~ U) ) T" L, Q) ().. 1l g ~ ~ Q) E o i:: <q' u C: G) ll Oregon: Violent Crime Rate Effect of 199 RTC Law 1 Years After AdopUon: -.6% ---"' / \ t \ \ \ -~ ~ 5 Ol a, T" - -!teated unit synthetic control unit l C-Ompositioo of SC: CA (f.4%), CO '42.%), H (6.2%), Ml {34.4%), MN (16.1 %) Sta\Gs Never Passi~g RTC lal\ s ncf u<led in Synlhetic C.Ontrcl: CA, H RTC-Adopling States 11\cluded n Synlllelic Control: CO {23), Ml (21), MN (23) ~ N 7 Li Deel

129 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:682.l!l C 4l U) "'C ';j Q) ~ in. ~ 't ~ v l.. Q a.!o ~~ Q).5 \,., O u t'),..1 't: ~ a U) Pennsylvania: Violent Crime Rate Effect of 1989 RTC Law 1 Years After Adoption: 26.5% 5N ~ ~ ~ treated unit synthetic control unit Cwpow:n ~ SC; Ol:{7.7%), H (1Q6'l'i,, tle ( i-3$;, NJ (17.~}. oti ll5.6ll~ " (~J%f Sim l~'a' Pa!shQ Rte l.3w-. Mlded n Synt"dl C1nt D:. H. llj RJC-~ ~ lnduded l1 Synlh~~: NE (27). OH (24), W (,1)12) 71 Li Deel

130 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page 3913 of 51 of 292 Page D #:683!l C QJ 'O - ~ O'.'.... ~ ~ O> r... Q) Cl. - Q... co ro a:: 8 Q) "" E co ' (.) / c <O j! 5 South Carolina: Violent Crime Rate Effect of 1997 RTC Law 1 Years After Adoption: 22.5% -~., \ \ \ \ ' \ ' ' ' ' ',----,... a ('j [-- - trealed unit ---- synthetic control unit Composition of SC: DE '17.9%}, L (82.1 %) Slatoo Never?as&lng RTC Laws lndl..led in S'JnlneUc Control: OE RTC-AdopLing Stales lnclooed ln Synthetic C<lnud: L (2ll14) 72 Li Deel

131 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page 4131 of 51 of 292 Page D #:684! C: Q) 'O.() 'iii (") Cl) a:: ~o (1),... L.() Cl) N a. ()..., ~N () E lc 1: T'" ()... C - South Dakota: Violent Crime Rate Effect of 1985 RTC Law 1 Years After Adoption: -1.6% /,,--/,..- -,,...,;/ / (l) T"" ~ ~ 5 l() co ), treated unit synthetic control unit CcmPQSitioo of SC: A (62.5%), W (37.5%) Slates Never Possi11g R1C Laws ncluded n sytilheuc ConlJol: RTC~Adc Jng Statas lncludoo n Synthetic Cooirol: A.(211}, W (212) / 73 Li Deel

132 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page of 51 of 292 Page D #:685.s C Q) 'O 'iii co D C( ~o oo r !) Cl. - Q) (... co ~ 8- E LO.: Do... t:.-,t / Tennessee: Violent Crime Rate Effect of 1997 RTC Law 1 Years After Adoption: 29.5% t"' --~\ "-- '\ \ \ : \ ',/ \ J,. /,,: ' ' \ \ ' ' '....,- '-;"" J!1 L, ~ > -- treated unh synthetic control unit! Coll'C)Ositioo of SC: DE (29.1%}, L (39.5%), A (31.4%) Slates Never Pa1>Sing RTC Law& lndude<l in Synlhe1lc c«,lrol: DE RTC-Adcptlng Slates ncluded n SynlheticConlrcl: L (214), A (211)... C\ 74 Li Deel

133 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:686 f C: ~ 'C 'ui QJ a: co ~ T"', OJ. OJ O - 1'~ cu (D ' X'. ll O / E o - / 'C O,,, (.) ""' C:.! ~ 5,... Si,- Texas: Violent Cnme Rate l:ffect of 1996 RTC Law 1 Years After Adoption: 16.6% U) Ol ) T"',- ( C\ ~- treated unit synthetic control unit j Ccmposiiion of SC: CA {67.S%}, ~E {8.6%), WJ (33.6%) Slates NeYef Passing RTC law""s lnduded SynlheUc Ccfllrol: CA RTC~Adopling Stales ncluded n Synthetic Colllrol: NE {27), W {212) 75 Li Deel

134 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #: ll C Cl) 'O Jl 'in (") Cl) a'.'. ~ : M Cl). Cl) oi cc Jl - Cl> N E 't:... C Utah: Violent Crime Rate Effect of 1995 RTC Law 1 Years After Adoption: -2.2%..,, \ \... ~N ~-~----~ 5 -- treated unit synthetic control unit Com~osition of SC: H (75.6%), KS (6.'~). RJ {1.8%), W1 (7.5%1! S.al.GS N8v8t Passirg Rl'C LaWD lndude<l in Synlhelic Conlrof: H, R RTC-AdcpUng St.!~ loc!uded n Syntheiic Control: KS (27), W (212) lo N 76 Li Deel

135 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:688.i!l C Gl O "CJ.iii,q' ~ r ~ O,- (")., (l) Cl. (l) - ~g al M E c u... C ll ~N t-,. 5 '- ),- Virginia: Violent Crime Rate Effect of 1995 RTC Law 1 Years After Adoption: -3.6% -- treated unit synlhetic control unit /') N CompositiCfl of SC: H (24.9%), KS (23.5%}, NE (15.7%), R {26.9%}, W {9.%) Slates Never Pas.ng R'rC Laws lndoosd in Synthetlc Conlto]; H. ru RTC-Adopting States ncluded n Synthetic Centro!: KS (27), NE (27), W1 (212) 77 Li Deel

136 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:689., (1) l() Q.. N! ~ Gl ~ E 'L (.) / LO / c""" ~ 5 West Virginia: Violent Crime Rate Effect of 199 RTC Law 1 Years After Adoption: 62.3%.., / /.,, /,,, / ;_...., 1; \ ' / l ~ m,.. \ /',../ \ N -- treated unit synthetic control unit j Composilfoo of SC: hi (.7%}, A (99.3%) Slates Ne 1ilf Passing R'rc Laws tndud~ rn S,,,nlhellc Cooltol: H RTC-AdcpOng States ncluded n Synlhefo Control: A {211) 78 Li Deel

137 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 46 of 51 Page D Case: , 1/2/218, D: , #:69 DktEntry: 17-9, Page 137 of 292 Ul... '.: ~g 'iii~ m r O ~ ~,--.._ O Ql C') a. j!l g & (') ~g 'i: N ()... CO. Wyoming: Violent Crime Rate Effect of 1995 RTC Law 1 Years After Adoption: 15.8% \ \ \ - "' ' "-, ' '\ ~N~ ~ ~ a > ~,- f) (\ -- treated unit synthetic control unit J Composition of SC: H (1.5%), R (55.5%), W (43.1%} Slates Never Passlng R'rC Laws nduded in Synthetic Control: H, RJ RTC AdopU~ States ncluded n Syntheti~ Control: W (212) 79 Li Deel

138 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:691 Appendix E: Data Sources ' '.~.. lilt:--- \", 111',- l'r;--;- :C Z,, lot,,:, -,-.,-.~\-,r-, -., l--1,-,-- "' " l 'i j ll1t lf tlf (\\\ i,o ' \ \ \.,, \11,, \.,1, - ""',t,...,,.,..h ot, 1 11\,., 1,1,),;: l lfh(l,,,,..,., l : ;1,ii,,1111,.,,,,,o1 ut:.1 1 l it tv,1-, t. 1.1,, r, l tr.. 11 \.,~... 11, \du,,,(,,,., 1 1 ~. lq'i,, r, 1, u!, hi.\ l...d.,-:,.,- t~, 1 1, t--. 1.,1.,l,, l, 1 i,,. 1111,r,.,,,. 111,,1.. i, 1.,., 1 1, 1r,, 1 t'. flf, 1,,.,J t H1 p~ 1,11!,.. "'11;.11,rh.,1,r 't,.t ul, 1 ~~ 1:1,,11,r t~,.,...,,:r1,,,,t., -', 11,,, 11., n 1.1n1.., \ t,,.._ l,,.t",f ~ ' 1.',l',1,h 1 1~ ' "" pji P u 11,t 111 ij~,! th..-' '.:, '"...,_,, ~!.,, j!, '1,-, -_,-,-,---+-,-,-,"'\\,--l-,\-,,..,+.,.,-,"1-1-r~,-.-,-,~l,- "'"l-1-,+,.c:.-"-,-,,-.-,-,-,.---,-,,-,:-, -11,-,-,-,-,-.\-,l-.,-,,.:.".;..., ',,..."- ' -.-,-,'-,-.-,,-,-,-,-,-,.-.,-."',-,-..,;.1,, 1.\1J, u,t il _,.q ~ H,t,,t 11, t,,, ''.,1 J' ii,, 11,.,! r 1.. ~H, i... +Ji,,,,hq,, l l \ \ J,Li, : t ~ 11, l... ' ' '"'j4 r ltll',\:, l r11t,~\1i.1h P,L,, 1t.c,J..,,,,, 1,1,~ u,.':\ )1r r,t11 1l l,\ 1\ d1~1,~lhi 1,,1 1!1:11 ;, t1,..;.., ''"!'-1., \ 11, 1,, 1... ~ \ 11,, ~,1 1 d t11,,11, r,11 1 1u 1 ' , 1 1., ;i.. \ ' >.,,., :,, 1 t';~,... '11: ~ 1 witr 1l,,,, n H1 11., \, ti,~ "', 1 \...,. \,: ; 1, ':1 r f iot;,",,, 4 1\j.l l \\ llh 11.i '-' ' p, i l'j711,:,, pi...11 l V,\,~, ( \ 1,:1.L \ 1 \J 11.,d. 1..,,,,, r~! r u '~'"-..,1,,, 1 1 o "l ii r \1 p ul \1 ~,J " '', t... d..,11 11,l,,,. ;.....,,.t l \,j / 1J' '!11, - j; 7--7-;-11-.-,- -1-,~,-,~\~l~l,-,~.+ ~;,-,,-,i., 11- i..,1, d,\,., 11,, \\11,11:,;-1 1,l;;.-.-, ,-,-, -,-,.-. -,-.-.-, -,,-,-":-, 11;~ \ :-.-, lln l \ f\l'., ,la11 \,1, \t'.\1 11,,,,, :... 1:,,,,,, :1,,.1,1..._, q d 111,.n 111'"' 1<1 ~,., 1..: r: 1 i:,~! 't!..:. _...!irj1,, h 1 ~, 111:.. r,., 1! 1,, i.,. 11, " ''' l ~f 11,, 11,1 \ u,,,,,,:, ~ 11j 11 \,1 ll H \ 11 \\\ '.,;j,,.,,j, '-- t,, j, \ n' l' ~" d i f!, ~';T" p.,...,,.:., 11,r,,,. '.l \ ', 1,1 1,, ~ ' l ~ ' 1, r " :. ~.,., " ',, ',,.,,,ii t ~, r ' 11,,.. h l \:,H l~ 11 1/j ' '" ll wt.r...,.,,.:,:, 1 l{,t, illol \, -ri;-;;:-.,. \ 1t.,.!,,, J. t1l11\ hi'- l-l 11.,. \ "t --', ' 1:7:,,,'11,; hid ",.,, 1t 1 9 1,f t l, 1 1,p : 111 1!. l.1\ 111 "' \!-..\. 1,, - -, 1,1i- t.- -,-.;-,'ii 1 jlt77 - q,. 111 ; 1.. '.J l l, ;;77 _.,.,, ~ "; : n '' "' l _'tjt ttl, 1';''; ~P 1 lj'; ~1,1 f' \ \\ \ l\1 \ ll,::w;,;1 1.1 >,\\\,,., ',1'..\! \ \ \t \ T, \\i ti< j.\\', ~.,~.,:...! 1 -nr ~-----~~ t 111.,,1,, "'... u, ~,,., 1, 1,,,,t ~.. 1 "t ,,,; - 11,1,,, q1\ 1 '1 1} 1 11" lit' ' d 1\1 1~, 9 ~1, f, "fl 111 ii \(. tl1,1tll lh 1.1 ~1.f u, ~ J,..,,t,f9,.;:1,~,r l,p 1...,!1! l..., h, : ' 11, t 'r t '",,-1 1 1, 1,i, 1 3;.. :,1, 1J., 1 u 1n, r 1, , \\ l 11 r! '" 1 11,,.,1, t,,., tt J:1 l~t.-. 11~, ht 1 11.,, H j...,,1 i \~ : 1 1t+f..,:, 1.. ;,, t 't 1' ,'-'',,, l :., 1 ~"' j l :,,:, l.-''1,111 L,;1:[.;-,,... h, 1: L t,... r..!, ~i.1,,1,.1 1. :1,+, " ',, d11;',,1,< 11 f. 1 1 ',... l,1,, ri-14 :. ' 11 1,, \', ' H,., J1 P'l ' l,d,, j, 1-11)"1 ' ,..,, 111, p:11,..,j J 11111'., fl, q'..f1 111 U,l,:), "",!111. J J,,1, '- iii 1,,..!, \, t ,-1.,-1,-,.-,-:,-,-.,-l,f-.-,..,td 11 ;,,t, a< ;oo r,,.,t, 1'1.!i.,, 1,11, l,..! t i ;, 1.,,:1,,! 111~... 1.,,._.~,,1 11,, 1- flt,,.,. p -., 111 ' 11!, i r:,:'."""1 1J J t 1, 11111,, 1, : 1".:--' "'l : 11 \,: ~, rn1 1,,.,... i.. " l ir 1 Lit,, 111.H' 1 ),1...,:,h ~~ J l ": :- 1.l " 'i i '1 i. 1! d -:, 1. "' 1 1, ' i J'' 111,,1. dt, d,! J,\ t 1t1 '- ~,:-,-,,-,.-,l-~!-,,-,-,-,--,-,.-,.~l,-~~, :--1:1,'- ''.: lh, \\J 111,u :, \A., 11 t (;,. ~-,,-,,-!- -L - -,-1-,.-,-, -,,-,,-,-,-,t-l tl,, ', " ' r. :, h ',H,!, ,1,1 11 l 'h \ fl tf l ",.,. t,,,. l ',,; t 1, 11 1 l n,.. \. 1\ r1" i,-1,11 'l';'j' 1,7 1.Ji 1~1!1 J l 1 ;,11 1,j i U,rr, -.,.ii.i \ 1!'... : 1, 11 11r "" r J,,,, ~ \ '!.,..,.. f..!.f,-,~,1~1-,,-,.-,,-.-.,-,-, Li Deel~

139 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 48 of 51 Page D Case: , 1/2/218, D: , #:692 DktEntry: 17-9, Page 139 of 292 A ppendix F: Methodology to choose the number of lags of the dependent variable to include as inputs in synthetic controls We use n cross validated npproacli to detennine the optimo.1 lag choice(s) to nclude a.q prcdictor(s) in the synthetic control model. We lls) this procedure to cbooec Rmong four potent.ill. lag cboicch u.scd in the synthetic control litera\;ure; these cl1oices involve ncluding lags of Llie dependent variable in CVl:lry pretrcat;ment yeu.r, three la.@! of the dependant v"rl'lll ble, r.g one lag which ill the average of t.ha dependent vadoble.in the prn-t.roa.tment pcrio<l, o.u<l. one lag which is the vnluo of the depemlcut wriablc in the year prior to :a:rc adoption. 57 'l'o implement tl1e croes va.lidaurm procedure, we fir.st define our training period as 1977 through the sixth year prior to RTC atloption, the vnlir.la.tfon period M the fifth yco.r prior to RTC adoption t.hrough ono ye11,r prior to Rl'C adoption,, tuuj Lhe full pre-treatment period l\ through one ycor prior to RTC adoption,. For each of ou;r 33 treatment units, dd.ta Tow the tra.ining period is used to determine tho composition of tho synthetic control. Specifically, for ea.ch of the :;l;j treatment unit.s, we assign the treatment 5 yea.re hefon:: t he tre.'ltru.cnt ll,(,1ually occnrrcd, a.nd thcu run the synthc\.ic control pcogra.m using thn standn.rd ADZ predicto1:11 defined in Aneja. et al. (211) and a S year reporting window. \Vo then ex.amino the fit dnring the trrunlog period, the vn.lldation period, and the 1mtirn pre-treatment period to see how cl=ly the synthetic control estimate matches tho value or.he depen.dcnt w.riable for different lag choices. 'Thblo11 All- A13 examine the fit or the 1:1ynthetlc c:ontrol e.11tin1!:lte during tha training pcric}d,,'alidntio11 l)<lriod, and the entire prc-troatment period using three dhfercnt losa functions. Table All dcfi.ncs the error using t.he mean aquared error belween the ~clual value of the dependant variable and the synthetic control cstirnato during a given period; Table A12 tii,a; the mea11 of the o.bsoh1~c value of t,l,c difference briltween the treated Vll.luc n.nd synthetic couttol t'.stimat.c;.finally, Tobie AlJ UBCS the CV of the RMSPE. For 'fables All-AUi, an un.weighw<l a.vera.ge of tl1h error for 1-.\J;h of the 33 t.rnatment st11.l~ i1:1 prescuw.cl. For 'Tobia.~ Al4-A16, a population weighted average of the CTor for each of the 33 tre.a.tmcnt statei; i1:1 prcsont-ed, where populat:on from the first year of \.he relevnnt period is used.sa The results from T11bles All-Al 6 provide Rtrong evidence that us~ ycady lai~ of tho dependent vnriable is the heat option. AR expected, acro.11:1 all six ta.blail, the error lu i.he training period is low~l using yc>..arly n.gs. How(:V()r, ycll!ly l~ aiao providel' the lowest r..r.rot in the vnlicl~tion pcrio<l, rngardlcss of how the crrot is defined or whether populallon wcighla \J'l:l used lo o.ggrer;at.c the nw\ilure of error oytjr all treatment stateil. n acidition, nerot1t1 all six te.blea, t.he error over the full plv-treatment JlBdod ill lowest utijng yearly ~19i. A potential concern with using Elli prcintcrvcnuon outcomes of the dcpcndo.o.t variable n.q syntbctic conlrol predict.ora rn that the syntuetic control 11nit will not clrlf:!e!ly l!jl;tch t1rn lreated imit 11 the non-agghcl predictorh during the p.c1hreatment period..5 9 But as Table Al 7 showa, we do nol fwd that tho synthetic conlrol unit's fit on the non-lagged predictot'; ih worae urlng yearly lngs. To generate thf! n.umbers in.tuble Al 7, foe each ~realmenl alo.l.c, we fl.rat take o. simple 11.vcro.gc of our prcdi<;lor of interest o,-or a.11 pro-t.rrotmenl yea.t'l'l (1977 th.tough the year prior to R.'rC!idoption). A population weighted average of the prodidur pre-tre.a.t ment means ih then t.uken ovar nll tr~tment r;1o.teti to reach the figures presented, which?ejyr~nt an nggrng11te r;a'l'be flrmt Lig ib the Vll,]11e of tho dependent ~rluble in 1977, tlm nocond lag ill th~ vnlue of the dapuudunt variable lu the yea: prior to RTC t\dop\ion, PJ.Dd the t,hird le.g 1.9 the vnlue of the dept.11do11t variable n ~h~ yoor t.hn.t ~ rn.idwl\y between the year corrcbpondlng t<'l tho fll\!l &rjd ~ce.ond lag. All rc11ul~& prcaool,cd ln Tl\blca All Lhroue:h To.blc Al6 1!Jlr. owm.ll vlolonl r.rimo o.a tbc dopon\lcnt vari11blo. 7Toe flrlit choice is W?OO, for exu.mple, n Bolm 1:~ al. {214), the oocond choice ls UHcd by Aba.dle et al. (21), anc.l lh1: third and fouru1 ('.hokcl5 are rngge,-.1.i:d by Kaul et lll (2 l!l). 58T he Ar8t ytt,.t of the training Md fuu pro-trce.tmcnt period ls 1977, whi!e tho i.rt year of the Vf\lidotion p1!1'locl hi t l,.o fifth yaar prior lo RTC adopi.l.mi. 1>SC1J Kaul ct nt. (216). 81 Li Deel

140 Case 2:16-cv-6164-JAK-AS Document 45-1 Filed 9/11/17 Page 49 of 51 Page D Case: , 1/2/218, D: , #:693 DktEntry: 17-9, Page 14 of 292 me.,suro of the pre-treatment predlc:tur mean.s. 6 B as!<! Qll the abaolute vn.lne of the difference between the aggregate trca.tod predictor mcn.ns and the aggregate synthetic control predictor n11.11:1, yearly lags has the ~econd best p erfonna.nce. The aggege.to synthetic control predictor me!.lb using yoarly lags comes c:losest or second closest to the trr.ntcd unit for 9/16 predictors. n coroparisoll, one lag t hat ia tho twcro.ge of the dependent va.rfable in the pre-treatment period comes clollll8t or second cloooa~ for 11/16 predicton1, one lag t hat ia the vnluc of the dependent vnrio.blo in the 1.ruit pro-treatment year comes cl()s()st or second closcbt for 7/16 pre<liclor.i, and thrco ~ for 5/16 prcdictoni. Vfu Lhu.s choose yearly l.tib'll uf thti dependent vu.riti.l>le aa our optima] lag choice for two rnnin rtla!lona. T hu first is tho.t yearly lags produces the lowest error not only n the training period, but e.lso in the validation period and the run pre-treatment patlod. 1'his statmnont s robuht t.o,'lll"iollil wayll of defining thu,mor and a.ggrege.~ the crl'or C"!~ treatment st.ates. The second is that the Hynth ctic e<introl unit.a using ycl:ltly le.es do 11. mirly good job, relativ1-, t.o the ot.her lag C'.hoice1;1, of matr..hing the pre--treatmcnt (nnn-l~ed) predictor mcana of t he treatment states. l!()uuloo: Tiilila; A11-.A1, where,ho treatment y-ea.r r!"jl: our 33 Btatffi or intcrca~ is amigne<l t.o C\vc ycan; before Lhc~.llci.u.nl year of TTC adoption, in Table A 17, the treatment yeiv s idcdtica.l to the ~r of rnc a.dopi.ion. For Table A 17, t.l1e etotcs eligible Lo bo lo 11. treated unll 's ~y11ihotic conlrol are i h<l6o states that elth11t no"yar pasaed FTC lawa, or pdeee<i rn~r,, than lo yea.re after t.bo trcatocl 1.1niL u<lt,ptod ntc law!j. lu ~unh-a11t, for Tables A 11- Al6, tho sta.teai u!igh.ifo to ba in a tru,1~c,;l unit 's ~ynt11,1tic con~rol are tbouc 11tv.tl>& that elther ru:vv, ~ H.''C law$, r.n: p&wc a oy yelll'!lftcr t.hc treated wilt wl Jptcd rrrc la\vll. 82 Li Deel

141 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page 5141 of 51 of 292 Page D #:694 Tobie All: Comparison of Fit AcrOOJ> varioue Lagchoicer,; Define flt using Mean Squared Enor tnl.wc'~t W-i"!-1~..bitJuz~ lroran[q.w'm~ ru,~~-~ tpu...s. ==""" c.' OTN~ 1, ,~!..tJ J.l:lrJ JU.-17 lip l..j'a::' l,tci~ *. 4-)5.. Jl cu~~ a,~ J,...'ll.. llln!!!j!l.6:ul~..._t"'"':' 1 1 U.4" 1.W~l f.tll N,...,. A~~ a-d fttw~~ u.~jfflnpt.~ kl.-t1w1., ll~,-,..._..jda,'hi:u.,rcbd~lfflwa1a Jll1',-c.. ~'D...Wt::DJC'('l~ - 1\brn.c:hf(reywe.r i 'l'ol,le A12: Comparu,on or Fit Across varlouh Lagchoiccs; Defh1e fit using Mcau A bsolutc Difference -----=""""'= ~ ~J.lllDia~l~~ tlra ii,. JD.t.!o.uu.,.., JU."17~ ii,.& ill!.! 11. i"j... Mil""""'».<M.t,. 41 U.A OO.l!f"!l;;m-~,-r ti.a,t)~f,, ' :"Mil Atw~~ --iil111tr..d,..,.. u.~--:r"'""'.,.. ~~ l flllit..._.j11! n..;.j,,q t~lr'ffl urnt:!um,<b::rrt1r- "' ', v~,~.jtn:u.ttrapw-:~.-"-" fl.'1'~-1 'Thble A13: Compn.rison of Fit Across vu.rinuz; Lagchoices; Dflfine fit using CVRl\.1.SPE ~M! ~,...-.tnat=d~ C\~ &>ew..11 tw Clll r-cj, kp G.) O C>.J.J'r ~"'fflr,p Q.?i >.Z u.ai;- ~3r~,.. "1.-,... e a.li c.3'. nh _ Hds J.1-1.'ftiltC 1~<ffbb.dl,.., us ~aonn J,...,..,.lmff,~ a:.;da-..nnnll'll 1)'UJJ.1~fra:s t.r;-, u.,np. w o, \, \~~h:ni Ktt;,_-. U"Ml.&.l~Jffl- l Table A14: Comparison of Flt Acroos various Lagchoices.; Define fit using Mean Squared F.rror - ---~- - ~ :_)'-~4711 ~M... V-!r9'!1;!En bl:~e-i... trffl~...,.._o a. aa:n,,..._,-w.., J-t-1,op l,ut.4f "u..q ~llul,-m...,. ",u,a ~tt111 v :t.n ~~~ ~Hl.91 ta,-.r. ~ -----~ ~ UM~1-.eilhi.r _ ii~u.. ~ ~--""a.,~p,ct,j5-lll~\lo... C'.J,-. ~"""""lw,.jl,:17'v ~11fN,-, t,,...~ta~"~..., Tobie Al5; Comparlson of }"it Across various Lagchoiccs; Define fit using Mean Absolute Difforence ~...,,...,";L:.u...&..t..a '::...iliij.. bra ~.\W... D=ti,.,..-.-4J'"'- 1:1.U All M..- _.._._ r 11 --~ w. ~ ~,._.~..,- a.n ---~ '4" A).fllll.l,l,,t.. #,l"1r~mfl>,t._...,... "'.._\oo_......n~t1ta..---.,-._...,...,.lffl:l,_4 ~---an:,-., ~C"O,... l 1~"-'-'-,1..._.~ -... Table AlO: Comparison of l!'it Across various Lagchoices; Define flt uning CVRMSPE '~,-...S..~"M... Lo._. '-l~j,,mj,a.t:'11'""*"',_...,. 114' OY :;-,,ilia l,-oil OU 1 t t.. ~..,_.,.... OJ li.1 ~ ---W'p a,: LO - Dt4 UJ -..,._ uw ~ _,...,".,. -'~1-.v:tn ~--- ~ "" - -.~-rl"'t,~ l'u.i.-...,rrr u...;. 1nt1,-.,... w1gvr,j'vol..._m...,:,- -'~" im:,...,, l'wl..\iliuol n.-~., - l'rll'll '- ' Li Deel

142 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , 45-1 Filed DktEntry: 9/11/ , Page Page of 51 of 292 Page D #:695 Table Al 7: Crime Predictor Means Before ltl'c Adoption ~-t-,j SytJlllf'JC: 5 lts) S~c:yuarly~ Syalbelf. : 1 ~ " 3 S) albaf.lc, 1 lag flnnl. ~'1i!eklleu!. >'l!ll: pr.,r,trtrtvwwnw 1,459, 1s:1 8,2, 13~ 8,4'r9, l27?, n~,6114 9, llll,91!8 l.joca,c_ Ul 1&'11.M l!j.l.n 1!7J.32 1Jl7.4 l_pcjjclnllll~~ ~ ll!,u m.s.s :.rm SR rpq1i lll,827.l lo4,aa2.7l u,,m.bl 14,,l~~.llO 14,41W.7fl rpcul M.iO lil.31 M.6'{ rpi:lw Q :b:!2.n 1!)2.14 2,,76 xpttpo l,~2'f,li3 1, ,,W..!l7 t, J,447.'ril l.illaclj~yltulu;,_ra,lt, B.17 ~.rn 6J) ~O\'l!r,7_;:w, u.e1 1:l.J '9 Ufft d.mjnl\y 12J,51 :/C(!.Jl :!31j..a1J :.16:Z.119 ~-bm...wl9 l.' ~ ll.112.7~ bm_ 21Yl!l Lil D,n.75 D.l!O.76 ag,,_bm_ail39 U.113 D.~ o.~ D.11~.57 ~ww Sl (1.2~ ~wm_'lll2 l,l '/,Oil 7.U 7.Hi -Wl.11_~1) ti..u ~ q N ""':.,,. encl!. troatmo:it tam, tl>lt p,-;.!jda- o f inl.fflftc Lo =snd "".Jl ~... L,~w.cl,=, (1977 through m'c Y ll l }, A pap.,j1<tl<,11 li<i;bkd a'"""w' c, t.b.&o dajinic ls ih"'1 LAlrn:J,,... all 1,.,.\-u:ims tai"" to rt!al'.h 1h11 t'jrunii< r=,..,..l.d.. 84 Li Deel

143 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 1 of of Page 292D #:696 EXHBT 9 Li Deel. Ex

144 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 2 of of Page 292D #:697 NBER WORKNG PAPER SERES RGHT-TO-CARRY LAWS AND VOLENT CRME: A COMPREHENSVE ASSESSMENT USNG PANEL DA TA AND A STATE-LEVEL SYNTHETC CONTROLS ANALYSS John J. Donohue Abhay Aneja Kyle D. Weber Working Paper NATONAL BUREAU OF ECONOMC RESEARCH 15 Massachusetts A venue Cambridge, MA June 217 We thank Dan Ho, Stefano DellaVigna, Rob Tibshirani, Trevor Hastie, Stefan Wager, and conference participants at the 211 Conference of Empi1ical Legal Studies (CELS), 212 American Law and Economics Review (ALER) Annual Meeting, 213 Canadian Law and Economics Association (CLEA) Annual Meeting, and 215 NBER Summer nstitute (Crime) for their comments and helpful suggestions. Financial support was provided by Stanford Law School. We are indebted to Alberto Abadie, Alexis Diamond, and Jens Hainmueller for their work developing the synthetic control algorithm and programming the Stata module used in this paper and for their helpful comments. The authors would also like to thank Alex Albright, Andrew Baker, Bhargav Gopal, Crystal Huang, saac Rabbani, Akshay Rao, and Vikram Rao, who provided excellent research assistance, as well as Addis O' Connor and Alex Chekholko at the Research Computing division of Stanford's nformation Technology Services for their technical support. The views expressed herein are those of the author and do not necessarily reflect the views of the National Bureau of Economic Research. NBER working papers are circulated for discussion and comment purposes. They have not been peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications. 217 by John J. Donohue, Abbay Aneja, and Kyle D. Weber. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including notice, is given to the source. Li Deel. Ex

145 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 3 of of Page 292D #:698 Right-to-Carry Laws and Violent Crime: A Comprehensive Assessment Using Panel Data and a State-Level Synthetic Controls Analysis John J. Donohue, Abhay Aneja, and Kyle D. Weber NBER Working Paper No June 217 JEL No. KO,Kl4,K4,K4,K42 ABSTRACT The 24 report of the National Research Council (NRC) on Firearms and Violence recognized that violent crime was higher in the post-passage period (relative to national crime patterns) for states adopting right-to-can-y (RTC) concealed handgun laws, but because of model dependence the panel was unable to identify the true causal effect of these laws from the then-existing panel data evidence. This study uses 14 additional years of panel data (through 214) capturing an additional 11 RTC adoptions and new statistical techniques to see if more convincing and robust conclusions can emerge. Our prefetted panel data regression specification (the "DAW model") and the Brennan Center (BC) model, as well as other statistical models by Lott and Mustard (LM) and Moody and Marvell (MM) that had previously been offered as evidence of crime-reducing RTC laws, now consistently generate estimates showing RTC laws increase overall violent crime and/or murder when run on the most complete data. We then use the synthetic control approach of Alberto Abadie and Javier Gardeazabal (23) to generate state-specific estimates of the impact of RTC laws on crime. Our major finding is that under all four specifications (DAW, BC, LM, and MM), RTC laws are associated with higher aggregate violent crime rates, and the size of the deleterious effects that are associated with the passage of RTC laws climbs over time. We estimate that the adoption of RTC laws substantially elevates violent c1ime rates, but seems to have no impact on property crime and murder rates. Ten years after the adoption of RTC laws, violent crime is estimated to be 13-15% percent higher than it would have been without the RTC law. Unlike the panel data setting, these results are not sensitive to the covariates included as predictors. The magnitude of the estimated increase in violent crime from RTC laws is substantial in that, using a consensus estimate for the elasticity of crime with respect to incarceration of.15, the average RTC state would have to double its prison population to counteract the RTC-induced increase in violent crime. John J. Donohue Stanford Law School Crown Quadrangle 559 Nathan Abbott Way Stanford, CA 9435 andnber donohue@law.stanford.edu Kyle D. Weber Department of Economics Columbia University kdw2126@columbia.edu Abbay Aneja Stanford Law School 559 Nathan Abbott Way Stanford, CA 9435 aanej a@stanford.edu Li Deel. Ex

146 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 4 of of Page 292D #:699 For nearly two decades, there has been a spirited academic debate over whether "shall issue" concealed carry laws (also known as right-to-carry or RTC laws) have an important impact on crime. The "More Guns, Less Crime" hypothesis originally articulated by John Lott and David Mustard (1997) claimed that RTC laws decreased violent crime (possibly shifting criminals in the direction of committing more property crime to avoid armed citizens). This research may well have encouraged state legislatures to adopt RTC laws, arguably making the pair's 1997 paper in the Journal of Legal Studies one of the most consequential criminological articles published in the last twenty-five years. The original Lott and Mustard paper as well as subsequent work by John Lott in his 1998 book More Guns, Less Crime used a panel data analysis to support their theory that RTC laws reduce violent crime. A large number of papers examined the Lott thesis, with decidedly mixed results. A number of studies, primarily using the limited data initially employed by Lott and Mustard for the period , supported the Lott and Mustard thesis, while a host of other papers were skeptical of the Lott findings. 1 t was hoped that the 25 National Research Council report Firearms and Violence: A Critical Review (hereafter the NRC Report) would resolve the controversy over the impact of RTC laws, but this was not to be. While one member of the committee-james Q. Wilson-did partially endorse the Lott thesis by saying there was evidence that murders fell when RTC laws were adopted, the other 15 members of the panel pointedly criticized Wilson's claim, saying that "the scientific evidence does not support his position." The majority emphasized that the estimated effects of RTC laws were highly sensitive to the paiticular choice of explanatory variables and thus concluded that the panel data evidence through 2 was too fragile to support any conclusion about the true effects of these laws. This paper begins by revisiting the panel data evidence to see if extending the data for an additional 14 years, thereby providing additional crime data for prior RTC states as well as on 11 newly adopting RTC states, offers any clearer picture of the causal impact of allowing citizens to caity concealed weapons. Across seven different permutations from four major sets of explanatory variables-including our preferred model (DAW) plus models used by the Brennan Center (BC), Lott and Mustard (LM), and Moody and Marvell (MM)-RTC laws are associated with higher rates of overall violent crime and/or murder. To answer the call of the NRC Report for new approaches to estimate the impact of RTC 1 n support of Lott and Mustard (1997), see Lott's 1998 book More Guns, Less Crime (and the 2 and 21 editions). Ayres and Donohue (23) and the 25 National Research Council repo1t Firearms and Violence: A Critical Review dismissed the Lott/Mustard hypothesis as lacking credible statistical support, as did Aneja, Donohue, and Zhang (211) (and Aneja, Donohue, and Zhang (214) further expanding the latter). Moody and Marvell (28) and Moody et al. (214) continued to argue in favor of a crime-reducing effect of RTC laws, although Zimmennan (214) concludes that RTC laws increase violent crime, as discussed in Section.B.6. 2 Li Deel. Ex

147 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 5 of 147 of Page 292D #:7 laws, we use a new statistical technique designed to address some of the weaknesses of panel data models, that has gained prominence in the period since the NRC Report's release (25). Using the synthetic controls methodology, we hope to present the type of convincing and robust results that can reliably guide policy in this area. 2 This synthetic controls methodology-first introduced in Abadie and Gardeazabal (23) and expanded in Abadie, Diamond, and Hainmueller (21) and Abadie, Diamond, and Hainmueller (214)-uses a matching methodology to create a credible "synthetic control" based on a weighted average of other states that best matches the pre-passage pattern of crime for each "treated" state, which can then be used to estimate the likely path of c1ime if RTC-adopting states had not adopted a RTC law. By comparing the actual crime pattern for RTC-adopting states with the estimated synthetic controls in the post-passage period, we derive year-by-year estimates for the impact of RTC laws in the ten years following adoption. 3 To preview our major findings, the synthetic controls estimate of the average impact of RTC laws across the 33 states that adopt between 1981 and 27 4 indicate that violent crime is substantially higher after ten years than would have been the case had the RTC law not been adopted. Essentially, for violent crime, the synthetic controls approach provides a similar portrayal of RTC laws as that provided by the DAW and BC panel data models and undermines the results of the LM and MM panel data models. According to the aggregate synthetic control models-whether one uses the DAW, BC, LM, or MM covariates-rtc laws led to increases in violent crime of percent after ten years, with positive but not statistically significant effects on property crime and murder. The median effect of RTC adoption after 1 years is 14.1 percent whether one considers all 31 states with ten years of data or limits the analysis to the 26 states with the most compelling pre-passage fit between the adopting states and their synthetic controls. Comparing our DAWspecification findings with the results generated using placebo treatments, we are able to reject the null hypothesis that RTC laws have no impact on aggregate violent crime. The structure of the paper proceeds as follows. Part ll discusses the panel data results for the four different models, showing that the DAW and BC models indicate that RTC laws have increased violent and property crime, while the LM and MM models provide evidence that RTC 2 Abadie, Diamond, and Hainmueller (214) identify a number of possible problems with panel regression techniques, including the danger of extrapolation when the observable characteristics of the treated area are outside the range of the con-esponding characteristics for the other observations in the sample. 3 The accuracy of this matching can be qualitatively assessed by examining the root mean square prediction error (RMSPE) of the synthetic control in Lhe pre-treatment period (or a variation on this RMSPE implemented in this paper), and the significance of the estimated treatment effect can be approximated by running a series of placebo estimates and examining the size of the estimated treatment effect in comparison to the distribution of placebo treatment effects. 4 Note that we do not generate a synthetic control estimate for ndiana, even though it passed its RTC law in 198, owing to the fact that we do not have enough pre-treatment years to accurately match the state with an appropriate synthetic control. We consider Lhe effect of making ndiana a treatment state as a robustness check and find that this change does not meaningfully change our results. Similarly, we do not generate synthetic control estimates for owa and Wisconsin (whose RTC laws went into effect in 211) and for llinois (214 RTC law), because of the limited post-passage data. 3 Li Deel. Ex

148 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 6 of of Page 292D #:71 laws have increased murder. We argue that the DAW set of explanatory variables are the most plausible and show that modest and advisable corrections to the LM and MM specifications also generate estimates that RTC laws increase violent crime. The remainder of the paper shows that the synthetic controls approach under all four sets of explanatory variables uniformly supports the conclusion that RTC laws lead to substantial increases in violent crime. Part ll describes the statistical underpinnings of the synthetic controls approach and specific detai s of our implementation of this technique. Part N provides our synthetic controls estimates of the impact of RTC laws, and Part V concludes with some thoughts on the mechanisms by which RTC laws increase violent crime.. Panel Data Estimates of the mpact of RTCLaws A. The No-Controls Model We follow the NRC Report by beginning with the basic facts about how crime has unfolded relative to national trends for states adopting RTC laws. Figure 1 depicts percentage changes in the violent crime rate over our entire data period for three groups of states: those that never adopted RTC laws, those that adopted RTC laws sometime between 1977 and 214, and those that adopted RTC laws prior to t is noteworthy that the nine states that never adopted RTC laws experienced declines (in percentage terms) in violent crime that are greater than four times the reduction experienced by states that adopted RTC either prior to 1977 or during our period of analysis. 5 The NRC Report presented a "no-controls" estimate, which is just the coefficient estimate on the variable indicating the date of adoption of a RTC law in a crime rate panel data model with state and year fixed effects. According to the NRC Report, "Estimating the model using data to 2 shows that states adopting right-to-carry laws saw 12.9 percent increases in violent crime-and 21.2 percent increases in property crime-relative to national crime patterns." We now estimate this same model using 14 additional years of data (through 214) and 11 additional adopting states (listed at the bottom of Table 8). Row l of Table 1 shows the results of 5 ver the same period, the states that avoided adopting RTC laws had substantially lower increases in their rates of incarceration and police employment. The nine never-adopting states increased their incarceration rate by 25 percent, while the incarceration rates in the adopting states rose by 262 and 259 percent, for those adopting RTC laws before and after 1977 respectively. Similarly, the rate of police employment rose by 16 percent in the never-adopting states and by 38 and 55 percent, for those adopting before and after 1977, respectively. 4 Li Deel. Ex

149 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 7 of of Page 292D #:72 Figure 1 The Decline in Violent Crime Rates has been Far Greater in States with No RTC Laws, Data Sources: UCR for crime rales; Census for state populations.,._ % l!!l 1977 D 214 J!l C: <D Q) "O.; Q) a:: <() ~ ;;; a. v Q) i;; a:: Q) (") E t'3 c Q) N > ~ Ra te; Slates 84.3M People -8.7% - 9.9% Rate; Slates 17.5M People States that have never adopted RTC Laws States that have adopted RTC laws between 1977 and 214 States that adopted RTC laws prior to 1977 this "no-controls" panel data approach using a dununy model, which just estimates how much on average crime changed after RTC laws were passed (relative to national trends). According to this model, the average post-passage increase in violent c1ime was 2.2 percent, while the comparable increase in property crime was 19.2 percent. Row 1 also reports the impact of RTC laws on the murder rate (Column 1) and the murder count using a negative binomial model (Column 2), which provide statistically insignificant estimates that RTC laws increase murder by 4-5 percent. 6 The NRC Report also presented a spline model to estimate how RTC adoption might alter the trend in crime for adopting states, which suggested violent crime and property declined relative to trend in the data through 2, while the trend in murder was unchanged. Row 2 of Table 1 recomputes this "no-controls" spline model on data through 214, which eliminates the earlier suggestion that RTC laws were associated with any drop (relative to trend) in violent or property crime, and reaffirms the null finding for murder. 7 n other words, more and better data have strengthened the 6 The dummy variable model reports the coefficient assoc iaced with an RTC variable that is given a value of zero if an RTC Jaw is not in effect in that year, a value of one if an RTC Jaw is in effect that entire year, and a value equal to the pot1ion of the year an RTC law is in effect otherwise. The date of adoption for each RTC state is shown in Appendix Table A 1. 7 The spline model repo11s results for a variable which is assigned a value of zero before the RTC law is in effect and a value equal to the portion of the year the RTC law was in effect in the year of adoption. A fter this year, the value o f the this vati able is incremented by one annually for states that adopted RTC laws between 1977 and 214. The spline model also includes a second trend variable representi ng the number of years that have passed since 1977 for the states adopting RTC laws over the sample period. s Li Deel. Ex

150 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 8 of of Page 292D #:73 dummy variable model finding that RTC laws increase violent crime, and eliminated the earlier spline model showing of possible declines in violent and property crime. Table 1: Panel Data Estimates Showing Greater ncreases in Violent and Property Crime Following RTC Adoption: State and Year Fixed Effects, and No Other Regressors, Murder Rate Murder Count Violent Crime Rate Property Crime Rate (1) (2) (3) (4) Dummy Variable Model 3.83 (8.79) 1.49 (.53) 2.21 *"* (6.83) 19.18*** (6.6) Spline Model -.28 (.61) 1.4 (.4).22 (.79).14 (.5) OLS estimations include year and state fixed effects and are weighted by state population. Robust standard errors (clustered at the state level) are provided next to point estimates in parentheses. ncidence Rate Ratios (RR) estimated using Negative Binomial Regression, where state population is included as a control variable, are presented in Column 2. The null hypothesis is that the RR equals 1. The source of all the crime rates is the Uniform Crime Rep1ts (UCR). * p <.1, ** p <.5, *** p <.1. All figures reported in percentage terms. While the Table 1 dummy model indicates that RTC states experience a worse -post-passage crime pattern, this does not prove that RTC laws increase crime. For example, it might be the case that some states decided to fight crime by allowing citizens to carry concealed handguns while others decided to hire more police and incarcerate a greater number of convicted criminals. f police and prisons were more effective in stopping crime, the "no controls" model might show that the crime experience in RTC states was worse than in other states even if this were not a true causal result of the adoption of RTC laws. As it turns out, though, RTC states not only experienced higher rates of violent crime but they also had larger increases in incarceration and police than other states. While the roughly 7 percent greater increase in the incarceration rate in RTC states is not statistically significant, the increases are large and statistically significant for police. Accordingly, Table 2 confirms that RTC states did not have declining rates of incarceration or total police employees after adopting their RTC laws that might explain their relatively bad crime performance. B. Adding Explanatory Variables We know from the analysis of the dummy model in the NRC Report and in Table 1 that RTC law adoption is followed by higher rates of crime (relative to national trends) and from Table 2 that the poorer crime performance after RTC law adoption occurs despite the fact that RTC states continued 6 Li Deel. Ex

151 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 9 of of Page 292D #:74 Table 2: Panel Data Estimates Showing Greater ncreases in ncarceration and Police Following RTC Adoption: State and Year Fixed Effects, and No Other Regressors, ncarceration Police Employment Per 1k ( l) (2) Dummy Variable Model 6.78 (6.22) 8.39m (3.15) Police Officers Per 1k (3) 7.8** (2.76) Estimations include year and state fixed effects and are weighted by state population. Robust standard errors (clustered at the state level) are provided next to point estimates in parentheses. The source of the police employment rate and the sworn police officer rate is the Uniforn1 Crime Reports (UCR). The source of the incarceration rate is the Bmeau of Justice Statistics (BJS) * p <.1, ** p <.5, *** p <. 1. All figures reported in percentage terms. to invest at least as heavily in prisons and actually invested more heavily in police than non-rtc states. While the theoretical predictions about the effect of RTC laws on crime are indeterminate, these two empirical facts based on the actual patterns of crime and crime-fighting measures in RTC and non-rtc states suggest that the most plausible working hypothesis is that RTC laws increase crime. The next step in a panel data analysis of RTC laws would be to test this hypothesis by introducing an appropriate set of explanatory variables that plausibly influence crime. The choice of these variables is impottant because any variable that both influences crime and is simultaneously correlated with RTC laws must be included if we are to generate unbiased estimates of the impact of RTC laws. At the same time, including irrelevant and/or highly collinear variables can also undennine effmts at valid estimation of the impact of RTC laws. At the very least, it seems advisable to control for the levels of police and incarceration because these are the two most important c1iminal justice policy instruments in the battle against crime. 1. The DAW Panel Data Model n addition to the state and year fixed effects of the no controls model and the identifier for the presence of a RTC law, our preferred "DAW model" includes an array of other factors that might be expected to influence crime, such as the levels of police and incarceration, various income, poverty and unemployment measures, and six demographic controls designed to capture the presence of males in three racial categories (Black, White, other) in two high-crime age groupings (15-19 and 2-39). The full set of explanatory variables is listed in Table 3, along with the regression models used in three other studies that have estimated the impact of RTC laws on crime. 8 8 While we attempt to include as many states in these regressions as possible, District of Columbia incarceration data is missing after the year 21. n addition, a handful of observations are also dropped from the LM and MM regressions owing to states that did not report any usable arrest data in various years. Our regressions are performed 7 Li Deel. Ex

152 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 1152 of 47 of 292 Page D #:75 'rhe DAW panel data model in Table 4 (run on data from ) is consistent with the same basic pattern observed in Table 1: 9 RTC laws on average increased violent crime by 9.5 percent and property crime by 6.8 percent in the years following adoption according to the dummy model, but again showed no statistically significant effect in the spline model. 1 As we saw in the no-controls model, the estimated effect of RTC laws in Table 4 on the murder rate is also not statisticaj l y significant. 2. The BC Panel Data Model Table 3 lists the variables used in the Brennan Center (BC) c1ime regression model, which differ in a few respects from the DAW model (although to a lesser degree than the LM and MM models) Roeder et al The BC model controls for both incarceration and police rates (as in DAW), but the BC model takes the log of both these rates. The BC model alone controls for the number of executions, and unlike DAW does not control for either the state poverty rate or the percentage of the state population living in a Metropolitan Statistical Area. Moreover, while DAW includes six demographic variables, BC uses three age groupings over the ages 15-29, and simply controls for the black percentage of the state population. The results of running the BC model over the period from are presented in Table 5, Panel A. With the exception that the BC dummy variable model estimate of the increase in violent crime is somewhat higher than that for DAW (1.98 percent increase versus 9.49 percent increase), the DAW and BC model estimates are almost identical in suggesting higher rates of violent and prope1ty crime (the dummy models) but no impact in the spline models. f we replace the four BC demographic variables with the 6 DAW demogrnphic vaiiables (Table 5, Panel B), the size of the estimated increases in violent crime and property crime (in the dummy models) are only modestly lower than the DAW results in Table The LM Panel Data Model Table 3 's recitation of the explanatory variables contained in the Lott and Mustard (LM) panel data model reveals two obvious omissions: there are no controls for the levels of police and incarcerwith robust standard enws that are clustered at the state level, and we lag the arrest rates used in both the LM and MM regression models. The rationales underlying both of these changes are desc1ibed in more detail in Aneja, Donohue, and Zhang (214). All of the regressions presented in this paper are weighted by state population. 9 The complete set of estimates for all explanatory variables (except the demographic variables) for the DAW, BC, LM, and MM dummy and spline models is shown in appendix Table A2. lodefensive uses of guns are more likely for violent crimes because the victim will clearly be present. For property crimes, the victim is typically absent, thus providing less opportunity to defend with a gun. t is unclear whether the many ways in which RTC laws could lead to more crime, which we discuss in Part V, would be more hkely to facilitate violent or property crime, but our intuition is that violent crime would be more strongly influenced, which is in fact what Table 4 suggests. 8 Li Deel. Ex

153 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:76 Table 3: Table of Explanatory Variables For Four Panel Data Studies Explanatory Variables Right to Carry Law Lagged Per Capita ncarceration Rate Lagged Log of Per Capita ncarceration Rate Lagged Police Staffing Per 1, Residents Lagged Log of Sworn Police Officers Per Resident Population Lagged Number of Executions Poverty Rate Unemployment Rate Per Capita Ethanol Consumption from Beer Percentage of the State Population living in Metropolitan Statistical Areas (MSAs) Real Per Capita Personal ncome Nominal Per Capita ncome (Median ncome in BC) Real Per Capita ncome Maintenance Real Per Capita Retirement Payments Real Per Capita Unemployment nsurance Payments Population Density Lagged Violent or Property Arrest Rate State Population Crack ndex Lagged Dependent Variable DAW X X X X X X X X BC X X X X X X X LM X X X X X X X X!\!M X X X X X X X X X X X X 6 Age-Sex-Race Demographic Variables -all 6 combinations of black, white, and other males in 2 age groups (15-19, 2-39) indicating the percentage of the population in each group X 3 Age-Group Percentages (15-19, 2-24, 25-29), and Black Percentage of Population X 36 Age-Sex-Race Demographic Variables -all possible combinations of black and white males in 6 age groups (1-19, 2-29, 3-39, 4-49, 5-64 and over 65) and repeating this all for females, indicating the percentage of the population in each group X X Note: The DAW model is advanced in this paper, while the other three models were previously published by the Brennan Center (BC), Lott and Mustard (LM), and Marvell and Moody (MM). See footnote 41 in Appendix B for an explanation of the differences in the retirement payments variable definition between the LM and MM specifi cations. The crack index variable in the MM specification is available only for Li Deel. Ex

154 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:77 Table 4: Panel Data Estimates Suggesting that RTC Laws increase Violent and Property Crime: State and Year Fixed Effects, DAW Regressors, Murder R ate Murder Count Violent Crime Rate Property Crime Rate (1) (2) (3) (4) Dummy Variable Model.3 (5.35) 1.5 (.52) 9.49*** (2.96) 6.76** (2.73) Spline Model -.3 l (.53) l.2 (.4).5 (.64).14 (.38) OLS estimations include year and state fi xed effects and are weighted by state population. Robust standard errors (clustered at the state level) are provided next to point estimates in parentheses. lncidence Rate Ratios (RR) estimated using Negative Binomial Regression, where state population is included as a control variable, are presented in Column 2. The null hypothesis is that the RR equals 1. The source of all the crime rates is the Uniform Crime Reports (UCR). Six demographic variables (based on different age-sex-race categories) are included as controls in the regression above. Other controls include the lagged incarceration rate, the lagged police employee rate, real per capita personal income, the unemployment rate, poverty rate, beer, and percentage of the population living in MS As. * p <.1, ** p <.5, *** p <.1. All figures reported in percentage terms. ation in each state, even though a substantial literature has found that these factors have a large impact on crime. ndeed, as we saw above in Table 2 both of these factors grew after RTC law adoption, and the increase in police employment after RTC adoption is substantively and statistically significant. A Bayesian analysis of the impact of RTC laws found that "the incarceration rate is a powerful predictor of future crime rates," and specifically faulted this omission from the Lott and Mustard model (Strnad 27: 21, fn. 8). Without more, then, we have reason to believe that the LM model is mis-specified, but in addition to the obvious omitted variable bias, we have discussed an anay of other infirmities with the LM model in Aneja, Donohue, and Zhang (214), including their reliance on flawed arrest rates, and highly collinear demographic variables. As noted in Aneja, Donohue, and Zhang (214), "The Lott and Mustard arrest rates... are a ratio of arrests to crimes, which means that when one person kills many, for example, the arrest rate falls, but when many people kill one person, the arrest rate rises since only one can be arrested in the first instance and many can in the second. The bottom line is that this "arrest rate" is not a probability and is frequently greater than one because of the multiple anests per crime. For an extended discussion on the abundant problems with this pseudo arrest rate, see Donohue and Wolfers (29)." The LM arrest rates are also econometrically problematic since the denominator of the arrest rate is the numerator of the dependent variable crime rate, improperly leaving the dependent variable 1 Li Deel. Ex

155 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:78 Table 5: Panel Data Estimates Suggesting that RTC Laws increase Violent and Property Crime: State and Year Fixed Effects, BC Regressors, Dummy Variable Model Panel A: BC Regressors ncluding 4 Demographic Variables Murder Rate Murder Count Violent Crime Rate Property Crime Rate (1) 3.45 (5.67) (2) 1.5 (.51) (3) l.98*** (3.65) (4) 6.86** (3.26) Spline Model -.49 (.51) 1.3 (.4).19 (.66).12 (.35) Panel B: BC Regressors with 6 DAW Demographic Variables Murder Rate Murder Count Violent Crime Rate Property Crime Rate ( 1) (2) (3) (4) Dummy Variable Model 1.88 (5.47) 1.57 (.51) 8.97*** (3.29) 5.57* (2.85) Spline Model -.33 (.48) 1.3 (.4).24 (.59).16 (.34) OLS estimations include year and state fixed effects and are weighted by state population. Robust standard errors (clustered at the state level) are provided next to point estimates in parentheses. ncidence Rate Ratios (RR) estimated using Negative Binomial Regression, where state population is included as a control vaiiable, are presented in Column 2. The null hypothesis is that the RR equals 1. The source of all the crime rates is the Uruform Crime Reports (UCR). Four demographic variables (percent black, percent aged 15-19, percent aged 2-24, and percent aged 25-29) are included in the Panel A regressions. The 6 DAW demographic variables are used in the Panel B regressions. Other controls include log of the lagged incarceration rate, lagged police employment per resident population, the unemployment rate, nominal per capita income, lagged number of executions, gallons of beer consumed per capita. * p <., ** p <.5, *** p <.. All figures reported in percentage terms. 11 Li Deel. Ex

156 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:79 on both sides of the regression equation. We lag the an est rates by one year to reduce this problem of ratio bias. Lott and Mustard's use of 36 demographic vaiiables is also a potential concern. With so many enormously collineai variables, the high likelihood of introducing noise into the estimation process is revealed by the wild fluctuations in the coefficient estimates on these vaiiables. For example, consider the LM explanatory variables "neither black nor white male aged 3-39" and the identical corresponding female category. The LM dummy vaiiable model for violent crime suggests that the male group will vastly increase crime (the coefficient is 211!), but their female counterparts have an enormously dampening effect on crime (with a coefficient of -255!). Both of those highly implausible estimates (not shown in Table A2) are statistically significant at the 1 percent level, and they are almost certainly picking up noise rather than revealing true relationships. BizruTe results are common in the LM estimates among these 36 demographic vai"iables. 11 Table 6, Panel A shows the results of the LM panel data model estimated over the period As seen above, the DAW model generated estimates that RTC laws raised violent and property crime (in the dummy model of Table 4), while having no obvious impact on murders. The LM model flips these predictions by showing strong estimates of increased murder (in the spline model) and no evidence of increased violent or property crime. We can almost perfectly restore the DAW Table 4 findings, however, by simply following the typical pattern of crime regressions by limiting the inclusion of 36 highly collinear demographic variables and including measures for police and incarceration. These results appear in Panel B of Table 6, and this modified LM dummy variable model suggests that RTC laws increase crime. This finding is similar but somewhat stronger than the DAW dummy variable model estimate of higher violent and property cnme. n summary, the LM model that had originally been employed using data through 1992 to ru gue that RTC laws reduce crime, no longer shows any statistically significant evidence of crime reduction. ndeed, using more complete data, the LM spline model (Panel A of Table 6) suggests that RTC laws increase the murder rate and count by about 6 or 7 percent after 1 years, which are the only statistically significant results in Panel A-no other crime category is affected. Those who are skeptical of these results because the LM specification is plagued by omitted variable bias, flawed pseudo-arrest rates, too many highly collinear demographic variables, and other problems, 11 Aneja, Donohue, and Zhang (214) test for the severity of the multicollinearity problem using the 36 LM demographic variables, and the problem is indeed serious. The Variance nflation Factor (VF) is shown to be in the range of 6 to 7 for the RTC variable in both the LM dummy and spjine models when the 36 demographic controls are used. Using the 6 DAW variables reduces the multicollinearity for the RTC dummy to a tolerable level (with VFs always below the desirable threshold of 5). ndeed, the degree of multicollinearity for the individual demographics of the black-male categories are astonishingly high with 36 demographic controls-in the neighborhood of 14,! This analysis makes us wary of estimates of the impact of RTC laws that employ the Lott-Mustard set of 36 demographic controls (as does the MM model). 12 Li Deel. Ex

157 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:71 Table 6: Panel Data Estimates of the mpact of RTC Laws: State and Year Fixed Effects, Using Actual and Modified LM Regressors, Dummy Variable Model Panel A: LM Regressors ncluding 36 Demographic Variables Murder Rate Murder Count Violent Crime Rate Property Crime Rate (1) (3.44) (2) L.3 (.32) (3) (3.15) (4) -.33 (l.71) Spline Model.65** (.33) 1.6** (.3).41 (.47).28 (.28) Panel B: LM Regressors with 6 DAW Demographic Variables and Adding Controls for ncarceration and Police Dummy Variable Model Murder Rate Murder Count Violent Crime Rate Property Crime Rate (1) (2) (3) (4) 3.6 (5.67) 1.58 (.54) 1.6** (4.54) 8.9** (3.63) Spline Model.3 (.43) 1.3 (.4).5 (.57).5 (.34) Estimations include year and state fixed effects and are weighted by state population. Robust standard errors (clustered at the state level) are provided next to point estimates in parentheses. n Panel A, 36 demographic variables (based on different age-sex-race categories) are included as controls in the regressions above. n Panel B, only 6 demographic variables are included and controls are added for incarceration and police. For both Panels, other controls include the previous year's violent or property crime arrest rate (depending on the crime category of the dependent variable), state population, population density, real per capita income, real per capita unemployment insurance payments, real per capita income maintenance payments, and real retirement payments per person over 65. * p <.1, ** p <.5, *** p <.1. All fi gures reported in percentage terms. 13 Li Deel. Ex

158 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:711 mjght prefer the estimates in Panel B, which simply limit the LM demographic variables from 36 to 6, and add the incarceration and police controls. These changes once again restore the Table 4 DAW dummy variable model result that RTC laws increase both violent and prope1ty c1ime. 4. The MM Panel Data Model Table 3 reveals that the Moody and Marvell (MM) model improves on the LM model in that it includes the key incarceration variable, but MM also omit the critical police measure found in the DAW specification. The MM model also contains the problematic pseudo-atest rates and over-saturated and highly collinear demographic variables that LM employ. 12 Panel A of Table 7 estimates the MM model for the period While MM's use of a potentially problematic lagged dependent variable control risks purging some of the effect of the RTC law, again we see evidence that RTC laws increase violent crime. The only other statistically significant estimate is for the murder rate in the spline model, which suggests that the murder rate would be roughly 4 percent higher ten years after RTC adoption. This finding is roughly similar to the Table 6, Panel A finding of increased murder in the LM model. Panel B of Table 7 mimics our previous critique of the LM model by including a measure of police and using more appropriate demographic controls. These modifications once again revive a dummy variable model estimate of increased violent crime. 5. The Lessons from the Panel Data Studies Estimated Over the Full Data Range All four models shown in Table 4 through Table 7 showed evidence that RTC laws increased murder and/or overall violent crime. DAW and BC showed almost identical increases in violent c1ime of 9-11 percent and property crime of 6-7 percent. The LM model (Table 6, Panel A)-the heart of the miginal More Guns, Less Crime hypothesis- estimates a sizeable and statistically significant increase in murder will follow RTC adoption. A similar finding emerges for the MM model (Table 7, Panel A), which also predicts an increase in violent crime. f we look at the modified versions of the LM and MM models in their respective Panel B 's, the LM model (Table 12 While our Table 6 MM panel data specification follows Moody and Marvell (28) in including lagged values of the dependent variable as a regressor, no analogous variable is explicitly included below in our synthetic control analysis featuring the Moody-Marvell predictor variables. Since all lagged vajues of the dependent variable are already included as predictors in the synthetic controls analysis, including the lagged dependent variable would be redundant. 13 rvi:m use the crack index of Fryer et al. (213), but this comes at the price of limiting the available data years for the MM panel data analysis to the years l We estimated the MM model on the data period from with and without the crack cocaine variable, which yielded virtually identical results. Therefore, in Table 7, we exclude the crack cocaine variable, which allows us to use 15 years of additional data to estimate the effect of RTC laws (from 1979, as well as 21 through 214). 14 Li Deel. Ex

159 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:712 Table 7: Panel Data Estimates of the mpact of RTC Laws: State and Year Fixed Effects, Using Actual and Modified MM Regressors without Crack Cocaine, Panel A: MM Regressors Without Crack Cocaine and ncluding 36 Demographic Variables Mmder Rate Murder Count Violent Crime Rate Property Crime Rate Dummy Variable Model (l) (1.85) (2) 1.22 (.27) (3).69 (.77) (4).48 (.69) Spline Model.38* (.16) 1.3 (.2).17** (.8).1 (.7) Panel B: MM Regressors Without Crack Cocaine, With 6 DAW Demographic Variables, and Adding a Control for Police Dummy Variable Model Murder R ate Murder Count Violent Crime Rate Property Crime Rate (1) (2) (3) (4) 1.22 (1.75) 1.31 (.35) 1.5*** (.53).52 (.53) Spline Model.24 (.1 7) 1.1 (.3).14 (.9).5 (.5) OLS estimations include year and state fi xed effects and are weighted by state population. Robust standard errors (clustered at the state level) are provided next to point estimates in parentheses. ncidence Rate Ratios (RR) estimated using Negative Binomial Regression, where state population is included as a control variable, are presented in Column 2. The null hypothesis is that the RR equals 1. n Panel A, 36 demographic variables (based on different age-sex-race categories) are included as controls in the regressions above. n Panel B, only 6 demographic variables are included and a control is added for police. For both panels, other controls include the lagged dependent variable, the previous year's violent or property crime arrest rate (depending on the crime category of the dependent variable), state population, the lagged incarceration rate, the poverty rate, the unemployment rate, real per capita income, real per capita unemployment insurance payments, real per capita income maintenance payments, and real per capita retirement payments. * p <.1, ** p <.5, * ** p <. l. All figures reported in percentage terms. 15 Li Deel. Ex

160 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 1816 of 47 of 292 Page D #:713 6) almost perfectly replicates the increased violent and property crime estimates of DAW and BC, while the MM model (Table 7) continues to show a statistically signifi cant increase in the violent crime rate. The strongest result to emerge from the seven panels across the fom sets of panel data specifications in Tables 4-7 is that 6 of these 7 panels show statistically significant evidence that RTC laws increase violent crime. The only exception (LM Panel A) shows statistically significant evidence of increases in murder. n other words, all 7 panels support the conclusion that RTC laws increase overall violent crime and/or murder. Across the 56 estimated effects in the seven panels, not one showed any evidence of a decrease in crime at the.5 level of significance. 6. The Zimmerman Model and Our 4 Panel Data Models Estimated for the Post-Crack Period Our previous discussion has focused on panel data estimates of the impact of RTC laws on crime over the full period from the late 197s through 214. Zimmerman (214) examines the impact of various crime prevention measures on crime using a state panel data set from He finds that RTC laws increased murder by 15.5 percent for the eight states that adopted RTC laws over the period he analyzed. The advantage of using this data period to explore the impact of RTC laws is that it largely avoids the problem of omitted variable bias owing to the crack phenomenon, since the crack effect had subsided by The disadvantage is that one can only gain estimates based on the eight states that adopted RTC laws over that twelve-year spell. 14 Zimmerman describes his finding as follows: "The shall-issue coefficient takes a positive sign in all regressions save for the rape model and is statistically significant in the murder, robbery, assault, burglary, and larceny models. These latter findings may imply that the passage of shall-issue laws increases the propensity for crime, as some recent research (e.g., Aneja, Donohue, & Zhang, 212) has suggested" (7 1). 15 n Table 8, we show the results of all four basic models that we discussed above-daw, BC, LM, and MM-when run over the period The DAW model mimics the Zimmerman finding of a large jump in the murder rate. The BC model weakly supports the increase in murder, and more strongly shows an 8 percent increase in the overall violent crime rate. The results for this 14 The relatively short time span makes the assumption of state fixed effects more plausible but it also limits the amount of pre-adoption data for an early adopter such as Michigan (21 ) and the amount of post-adoption data for the late adopters Nebraska and Kansas (both in 27). 15 Aneja, Donohue and Zhang (211) also ran the ADZ model over the same O period that Zimmerman employs, which generated an estimate that murder rates rose about 1.5 percentage points each year that a RTC law was in effect. 16 We started this time period in 2 because the sharp crime decreases of the 199s ended by then and crime starting in 2 was more stable for the remainder of our data period than it had previously been. 16 Li Deel. Ex

161 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:714 shortened period using the LM and MM models are never statistically significant at the.5 level. 17 Li Deel. Ex

162 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 2162 of 47 of 292 Page D #:715 Table 8: Panel Data Estimates of the mpact of RTC Laws Using DAW, BC, LM, and MM specifications, Panel A: Panel Data Estimates Suggesting that RTC Laws increase Murder: State and Year Fixed Effects, DAW Regressors, Murder Rate Murder Count Violent Crime Rate Property Crime Rate Dummy Variable Model ( ) (2) (3) (4) 5.49 (3.58) 1.2 (.4) 4.98 (3.51) (2.27) Spline Model 1.4" (.57) 1... (.1).52 (1.11).4 (.43) Panel B: Panel Data Estimates S11ggesting that RTC Laws increase Violent Crime: State and Year Fixed Effects, Brennan Center Regressors, Dummy Va1iable Model Murder Rate Murder Count Violent Cri me Rate Property Crime Rate (1) (2) (3) (4) 7.14' (4.8) 1.3 (.4) 8.*' (3.58) - l.85 (2.47) Spline Model.89 (.66) 1.1 (.1 ).59 (1.3).36 (.46) Panel C: Panel Data Estimates With 36 Collinear Demographic Variables Show No Effect of RTC Laws: State and Year Fixed Effects, LM Regressors, Dummy Variable Model Murder Rate Murder Count Violent Crime Rate Property Crime Rate ( l) (2) (3) (4) 2.74 (3.64) 1.3 (.4) -.87 (3.35) (1.94) Spline Model.85 (.78) 1. (.1) -.6 (.73) -.32 (.5 1) Panel D: Panel Data Estimates With 36 Collinear Demographic Variables Show No Effect of RTC Laws: State and Year Fixed Effects, MM Regressors witho11t Crack Cocaine, Dummy Variable Model Murder Rate Murder Count Violent Crime Rate Property Crime Rate (!) (2) (3) (4) 2.44 (3.33) 1.2 (.4).14 (1.49) -1.44* (.86) Spline Model.68 (.81) 1. (. l ).3 (.33). 17 (. 18) Estimations include year and state fixed effects and are weighted by state population. Robust standard errors (clustered at the state level) are provided next to point estimates in parenthesis. Panels A, B, C, and D replicate the standard specifications on data. To allow for estimation in this period for the MM model, the crack index variable is dropped. The following 11 states adopted RTC Laws during the period of consideration: CO (23), A (211), l, (214), KS (27), M (2 1), MN (23), MO (24), NE (27), NM (24), OH (24), and W (211) * p <.!, ** p <.5, *** p <.1. All figures reported in percentage terms. 18 Li Deel. Ex

163 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #: Summary of Panel Data Analysis The uncertainty about the impact of RTC laws on crime expressed in the NRC Report was based on an analysis of data only through 2. The preceding evaluation of an a.tay of different specifications over the full data period from the late 197s through 214 has eliminated any suggestion of benign effects on crime from the adoption of RTC laws and consistently shown evidence that RTC laws increase murder and/or overall violent crime. Three of five models estimated on postcrack-era data (Zimmerman, DAW, and BC) provide further support for this conclusion. Durlauf, Navarro, and Rivers (216) attempts to sort out the different specification choices in evaluating RTC laws by using a Bayesian model averaging approach using county data from Applying this technique, the authors find that in their prefened spline (trend) model, RTC laws elevate violent crime in the three years after RTC adoption: "As a result of the law being introduced, violent crime increases in the first year and continues to increase afterwards" (5). By the third year, their prefeted model suggests a 6.5 percent increase in violent crime. Since their paper only provides estimates for three post-passage years, we cannot draw conclusions beyond this but note that their finding that violent crime increases by over 2 percent per year owing to RTC laws is a substantial crime increase. Moreover, the authors note that "For our estimates, the effect on crime of introducing guns continues to grow over time" (5). Despite the substantial panel data evidence in the post-nrc literature that supports the finding of the pernicious influence of RTC laws on crime, the NRC suggestion that new techniques should be employed to estimate the impact of these laws is fitting. The important paper by Strnad (27) used a Bayesian approach to argue that none of the published models used in the RTC evaluation literature rated highly in his model selection protocol when applied to data from Moreover, one member of the NRC panel (Joel Horowitz) doubted whether a panel data model could ever convincingly establish the causal impact of RTC laws: "the problems posed by high-dimensional estimation, misspecified models, and lack of knowledge of the correct set of explanatory variables seem insurmountable with observational data" (NRC 25: 38). But owing to the substantial challenges of estimating effects from observational data, it will be useful to see if a different statistical approach that has different att1ibutes from the panel data methodology can be brought to bear on the issue of the impact of RTC laws. The rest of this paper will present this new approach. 19 Li Deel. Ex

164 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:717. Estimating the mpact of RTC Laws Using Synthetic Controls The synthetic controls methodology, which is becoming increasingly prominent in economics and other social sciences, is a promising new statistical approach for addressing the impact of RTC laws. 17 A number of papers have used the synthetic control technique to evaluate various influences on crime. Rudolph et al. (215) construct a synthetic control for the state of Connecticut yielding evidence that the state's firearm homicide rate (but not its non-firearm homicide rate) fell appreciably after the implementation of a permit-to-purchase handgun law. Munasib and Guettabi (213) use this methodology to examine the effect of Florida's "Stand Your Ground" law, concluding that this law was associated with an increase in overall gun deaths. Similarly, Cunningham and Shah (217) study the effect of Rhode sland's unexpected decriminalization of indoor prostitution on the state's rape rate (among other outcome variables); Lofstrom and Raphael (213) estimate the effect of California's public safety realignment on c1ime rates; and Pinotti (213) examines the consequences of an influx of organized crime into two talian provinces in the late 197s. While these papers focus on a single treatment in a single geographic region, we look at 33 RTC adoptions throughout the country. For each adopting (treated) state we will find a weighted average of other states designed to serve as a good counter-factual for the impact of RTC laws, because this "synthetic control" had a similar pattern of crime to the adopting state prior to RTC adoption. By comparing what actually happened for the adopting state post-passage to the crime pe1formance of the synthetic control over the same period, we generate estimates of the causal impact of RTC laws on crime. 18 A. The Basics of the Synthetic Control Methodology The synthetic control method attempts to generate representative counterfactual units by comparing a treatment unit (i.e., a state adopting a RTC law) to a set of control units across a set of explanatory variables over a pre-intervention period. The algorithm searches for similarities between the treatment state of interest and the control states during this period and then generates a synthetic counte1factual unit for the treatment state that is a weighted combination of the compo- 17 The synthetic control methodology has been deployed in a wide variety of fields, including health economics (Engelen, Nonnemaker, and Shive 2 11). immigration economics (Bohn, Lofstrom, and Raphael 214), political economy (Keele 29), urban economics (Ando 215), the economics of natural resources (Mideksa 213), and the dynamics of economic growth (Cavallo et al. 213). 18 For a more detailed technical description of this method, we direct the reader to Abadie and Gardeazabal (23), Abadie, Diamond, and HainmueUer (21), and Abadie, Diamond, and Hainmueller (214). 2 Li Deel. Ex

165 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:718 nent control states. 19 Two conditions are placed on these weights: they must be non-negative and they must sum to one. n general, the matching process underlying the synthetic control technique uses pre-treatment values of both the outcome variable of interest and other predictors believed to influence this variable. 2 As justified in Appendix H, we use every lag of the dependent variable as predictors in the DAW, BC, LM, and MM specifications. 21 Once the synthetic counte1factual is generated and the weights associated with each control unit are assigned, the synth program then calculates values for the outcome variable associated with this counte1factual and the root mean squared prediction e1rnr (RMSPE) based on differences between the treatment and synthetic control units in the pre-treatment period. The effect of the treatment can then be estimated by comparing the actual values of the dependent vmiable for the treatment unit to the corresponding values of the synthetic control. B. Generating Synthetic Controls for 33 States Adopting RTC Laws During our Data Period To illustrate the procedure outlined above, consider the case of Texas, whose RTC law went into effect on January 1, The potential control group for each treatment state consists of all nine states with no RTC legislation as of the year 214, as well as states that pass RTC laws at least ten years after the passage of the treatment state (e.g., in this case, those states passing RTC laws after 26, such as Nebraska and Kansas, whose RTC laws went into effect at the beginning of 27). Since we estimate results for up to ten years post-passage, 22 this restriction helps us avoid t 9 our analysis is done in Stata using the synth Software package developed by Alberto Abadie, Alexis Diamond, and Jens Hainmueller. 2 Roughly speaking, the algorithm that we use finds W (the weights of the components of the synthetic control) that minimizes j(x1 - Xo W)'V(X1 -Xo W), where Vis a diagonal malrix incorporating information about the relative weights placed on different predictors, Wis a veclor of non-negative weights that sum to one, X 1 is a vector containing pre-treatment information about the predictors associated with the treatment unit, and Xo is a matrix containing pretreatment infonnation about the predictors for all of the control units. For our main analysis, we use the nested option in Stata to generate the relevant weights. This option uses standard optimization techniques to find the weights associated with each predictor that minimize the pre-treatment RMSPE of the resulting synthetic control. The Stata module that we use also can generate the relevant weights using a less computationally expensive regression-based technique. Owing to computational constraints, we use this approach in our placebo analysis. 21 We considered using one lag, three lags, and yearly lags as predictors and we eventually chose to use yearly pre-treatment crime rates since that option minimized the average coefficient of variation of the RMSPE during the validatio n period. t is worth noting that the estimated treatment effect associated with the passage of a state-level RTC law remains similar for violent crime regardless of whether one, three, or every possible lag is included along with the DAW, BC, LM, and MM predictors (or whether these lags are excluded from the list of predictors entirely). 22 ur choice of ten years in this context is informed by the tradeoffs associated with using a different time frame. Using a longer post-passage period would enable us to estimate the impact of RTC laws for states in which there were more than ten years of post-passage data, but it would likely reduce the accuracy of our estimates of the effect of the treatment in earlier periods. This degradation would occur owing to the exclusion of additional control states from consideration in the composition of our synthetic control, which would tend to reduce the quality of our synthetic 21 Li Deel. Ex

166 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:719 including states with their own permissive concealed cai1-y laws in the synthetically constructed unit. After entering the necessary specification information into the synth program (e.g., treatment unit, list of control states, explanatory variables, etc.), the algorithm proceeds to construct the synthetic unit from the list of control states specific to Texas and generates values of the dependent variable for the counterfactual for both the pre-treatment and post-treatment periods. The rationale behind this methodology is that a close fit in these time series of crime between the treatment state and synthetic control in the pre-passage period generates greater confidence in the accuracy of the constructed counte1factual. Computing the post-treatment difference between the dependent variables of the treatment state and the synthetic control unit provides the synthetic controls estimate of the treatment effect attributable to RTC adoption in that state. 1. Synthetic Controls Estimates of Violent Crime in Four States Figure 2 shows the synthetic controls graph for violent crime in Texas over the period from 1977 through 26 (ten years after the adoption of Texas's RTC law). The solid black line shows the actual pattern of violent crime for Texas, and the vertical line indicates when the RTC law went into effect. mplementing the synthetic control protocol identifies three states that generate a good fit for the pattern of crime experienced by Texas in the pre-1996 period. These states are California, which gets a weight of 57.8 percent owing to its similai attributes to Texas, Nebraska with a weight of 8.6 percent, and Wisconsin with a weight of 33.6 percent. One of the advantages of the synthetic controls methodology is that one can assess how well the synthetic control (call it "synthetic Texas," which is identified in Figure 2 by the dashed line) matches the pre-rtc-passage pattern of violent crime to see whether the methodology is likely to generate a good fit in the ten years of post-passage data. Here the fit looks rather good in mimicking the rises and falls in Texas violent crime from This pattern increases our confidence that synthetic Texas will provide a good prediction of what would have happened in Texas had it not adopted a RTC law. Another advantage of the synthetic controls protocol is that one can consider the attributes of the three states that make up synthetic Texas to see if they plausibly match the features that generate c1ime rates in states across the country. Looking at Figure 2, we see that while both Texas and synthetic Texas (the weighted average violent crime performance of the three mentioned states) show declining crime rates in the postpassage decade after 1996, the crime drop is substantially greater in synthetic Texas, which had no RTC law over that period, than in actual Texas, which did. As Figure 2 notes, ten years after control estimates for the earlier portion of the post-treatment period. Using a shorter post-passage period risks failing to capture effects of RTC laws that take a decade to unfold. 22 Li Deel. Ex

167 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:72 adopting its RTC law, violent crime in Texas was 16.6 percent higher than we would have expected had it not adopted a RTC law. 23 Figure 2 also illustrates perhaps the most important lesson of causal inference: one cannot simply look before and after an event to determine the consequence of the event. Rather, one needs to estimate the difference between what did unfold and the counterfactual of what would have unfolded without the event. The value of the synthetic controls methodology is that it provides a highly transparent estimate of that counterfactual. Thus, when Lott (21) quotes a Texas District Attorney suggesting that he had reversed his earlier opposition to the state's RTC law in light of the perceived favorable experience with the law, we see why it can be quite easy to draw the inaccurate causal inference that Texas' crime decline was facilitated by its RTC law. The public may perceive the falling crime rate post-1996 (the solid black line) but our analysis suggests that Texas would have expe1ienced a more sizable violent crime decline if it had not passed a RTC law (the dotted line). More specifically, Texas experienced a 19.7 percent decrease in its aggregate violent crime rate in the ten years following its RTC law (between 1996 and 26), while the state's synthetic control experienced a larger 3.8 percent decline. This counterfactual would not be apparent to residents of the state or to law enforcement officials, but our results suggest that Texas's RTC law imposed a large social cost on the state. The greater transparency of the synthetic controls approach is one advantage of this methodology over the panel data models that we considered above. Figure 2 makes clear what Texas is being compared to, and we can reflect on whether this match is plausible and whether anything other than RTC laws changed in these three states during the post-passage decade that might compromise the validity of the synthetic controls estimate of the impact of RTC laws. Specifically, if one agreed with some of John Lott's written work that the death penalty is a powerful deten-ent one might be concerned that Texas's far greater use of the death penalty during the post-passage period than in the states comprising synthetic Texas might undermine the prediction that RTC laws increased c1ime by 16.6 percent in Texas. 24 But the death penalty, 23 Texas' violent crime rate ten years post-adoption exceeds that of "synthetic Texas" by 51 ~3ri 3 x 1 = 2.2%. Figure 2 shows the estimated violent crime increase in Texas of 16.6 percent, which comes from subtracting from 2.2 percent, the amount by which Texas' violent crime rate exceeded that of synthetic Texas in 1996 = 6 ~ 2 ~ 21 x 1 = 3.7%. (See footnote 31 for further discussion of this calculation.) 24 Texas executed 275 convicts during the post-passage decade while California executed 11, Nebraska 2, and Wisconsin executed no one (Death Penalty nformation Center 215). Other key explanatory variables might be the incarceration and police rates. Because Texas had an enom1us jump in its incarceration rate from , the growth in the Texas incarceration rate from was only 7.6 percent, while for "synthetic Texas" the growth rate was 22. percent. ndividually, the growths for the three synthetic control states were 6.8 percent (CA), 69. percent (W), 26.6 percent (NE). The growth rate in the Texas police employment rate over the decade was -.6 percent, while for "synthetic Texas" the growth rate was 7.8 percent. ndividually, the growths for the three synthetic control states were 9. percent (CA), 5.5 percent (W), 8.2 percent (NE). Using plausible crime elasticities for police and prison suggests that accounting for these two factors could conceivably shrink the estimated impact on violent crime by 3 percent. Even th is would suggest that the Texas RTC law increased violent crime in the tenth year by 23 Li Deel. Ex

168 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:721 Figure 2 g "' - Treated Unit - - Synl/1etic Control Unit Texas: Violent Crime Rate Effect of 1996 RTC Law 1 Years After Ado lion: 16.6% al 2 C: a, "C ;; Q) D "'..._ "' ~ :;;.. ~ "'., E "' 8 c J!! 5 "',.,.,, ' ' \,: ~,, : \ l Note: DAW Variables and yearly lags are used as predictors Composition of SC: CA (.578); NE (.86); W (.336) CVRMSPE:.6 (1 of 33 states, where 1 denotes the state with the best pre-passage fit. ) Slates Never Passing RTC Laws ncluded in Synthetic Control: CA RTC Adopting States ncluded in Synthetic Control: NE (27); W1 (212) ' ' ' 26 according to Lott, depresses crime, so to the extent the death penalty played a greater role in Texas than in synthetic Texas during the post-passage period (relative to the pre-passage period), then our estimate of the increase in violent crime generated by the RTC law would actually understate the true increase. Figure 3 shows our synthetic controls estimate for Pennsylvania, which adopted a RTC law in 1989 that did not extend to Philadelphia until a subsequent law went into effect on October 11, n this case, synthetic Pennsylvania is comprised of eight states and the pre-passage fit is nearly perfect. Following adoption of the RTC laws, synthetic Pennsylvania shows substantially better performance than actual Pennsylvania after the RTC law is extended to Philadelphia in late 1995, as illustrated by the second vertical line at The synthetic controls method estimates roughly 1 percent. As we noted in Table 2, the overall increases in incarceration and police rates were greater in RTC stales than in non-rtc states, which would tend to make our estimates of the increased violence in RTC states conservative. 25n our synthetic controls approach, we treat the year of passage to be the first year in which a RTC law was in effect for the majority of that year. Accordingly, we mark Philadelphia's law passage in 1996, as documented in 24 Li Deel. Ex

169 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:722 that RTC laws in Pennsylvania increased its violent crime rate by 26.5 percent. Treated Unit Synthetic Control Unit Figure 3 Pennsylvania: Violent Crime Rate Effect of 1989 RTC Law 1 Years After Ad tion: 26.5% "' a.. '".!! er 8 (.) V 'E _gi 5 g "',,, J '... _-! ~':\ : : : \ _,, '4 '"" \ : \ Note: DAW Variables and yearly lags are used as predictors 1995 Composition of SC: DE (.77); H (. 16); NE (.43); NJ (.17); OH (.256); W1 (.348) CVRMSPE:.1 8 (1 of 33 states, where 1 denotes the state with the best pre-passage fil) Slates Never Passing RTC Laws ncluded in Synthetic Control: DE : Ht ; NJ RTC Adopting States ncluded in Synthetic Control: NE (27); OH (24); W (212) \ : \ ' \ \ : \ ' \ \ ' ' 1999 Figures 4 and 5 show the comparable synthetic controls matches for North Carolina and Mississippi. Again both states show good pre-passage fit between the violent crime rates of the treatment state and the synthetic control. The methodology estimates that RTC laws led to an increase in violent crime in North Carolina of 18.3 percent and in Mississippi of 34.2 percent State-Specific Estimates Across all RTC States Because we are projecting the violent crime experience of the synthetic control over a ten-year pe1iod, there will undoubtedly be a deviation from the "true" counterfactual and our estimated Appendix B.. 26 n Appendix F we include all 33 graphs showing the path of violent crime for the treatment states and the synthetic controls, along with infonnation about the composition of these synthetic controls, the dates of RTC adoption (jf any) for states included in these synthetic controls. and the estimated treatment effect (expressed in terms of the percent change in a par tic ular crime rate) ten years after adoption (or 7 years after adoption for 2 states that adopt RTC laws in 27, since our data ends in 214). 25 Li Deel. Ex

170 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 2817 of 47 of 292 Page D #:723 Figure 4 North Carolina: Violent Crime Rate Effect of 1996 RTC Law 1 Years After Ado lion: 18.3% - Treated Unit - - Synthetic Control Unit ',-.. c "'."S'. "' a: "' 8 CD "' 8 ~ ll.!!! a: "' > E ;'.i c Q) "' 5 8 V /, ' '! ' l ' : ' ' ~--- i ' --~ Note: DAW Variables and yearly lags are used as predictors Composition of SC: DE (.92); L (.396); NE (.512) CVRMSPE:.49 (4 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: DE" RTC Adopting States ncluded in Synthetic Control: L (214); NE (27) 26 counterfactual. One of the advantages of our task is that we have a large number of states adopting RTC laws so that the over-estimates and under-estimates will tend to wash out in our mean treatment estimates. Figure 6 shows the synthetic control estimates on violent crime for all 31 states for which we have ten years of post-passage data. For 23 of the 31 states adopting RTC laws, the increase in violent crime is noteworthy. While three states were estimated to have crime reductions greater than the -1.6 percent estimate of South Dakota, if one averages across all 31 states, the (population-weighted) mean treatment effect after ten years is a 15.l percent increase in violent crime. f one instead uses an (unweighted) median measure of central tendency, RTC laws are seen to increase crime by 14.1 percent. 3. Less Effective Pre-Passage Matches Section 1 above provided four examples in which the synthetic controls approach generated synthetic controls that matched the crime of the treatment states well in the pre-passage period, but this 26 Li Deel. Ex

171 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:724 Figure 5 Mississippi: Violent Crime Rate "' - Treated Unit -- Synthetic Control Unit "' "' ~ Q) - ;;; "' Q) "' U) "' ~... :;;.. Q).; "' Q) E....:: u c <) 9 > "' 8 "' "' N,,., ~ ' '' ' ' ' ' ' ',, ' ; \ ' , \ ~-, \ Nole: DAW Variables and yearly lags are used as predictors Composition of SC: H (.721 ): A (.16); NE (.11 ); OH (.252) CVRMSPE:.5 (6 of 33 states, where 1 denotes lhe stale with the besl pre-passage fil.) Stales Never Passing RTC Laws ncluded in Synthetic Control: Ht RTC Adopting Stales ncluded in Synlhelic Control: A (211 ); NE (27): OH (24) 2 does not always happen. Again, one advantage of the synthetic controls approach is that one can assess the nature of this fit in the pre-passage period in order to determine how much confidence one can have in the post-passage prediction. Two states for which we would have considerably less confidence in the qual ity of the synthetic controls estimate are South Dakota and M aine, both of which happen to show declines in crime after RTC adoption. ndeed, these are two of the eight states showing improvements in crime following RTC adoption as indicated in Figure 6. An examination of Figures 7 and 8 showing the synthetic controls estimates for these two states provides dramatic visual confirmation that the methodology has fai led to provide a good pre-passage fit between the crime peliormance of the treatment states and a suitable synthetic control. For South Dakota, one sees that the synthetic control and the state violent crime performance diverged long before RTC adoption in 1985, and that, by the date of adoption, synthetic South Dakota had a far higher violent crime rate that was rising while actual South Dakota had a violent crime rate that was falling in A similar pattern can be seen for Maine, which again 27 Li Deel. Ex

172 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 3172 of 47 of 292 Page D #:725 Figure 6 The Effect of RTC Laws on Violent Crime after 1 Years Synthetic Control Estimates for 31 States ( ) Q) E 8 i:: Q) a >.!: Q) Cl C ex, ) "' -q-.c u N '$. N Year of RTC Passage D (5) ii 9-94 (6) a (14) a -4 (6) SC NV AR npa 1 TN MSn Mean= ""MnAKNMLAWY~~ ,...,.-,,--.,..,--,= M:: ~: ~ _.. rn iii i D VA socoohmnor DME UT WV MT undermines confi dence in the synthetic controls estimates for these two states. The difficulty in generating good pre-passage matches for South Dakota and Maine sterns from their unusually low violent c1ime in the pre-passage period. Figure 9 reproduces Figure 6 while leaving out the four states for which the quality of prepassage fit is clearly lower than in the remaining 27 states. 27 This knocks out ND, SD, MT, and WV, leaving a slightly lower estimated mean but the same median effect of RTC laws. As Figure 9 shows, the (weighted) mean increase in crime across the listed 27 RTC-adopting states is 11.3 percent while the (unweighted) median increase is 14.1 percent. ncreases in violent crime of this magnitude are troubling. Consensus estimates of the elasticity of crime with respect to incarceration hover around.15 today, which suggests that to offset the increase in crime caused by RTC adoption, the median RTC state would need to approximately double its prison population. 27 n particular, for these four states, the pre-passage CVRMSPE-that is, the RMS PE transformed into a coefficient of variation by dividing by average pre-passage crime- was significantly greater than for the other 27. See Footnote 33 for further discussion of this statistic. 28 Li Deel. Ex

173 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:726 Figure 7.., 8 South Dakota: Violent Crime Rate Effect of 1985 RTC Law 1 Years After Ado lion: -1.6% - Treated Unit - - Synthetic Control Unit tl C Q) ~ Q) o::'.!,,:'. "' ~ ~.. ~ o::'. N "' Q) E 8 c.91 5 N / 1,',,"'.,,,"'l. ",,. ',.,,, l, ~ ' ~ Note: DAW Variables and yearly lags are used as predictors Composition of SC: A (.625); W (.375) CVRMSPE:.436 (32 of 33 stales, where 1 denotes the state with the best pre-passage fit.) Stales Never Passing RTC Laws ncluded in Synthetic Control: RTC Adopting States ncluded in Synthetic Control: A (211 ); W (212) Aggregation Analysis Using Synthetic Controls A small but growing literature applies synthetic control techniques to the analysis of multiple treatments. 28 We estimate the percentage difference in violent crime between each b eatment (RTCadopting) state and the corresponding synthetic control in both the year of the treatment and in the ten years following it (we obviously use data from fewer post-treatment years for the two treatment 28 The closest paper lo the present study is Arindrajit Dube and Ben Zipperer (2 13), who introduce their own methodology for aggregating mulliple events into a single estimated treatment effect an<l calculating its significance. Their study centers on the effect of increases in the minimum wage on employment outcomes, and, as we do, the authors estimate the percentage difference between the treatment and the synthetic control in the post-treatment period. While some papers analyze multiple treatments by aggregating the areas affected by these treatments into a single unit, this approach is not well-equipped to deal with a case such as RTC law adoption where treatments affect the majority of panel units and more than two decades separate the dates of the first and last treatment under consideration, as highlighted in Figure Li Deel. Ex

174 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:727 ~ - Treated Unit - - Synthetic Control Unit Figure 8 Maine: Violent Crime Rate Effect of 1966 RTC Law 1 Years After Ado lion: -16.5% V c "' "O ;;; " a:: " " 8 a, a. ~ a:: "' " E c: (..) c " 5 "' M M :il "' "' ~ /,.,, (," / /,,, / : / ' _,,.,, :, ' / ',! l---',,',/,,..,,--', ', ~..,_, \ \ \ \ \,1 ~ Nole: DAW Variables and yearly lags are used as predictors Composition of SC: H (.16); A (.84) CVRMSPE:. 196 (28 of 33 states. where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: H RTC Adopting States ncluded in Synthetic Control: A (211 ) 1996 states 29 that bad RTC laws that took effect less than ten years before the end of our sample). 3 We could use each of these ten percentage differences as our estimated effects of RTC laws on violent crime for the ten post-passage years, but we make one adjustment to these figures by subtracting from each the percentage difference in violent crime in the adoption year between the treatment and synthetic control states. n other words, if ten years after adopting a RTC law, the violent crime rate for the state was 44 and the violent crime rate for the synthetic control was 4, one estimate of the effect of the RTC law could be 1 percent(= 44 ~ g ). Rather than use this estimate, however, we would subtract from this figure the percentage difference between the synthetic and treatment states in the year of RTC adoption. f, say, that value were 2 percent, we 29 These two states are Kansas and Nebraska, which adopted RTC laws in 27. See footnote 4 discussing the states for which we cannot estimate the impact of RTC laws using synthetic controls. 3 Dube and Zipperer (213) average the estimated post-treatment percentage differences and convert this average into an elasticity. Thus, our work reports separate average treatment effects for ten yearly intervals following the time of the treatment, while Dube and Zipperer (213) emphasize an average treatment effect (expressed as an elasticity) estimated over the entire post-treatment pe,iod. 3 Li Deel. Ex

175 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:728 Figure 9 The Effect of RTC Laws on Violent Crime after 1 Years Synthetic Control Estimates for 27 States ( ) M Year of RTC Passage D (3) l!.' 9-94 (4) (14) -4 (6) MS FL Q) N E c'.5 c Q) ~ 5 J; Q) Ol C ro,r:. u cf. ' N Median= 14.7 Mean= ME UT - rn< would subtract 2 from 1 to obtain an estimated effect of RTC laws in year 1 of 8 percent. 3 L We then aggregate all the state-specific estimates of the the impact of RTC laws on violent crime and test whether they are significantly different from zero. 32 As we saw in Figures 2-5 and 7-8, the validity of using the post-treatment difference between crime rates in the treatment state (the particular state adopting a RTC law that we are analyzing) and its cotesponding synthetic control as a measure of the effect of the treatment depends on the 31 Both approaches should generate similar estimates on average, and in fact our mean estimated effects are more conservative with our preferred approach. The intuitive rationale for our choice of outcome variable was that pretreatment differences between the treatment state and its synthetic control at the time of RTC adoption likely reflected impetfections in the process of generating a synthetic control and should not contribute to our estimated treatment effect if possible. n other words, if the treatment state had a crime rate that was 5 percent greater than that of the synthetic control in both the pre-treatment and post-treatment period, it would arguably be misleading to ignore the pre-treatment difference and declare that the treatment increased crime rates by 5 percent. For violent crime, the mean (median) percentage difference between the treatment state and the synthetic control in the year of the treatment was 4.9 percent (1.3 percent), with 18 treatment states showing greater crime rates than their synthetic controls and 15 treatment states showing less crime than their synthetic control in that year. As a result, our estimate of the amount by which RTC laws increased violent crime was lower than would have been the case under the alternative appraoch. 32 This test is performed by regressing these differences in a model using only a constant term and examining whether that constant is statistically significant. These regressions are weighted by the population of the treatment state in the post-treatment year under consideration. Robust standard errors corrected for heteroskedasticity are used in this analysis. 31 Li Deel. Ex

176 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:729 strength of the match between these two time series in the pre-treatment period. To generate an estimate of pre-treatment fit that takes into account differences in pre-treatment crime levels, we estimate the coefficient of variation for the root mean squared prediction error (RMSPE), which is the ratio of the synthetic control's pre-treatment RMSPE to the pre-treatment average level of the outcome variable for the treatment state. 33 After generating the aggregate synthetic controls estimates of the crime impact of RTC laws described in the paragraph above using the full sample, we consider two subsamples of treatment states: states whose coefficients of variation are less than two times the average coefficient of variation for all thirty-three treatments and states whose coefficients of variation are less than thi s average. We then re-run our synthetic controls protocol using each of these two subsamples to examine whether restricting our estimation of the average treatment effect to states for which a relatively "better" synthetic control could be identified would meaningfully change our findings. A. RTC Laws ncrease Violent Crime We now turn our attention to the aggregated results of our synthetic control analysis using predictors derived from the DAW specification. Table 9 shows our results on the full sample examining violent crime. 34 Our estimates suggest that states that passed RTC laws experienced more deleterious changes in violent criminal activity than their synthetic controls in the ten years after adoption. On average, treatment states had aggregate violent crime rates that were around 7 percent higher than their synthetic controls five years after passage and almost 15 percent hjgher ten years after passage. Table 9 suggests that the longer the RTC law is in effect (up to the tenth year that we analyze), the greater the cost in terms of increased violent crime. Table 1 repeats the Table 9 analysis while dropping the four states with a CV of the RMSPE that is above twice the average of the sample. Table 11 uses a more stringent measure of assessing 33 While the RMSPE is often used to assess thjs fi t, we believe that the use of this measure is not ideal in the present context owing to the wide variation that exists in the average pre-treatment crime rates among the 33 treatment states that we consider. For example, the pre-treatment RMPSE associated with our synthetic control analysis using the DAW predictor variables and aggregate violent clime as the outcome variable is simjlar for Colorado (36.1) and Maine (36.4), but the pre-treatment levels of Colorado's aggregate violent clime rate are far greater than Maine's. To be more specific, Colorado's average violent crime rate prior to the implementation of its RTC law (from 1977 through 22) was.+67 violent crimes per 1, residents, while the con-esponding figure for Maine was 186 violent crimes per, residents. For this reason, we have greater confidence in our estimates that in the tenth year, Colorado's RTC law had decreased violent crime by 1.23 percent than we do in an estimate that Mmne's law had decreased violent crime by 16.5 percent, s ince the percentage imprecision in our synthetic pre-treatment match for Maine is so much greater than for Colorado. 34 We discuss the synthetic controls estimates for murder and prope11y crime in section G below. n all cases, the tenth-year effect is always positive-suggesting that RTC laws increase crime-but not statistically significant. The point estimates across the four models suggest RTC laws increase the murder rate by 5 to 6 percent, and the property crime rate by l -3 percent after ten years. 32 Li Deel. Ex

177 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:73 Table 9: The mpact of RTC Laws on the Violent Crime Rate, DAW covariates, () (2) t3) (4) (5) (6) (7) (8) (9) [ D) Average. Normaliud TEP '" 3,676" " 6.989'"" ''" J.224''" 14.44Y ( ) ( J3 1) ( 1.862) (.986) (2.524) (3.57) (2.739) (J.752) (3.577) (2.957) N Pseudo P-Vulue ()24.32 Proporti on of Corresponding Placebo Estimntes Significant at. O Level HS Proportlon ofcorrc:ipnding Placebo EstimJ.tcs Signific:rnt at.5 Level ,12 Proportion of Corresponding Placebo Estimates Significant at. Levd DJS S1nndard errors in p;u1!nthcses Column numbers inditate po.st-passage ye.tr under considcrmion: N = number uf slates in s ample Dependenl variable is Ule difference ~ t\!jeen the percentage difference in the violent crime rnte. rn lreatmenl and synthetic conlfol states at given po!il-lre:umenl inierval untl at Lime of the 1re31mem Rts\lhS reported fonhe conscfilll 1em1 resuhing fro1 lhi i; rt!gression States in gmup: AK AR AZ CO FL GA D KS KY LA ME Ml MN MO MS MTNC NO NE NM NV OH OK OR PA SC SD TN TX UT VA WV WY The sy111hc1ic conuol.s used to gener.n.t the placebo estim:i.tes in th.: 1.able ubow were generated using lhe regression methodology described in the mnin 1exi. 'p < O.O,.. p <.5,... p <.1 how well the synthetic control fits the pre-passage data by dropping the six states with an above average CV for the RMSPE. t is striking how all three tables show roughly identical conclusions: RTC laws are consistently shown to increase violent c1ime starting three years after passage. Table 1: The mpact of RTC Laws on the Violent Crime Rate, DAW covariates, < 2x Average Coefficient of Variation of the RMSPE, () (2) (3) (4) (5) (6) (7) (8) (9) (O) Avernge Nomrnlitt:d TEP ' 3.737" 4.91'" ] """ ,853""' i l.148) ( 1.354) ( 1.98) (2.38) (2.574) (3.151) (2.823) (3.874) (3.67) (2.999) N Pseudo P-Value Proponion of Correspondin!:? Pl!lcebo Estim:lles Significant at. O Level g Proponion of Corresponding Pl:iceho Esunuues Significant at.5 Level J4.144 Proponion of Corresponding Pl;1cebo Estim,ues Significant at. Level Standard errors in p:irr:ntheses Column numbers ndicate post-passage year under consideration: N = number of st31cs in sumpk Dependent vana.ble: i5 the ditcrencc bc1wce.n the pen.:cntuge difference in the violent crime r:.11e in lfc:3tmenl and synthetic control 5t:ites at gi\ Cn pushreauneni interval and :11 time of the treatment Resullli reported for Lllc cons1:mt term re!.ulling from this rtgre:.sinn Stote.s in iroup: AK AR AZ CO FL GA fd KS KY LA ME Ml MN MO MS NC NE NM NV OH O K OR PA SC TN l 'X UT VA WY Staics excluded for poor pre-1rcatmcnt ti1: MT ND SD WV llte synlhetic con1rol1, used to generate the placebo estimates in the Lnble above wen: iener:1tcd using ihc n:gre_ssion methodology described in the main ten. p < O.O. p <.5... p <.1 B. The Placebo Analysis Our ability to make valid inferences from our synthetic control estimates depends on the accuracy of our standard e1rnr estimation. To test the robustness of the standard en-ors that we present under the first row of Tables 9-11 and to get a sense of whether the coefficients that we measure are qualitatively large compared to those that would be produced by chance, we incorporate an analysis using placebo treatment effects similar to Ando (215). 35 For this analysis, we generate 5 sets 35 Ando (215) examines the impact of constructing nuclear plants on local real per capita taxable income in Japan by generating a synthetic control for every coastal municipality that installed a nuclear plant. While the average treatment effect measured in our paper differs from the one used by Ando, we follow Ando in repeatedly estimating 33 Li Deel. Ex

178 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:731 Table 11: The mpact of RTC Laws on the Violent Crime Rate, DAW covariates, < lx Average Coefficient of Variation of the RMSPE, ( ) (2) (3) (4) (5) (6) (7) (S) (9) (O) Av~r.igc- Normalized TEP " 3.942" 5.192"" 7_ J ' "" ( ) (1.379) (1.934) (2.59) (2.62) (3.191) (2.848) (3.876) (J.667) (3.2) N Pseudo P~ Vuluc U.3.3 Proponion orcorre!-ponding Placebo Estimates Significant at.1 Level Proportion of Corresponding Plai:ebo Eiairnates Significant at.5 Level Proponinn of Corresponding. Placebo Eslimatcs Significam al.1 Level b R.6.56 Standard errors in paremhe.,;es C olumn numbc~ imlicole JXlSl-passage yea.r under co11sidcr..1tion; N:::; numher or st.jlcs in sample Dcpe1u.h:nt vamible is the: difference bc:1wccn lhc: ~rcemage difference in lhe, iulent trim c= r..1te in o-t;ume nl,m<l synthetic control s1;th$ at gh cn post-trentment intcrv11 and at time or 1c trl! Hmem ResulL'i repon.:d for tj1e c:onsurnt tem1 resulting from tl1is rcgn:ssmn States in group: AK AR AZ CO FL GA t KS KY LA Ml MN MO MS NC NM NV OH OK OR PA SC TN TX UT YA WY States c:tcludc<. for poor pre-t.rl'.'aunc:m lit ME MT NO ~E.SD WV 111c: synlhet.ic cqntrols usc"d to geni:rute the placebo estimates in the tahle abo\'c were generated using lhc rc:grcss1on methodology dc.!icri bcd in the main tc':tt p <.1.. p <.5... p <.1 of randomly generated RTC dates that are designed to resemble the distribution of actual RTC passage dates that we use in our analysis. 36 For each of the 5 sets of randomly generated RTC dates, we then use the synthetic control methodology and the DAW predictors to estimate thirtythree synthetic controls for each state whose randomly generated adoption year is between 1981 and 21. We use this data to estimate the percentage difference between each placebo treatment and its COTesponding synthetic control during both the year of the treatment and each of the ten post-treatment years (for which we have data) that follow it. We then test whether the estimated treatment effect for each post-treatment year is statistically significant (using the methodology described in footnotes 23 and 31 ). We also repeat our estimation of the average treatment effect associated with each of the ten post-treatment years after excluding states whose coefficient of variation is either one or two times the average observed for all (placebo) treatment states, leaving us with 3 coefficients and p-values corresponding to each of the 5 sets of randomly generated placebo treatments that we consider. At the bottom of Table 9, we list the proportion of each post-treatment year's placebo regressions that were "significant" at the.1 level,.5 level, and.1 level. We provide these proportions to give the reader an intuitive sense of the possible bias associated with our standard error estiaverage placebo effects by randomly selecting different areas to serve as placebo treatments. (The sheer number of treatments that we are considering in this analysis prevents us from limiting our placebo treatment analysis to states that never adopt RTC laws, but this simply means that our placebo estimates will likely be biased against finding a qualitatively significant effect of RTC laws on crime, since some of our placebo treatments will be capturing the effect of the passage of RTC laws on crime rates.) The actual average treatment effect can then be compared to the disttibution of average placebo treatment effects. Heersink and Peterson (214) also perform a similar randomization procedure to estimate the significance of their estimated average treatment effect. Cavallo et al. (213) perfonn a similar test to examine how the average of different placebo effects compares to the average treatment effect that they measure using synthetic control techniques, although their randomization procedure differs from ours by restricting the timing of placebo treatments to the exact dates when actual tt eatments took place. 36 More specifically, we randomly choose eight states to never pass RTC laws, six states to pass RTC laws before 1981, 33 states to pass RTC laws between 1981 and 21, and three states to pass their RTC laws between 211 and 214. (Washington, D.C. is not included in the placebo analysis since it is excluded from our main analysis.) These figures were chosen to mirror the number of states in each of these categories in our actual data se t. 34 Li Deel. Ex

179 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:732 mation, although (for the reason noted in footnote 35) it is likely that these placebo estimates are capturing some of the effect of RTC laws. Table 9 shows that the placebo results appear to be significant at the.1 level 2.2 percent of the time for our first year after passage to 5.8 percent in the tenth year. n other words, the standard errors we repmt at the top of Table 9 are potentially underestimated, as our placebo averages are statistically significant more often than would be expected by chance. 37 As another check on the statistical significance of our results, we compare each of the ten coefficient estimates in Table 9 with the distribution of the 5 average placebo treatment effects that use the same crime rate, post-treatment year, and sample as the given estimate. To assist in this comparison process, we repmt a pseudo p-value which is equal to the proportion of our placebo treatment effects whose absolute value is greater than the absolute value of the given estimated treatment effect. This pseudo p-value provides another intuitive measure of whether our estimated average treatment effects are qualitatively large compared to the distribution of placebo effects. Our confidence that the treatment effect that we are measuring for RTC laws is real increases if our estimated treatment effect is significantly greater than the vast majority of our estimated average placebo treatment effects. 38 Examining our pseudo p-values in Tables 9-11, we see that our violent crime results are always statistically significant in comparison to the distribution of placebo coefficients at the.5 level after seven years or more have passed since the treatment date. V. Conclusion The extensive array of panel data and synthetic controls estimates of the impact of RTC laws that we present uniformly undermine the "More Guns, Less Crime" hypothesis. There is not even the slightest hint in the data that RTC laws reduce violent crime. ndeed, the weight of the evidence 37 n general, the difference between the proportion of placebo results significant at a given level and the significance level itself varies across crime rates and treatment selection crite1ia. We do not observe any consistent tendency for the significance levels and proportion of placebo results s ignificant at those levels lo converge when restricting the sample to states with a relatively low RMSPE. 38 Because of the computational demands required to perform this analysis using the maximum likelihood estimation technique-the nested option that we employed in our main analysis (mentioned in footnote 2)-we perform this placebo analysis using the synth module's default regression-based technique for estimating the weights assigned to each predictor when constructing the synthetic control. This change should bias our estimates against finding a significant effect ofrtc laws on crime. Since the nested option lends lo improve the matching process between states, we would expect larger deviations between our placebo treatments and their synthetic controls when this option is not utilized. This would suggest more dispersion in our estimated placebo treatment effects and a greater likelihood of estimating that our actual treatment effects were not significantly different from the distribution of placebo effects. ndeed, when performing an earlier version of this placebo analysis, we found that our estimated pseudo p-values were conservatively estimated when comparing our non-nested pseudo p-values with those produced using the nested function. 35 Li Deel. Ex

180 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 3818 of 47 of 292 Page D #:733 from the panel data estimates as well as the synthetic controls analysis best supports the view that the adoption of RTC laws substantially raises overall violent crime in the ten years after adoption. n our initial panel data analysis, our prefen-ed DAW specification as well as the BC specification predicted that RTC laws have led to statistically significant and substantial increases in violent crime. When the LM and MM models were approp1iately adjusted, they generated the same findings, but even without adjustment, these models showed RTC laws increased murder significantly. We then supplemented our panel data results using our synthetic control methodology, again using the DAW, BC, LM, and MM specifications. Now the results were urnform: for all four specifications, states that passed RTC laws experienced percent higher aggregate violent clime rates than their synthetic controls after 1 years (results that were significant at either the.5 or. level after five years). The synthetic controls estimates for the impact of RTC laws on murder and property crime were also uniformly positive after ten years (but not statistically significant). f one adjusts the synthetic controls estimates for the increased rates of police and incarceration that follow RTC adoption, the RTC-induced increases in murder become large and statistically significant. The synthetic controls effects that we measure represent meaningful increases in violent crime rates following the adoption of RTC laws, and this conclusion remained unchanged after restricting the set of states considered based on model fit and after considering a large number of robustness checks. While our placebo analysis suggests that the standard errors associated with some of these estimates may have been biased downward, the size of our average estimated treatment effect in comparison to the distribution of placebo effects indicates that the deleterious effects associated with RTC laws that we estimate for aggregate violent crime are qualitatively large compared to those that we would expect to observe by chance. The consistency across different specifications and methodologies of the finding that RTC elevates violent crime enables far stronger conclusions than were possible back when the NRC Report was limited to analyzing data only through 2 with the single tool of panel data evaluation. Nonetheless, estimation using observational data always rests on numerous assumptions, so one must always be alert to potential shortcomings. For example, if states that were expected to experience future increases in clime were more likely to adopt RTC laws, then we might exaggerate the detrimental effect of RTC laws on crime. Given the very limited ability of politicians, pundits, and even academic experts to con-ectly predict crime trends over this period, though, this problem of endogeneity is unlikely to mar our results. Panel data analysis can be susceptible to problems of omitted variable biase, but the synthetic controls approach was designed to better address that concern. The results presented in this paper also help to explain the longstanding discrepancy that has existed between the econometric results suggesting that RTC laws increase clime and the percep- 36 Li Deel. Ex

181 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:734 tion "on the ground" that RTC laws are not associated with a contemporaneous increase in crime rates. The conflict between these findings is resolved when one realizes that since the crime spikes of the late 198s and early 199s, most states experienced large and imp1tant crime decreases, including those adopting RTC laws. However, our analysis suggests that had states avoided adoption of RTC laws, they would have experienced greater drops in violent crime. ndeed, as Figure 1 illustrated, while RTC states have now fallen below their violent crime rates of 1977 by about 9 or 1 percent, the states that did not adopt RTC laws enjoyed violent crime drops from the late 197s of over 4 percent. Finally, while this paper has focused on the statistical estimation of the impact of RTC laws, it is useful to consider the mechanisms by which RTC laws would lead to net increases in violent crime; that is, the statistical evidence shows us that whatever beneficial effects RTC laws have in reducing violence, they are outweighed by greater harmful effects. The most obvious mechanism is that the RTC permit holder may commit a crime that he or she would not have committed without the permit. A number of high profile crimes by RTC permit holders would seem to follow this pattern: George Zimmerman, the popcorn killer at a Florida movie theater who was angry at a father texting a babysitter, and the angry gas station killer (sho.oting a black teen for playing loud rap music) are all individuals who would likely never have killed anyone had they not had an RTC pennit (Trotta 212; Luscombe 214; Robles 214). Of course, aggravated assaults are far more common than murder (albeit far less visible to the public), so the same impulses that generate killings also work to stimulate aggravated assaults (and hence overall violent crime). Some have questioned whether permit holders commit enough crime to substantially elevate violent criminality, citing apparently low rates of official withdrawals from permit holders convicted of crimes. Two points need to be made in response to this claim. First, official withdrawals clearly understate criminality by permit holders. For example, convictions for violent crime are far smaller than acts of violent crime, so many permit holders would never face official withdrawal of their permits even if they committed a violent c1iminal act that would warrant such termination. Moreover, official withdrawals will be unnecessary when the offending permit holder is killed. n the nightmare case for RTC, two Michigan permit holding drivers pulled over to battle over a tailgating dispute in September of 213 and each shot and killed the other. Again, without permits this would likely have not been a double homicide, but note that no official action to terminate permits would ever be recorded in a case like this (Stuart 213). The second critical point is that RTC laws also increase crime by individuals other than permit holders in a variety of ways. First, the culture of gun carrying can promote confrontations. Presumably, George Zimmerman would not have hassled Trayvon Martin if Zimmerman had not had a gun. f Martin had assaulted Zimmerman, the gun permit then could have been viewed as a stimulant to crime ( even if the permit holder was not the ultimate perpetrator). The messages of the gun 37 Li Deel. Ex

182 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 4182 of 47 of 292 Page D #:735 culture can promote fear and anger, which are emotions that can invite more hostile confrontations leading to more violence. This attitude may be reinforced by the adoption of RTC laws. When Philadelphia permit holder Louis Mockewich shot and killed a popular youth football coach (another permit holder can-ying his gun) over a dispute concerning snow shoveling in January 2, the bumper sticker on Mockewich's car had an NRA bumper sticker reading "Armed with P1ide" (Gibbons and Moran 2). f you are an angry young man, with somewhat of a paranoid streak, and you haven't yet been convicted of a crime or adjudicated to be a mental defective, it is likely that the ability to carry a gun will both be more attractive and more likely in a RTC state. That such individuals will, therefore, be more likely to be aggressive once armed and hence more likely to stimulate violence by others should not be surprising. Second, individuals who can-y guns around are a constant source of arming criminals. When Sean Penn obtained a permit to carry a gun, his car was stolen with two guns in the trunk. The car was soon recovered, but the guns were gone (Donohue 23). n July 215 in San Francisco, the theft of a gun from a car in San Francisco led to a killing of a tourist on a city pier that almost certainly would not have occurred if the lawful gun owner had not left it in the car (Ho 215). Just a few months later, a gun stolen from an unlocked car was used in two separate killings in San Francisco in October 215 (Ho and Williams 215). According to the National Crime Victimization Survey, in 213 there were over 66, auto thefts from households. The more guns being carried in vehicles by permit holders, the more criminals will be walking around with the guns taken from the car of some permit holder. Of course, the San Francisco killer did not have a RTC permit; although the owner of the gun used in the killing did (Ho 215). Lost, forgotten, and misplaced guns are another dangerous by-product of RTC laws, as the growing TSA seizures in carry-on luggage attest. 39 Third, as more citizens can-y guns, more criminals will find it increasingly beneficial to carry guns and use them more quickly and more violently to thwart any potential armed resistance. Fourth, the passage of RTC laws normalizes the practice of can-ying guns in a way that may enable criminals to carry guns more readily without prompting a challenge, while making it harder for the police to know who is and who is not allowed to possess guns in public. Having a "designated permit holder" along to take possession of the guns when confronted by police seems to be an attractive benefit for criminal elements acting in concert (Fernandez, Stack, and Blinder 215; Lnthern 215). Fifth, it almost certainly adds to the burden of a police force to have to deal with armed citizens. A policemen trying to give a traffic ticket has far more to fear if the driver is armed. When a gun is found in a car in such a situation, a greater amount of time is needed to ascertain the driver's status as a permit holder. Police may be less enthusiastic about investigating certain suspicious activities given the greater risks that widespread gun can ying poses to them. Police 39 See Williams and Waltrip (24). 38 Li Deel. Ex

183 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:736 resources used to process gun permits could instead be more efficiently used to directly fight crime. All of these factors are a tax on police, and therefore one would expect law enforcement to be less effective on the margin, thereby contributing to crime. ndeed, this may in part explain why RTC states tend to increase the size of their police forces (relative to non-adopting states) after RTC laws are passed. The fact that two different types of statistical data-panel data regression and synthetic controls- with varying strengths and shortcomings and with different model specifications both yield consistent and strongly statistically significant evidence that RTC Jaws increase violent crime constitutes persuasive evidence that any beneficial effects from gun carrying are likely substantially outweighed by the increases in violent crime that these laws stimulate. 4 4 t should be noted that, even with the enormous stock of guns in the U.S., the vast majority of the time that someone is threatened with violent crime no gun will be wielded defensively. A five-year study of such violent victimizations in the United States found that victims failed to defend or to threaten the criminal with a gun 99.2 percent of the time -this in a country with 3 million guns in civilian hands (Planty and Truman 213). 39 Li Deel. Ex

184 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:737 References Abadie, Alberto, Alexis Diamond, and Jens Bainmueller. 21. "Synthetic Control Methods for Comparative Case Studies: Estimating the Effect of California's Tobacco Control Program." Journal of the Am.erican Statistical Association, 15(49): Abadie, Alberto, Alexis Diamond, and Jens Hainmueller "Comparative Politics and the Synthetic Control Method." American Journal of Political Science, 59(2): Abadie, Alberto, and Javier Gardeazabal. 23. "The Economic Costs of Conflict: A Case Study of the Basque Country." American Economic Review, 93(1): Ando, Michihito "Dreams of Urbanization: Quantitative Case Studies on the Local mpacts of Nuclear Power Facilities Using the Synthetic Control Method." Journal of Urban Economics, 85: Aneja, Abbay, John J Donohue, and Alexandria Zhang "The mpact of Right to CaiTy Laws and the NRC Rep1t: The Latest Lessons for the Empirical Evaluation of Law and Policy." American Law and Economics Review, 13(2): Aneja, Abbay, John J. Donohue, and Alexandria Zhang "The mpact of Right to Can-y Laws and the NRC Report: The Latest Lessons for the Empirical Evaluation of Law and Policy." National Bureau of Economic Research Working Paper Ayres, an, and John J Donohue. 23. 'The Latest Misfires in Support of the "More Guns, Less Crime" Hypothesis." Stanford Law Review, 55: Bohn, Sarah, Magnus Lofstrom, and Steven Raphael "Did the 27 Legal Arizona Workers Act Reduce the State's Unauthorized mmigrant Population?" The Review of Economics and Statistics, 96(2): Cavallo, Eduardo, Sebastian Galiani, lan Noy, and Juan Pantano "Catastrophic natural disasters and economic growth." Review of Economics and Statistics, 95(5): Chalfin, Aaron, and Justin Mccrary "The Effect of Police on Crime: New Evidence From U.S. Cities, " National Bureau of Economic Research Working Paper Cunningham, Scott, and Manisha Shah "Decriminalizing ndoor Prostitution: mplications for Sexual Violence and Public Health." Review of Economic Studies. Revise and Resubmit (third round). 4 Li Deel. Ex

185 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:738 Death Penalty nformation Center "Executions by State and Year." Accessed: Donohue, John J. 23. "The Final Bullet in the Body of the More Guns, Less Crime Hypothesis." Criminology and Public Policy, 2(3): Donohue, John J., and Justin Wolfers. 29. "Estimating the mpact of the Death Penalty on Murder." American Law and Economics Review, 11(2): Dube, Arindrajit, and Ben Zipperer "Pooled Synthetic Control Estimates for Recurring Treatments: An Application to Minimum Wage Case Studies." Durlauf, Steven N., Salvado Navarro, and David A. Rivers "Model uncertainty and the effect of shall-issue right-to-carry laws on crime." European Economic Review, 81: Fernandez, Manny, Liam Stack, and Alan Blinder "9 Are Killed in Biker Gang Shootout in Waco." New York Times. Fryer, Roland G, Paul S Heaton, Steven D Levitt, and Kevin M Murphy "Measuring crack cocaine and its impact." Economic nquiry, 51(3): Gibbons, Thomas, and Robert Moran. 2. "Man Shot, Killed in Snow Dispute." Philadelphia nquirer. Heersink, Boris, and Brenton Peterson "Strategic Choices in Election Campaigns: Measuring the Vice-Presidential Home State Advantage with Synthetic Controls." Available at SSRN Ho, Vivian "Gun linked to pier killing stolen from federal ranger." San Francisco Chronicle. Ho, Vivian, and Kale Williams "Gun in 2 killings stolen from unlocked car in Fisherman's Wharf, cops say." San Francisco Chronicle. Kaul, Ashok, Stefan Klobner, Gregor Pfeifer, and Manuel Schieler "Synthetic Control Methods: Never Use All Pre-ntervention Outcomes as Economic Predictors." Keele, Luke. 29. "An observational study of ballot initiatives and state outcomes." Working paper. Lofstrom, Magnus, and Steven Raphael "ncarceration and Crime: Evidence from California's Public Safety Realignment Refo1m." nstitute for the Study of Labor (ZA) ZA Discussion Papers Li Deel. Ex

186 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:739 Lott, John R. 21. More Guns, Less Crime: Understanding Crime and Gun Control Laws. University of Chicago Press. Lott, John R, and David B Mustard "Crime, dete1tence, and right-to-carry concealed handguns." The Journal of Legal Studies, 26( 1): Luscombe, Richard "Florida man accused of killing unarmed teen 'lost it' over loud rap music." The Guardian. Luthern, Ashley "Concealed carry draws opposite views - and a murky middle." Milwaukee Wisconsin Journal Sentinel. Mideksa, Torben K "The economic impact of natural resources." Journal of Environmental Economics and Managenient, 65(2): Moody, Carlisle E, and Thomas B Marvell. 28. "The debate on shall-issue laws." Econ Journal Watch, 5(3): Moody, Carlisle E, Thomas B Marvell, Paul R Zimmerman, and Fasil Alemante "The mpact of Right-to-Carry Laws on Crime: An Exercise in Replication." Review of Economics & Finance, 4: Munasib, Abdul, and Mouhcine Guettabi "Florida Stand Your Ground Law and Crime: Did t Make Fl1idians More Trigger Happy?" Available at SSRN National Research Council. 25. Firearms and Violence: Academies Press. A Critical Review. National Nonnemaker, James, Mark Engelen, and Daniel Shive "Are methamphetamine precursor control laws effective tools to fight the methamphetamine epidemic?" Health economics, 2(5): Pinotti, Paolo "Organized Crime, Violence, and the Quality of Politicians: Evidence from Southern taly.", ed. Philip J. Cook, Stephen Machin, Marie Olivier and Mastrobuoni Giovanni, Chapter 8, MT Press. Planty, Michael, and Jennifer Truman "Firearm Violence, " U.S. Department of Justice Bureau of Justice Statistics BJS Special Report Robles, Frances "Man Killed During Argument Over Texting at Movie Theater." New York Times. 42 Li Deel. Ex

187 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:74 Roeder, Oliver K., Lauren-Brooke Eisen, Julia Bowling, Joseph E. Stiglitz, and nimai E. Chettiar "What Caused the Crime Decline?" Columbia Business School Research Paper No Rudolph, Kara E., Elizabeth A. Stuart, Jon S. Vernick, and Daniel W. Webster "Association Between Connecticut's Permit-to-Purchase Handgun Law and Homicides." American Journal of Public Health, 15(8): e49-e54. Strnad, Jeff. 27. "Should Legal Empiricists Go Bayesian?" American Law and Econoniics Review, 9(1): Stuart, Hunter "2 Concealed Carry Holders Kill Each Other n Road Rage ncident." Hujfington Post. Trotta, Daniel "Trayvon Maitin: Before the world heard the cries." Reuters. Williams, Clois., and Steven Waltrip. 24. Aircrew Security: A Practical Guide. New York, NY: Ashgate Publishing. Zimmerman, Paul R "The deterrence of crime through private security efforts: Theory and evidence." nternational Review of Law and Econoniics, 37: Li Deel. Ex

188 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:741 Right-to-Carry Laws and Violent Crime: A Comprehensive Assessment Using Panel Data and a State-Level Synthetic Controls Analysis By John J. Donohue, Abhay Aneja, and Kyle D. Weber Appendix Appendix A: RTC Adoption Dates An RTC adoption year of O indicates that a state did not adopt a right-to-cany (RTC) law between 1977 and the early months of 214. f the fraction of year in effect is less than.5, the RTC date used in the synthetic control analysis is the following year. RTC dates before the year 1977 may not be exact, since differences between these dates would neither affect our regression results nor our synthetic control tables. For example, we only read Vermont's statutes up to the year 197 to confirm there were no references to blanket prohibitions on carrying concealed weapons up to the year 197, although it appears given widespread public commentary on this point that Vermont never had a comprehensive prohibition of the carrying of concealed weapons. We follow earlier convention in the academic literature on the RTC issue in assigning RTC adoption dates for Alabama and Connecticut. 44 Li Deel. Ex

189 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 47 of 292 Page D #:742 Table Al: RTC Adoption Dates State Effective Date of RTC Law Fraction of Year n Effect Year of Passage RTC (Date in Synthetic Controls Analysis) Alabama A laska Arizona 7/17/ Arkansas 7/ Cali fornia NA Colorado 5/ Connecticut Delaware NA District of Columbia NA Florida 111/ Georgia 8125/ Hawaii NA ldaho 711/ llinois 1/5/ ndiana 1115/ owa 1/ Kansas 111/ Kentucky 111/ Louisiana 4/19/ Maine 9119/ Maryland NA Massachusetts NA Michigan 71 1/ Minnesota Mississippi 7/ Missouri Montana 1/1/ Nebraska 1/ Nevada 111/ New Hampshire New Jersey NA New Mexico 1/1124 l. 24 New York NA North Carolina North Dakota 8/! Ohio 418/ Oklahoma 1/l/ Oregon 1/1/ Pennsylvania Philadelphia Rhode sland NA South Carolina South Dakota 7/1/ Tennessee 111/ Texas 1/ l Utah 5/l/l Vem1ont Vrrginia 515/ Washington West Virginia 7/7/ Wisconsin 1111/ Wyoming 1/1/ Li Deel. Ex

190 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 1 of of Page 292D #:743 Appendix B: Complete Regression Output Table A2: Panel Data Violent Crime Coefficients using DAW, BC, LM, and MM models, State and Year Fixed Effects Pa11el A: Dummy Variable Model Results (Table 4) (Table 5.A) (Table 6.A) (Table 7.A) DAW Model BC Model LM Model MM Model (!) (2) (3) (4) Right-to-Carry Law 9.48 '" (2.96) (3.65) (3.16).69 (.77) Lagged ncarceration Rate.4' (.2) -. (.) Lagged Log of Per Capi[a lncarceralion Rate 24.7" (9.56) Lagged Police Employee Rate -.5 (.4) Lagged Log of Sworn Police Officers Per Resident Population 3.18 (13.59) Lagged A rrest Rate for Violent Crimes -.16" (.8) -.4" " (.2) Lagged Dependent Variable 87.12'" (1.45) Real Per Capita Personal ncome. (.) o.oo (O.OOJ. (.) Real Per Capita Unemployment nsurance. (.1) o.oi co.1) Real Per Capita ncome Maintenance.4 (.3).2 (.1) Real Per Capita Retireme nt Payments and Other (Lott version). (.1) Real Per Capita Retirement Payments and Other (MM version) - o.oo co.oo) Nominal Per Capita ncome -. (.) Unemployment Rate.16 (.77) - 1. (.67) -.37 (.23) Poverty Rate -.29 (.49) -.12 (.9) Lagged Number of Executions. 11 (. 16) Beer " " (17.59) 71.97'" (18.23) Population. (.) -. (.) Percent of the population living in MSAs.95'" (.29) Population Density -. 1 (.2) Observations Panel B: Spline Model Results (Table 4) (Table 5.A) (Table 6.A) (Table 7.A) DAW Model BC Model LM Model MM Model (! ) (2) (3) (4) Right-to Carry Law.5 (.64).19 (.66).41 (.47).17" (.8) Trend for Changer States.93' (.49).96" (.53).12 (.39) -.7 (.8) Lagged ncarceration Rate.3" (.2) -. (.) Lagged Log of Per Capita lncarceraiion Ra1e (8. 12) Lagged Police Employee Rate -.5 (.4) Lagged Log of Sworn Police Officers Per Resident Population 2.25 (13.56) Lagged Arrest Rate for Violent Crimes -. 1 T' (.8) -.4"" (.2) Lagged Dependent Variable 86.65'.. (1.46) Real Per Capita Personal ncome. (.)." (.). (.) Real Per Capila Unemployment nsurance -. (.2).1 (.1) Real Per Capita ncome Maintenance.3 (.3).1 (.1) Real Per Capiia Retirement Payments and Other (Lou version). (.1) Real Per Capita Retirement Payments and Other (MM version) -." (.) Nominal Per Capita ncome. (.) Unemployment Rate.7 (.87) -.29 (.81) -.28 (.23) Poverty Rate -.41 (.5) (.1) Lagged Number of Executions. 16(.17) Beer " (16.33) 67.42'" (15.43) Population. (.) -o.oo (O.OOJ Percent of the population living in MSAs.78"" (.28) Population Density. (.2) Observations Estimations include year and state fixed effects and are weighted by state population. Coefficients on demographic variables and the constant omitted. Robust standard errors (clustered at the srate level) are provided next to point estimates in parentheses. The source of all ihe crime rates is the Uniform Crime Reports (UCR). * p <., p <.5. p <.1. All figures reported in percentage 1enns. The DAW model is run on data from , the BC model from , the LM model from , and the MM model (without the crack cocaine index) from Li Deel. Ex

191 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 2 of of Page 292D #:744 Appendix C: Synthetic Control Estimates of the mpact of RTC Laws on Murder and Property Crime for 4 Different Models Table A3: The mpact of RTC Laws on the Murder Rate, DAW covariates, () (2) (3) (4) (5) (6) (7) (8) (9) (1) Average Nonnalized TEP (2.6) (4.9) (4.415) (4.534) (5.136) (4.478) (5.39) (4.739) (4.28) (3.692) N Standard errors in parentheses Co lumn numbers indicarc post-passage year under consideration : N = number o f slates in sample Dependent variable is the difference between the pcrccnrage difference in the murder rate in treatment and syn1hc1ic control states a1 given poshreatmcnt i111crval and at rime of the trcalmcnt Rcsulcs reported for the constant term resulting from this regression Stales in group : AK AR AZ CO FL GA D KS KY LA ME M l MN MO MS MT NC ND NE NM NV O H OK OR PA SC SD TN TX UT VA WV WY p <. 1. p<.5. p <.1 Table A4: The mpact ofrtc Laws on the Property Crime Rate, DAW covariates, ( ) (2) (3) (4) (5) (6) (7) (8) (9) (1) Average Nonnalized TEP (.993) ( 1.219) (2.549) (2.693) (2.748) (2.593) (2.538) (2.341) (2.367) (2.3) N Stam.Jani errors in parentheses Colum n numbers indicate post-passage year under considcrnlion; N = number of slates in sample Dependent variable is the difference between th e percentage difference in the propeny crime rme in treatme nt and synthetic comrol stmes at given post-1rea1ment interval and at time of the treatment Results reported for the constant term resulting from this regression S1a1cs in group: AK AR AZ CO FL GA D KS KY LA ME Ml MN MO MS MT NC ND NE NM NV OH OK OR PA SC SD TN T X UT VA WV WY p <.1... p <.5. p <.1 Table AS: The mpact of RTC Laws on the Murder Rate, BC covariates, () (2) (3) (4) (5) (6) (7) (8) (9) (1) Average Nonnalized TEP (2.44) (3.874) (4.367) (4.412) (5.38) (4.535) (4.85) (4.53) (3.698) (3. 169) N Standard errors in parentheses Column nu mbers indicale po st-passage year under consideration ; N = number of stales in sample Dependent variable is the difference between the percentage difference in the murder r:.1tc in treatment and synthetic con1rol s1a1es at given post-treatment interval and al time of the ucatmcnt Results reported for the constant term resulting from lhis regression S1a1es in group: AK AR AZ CO FL GA D KS KY LA ME Ml MN MO MS MT NC ND NE NM NV OH OK OR PA SC SD TN TX UT VA WV WY p <.1. p <.5... p <. Table A6: The mpact of RTC Laws on the Property Crime Rate, BC covariates, ( ) (2) (3) (4) (5) (6) (7) (8) (9) (1) Average Nonnalizcd TEP (1.36) (1.254) (2.558) (2.71 ) (2.755) (2.562) (2.549) (2.375) (2.372) (2.38) N St!lndanJ errors in parentheses Column numbers indicate post-passage year under consi<leralion; N = number of states in sample Dependent variable is the difference between the percentage difference in 1he propen y crime rnte in treatment and synthe1ie control states at given pos1-treatment interval and nr time of the treatment Results reported for the constant term resuhing from this regression Simes in group: AK AR AZ CO FL GA D KS KY LA ME Ml MN MO MS MT NC ND NE NM NV OH OK OR PA SC SD TN TX UT VA WV WY p <.1. p < OOS.'" p < Li Deel. Ex

192 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 3 of of Page 292D #:745 Table A7: The mpact of RTC Laws on the Murder Rate, LM covariates, () (2) (3) (4) (5) (6) (7) (8) (9) (1 ) Average Nonnalized T EP (1.713) (4.166) (4.51) (4.66 1) (5.313) (5.155) (5.484) (5.5 12) (4.968) (4.141 ) N Standard e rrors in parentheses Column numbers indicate post-passage year under consideration; N = number of slates in sample Dependent variable is the difference between the percentage difference in 1hc murder rate in treatment and synthetic control states ar given pos1-trca1mcn1 interval and at time of the 1rcat mc111 Rcsuhs reported for the constant term resulting fmm this regression S<a<cs in group: AK AR AZ CO FL GA D KS KY LA ME Ml MN MO MS MT NC ND NE NM NV OH OK OR PA SC SD TN TX UT VA WV WY p < p <.5... p <.1 Table AS: The mpact of RTC Laws on the Property Crime Rate, LM covariates, () (2) (3) (4) (5) (6) (7) (8) (9) (1) Average Normalized TEP (1.5) ( ) (2.616) (2.688) (2.719) (2.575) (2.387) (2.374) (2.715) (3.5) N Standard erro'"" in paremheses Column numbers indic;nc post-passage year under consideration; N = number of states in sample Dcpcndem variable is the difference becween the percentage difference in lhe propeny crime rate in Lreatment and synthetic control stales at given pos1-trea1ment imerval and at time of the trealmen( Results reported for lhe constant 1em1 resulting from this regression Sm1es in &<roup: AK AR AZ CO FL GA D KS KY LA ME Ml MN MO MS MT NC ND NE NM NV OH OK OR PA SC SD TN TX UT VA VV WY p <.1,.. p < -5.. p <.1 Table A9: The mpact of RTC Laws on the Murder Rate, MM covariates, ( ) (2) (3) (4) (5) (6) (7) (8) (9) (1) Average Nonnalized T EP (1.774) (3.987) (4. 179) (4.266) (4.559) (4.428) (4.431) (4.44) (4.86) (4.71) N S1andard errors in parentheses Column nu mbers indic;he post passage year under consideration; N = number of states in sample Dependent variable is the difference between the percentage difference in the murder rate in treatment and synthetic conlrol stales al given post lreaoncnl interval and al time of the Lrca1men1 Results repo rted for the constant tcnn resulting from lhis regression S1a<cs in group: AK AR AZ CO FL GA D KS KY LA ME Ml MN MO MS MT NC ND NE NM NV OH OK OR PA SC SD TN TX UT VA WV WY p <. O. p <.5. p <.1 Table AlO: The mpact ofrtc Laws on the Property Crime Rate, MM covariates, () (2) (3) (4) (5) (6) (7) (8) (9) (1) Average Nonnalizcd TEP (.941) ( 1.157) (2.526) (2.637) (2.596) (2.381) (2.44) (2.334) (2.31 2) (2.342) N Standard errors in parentheses Column numbers indicate post~passage year under consideration: N = number of Slates in sample Dependent variable is the difference between the percemage difference in the property crime rate in treatmem and synthetic control states at given post UCatment interval and at time of the treatment Resuhs reported for the consiam 1enn resuh ing from chis regression S<a<cs in group: AK AR AZ CO FL GA D KS KY LA ME Ml MN MO MS MT NC ND NE NM NV OH OK OR PA SC SD TN T X UT VA WV WY " p <. 1.. p <.5... p <.1 48 Li Deel. Ex

193 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 4 of of Page 292D #:746 Appendix D: Data Methodologies. Data ssues The state-level data set used in this paper updated through 214 earlier data sets used in Aneja, Donohue, and Zhang (214) and Aneja, Donohue, and Zhang (211). We further update this data set to incorporate changes to the various primary sources that have occurred since first released, and to include the additional predictor variables that are featured in the DAW and BC models. All variables are collected for the years unless otherwise noted. 41 Annual state-level crime rates are taken from the FB's Uniform Crime Reporting program. 42 Four state-level income variables (personal income, income maintenance payments, retirement payments, and unemployment insurance payments) are taken from the BEA's Regional Economic Accounts. The personal income, income maintenance, and unemployment insurance payment variables are estimated in real per capita terms (defined using the CP). The the LM and MM specifications use alternative versions of the retirement variable that are described in footnote 41. Statelevel population is generated using the Census Bureau's intercensal population estimates, while the proportional size of LM's 36 age-race-sex demographic groups are estimated using state-level population by age, sex, and race gathered by the Census. (n cases where the most recent form of these data were not easily accessible at the state level, state-level figures were generated by aggregating the Census Bureau's county-level population estimates by age, sex, and race.) Population density is estimated by dividing a given observation's population by the area of that state reported in the previous decennial census. State-level unemployment rate data is taken from the Bureau of Labor Statistics, while the poverty rate is taken from two Census series (the 1979 state-level poverty rate is derived from the Decennial Census and the poverty rates are generated using the Current Population Survey). A measure of incarceration (incarcerated individuals per 1, 41 Many of the data sources that we used in our earlier analysis are revised continuously, and we use a newer version of these data series in this paper than we did in our earlier ADZ analysis. We sometimes made data changes during the data cleaning process. For instance, a detailed review of the raw data underlying a.test statistics uncovered a small number of agencies which reported their police staffing levels twice, and we attempted to delete these duplicates whenever possible. Moreover, we sometimes use variables that are defined slightly differently from the corresponding variable used in Lott and Mustard (1997) or Moody and Marvell (28). For example, after examining the extension of Lott's county data set to the year 2, we found that our estimates more closely approximated Lott's per capita retirement payment variable when we (a) used the total population as the denominator rather than population over 65 and (b) used as our numerator a measurement that includes retirement payments along with some other forms of government assistance. As a result, we use a modified retirement variable that incorporates these changes in the MM specification. Our retirement variable in the LM specification, in contrast, uses the population over 65 as a denominator and uses a tighter definition of retirement payments. 42 For our main analysis, we formulate our crime rates by dividing FB reported crime counts by FB reported state-level populations. As a robustness check we used the rounded state-level crime rates reported by the FB while using the DAW regressors and aggregate violent crime as an outcome variable. We find that this alternative crime rate definition does not qualitatively affect our findings. 49 Li Deel. Ex

194 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 5 of of Page 292D #:747 state residents) is calculated from tables published by the Bureau of Justice Statistics counting the number of prisoners under the jurisdiction of different state penal systems. Our primary estimates for crime-specific state-level arrest rates are generated by adding together estimates of arrests by age, sex, and race submitted by different police agencies. We then divided this variable by the estimated number of incidents occurring in the same state (according to the UCR) in the relevant crime category. 43 We also use the index of crack cocaine usage constructed by Fryer et al. (213) for our analysis, which is only available between the years 198 and 2, and therefore we drop this variable from the MM model when we estimate this model on data through 214. Since we already include controls that incorporate information on the racial composition of individual states in our analysis, we use the unadjusted version of the crack index instead of the version that is adjusted to account for differences in state racial demographics. No data for the crack cocaine index that we use was available for the District of Columbia, and our matching methodology does not allow the District of Columbia to be included in our analysis in specifications that include this variable as a predictor. After considering several different ways to confront this issue, we ultimately decided to exclude the District of Columbia from the synthetic controls analysis owing to its status as a clear outlier whose characteristics are less likely to be meaningfully predictive for other geographic areas. Abadie, Diamond, and Hainmueller (21) emphasize that researchers may want to "[restrict] the comparison group to units that are similar to the exposed units [in terms of the predictors which are included in the model]" (496). Given that the District of Columbia had the highest per capita personal income, murder rate, unemployment rate, poverty rate, and population density at various points in our sample, Abadie's admonition would seem to support omitting the District as one of our potential control units. 44 We should note that even if we include DC in the synthetic controls estimates, it still shows RTC laws increase violent crime by 13.2 percent in the tenth year (as opposed to the 14.7 percent figure shown in Table 9). We consider two separate police measures for the purposes of our analysis. Our reported results are based on the same police variable that we used in Aneja, Donohue, and Zhang (214). To construct this variable, we take the most recent agency-level data provided by the FB and use this information to estimate the number of full-time police employees present in each state per 1, 43 We chose this variable as the primary one that we would use in this analysis after confirming that this variable was more closely cotelated with Lott's state-level atest variables in the most recent data set published on his website (a data set which runs through the year 25) than several alternatives that we const:mcted. 44 Another advantage of excluding the District of Columbia from our sample is that the Bureau of Justice Statistics stops estimating the incarcerated population of the District of Columbia after the year 21 owing to the transfer of the district's incarcerated population to the federal p1ison system and the DC Jail. While we have tried to reconstruct incarceration data for DC for these years using other data sources, the estimates resulting from this analysis were not, in our view, plausible substitutes for the BJS estimates we use for all other states. The raw data set that we use to gather information about state-level a.ttest rates is also missing a large number of observations from the District of Columbia's main police department, which further strengthens the case for excluding DC from our data set. 5 Li Deel. Ex

195 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 6 of of Page 292D #:748 residents. We fill in missing observations with staffing data from previous years in cases where the FB chose to append this information to their agency entries, and we divide the resulting estimate of the total number of police employees by the population represented by these agencies. This variable, which was originally constructed for our regression analysis, has the advantage of not having any missing entries and is closely correlated (r =.96) with an alternative measure of police staffing generated by extrapolating missing police agency data based on the average staffing levels reported by agencies in the same year and type of area served (represented by a variable incorporating nineteen categories separating different types of suburban, rural, and urban developments.) As an alternative, we use data published by the Bureau of Justice Statistics on the number of fulltime equivalent employees working for police agencies (figures that were also included in the data set featured in John R Lott and David B Mustard (1997)). (We do not rely on this variable in our main analysis owing to the large number of missing years present in this data set and owing to discrepancies in the raw data provided by the BJS, which sometimes needed to be corrected using published tables.) We find that our estimated average treatment effects for aggregate violent crime and the conclusions that we draw from these averages are qualitatively unaffected by substituting one police employment measure for another, which suggests that measurement error associated with our estimates of police activity is not driving our results.. The Dates of Adoption of RTC Laws We use the same effective RTC dates used in Aneja, Donohue, and Zhang (214) with one small modification. Owing to the fact that we are using annual panel data, the mechanics of the synthetic control methodology require. us to specify a specific year for each state's RTC date. To take advantage of the information we have collected on the exact dates when RTC laws went into effect in each state, each state's effective year of passage is defined as the first year in which a RTC law was in effect for the majority of that year. 45 This causes some of the values of our RTC variable to shift by one year (for instance, Wisconsin's RTC date shifts from 211 to 212, since the state's RTC law took effect on November 1, 211). 46 While there have been numerous disagreements about the exact laws that should be used to determine when states made the transition from a "may issue" to a "shall issue" state, we believe that the dates used in this paper accurately reflect the year when different states adopted their RTC law. We supplemented our analysis of the statutory history of RTC laws in different states with 45 A table showing each state's original adoption date and adjusted adoption date is shown in Table A of Appendix A. 46 By default, we also take this adjustment into account when deciding which states adopt RTC laws within ten years of the treatment state's adoption of the given law. As a robustness check, we re-ran our aggregate violent crime codes under the DAW specification without considering the modified RTC dates in our selection of control units, finding that this change did not affect our qualitative findings meaningfully. 51 Li Deel. Ex

196 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 7 of of Page 292D #:749 an extensive search of newspaper archives to ensure that our chosen dates represented concrete changes in concealed caity policy. We extensively document the changes that were made to our earlier selection of right-to-carry dates and the rationales underlying these changes in Appendix G of Aneja, Donohue, and Zhang (214). t is important to note that the coding of these dates may not reflect administrative or logistical delays that may have prevented the full implementation of a RTC law after authorities were legally denied any discretion in rejecting the issuing of RTC permits. deally, a researcher would be able to control for the actual level of RTC permits in existence each year for each state. Although this data would be preferable to a mere indicator variable for the presence of an RTC law, such comprehensive information unfortunately is not available. 52 Li Deel. Ex

197 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 8 of of Page 292D #:75 Appendix E: Replicating Our Analysis One issue which is rarely addressed directly in the existing literature surrounding the application of the synthetic control technique is the sensitivity of the selection of the synthetic control to seemingly inconsequential details when using maximum likelihood to select the weights associated with different predictors in our analysis. More specifically, when using the excellent "synth" package for Stata created by Abadie, Hainmueller, and Diamond along with the nested option (which implements the optimization technique described in footnote 2), both the version of Stata (e.g., SE vs. MP), the specifications of the computer running the command, and the order in which predictors are listed can affect the composition of the synthetic control and by extension the size of the estimated treatment effect. The root cause of the differences between Stata versions is explained by a 28 StataCorp memo, which noted that: "When more than one processor is used in Stata/MP, the computations for the likelihood are split into pieces (one piece for each processor) and then are added at the end of the calculation on each iteration. Because of round-off error, addition is not associative in computer science as it is in mathematics. This may cause a slight difference in results. For example, al +a2+a3+a4 can produce different results from (al+a2)+(a3+a4) in numerical computation. When changing the number of processors used in Stata, the order in which the results from each processor are combined in calculations may not be the same depending on which processor completes its calculations first. " 47 Moreover, this document goes on to note that the differences associated with using different versions of Stata can be minimized by setting a higher threshold for nrtolerance( ). This optimization condition is actually relaxed by the synth routine in situations where setting this threshold at its default level causes the optimization routine to crash, and we would therefore expect the results of Stata SE and MP to diverge significantly whenever this occurs. n our analysis, we use the UNX version of Stata/MP owing to the well-documented performance gains associated with this version of the software package. Another discrepancy that we encountered is that memory limitations sometimes caused our synthetic control analyses to crash when using the nested option. When this occurred, we would generate our synthetic control using the regression-based technique for determining the relative weights assigned to different predictors. We encountered this situation several times when running our Stata code on standard desktop computers, and these errors occurred less often when using 47 This memo can be found at the following link: 53 Li Deel. Ex

198 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 9 of of Page 292D #:751 more powe1ful computers with greater amounts of memory. For this reason, to replicate our results with the greatest amount of precision, we would recommend that other researchers run our code on the same machines that we ran our own analysis: a 24-core UNX machine with 96GB of RAM running Stata/MP. One final discrepancy that we are still in the process of investigating is the effect of changing the variable order in the synthetic control command on the composition of the synthetic control when using the nested option. Unfortunately, the large number of predictors included in the LM and MM specifications make it difficult to use a fixed criteria (e.g., minimizing the average coefficient of variation of the RMSPE) for determining the order in which vaiiables should be listed. While we have not modified the order in which predictors were listed in our models after observing the results that we de1ived from that variable order, it is useful to be awai e that different vai iable orders can alter estimates slightly. However, the observation that our synthetic controls estimates for violent crime results are essentially unchanged after trying multiple specifications featuring different sets of predictors gives us greater confidence that our conclusions about these specifications are robust to changes in variable order as well. 54 Li Deel. Ex

199 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 1199 of 56 of 292 Page D #:752 Appendix F: Synthetic Control Graphs Estimating mpact of RTC Laws On Violent Crime Using the DAW Model 48 Figures Al-A33 Treated Unit Synthetic Control Unit Alaska: Violent Crime Rate Effect of 1995 RTC Law 1 Years After Ado lion: 14.4% ~ ~ c., ~., :: "" ~ iii (l_,_ Q) 1ij a::., E 8 c.!!! > <O 8 "', - -, \ \ -_, \ \,, ~----.., Note: DAW Variables and yearly lags are used as predictors Composition of SC: MD (.519); NE (.481 ) CVRMSPE:.119 (21 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: MD RTC Adopting States ncluded in Synlhetic Control: NE (27) Recall that each state's effective year of passage is defined as the first year in which a RTC law was in effect for the majority of that year. 55 Li Deel. Ex

200 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 112 of 56 of 292 Page D #:753 Treated Unit Synthetic Control Unit Arkansas: Violent Crime Rate Effect of 1996 RTC Law 1 Years After Ado lion: 23.7% 8 U) ~ "O " ~ "' " ~ :;; Q. ~ "' ". c " 5 "'...,, / ~~ ' ' \ "' Note: DAW Variables and yearly lags are used as predictors Composition of SC: DE (.186); L (.231); A (.584) CVRM SPE:.13 (2 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: DE RTC Adopting States ncluded in Synthetic Control: A (211); L (214) 26 Li Deel. Ex

201 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 1221 of 56 of 292 Page D #:754 <D Treated Unit Synthetic Control Unit Arizona: Violent Crime Rate Effect of 1995 RTC Law 1 Years After Ado tion: 8.8% ~ C: "O ;;; " "' " "' :;; n. 1ii " "' " E g 8 ~ g 5 \ \,...,---, ' ' ' '--- g., Note: DAW Variables and year1y lags are used as predictors Composition of SC: CA (.327); H (.49); MD (.229); NE (.35) CVRMSPE:.58 (9 of 33 states, where 1 denotes the state with the best pre-passage fil) States Never Passing RTC Laws ncluded in Synthetic Control: CA. H ; MD RTC Adop ting States ncluded in Synthetic Control: NE (27) 25 Li Deel. Ex

202 ~ Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 1322 of 56 of 292 Page D #:755 Treated Unit Synthetic Control Unit Colorado: Violent Crime Rate Effect of 23 RTC Law 1 Years After Ado tion: -1.2% g "' C.il ;;; Q'. " "' g ~ a. " Q'. " E " 8 C: " 5... g "' Note: DAW Variables and year1y lags are used as predictors Compositi on of SC: H (.499); NY (.293); R (.28) CVRMSPE:.77 (16 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: H ; NY ; R RTC Adopting States ncluded in Synthetic Control: 213 Li Deel. Ex

203 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 1423 of 56 of 292 Page D #:756 Treated Unit Synthetic Control Unit Florida: Violent Crime Rate Effect of 1988 RTC Law 1 Years After Ado lion: 34.8% 8 N w c ;; O'. " t Q. ~ O:'. " E c u c " 5 \ \ \ \ \ \ \ \ \ \ \ ' \ ' ', ' ' ' ' ' ' 8 <O Note: DAW Variables and yearly lags are used as predictors Composition of SC: CA (.223); Ml (.11); NY (.667) CVRMSPE:.56 (8 of 33 states, where 1 denotes the state with the best pre-passage fit.) Stales Never Passing RTC Laws ncluded in Synthetic Control: CA ; NY RTC Adopting States ncluded in Synthetic Control: Ml (21) Li Deel. Ex

204 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 1524 of 56 of 292 Page D #:757 g <O Treated Unit Synthetic Control Unit Georgia: Violent Crime Rate Effect of 199 RTC Law 1 Years After Ado lion: 6.6% r--.. " ;;; " :: " E 8 E " 5 ' \ \ ~ \ \ \ \ \ \ \ \ "" Note: DAW Variables and year1y lags are used as predictors Composition of SC: CA (.31); A (.452); MO (.97); NY (.15) CVRMSPE:.68 (12 of 33 states, where 1 denotes the state w,th the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: CA, NY RTC Adopting States ncluded in Synthetic Control: A (21 1); MO (24) 2 Li Deel. Ex

205 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 1625 of 56 of 292 Page D #: Treated Unit Synthetic Control Unit daho: Violent Crime Rate Effect of 199 RTC Law 1 Years After Ado lion: 5.3% a. ".; " Cl'. g "., E c u C: " 5 \,, ' N Note: DAW Variables and yearly lags are used as predictors Composition of SC: H (.961): A (.39) CVRMSPE:.91 (19 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: H RTC Adopting States ncluded in Synthetic Control: A (211) 2 Li Deel. Ex

206 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 1726 of 56 of 292 Page D #:759 g "' Treated Unit Synthetic Control Unit Kansas: Violent Crime Rate Effect of 27 RTC Law 7 Years After Ado lion: -6.3% 5l 1ii " a: " E c: u C: " >... \ \ \ \ \ \ \ '/ \ \ \ \ Note: DAW Variables and yearly lags are used as predictors Composition of SC: DE (.183): H (.583): MA (.235) CVRMSPE:.69 (13 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: DE : H ; MA RTC Adopting States lnduded in Synthetic Control: 214 Li Deel. Ex

207 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 1827 of 56 of 292 Page D #:76 Treated Unit Synthetic Control Unit Kentucky: Violent Crime Rate Effect of 1997 RTC Law 1 Years After Ado lion: 3 9% g g...,, \ \ v" :, ' ' \ : \ : \ ' \ : \. ',,',,, ',' ' ' g N Note: DAW Variables and year1y lags are used as predictors Compositio n of SC: L (. 198); W1 (.82) CVRMSPE:.119 (22 of 33 states. where 1 denotes the state with the best pre-passage fit. ) States Never Passing RTC Laws ncluded in Synthetic Control: RTC Adopting States ncluded in Synthetic Control: L (214); W1 (212) 27 Li Deel. Ex

208 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 1928 of 56 of 292 Page D #:761 Treated Unit Synthetic Control Unit Louisiana: Violent Crime Rate Effect of 1996 RTC Law 1 Years After Ado tion: 8 cl'..; " a: " E c: u C: " 5 8 <D,,---\, \ \ \ ' ' ' ' ' ' ' ' ' ' ' ~ Note: DAW Variables and yearly lags are used as predictors Composition of SC: CA (.38); DE (.27); L (.756) CVRMSPE:.8 (17 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: CA ; DE RTC Adopting States ncluded in Synthetic Control: L (214) 26 Li Deel. Ex

209 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 229 of 56 of 292 Page D #:762 "' " Treated Unit Synthetic Control Unit Maine: Violent Crime Rate Effect of 1986 RTC Law 1 Years After Ado tion: -16.5% " c i ;;; oc " "'... >'.... a. " ~ oc " E "' 8 c " 5 "'...:-- J ' ' i r - - ', / './, \ \ \ \..., \ \,, "' ~ Note: DAW Variables and yearty lags are used as predictors Composition of SC: H (.16); A (.84) CVRMSPE:.196 (28 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: H RTC Adopting States ncluded in Synthetic Control: A (21 1) 1996 Li Deel. Ex

210 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 2121 of 56 of 292 Page D #:763 g a, Treated Unit Synthetic Control Unit Michigan: Violent Crime Rate Effect of 21 RTC Law 1 Years After Ado lion: 8.8%..., c.., ~ "' "' "' ~.. " 2,.._ "' " E <D 8 ~ 5, f ' ' ' ' g Note: DAW Variables and year1y lags are used as predictors Composition of SC: MD (.362): NY (.334); W (.33) CVRMSPE:.75 (15 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: MO ; NY RTC Adopting States ncluded in Synthetic Control: W (212) 211 Li Deel. Ex

211 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:764 Treated Unit Synthetic Control Unit Minnesota: Violent Crime Rate Effect of 23 RTC Law 1 Years After Ado lion: -.7% ' "' ' ' ' 1ii " :: " E c: () C " 5 M ~ N, ~., \ \,,,,, ',:,, ' \, \ \ \,,,, \ \ 8 N Note: DAW Variables and yearly lags are used as predictors Composition of SC: DE (.84): H (.916) CVRMSPE:.166 (27 of 33 states, where 1 denotes the state with the best pre-passage fit. ) States Never Passing RTC Laws ncluded in Synthetic Control: DE. H RTC Adopting States ncluded in Synthetic Control: 213 Li Deel. Ex

212 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:765 Treated Unit Synthetic Control Unit Missouri: Violent Crime Rate Effect of 24 RTC Law 1 Years After Ado tion: 14.1% 8 "' 8 r- ii o._ ~ "' o:'. <D V E 8 E V 5 8 "' ' ' ' ' Note: DAW Variables and year1y lags are used as predictors Composition of SC: CA (.41); DE (.265); H (.333) CVRMSPE:.62 (11 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: CA ; DE ; H RTC Adopting States ncluded in Synthetic Control: 214 Li Deel. Ex

213 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:766 Treated Unit Synthetic Control Unit Mississippi: Violent Crime Rate Effect of 199 RTC Law 1 Years After Ado lion: 34.2% "' "' a. ".; " Cl'. " g E " B ~ 5,,,,,', \,' ', ' \ ' ' ' ',, \ ' \ \ \ ' / ' /,,, Note: DAW Variables and year1y lags are used as predictors Composition of SC: H (.721): A (.16); NE (.11); OH (.252) CVRMSPE:.5 (6 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws lnduded in Synthetic Control: H RTC Adopting Stales ncluded in Syntheti c Control: A (211 ); NE (27); OH (24) 2 Li Deel. Ex

214 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:767 Treated Unit Synthetic Control Unit Montana: Violent Crime Rate Effect of 1992 RTC Law 1 Years After Ado lion: 9.9% 8 "' ~ C ~.,. "C ~ "' ' ~,; Q. M ~ ' ~. () C.S! 5,..., \ \ \,', - N,, Note: DAW Variables and year1y tags are used as predictors Composition of SC: H (.121); W (.879) CVRMSPE:.49 (31 of 33 states, where 1 denotes the state with the best pre-passage fit. ) States Never Passing RTC Laws ncluded in Synthetic Control: H RTC Adopting States ncluded in Synthetic Control: W (212) 22 Li Deel. Ex

215 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:768 Treated Unit Synthetic Control Unit North Carolina: Violent Crime Rate Effect of 1996 RTC Law 1 Years After Ado lion: 18.3%.. " a. " Q'.'. " E i5 E " 5 8 "',, ' ' Note: DAW Variables and yearly lags are used as predictors Composition of SC: DE (.92); L (.396); NE (.512) CVRMSPE:.49 (4 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws lnduded in Synthetic Control: OE RTC Adopting States ncluded in Synthetic Control: L (214); NE (27) 26 Li Deel. Ex

216 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:769 "' Treated Unit Synthetic Control Unit North Dakota: Violent Crime Rate Effect of 1986 RTC Law 1 Years After Ado tion: 13.4% ~ ~ c M 'O.;; " a: " "' "' N ~ ~ Cl. 1ij " " a: N. u C.!!! 5 ~..,..,,.,,.' ( ~ 8 "' Note: DAW Variables and yea~y lags are used as predictors Composition of SC: W (1) CVRMSPE: 2.43 (33 of 33 states, where 1 denotes the state with the best pre-passage fit) States Never Passing RTC Laws ncluded in Synthetic Control: RTC Adopting Slates ncluded in Synthetic Control: W (212) 1996 Li Deel. Ex

217 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:77 Treated Unit Synthetic Control Unit Nebraska: Violent Crime Rate Effect of 27 RTC Law 7 Years After Ado lion: 9.7% g "'... ll. " ;;; " O'.'. " E 8 E " 5 "' J J J - ', ' \/ \ \ \ \ ' \ \,, \ \......, ',, } ' ' ' \ \ \ ' N Note: DAW Variables and yearty lags are used as predictors Composition of SC: DE (.173); H (.827) CVRMSPE:.28 (29 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: DE ; H RTC Adopting States lnd uded in Synthetic Control: 214 Li Deel. Ex

218 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:771 Treated Unit Synthetic Control Unit New Mexico: V iolent Crime Rate Effect of 24 RTC Law 1 Years After Ado lion: 14.7% "' g ( ',_, ' ' \ ' \ r-, ',/ \ : / \ \ : \ V \\ "' Note: DAW Variables and yearly lags are used as predictors Composition of SC: CA (.485): DE (.515) CVRMSPE:.122 (23 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws lnciuded in Synthetic Control: CA : DE RTC Adopting States ncluded in Synthetic Control: 214 Li Dee l. Ex

219 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 3219 of 56 of 292 Page D #:772 ~ Treated Unit Synthetic Control Unit Nevada: Violent Crime Rate Effect of 1996 RTC Law 1 Years After Ado lion: 23.7% g ~ J,@ a>.,, ;;; " a: " "'., ~ : o._ a: * E r-. 8 c " 5 <D :il ~ Note: DAW Variables and yearly lags are used as predictors Composition of SC: H (.167); MD (.833) CVRMSPE:.152 (25 of 33 states. where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: H; MD RTC Adopting States ncluded in Synthetic Control: 26 Li Deel. Ex

220 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 3122 of 56 of 292 Page D #:773 Treated Unit Synthetic Control Unit Ohio: Violent Crime Rate Effect of 24 RTC Law 1 Years After Ado lion: -.8% <O "' C <l> " io <l> :: " l, a. <l> <O :: <l> E c u C <l> 5 "' ~ g.., Note: DAW Variables and year1y lags are used as predictors Composition of SC: CA (.195): H (.27); R (. 598) CVRMSPE:.4 (2 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: CA. H : R RTC Adopting States ncluded in Synthetic Control: 214 Li Deel. Ex

221 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:774 Treated Unit Synthetic Control Unit Oklahoma: Violent Crime Rate Effect of 1996 RTC Law 1 Year.; After Ado lion- 9.7% f 1', \ c Q) 5 g...,--,. l ' g-'---, t ~ j Note: DAW Variables and yeariy lags are used as predictors Composition of SC: DE (.27); L (.246): NE (.484) CVRMSPE:.55 (7 of 33 states. where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws lnduded in Synthetic Control: DE RTC Adopting States ncluded in Synthetic Control: L (214); NE (27) 26 Li Deel. Ex

222 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:775 Oregon: Violent Crime Rate Effect of 199 RTC Law 1 Years After Ado tion: -.6%,.._ Treated Unit Synthetic Control Unit.. C,, a, " a: " "'.. Q. " a: " E 23 C " 5 <D g U} /,,.----,, ;"' \ /,: : '\ /! \ \ \ \ Note: DAW Variables and year1y lags are used as predictors Composition of SC: CA (.14); C(.42); H (.62); Ml (.344); MN (.161 ) CVRMSPE:.49 (5 of 33 states. where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: CA ; H RTC Adopling States ncluded in Synthetic Control: CO (23); Ml (21); MN (23) 2 Li Deel. Ex

223 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:776 Pennsylvania: Violent Crime Rate Treated Unit Synthetic Control Unit ~ ~ :: " E ~... 1ii " :: " E c: g... \ \ / / / j, Note: DAW Variables and yearty tags are used as predictors Composition of SC: DE (.77); H (.16): NE (.43); NJ (.17); OH (.256): W (.348) CVRMSPE:.18 (1 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: DE ; H ; NJ RTC Adopting States ncluded in Synthetic Control: NE (27): OH (24); W (212) 1999 Li Deel. Ex

224 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:777 Treated Unit Synthetic Control Unit South Carolina: Violent Crime Rate Effect of 1997 RTC Law 1 Years After Ado lion: 22.5% 8 c "'.., a; "' a: "' " 8,-- / \ ' --k : \ ' \ : : : : : \ \ ' ' ' ' ' ' ', ',... " Note: DAW Variables and year1y lags are used as predictors Composition of SC: DE (.179); L (.821 ) CVRMSPE:.88 (18 of 33 states. where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: DE RTC Adopting States ncluded in Synthetic Control: L (214) 27 Li Deel. Ex

225 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:778 Treated Unit Synthetic Control Unit South Dakota: Violent Crime Rate Effect of 1985 RTC Law 1 Years After Ado tion: -1.6% g :;;. ni " ~ a: "' E " i:: /,... r g "' "' ---,, ',,,,.,..,. '...-,,' g Note: DAW Variables and yearly fags are used as predictors Composition of SC: A (.625); W (.375) CVRMSPE:.436 (32 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: RTC Adopting States ncluded in Synthetic Control: A (211 ); W1 (212) 1995 Li Deel. Ex

226 ~ Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:779 8 "' Treated Unit Synthetic Control Unit Tennessee: Violent Crime Rate Effect of 1997 RTC Law 1 Years After Ado lion: 29.5% ~ ~ a: " "' l 1ii a: " " E c: c " 8 "' / , r \ \,, '", '... \., \ / l '\ ' : \ : \ \ \ \ ' \ ' Note: DAW Variables and year1y lags are used as predictors Composition of SC: DE (.291): L (.395): A (.314) CVRMSPE:.123 (24 of 33 states, w here 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws lnduded in Synthetic Control: DE RTC Adopting States lnduded in Synthetic Control: A (211): L (214) 27 Li Deel. Ex

227 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:78 a, Treated Unit Synthetic Control Unit Texas: Violent Crime Rate Effect of 1996 RTC Law 1 Years After Ado lion: 16.6% g "' J'.... " :: "' E c: u C: "' ~ "' g "' ' ' \'., ~ ' \ \ \ \ \ Note: DAW Variables and yearly lags are used as predictors Composition of SC: CA (.578); NE (.86); W (.336) CVRMSPE:.6 (1 of 33 states, where 1 denotes the state with the best pre-passage fil) States Never Passing RTC Laws ncluded in Synthetic Control: CA RTC Adopting States ncluded in Synthetic Control: NE (27); W (212) 26 Li Deel. Ex

228 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:781 Treated Unit Synthetic Control Unit Utah: Violent Crime Rate Effect of 1995 RTC Law 1 Years After Ado lion: -2.2% "' C " ~ a: " a. " ~ a:. " ~ 5 M "' N N Note: DAW Variables and year1y lags are used as predictors Composition of SC: H (.756); KS (.6): R (.18); wt (.75) CVRMSPE:.72 (14 of 33 sta tes, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: H ; R RTC Adopting States ncluded in Synthetic Control: KS (27): wt (212) 25 Li Deel. Ex

229 ~ Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 4229 of 56 of 292 Page D #:782 Treated Unit Synthetic Control Unit Virginia: Violent Crime Rate Effect of 1995 RTC Law 1 Years After Ado lion: -3.6% ~ ~ "' " " cf ;;; " "' " E 8 <= " 5 ' ' ' \ \ \ Note: DAW Variables and year1y lags are used as predictors Composition of SC: H (.249); KS (.235); NE (.157); R (.269); W (.9) CVRMSPE:.44 (3 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: H ; R RTC Adopting States ncluded in Synthetic Control: KS {27); NE (27); W (212) 25 Li Deel. Ex

230 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page 4123 of 56 of 292 Page D #:783 Treated Unit Synthetic Control Unit West Virginia: Violent Crime Rate Effect of 199 RTC Law 1 Years After Ado lion: 62.3% g "' ' ' "' C "' ;;; "' " ". " ~ "' ". ~ 5 "' <') g N / / _,, { - -\,' / 1 \ / : \ ' / : \ ',1 \ \ \ \/ Note: DAW Variables and year1y lags are used as predictors Composition of SC: H (.7); A (.993) CVRMSPE:.377 (3 of 33 states. where 1 denotes the state with the best pm-passage fit.) States Never Passing RTC Laws ncluded in Synthetic Control: H RTC Adopting States ncluded in Synthetic Control: A (211) 2 Li Deel. Ex

231 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:784 Treated Unit Synthetic Control Unit Wyoming: Violent Crime Rate Effect of 1995 RTC Law 1 Years After A do tion- 15.8% "' ~ u g " a; " a:: " "' " :;; n_.. ~ g M a:: E B C " 5 g g "',, \ \ \ \.. '--~ : \ ' "' Note: DAW Variables and year1y lags are used as predictors Composition of SC: H (.71); R (.525): W (.44) CVRMSPE:.166 (26 of 33 states, where 1 denotes the state with the best pre-passage fit.) States Never Passing RTC Laws lnciuded in Synthetic Control: H ; R RTC Adopting States ncluded in Synthetic Control: W (212) 25 Li Deel. Ex

232 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:785 Appendix G: Data Sources Years Variable(s) Source Model(s) Notes Available Statutes researched via Westlaw and HeinOnline. See footnotes 6 and 7 for explanations of State DAW, RTC Variables these variables' constructions. Note that the spline variable is coded as O in all years for states session BC, LM, (shall/ & aftr) 214 that passed before the data period, which depends on the model under consideration. For laws MM example, fo r the DAW model ( ), it is coded as O for states that passed before Police Staffing Agency-year-level police employment data were acquired from the FB and aggregated to the DAW, state-year level. The police employee rate is the total number of employees, divided by the FB 214 BC population as given in the same dataset. n the BC model, this variable is the one-year lag of logged police staffing per capita. DAW, UCR Data Tool for data through 21 3; Table 4 of 215 crime report for data in 214. Each Crime FB[ BC, LM, crime rate is the corresponding crime count, divided by the population metric used by the FB[, 214 MM times 1,. DAW, ntercensal estimates are used, except in 197 and 198, for which decadal-census estimates are Population Census BC, LM, used. All models weight regressions by population; the LM and!\,m models also incl ude it as a 214 MM covariate. Population by DAW, Age, Sex, and Census BC, LM, ntercensal estimates are used. 214 Race MM DAW, [ncludes personal income, unemployment insurance, retirement payments and other, and [ncome Metrics BEA BC, LM, 214 income mai ntenance payments. All 4 measures are divided by the CP to convert to real terms. MM ndex 214 DAW, BLS BC, LM, CP varies by year but not by state. MM The number of prisoners under the j urisdiction of a state as a percentage of its intercensal DAW, ncarcerations BJS population. n the BC model, this variable is the one-year lag of the log of year-end 214 BC, MM jurisdictional population per capita Land area over a given decade is taken from the most recent decadal Census. The density Land Area Census LM 214 variable is intercensal population divided by land area DAW, Poverty Rate Census The Census directly reports the percentage of the population earning less than the poverty line. 214 MM Unemployment DAW, BLS Rate 214 BC,LM Agency-month-year-level arrests data, separated by age, sex, race, and crime category, were acquired from the FB and aggregated to the state-year level. For each crime category, the arrest Arrests FB LM,MM 214 rate is the number of arrests for that crime as a percentage of the (VCR-reported) number of cnmes. Consumer Price Prof Crack ndex Roland MM Following the MM model, we use the unadjusted version of the index. 2 Fryer Beer Population in Metropolitan Statistical Areas DAW, The NH reports per-capita consumption of ethanol broken down by beverage type, including NTH 2 14 BC beer FB / MSA population counts obtained from CPSR-provided UCR arrests data values are DAW 2 14 CPSR linearly extrapolated Executions 2 14 BJS BC All variables are at the state-year level unless otherwise noted. Variable creation scripts are available from the authors upon request. 88 Li Deel. Ex

233 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:786 Appendix H: Methodology to Choose the Number of Lags of the Dependent Variable to nclude as Predictors in Synthetic Controls We use a cross validated approach to determine the optimal lag choice(s) to include as predictor(s) in the synthetic control model. We use this procedure to choose among four potential lag choices used in the synthetic control literature; these choices involve including lags of the dependent vaiiable in every pre-treatment yeai, three lags of the dependent vaiiable, 49 one lag which is the average of the dependent vai iable in the pre-treatment period, and one lag which is the value of the dependent variable in the year prior to RTC adoption. 5 To implement the cross validation procedure, we first define our training period as 1977 through the sixth year prior to RTC adoption, the validation period as the fifth year prior to RTC adoption through one year prior to RTC adoption, and the full pre-treatment period as 1977 through one year prior to RTC adoption. For each of our 33 treatment units, data from the training period is used to determine the composition of the synthetic control. Specifically, for each of the 33 treatment units, we assign the treatment 5 years before the treatment actually occurred, and then run the synthetic control program using the standard ADZ predictors defined in Aneja, Donohue, and Zhang (211) and a 5 yeai reporting window. We then examine the fit during the training period, the validation period, and the entire pre-treatment period to see how closely the synthetic control estimate matches the value of the dependent variable for different lag choices. Tables A 11-A 13 examine the fit of the synthetic control estimate during the training period, validation period, and the entire pre-treatment period using three different loss functions. Table A 11 defines the error using the mean squared error between the actual value of the dependent variable and the synthetic control estimate during a given period; Table A12 uses the mean of the absolute value of the difference between the treated value and synthetic control estimate; finally, Table A13 uses the CV of the RMSPE. For Tables Al 1-A13, an unweighted average of the error for each of the 33 treatment states is presented. For Tables A14-A16, a population weighted average of the error for each of the 33 treatment states is presented, where population from the first year of the relevant period is used The first lag is the value of the dependent variable in 1977, the second lag is the value of the dependent variable in the year prior to RTC adoption, and the third lag is the value of the dependent variable in the year that is midway between the year co1tesponding to the first and second Jag. All results presented in Tables A 11 through Table A 16 use overall violent crime as the dependent variable. 5 The first choice is used, for example, in Bohn, Lofstrom, and Raphael (214), the second choice is used by Abadie, Diamond, and Hainmueller (21), and the third and fourth choices are suggested by Kaul et al. (216). 51 The first year of the training and full pre-treatment period is 1977, while the first year of the validation period is the fifth year prior to RTC adoption. 89 Li Deel. Ex

234 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:787 The results from Tables Al 1-Al6 provide strong evidence that using yearly lags of the dependent variable is the best option. As expected, across all six tables, the error in the training period is lowest using yearly lags. However, yearly lags also provides the lowest error in the validation period, regardless of how the error is defined or whether population weights are used to aggregate the measure of error over all treatment states. n addition, across all six tables, the error over the full pre-treatment period is lowest using yearly lags. A potential concern with using all preintervention outcomes of the dependent variable as synthetic control predictors is that the synthetic control unit will not closely match the treated unit on the non-lagged predictors during the pre-treatment period. 52 But as Table A17 shows, we do not find that the synthetic control unit's fit on the non-lagged predictors is worse using yearly lags. To generate the numbers in Table A 17, for each treatment state, we first take a simple average of our predictor of interest over all pre-treatment years (1977 through the year prior to RTC adoption). A population weighted average of the predictor pre-treatment means is then taken over all treatment states to reach the figures presented, which represent an aggregate measure of the pre-treatment predictor means. 53 Based on the absolute value of the difference between the aggregate treated predictor means and the aggregate synthetic control predictor means, yearly lags has the second best performance. The aggegate synthetic control predictor means using yearly lags comes closest or second closest to the treated unit for 9/16 predictors. n comparison, one lag that is the average of the dependent variable in the pre-treatment period comes closest or second closest for 11/16 predictors, one lag that is the value of the dependent variable in the last pre-treatment year comes closest or second closest for 7 /16 predictors, and three lags for 5/16 predictors. We thus choose yearly lags of the dependent variable as our optimal lag choice for two main reasons. The first is that yearly lags produces the lowest error not only in the training period, but also in the validation period and the full pre-treatment period. This statement is robust to various ways of defining the error and aggregating the error across treatment states. The second is that the synthetic control units using yearly lags do a fairly good job, relative to the other lag choices, of matching the pre-treatment (non-lagged) predictor means of the treatment states. 52 See Kaul et al. (216). 53 Unlike Tables A 11-A 16, where the treatment year for our 33 states of interest is assigned to five years before the actual year of RTC adoption, in Table A 17, the treatment year is identical to the year of RTC adoption. For Table A 17, the states eligible to be in a treated unit's synthetic control are those states that either never passed RTC laws, or passed more than 1 years after the treated unit adopted RTC laws. n contrast, for Tables A 11-A 16, the states eligible to be in a treated unit's synthetic control are those states that either never passed RTC laws, or passed any year after the treated unit adopted RTC laws. 9 Li Deel. Ex

235 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:788 Table All: Comparison of Fit Across Various Lagchoices - Define Fit Using Mean Squared Error three lags yearly lags one lag 3\'Cragc one.lg final prc-tn:atmcnt year lr.lining period;.\kan S4uarcd En or 2, , '.?,63.-l-t validation period; :\lean S<1uarcd Error 7, ,89).2 7, foll prc-1rcatmcm period: ;\lean Sq11:ircJ Error 3, , ) J ~oles: After getting a mcasul't' of fi.1 for each slate, an unweighted a\ crai;c is iakcn to arrl\'c at a single m easure of fil.t raining Period from 1977 lhrnugh RTC ycur - 6: Validation Period from RTC ycar - 5 through RTC year- Table A12: Comparison of Fit Across Various Lagchoices - Define Fit Using Mean Absolute Difference three lags yearly lags one lag a\'cragc one lag fimd pre-treatment year lrjining pcrioll: :\ikan Absolu1c Oiffcn-ncc valid:uton period: :-.kan Absolu1c Difference fu ll prc-ln:atmcnl period:.\lean Absolute Diffcrrncc :'\lo[es: Af1er r cuing a measure nf fit frir i:ach SliUC, an unweighted a, eragc is 1a.ken O amvc at a singli: m c,isurc or fit.trainm g Period frnm 1977 t.hm ugh RTC ye;1r. 6: ValirJauon Period frnm RT\ year. 5 thrnugh RT(" yc.ir Table A13: Comparison of Fit Across Various Lagchoices - Define Fit Using CVRMSPE training period: CVR.\SPE three lags. 12 yearl y lags O. O one lag a, erage. 15 one l;1g fin.ii prc-u-c:umcm yi:ar O. J v:ilid:u ion period: CVR.\ SPE.25.'.?3.26.'.?-1 full pre-tre:mucnt period: C VR~,,SPE Notes: Arter gelling a measure or fit for each Male. an unweighted a, crage ls 1aken lo an ivc at a sin~lc measure of fit.training Pcrind from 1977 dtrough RTC year. 6; Valid::uiun Pcnod from RTC yc;tr - S du ough RTC year. t U. 18 Table A14: Comparison of Fit Across Various Lagchoices - Define Fit Using Mean Squared Error thrtt b 1~ ye;uly bg, oirbg :1, c:r.ii:c:.,...,:1 ii lpr~-wc:i m<nc ; c:u lr.lini11; p,,ncd: :\k1... Squ::iml EmH-.-::aliwllo n penal; :\n n Squ.,,rc,d Error US7.JJ.SS9.6J -l.21& li.:67.6.l 6.!!2.9S J..l9!.ll fullpn:-trulme:11tpcriocl: :\k:111 Squ1rrdEnw ;'} J.S.lX.9 fi. llu2 S.716.! l Table AlS: Comparison of Fit Across Various Lagchoices - Define Fit Using Mean Absolute Difference Lr.uru 11!f p~rioj:.\t~;u Abrnlul~ D,U~renc:~ v;:ilid::i.l.um pc,nod: :\le::1.11 Ab,olute Di1l ~n::ice Table A16: Comparison of Fit Across Various Lagchoices - Define Fit Using CVRMSPE tr.umng pc:,iod; CV R;\Sl'E n l1dj.ufl!1 p:,rioj: CVR.\SPf.: full pn:-tn::1unc1n penod: CVR.\ SPE lhrc:e l::i.~, y,:::11ly b JS on.r ll1, 1t r.i1e oa~ b g lhu t pre-i.-u1m~111 )"~M.7.1 O.lJ O.? Li Deel. Ex

236 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page of 56 of 292 Page D #:789 Table A17: Crime Predictor Means Before RTC Adoption treated Synthetic: 3 lags Symhetic: yearly lags Synthetic: lag avg Synthetic: lag fina l pre-treatment year popstalecensus 7,459, ,26, ,479, ,594. l_incarc_rate l_policeemployeerateo rpcpi 12, , , ,439.3 rpcui rpcim rpcrpo 1, , , , unemployment_rale poverty _rate density age_bm_ l age_bm_ age_bm_ age_wm_ age_wm_ age_wm_ For each treatment state, the predictor of interest is averaged over all pre-treatment years ( 1977 through RTC year - ) a population weighted averageof this statistic is then taken over all treatment states to reach the figures presented 9, 16 1, , , Appendix : Synthetic Control Estimates Using Other Sets of Explanatory Variables. Synthetic Control Estimates Using the BC Explanatory Variables Table A18 provides synthetic control estimates of the impact of RTC laws on violent crime using the BC model's set of predictors. 54 This model estimates that RTC laws increase violent crime consistently after adoption, rising to 13.3 percent after ten years (significant at the.1 level). This tenth-year effect is also quite close to the corresponding DAW model's synthetic control estimate (Table 9), as well as the DAW and BC panel data models' dummy variable coefficients (Tables 4-5). Table A18: The mpact of RTC Laws on the Violent Crime Rate, BC covariates, ( ) (2) (3) (4) (5) (6) (7) (8) (9) (1) Average Normalized TEP [4 ' "' ' " " ( 1. 17) ( 1.488) (1.99) (2.16) (2.458) (3.8) (3.184) (4.46) (3.828) (3.42) N Standard errors in parentheses Column numbers indica1c posl pass.igc year un<ler consiucration: N = number of states in sample Dcpcndenl \'ariablc is the difference he1wecn the percentage difference in 1hc vio lcnl crime ralc in treatment and synthetic control!>lates 31 gi\'cn post-crca.1mcn1 imcrval and 31 time of the treatment Results rcponcd for the constant tcnn resulting from this regression Stares in group: AK AR AZ CO FL GA D KS KY LA Mc Ml MN MO MS MT NC ND NE NM NV OH OK OR PA SC SD TN TX UT VA WV WY p <. lo.. p <.5... p <.1 54 For certain treatment states with O executions prior to RTC adoption, the synthetic control program is unable to generate a counterfactual unit. To resolve this problem, and to maintain consistency i.n the process of generating a counterfactual unit for the 33 treatment states, the executions variable is dropped from the BC model in the synthetic controls analysis. 92 Li Deel. Ex

237 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page of 56 of 292 Page D #:79. Synthetic Control Estimates Using the LM Explanatory Variables n our Part panel data analysis, we saw that RTC laws were associated with significantly higer rates of violent crime in the DAW model (Table 4), the BC model (Table 5, Panel A), and the MM model (Table 7, Panel A), but not in the LM model (Table 6, Panel A), although both the LM and MM models did show RTC laws increased murder. Table A19 estimates the impact of RTC laws on violent crim~ using the LM specification. 55 The detrimental effects of RTC laws on violent crime rates are statistically significant at the.5 level starting five years after the passage of a RTC law, and appear to increase over time. The treatment effects associated with violent crime in Table Al9 range from 11. percent in the seventh post-treatment year to 12.8 percent in the tenth posttreatment year. Remarkably, the DAW, BC, and LM synthetic control estimates of the impact of RTC laws on violent crime are nearly identical (compare Tables 9, Al8, and Al9), and this is true even when we limit the sample of states in the manner described in Tables Table A19: The mpact of RTC Laws on the Violent Crime Rate, LM covariates, ( ) (2) (3) (4) (5) (6) (7) Averoge Normalized TEP " 4.599' 7.97" 7.687" 1.984"' (1.247) (1.623) (2.77) (2.298) (2.61 8) (3.2 11) (3.185) N (8) "' (3.86-1) 3 1 (9) "' (3.699) 3 1 (1) "' (2.723) 3 1 Slandard errors in parcn1hcses Column numbers indicate pos1-passagc year under consideration: N = number of slates in sample Depende nt variable is the diffcrcnt..'c bc1wccn the pcrccn1 agc difference in the violent crime r.llc in treatment and synthetic control s1a1cs al gi\ cn post-treatment interval and at ti me of Lhc trea1mcn1 Results rcponcd for the constant tcnn rcsuhing from 1his regression S1a1cs in group: AK AR AZ CO FL GA 1 KS KY LA ME Ml MN MO MS MT NC NO NE NM NV OH OK OR PA SC SO TN TX UT VA WV WY ' p < p<.5,"' p <.1. Synthetic Control Estimates Using the MM Explanatory Variables Table A2 provides synthetic control estimates of the impact of RTC laws on violent crime using the MM predictors. 57 The table reveals that RTC states experienced overall violent crime rates that were roughly 15 percent greater than those of their synthetic controls ten years after passage, which was statistically significant at the.1 level. The similarity of the DAW, BC, LM, and MM 55 The modified panel data analyses of LM and MM, shown in Panel B of Tables 6 and 7, did find RTC laws increase violent crime. n conducting the LM panel data analysis, we used the violent and property arrest rates rather than the crime-specific arrest rates described by Lott and Mustard (1 997) owing to the fact that th is would essentially (and improperly) place the same variable on both sides of the regression model. This objection is less important under the synthetic control framework. For this reason, we use their contemporaneous crime-specific arrest rates in our synthetic control model using the Lott and Mustard ( L 997) control variables. 56 The tenth-year effect in the synthetic controls analysis using the LM variables is 12.5 percent when we eliminate the states with more than twice the average CY of the RMSPE. Knocking out the six states with above-average values of this CV generates an almost identical 12.6 percent effect. We also estimated the impact of RTC laws on violent crime using the synthetic controls approach and the LM model modified to use six DAW demographic variables. This change increased the esti mated tenth-year increase in violent crimes from 12.8 percent to 15.3 percent. 57 For the same reasons described in footnote 55, we use the lagged violent or property crime an est rate in our regression tables but use the contemporaneous violent or property crime aitest rate as a predictor in our synthetic controls code for the MM specification. 93 Li Deel. Ex

238 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:791 synthetic controls estimates of the impact of RTC laws on crime is striking. Moreover, these four sets of estimates are remarkably consistent with the DAW and BC panel data estimates of the impact of RTC laws, which bolsters the case that the DAW and BC panel data specifications provide more reliable estimates of the impact of RTC laws on violent crime than either the LM or MM models. 58 Table A2: The mpact of RTC Laws on the Violent Crime Rate, MM covariates, ( l) (2) (3) (4) (5) (6) (7) (8) (9) (1) Average Nonnalized TEP.67 l " 4.78" 7.575"" 8.196"" " (l.186) (l.535) (l.833) (2.366) (2.832) (3. 171) (3.236) (3.999) (4.246) (3.796) N Standard errors in parcnthc-.cs Co lumn numbers indicate post-passage year under consideralion: N = number of slates in sample Dependent variable is the difference between the percentage difference in the violent crime r:ite in trc.1tmcnt and synthetic control slates at given post-treatment interval and al time of the tr~atmcnt Results reported for the constanl term resulting from this regression States in group; AK AR AZ CO FL GA D KS KY LA M E Ml M N MO MS MT NC ND NE NM NV OH OK OR PA SC SD TN TX UT VA WV WY. p <.1. p <.5. /J <.1 Turning our attention to property cnmes, we find little systematic evidence that RTC laws influence property crime in the synthetic control approach, as our aggregate property crime results are never significant. 58 As we have seen previously, leaving out states with larger CVRMPSEs barely changes the results: Eliminating states with twice the average CVRMSPE leads to an estimated tenth-year effect using MM variables of 15. percent, and eliminating those with above-average CVRMSPE values leads to an estimated effect of 14.7 percent. We also estimated the impact of RTC laws on violent crime using the synthetic controls approach and the Mt\tl model modified to use six DAW demographic variables. This change increased the estimated tenth-year increase in violent crimes from 15.3 percent to 15.4 percent. 94 Li Deel. Ex

239 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 5239 of 56 of 292 Page D #:792 Appendix J: The Contributions of Donor States to the Synthetic Controls Estimates - Evaluating Robustness One of the key elements of the synthetic controls approach is that, for each state adopting a RTC law in year X, the approach searches among states that do not have RTC laws through at least ten years after X-including never-adopting states-to select a plausible set of control states for the adopting state. Figure A34 lists all the states that are eligible, under this criterion, to serve as synthetic controls for one or more of the 33 adopting states, and shows how often they are in fact selected. The horizontal length of each bar tells us how much, on average, that state contributed to the synthetic controls in our violent crime estimates. As the Figure indicates, Hawaii appears most frequently-contributing to a synthetic control 18 of the 33 times it is eligible-and it has the largest average weight in the synthetic controls, of percent. Figure A34 Frequency of Potential Donor States to Appear as Synthetic Controls in Violent Crime Estimates TX (1996) 11 TN {1997) /3 SC (1997) /3 OK {1996) Oi1 N V (1996) 11 NM (24) /11 NC (1996) /1 LA (1996) /1 KY (1997) /3 AR (1996) 11 NJ(NA) 1133 MA (NA) 1133 MO (2') KS (27) P?1 2/22 MN {23).,... 1,11 co (23) NY (NA) 11 (21) OH (24} Rl (NA) MO{NA) OE(NA) CA(NA) NE (27) L (214 ) A (211 ) W (212) Hl (NA) / / / 1/ / / D Adopted ta Adopted Adopted Never Adopted Average Percent Contribution to Synthetic Controls The numeralor of the fraction counts lhe number of instances a state appears as a contra& unit The denominator of the fraction counts tho number of instances a stalo is elfgible to bo in lhe contr ' unit The cok)r codes Klentify if and when a Slate appe.;iring a, a synlhetic control went on to subsequentty adopl a RTC law 25 Given that Hawaii makes such a large contribution as a donor state in the synthetic controls estimates, and this small state might be unrepresentative of the states for which it is used as a control, one might be concerned that it might be unduly skewing the estimates of the impact of RTC laws on violent crime. To address this, as well as the analogous concern for other control states, we generated 18 additional TEP estimates, with each one generated by dropping a single one of the 18 states that appears as an element of our synthetic controls analysis (as identified in Figure A34). The results of this exercise are presented in Figure A35, which shows that our estimated increase in violent crime resulting from the adoption of a RTC law is extremely robust: All 18 estimates remain statistically significant at the 1 percent level, and the smallest TEP, which 95 Li Deel. Ex

240 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page Page 5124 of 56 of 292 Page D #:793 comes from dropping New York as a control state, is 11.9 percent. Figure A35 Estimated ncrease in Violent Crime 1 Years After RTC Adoption, Dropping One Donor State at a Time N Estimated ncrease with No States Dropped: 14.7% Q) LO E...:: (.)... C: Q) > C: -... Q) en ro Q)... (.) C: -;ft LO...,..._ H W A L NE CA DE MD R OH Ml NY CO MN KS MO MA NJ State Dropped from Donor Pool This graph shows the overall synthetic-controls estimate of the impact of RTC laws on violent crime ten years after adoption when baning individual states from inclusion in the synthetic control. (The horizontal line shows the estimate when no states are barred.) The states are atrnnged in declining order of average cont.ibution to synthetic controls (see Figure A34), from a high of 21.5 percent for Hawaii to a low of.5 percent for New Jersey. 96 Li Deel. Ex

241 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:794 Appendix K: Does Gun Prevalence nfluence the mpact of RTC Laws? The wide variation in the state-specific synthetic control estimates that was seen in Figures 6 and 9 suggests that greater confidence should be reposed in the aggregated estimates than in any individual state estimate, as averaging across a substantial number of states will tend to eliminate the noise in the estimates. Another way to distill the signal from the noise in the state-specific estimates is to consider whether there is a plausible explanatory factor that could explain underlying differences in how RTC adoption influences violent crime. One possible mechanism could be that RTC laws will influence c1ime differently depending on the level of gun prevalence in the state at the time of adoption. Figure A36 The mpact of Gun Prevalence on the ncrease in Violent Crime Due to RTC Laws (Synthetic Control Estimates, ) FL TN "' SC AR MN Ott "" Ml CO SD JiJ'- VA 1 GA "' MO Al( w, LA ME UT Average Fraction of Suicides Committed w ith a Firearm in th e 3 Years Prior lo RTC Adoption Nole: The Synthetic Control Treatment Effect displayed is for lhe max{7lh, 8th. 9th, 1th) year after RTC adoption Treatment Effect = Gun Prevalence. l = 1.54 ; R'2 =.7. Regression weighted by population Figure A36 shows the scatter diagram for 33 RTC-adopting states, and relates the estimated impact on violent crime to a measure of gun prevalence. (Gun prevalence is proxied by the commonly used measure showing the fraction of suicides in a state that are committed with guns.) The last line of the note below the Figure provides the regression equation, which shows that the gun prevalence proxy is positively related to the estimated increase in crime, but the coefficient is not statistically significant (t = 1.54) and the R 2 value is very low. 59 The population-weighted 59 A bivariate regression that weights by the inverse of the CV of the RMS PE, rather than by state population yields results substantively identical to those in Figure A36. We also repeat this analysis when dropping the 5 states with the worst pre-passage fit (NE, WV, MT, SD, and ND), and this modification again does not substantively change the Figure A36 regression results. 97 Li Deel. Ex

242 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:795 mean gun proxy level across our 33 states is.64 (roughly the level of Montana), which would be associated with a 14 percent higher rate of violent crime 1 years after RTC adoption. 98 Li Deel. Ex

243 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:796 Appendix L: The Murder and Property Crime Assessments with Synthetic Controls Because the synthetic controls estimates of the impact of RTC laws on violent crime uniformly generate statistically significant estimates, we have heretofore focused on that analysis. Our synthetic control estimates of the impact of RTC laws on murder and property crime appear in Tables A3-Al of the appendix. While in all cases the tenth-year effect for these crimes is positive, in no case is it statistically significant at even the.1 level. For murder, the point estimates suggest an increase of 4-5 percent, and for property crime, the point estimates range from 1-4 percent mcreases. The relatively smaller impact of RTC laws on property crime is not surprising. Much property c1ime occurs when no one is around to notice, so gun use is much less potentially relevant in property crime scenarios than in the case of violent crime, where victims are necessarily present. Most of the pernicious effects of RTC laws-with the exception of gun thefts-are likely to operate far more powerfully to increase violent crime rather than property crime. The fact that the synthetic controls approach confirms the DAW panel data estimates showing that RTC laws increase violent crime while simultaneously showing far more modest effects on property crime (thereby undermining the DAW panel data estimate showing substantial increases in property crime) may be thought to enhance the plausibility of the synthetic controls estimates. But then what are we to make of the relatively small estimated impact of RTC laws on murder? This might seem to be at odds with our theoretical expectations, and in conflict with the estimated increases in overall violent crime since one might expect violent crime and murder to move together. Part of the explanation is that we are able to get more precise estimates of the impact of RTC laws on violent crime then for the far less numerous, and hence much more volatile, crime of murder. ndeed, the standard errors for the synthetic controls estimate of increased mmder in the tenth year is 25 percent higher than the comparable standard error for violent crime (compare Table 9 with Table A3). But a second and more important fact is also at work that likely suppresses the true estimated impact of RTC laws on the murder rate. We know from Table 2 that RTC states increased police employment by 8.39 percent more in the wake of RTC adoption than did non-rtc states. This suggests that our estimates of the crime-increasing impact of RTC laws are biased downward, but since police are more effective in stopping murder than either overall violent or property crime, the extent of the bias is greatest for the crime of murder. n other words, the greater ability of police to stop murders than overall violent ( or property) crime may explain why the synthetic controls estimates for murder are weaker than those for violent crime. An increase in police employment of 8.39 percent would be expected to suppress murders in RTC states (relative to non-rtc states) 99 Li Deel. Ex

244 Case 2:16-cv-6164-JAK-AS Case: , 1/2/218, Document D: , Filed DktEntry: 9/11/ , Page of 56 of 292 Page D #:797 by about 5.6 percent. 6 Since the synthetic controls approach does not control for the higher police employment in the post-adoption phase for RTC states, it may be appropriate to elevate the synthetic controls estimates on murder to reflect the murder- dampening effect of their increased police presence. To adjust our synthetic control estimates of the impact of RTC laws on murder to reflect the post-adoption changes in the rates of police employment and incarceration, we can compare how these crime-reducing elements changes in the wake of adoption for our RTC-adopting state and for the synthetic control. Consistent with the panel data finding of Table 2 that police and incarceration grew more post-rtc- adoption, we found that, over the 33 models using the DAW covariates and murder rate as the dependent variable, the population-weighted average percent change in the incarceration rate from the year of adoption to the 1th year after adoption (the 7th year after adoption for Kansas and Nebraska) is 28 percent for the treated unit and 19 percent for the synthetic control unit. For the police employee rate, the analogous numbers are 9.1 percent for the treated unit and 7.2 percent for the synthetic control unit. 61 We correct for this underestimation by restricting the synthetic control unit to have the same growth rate in incarceration and police as the treated unit. 62 Once we have computed an adjusted murder rate for the 31 synthetic control units in the 1th year after adoption, we then use the formula described in part V to construct an adjusted aggregate treatment effect. 63 The impact of controlling for police and incarceration are substantial: the 1th year impact of RTC laws rises from 4.68 percent (t = 1.28) to 9.75 percent (t = 1.98). 64 n other words, the ostensible puzzle that 6 The important recent paper by Professors Aaron Chalfin and Justin McCrary concludes that higher police employment has a dampening effect on crime, and, most strikingly, on murder. Specifically, Chalfin and McCrary (213) find elasticities of -.67 for murder but only -.34 for violent crimes and -.17 for property crimes of the 33 states experienced growth in the incarceration rate ( 17 /33 for police employee rates) that was greater than their respective synthetic controls growth rate. 62 By comparing the synthetic control unit's adjusted police/incarceration figures with its actual police/incarceration figures, and by applying standard estimates of the elasticity of murder with respect to police (-.67) and incarceration (-.15), we can create an adjusted version of the control unit's murder rate for each year after RTC adoption. For example, if the adjusted police and incarceration rates for the synthetic control unit were both 1 percent greater than the actual rates in the 1th year after adoption for a RTC-adopting state, we would adjust the murder rate for the synthetic control unit downwards by.67* * 1 = 8.2 percent (thereby elevating the predicted impact of RTC laws on murder). 63 Kansas and Nebraska, both 27 adopters, have no comparable data for 1 years after adoption and are thus not included in this calculation. 64 f one only con-ects for the larger jump in police expe1ienced by the treatment states, the 1th year effect jumps from 4.68 percent (t = 1.28) to 7.77 percent (t = 1.7). The 9.75 percent estimated jump in the murder rate in the text results from restricting the synthetic control unit to have the same post-adoption year growth rate in police and incarceration as the treated unit. One can also try to control for differential post-adoption movements in police and incarceration by focusing on the post-adoption change in the levels of the police employee rate and the incarceration rate. When we constrain the post-adoption change in police and incarceration between the treated and synthetic control unit to be the same O yea.rs thereafter, the aggregate 1th-year effect is 9.94 percent (t = 2.8). Using this second technique, if one only con-ects for the larger jump in police experienced by the treatment states, the 1th-year effect is 8.6 (t = 1.83). 1 Li Deel. Ex