What Do Prosecutors Maximize? An Analysis of the Federalization of Drug Crimes

Transcription

1 What Do Prosecutors Maximize? An Analysis of the Federalization of Drug Crimes Edward L. Glaeser, Harvard University and NBER Daniel P. Kessler, Stanford University and NBER and Anne Morrison Piehl 1 Harvard University and NBER Address correspondence to Anne Piehl: 79 John F. Kennedy Street, Cambridge, MA, Phone: (617) Fax: (617) anne_piehl@harvard.edu. Note: This is a preprint of an article published in American Law and Economics Review, vol. 2, no. 2, Fall 2000, pp

2 Abstract Recent legislation has expanded the jurisdiction of the federal government over crimes that were traditionally prohibited only by state law. Previous work has neglected the potentially important role of prosecutors incentives in allocating cases to the federal and state systems. We model the decision-making process of state and federal prosecutors, and, using data from state and federal inmates incarcerated for drug crimes, we estimate empirically the importance to the federal/state jurisdictional decision of the characteristics of defendants, states legal environments, and lawyers labor markets. We find that individuals who hire private attorneys and who are high human capital and successful in the legitimate sector are more likely to end up in the federal system. This is consistent with the model in which prosecutors maximize both the payoffs from eliminating crime and their private human capital. 1

3 I. Introduction Starting at the turn of the century, and at an increasing pace since the 1970s, the U.S. Congress has expanded the jurisdiction of the federal government over crimes that were traditionally prohibited only by state law (e.g., Beale, 1996; Brickey, 1995, 1996). In particular, virtually any drug crime can now be prosecuted federally, regardless of the amount seized or the conduct of the defendant (Guerra, 1995). Because federal penalties for drug crimes are generally more harsh than state penalties (Guerra, 1995; Heller, 1997), the allocation of cases to the federal versus state court systems may have implications for deterrence and also for both vertical and horizontal equity (Curtis, 1996). Indeed, the extensive debates in two recent conference volumes (Hastings Law Review, 1995; Strazella, 1996) attest to the importance of the federal/state jurisdictional decision. This expansion of jurisdiction has generally taken the form of concurrent federal/state jurisdiction under which crimes can be prosecuted in either federal or state court. Although the double jeopardy clause of the U.S. Constitution does not prohibit federal prosecution for conduct already prosecuted by a state, very few cases are subjected to dual federal and state prosecution. For example, the 94 U.S. Attorney s Offices and litigating divisions typically bring fewer than 150 dual prosecutions of all types per year (Litman and Greenberg, 1996). Anecdotal evidence suggests that this jurisdictional decision over crimes which could conceivably fall within either jurisdiction is largely in the control of federal prosecutors, due to the growing federal share of prosecutorial resources (e.g., Jeffries and Gleeson, 1995). Despite written U.S. Department of Justice Guidelines on this topic (U.S. Department of Justice 1992, Sections 9-27 and for drug crimes), even members of the federal executive branch acknowledge the wide discretion of federal and state prosecutors (Gorelick and Litman, 1995). Other commentators, including two federal judges (Miner, 1989; Jeffries and Gleeson, 1995), emphasize the influence of federal prosecutors. Some legal scholars go even further, describing the jurisdictional discretion of U.S. attorneys as virtually unreviewable and almost unchecked (see Hollon 1996, p. 501 and Heller 1997, p. 1314, respectively). 2 2

4 This paper extends the growing literature that understands that legal actors, including prosecutors, may not simply maximize social welfare (e.g., Schulhofer, 1988; Posner, 1995). The incentives facing prosecutors certainly lead them to take actions to minimize the social costs of crime. However, prosecutors also may seek to further their careers, both in the private and the public sector. For example, prosecutors may seek to try highprofile criminals, either to obtain political benefits, or to develop their legal skills and connections in the private sector by arguing against skilled private attorneys. Our primary goal is to understand systematically the implications of prosecutorial maximization for federalization in a system in which federal prosecutors have greater resources than their state counterparts. We begin with a model of how state and federal prosecutors make jurisdictional decisions. In the model, prosecutors select which cases to pursue. Both types of prosecutors seek to reduce crime, perhaps because of direct incentives or perhaps because a reputation as a crime fighter is helpful in running for higher office. Both types also care about potential careers in the private and public sectors. These different incentives create a desire among prosecutors on the one hand to pursue the most dangerous criminals, and on the other hand, to pursue criminals who will bring them private career returns. We refer to these two types of incentives and two types of cases as crime reduction and career concerns cases and incentives. 3 The model illustrates that asymmetries in resources in this context can lead federal prosecutors to try a disproportionate share of career concerns cases. Then, we explore this and other implications of the model with an empirical analysis of the determinants of prosecutors decisions to allocate drug cases to states versus the federal systems. We use two separate data sources. First, we use data from the 1991 Surveys of Inmates of Correctional Facilities. These surveys enable us to examine a large set of recent prison admissions in both state and the federal systems. With these data, our objective is to determine the characteristics of individual inmates associated with confinement in a state prison as opposed to in the federal system. Second, we use the 1991 National Corrections Reporting Program (NCRP) data to investigate the characteristics of states associated with the federalization of criminal punishment. Section II presents our model of incentives and prosecutor behavior, and suggests how crime reduction and career concerns affect prosecutors jurisdictional decisions. Section 3

5 III describes our data sources and empirical models, and presents our empirical results. Section IV concludes with our observations about the importance of various determinants of prosecutors decision-making processes and their implications for the consequences of the federalization of drug crimes. II. Model What determines the allocation of cases, drug cases in particular, between U.S. and state attorneys when criminals can be tried under either Federal or state law? Federal, State, and local agencies share responsibility for enforcing drug laws, although most arrests are made by state and local authorities (Bureau of Justice Statistics, 1995). This means that the state or local attorneys generally have the option to prosecute many more cases than they actually do. Our framework models the demand and supply conditions that determine the willingness of state attorneys to cede cases and the willingness of Federal attorneys to accept them. We assume that there are two prosecutors: a U.S. attorney and a state attorney. Both attorneys have responsibility for and control over an initial allocation of cases. Each attorney can allow the other attorney to prosecute some of his cases, but he must ensure that his initial allotment of cases is prosecuted by some attorney. 4 We assume there are no dual federal/state prosecutions. 5 We have little to say about the initial allocation of cases, and technically we will assume that the characteristics of Federal and state arrestees have the same characteristics. More realistically, we suspect that the allocation of arrests is the result of the same career concerns on the part of arresting agents that we analyze for prosecutors. The Attorneys Objective Function: Both attorneys internalize some of the social gains from reducing crime, because they face incentives within the government to reduce crime (from either voters or their superiors). Expected crime is denoted G G 1 where G 0 and G 1 are parameters, π S 0 j j j, π j is the probability the case j will be successfully 4

6 prosecuted and Sj represents the marginal impact of any particular case of the level crime. This case-specific effect of crime is meant to include both the deterrence and incapacitation benefits from incarceration. 6 Total utility declines by θ times the crime rate. There are also private career benefits from each case, which are denoted Cj; attorneys receive these benefits only if they successfully prosecute the case. Career capital is meant to include building a reputation (which has value in both the public and the private sector), exposure to private attorneys, and skill. The probability of success is a function of the average amount of resources the attorney possesses per case or Ω / N, where Ω represents total resources available to the attorney. The returns from career capital will equal R so total expected utility is: (1) Ω Ω π RC Your Cases j θ G G1 π S All Cases N N 0 j. We will only consider the transfer of cases from the attorney with fewer resources (the state attorney) to the attorney with more resources (the U.S. attorney). A state attorney is willing to give up a case z if: (2) Ω π N S S RC z < π F S θg S + δ RC + 1 z State Cases S ( j θ 1 G S j ), where δ i Ωi Ωi = π π N i 1 N (the marginal increase in probability of success from i reducing the caseload) and π F S Ω F Ω S = π π N F + 1 N > 0 (the increase in the S probability of success from transferring the case to the federal attorney). 5

7 The cost of foregoing the case is the lost career returns on the left hand side of (2). The benefits are that the state attorney conserves more resources for other cases and that the case has a higher probability of being successfully prosecuted under the federal attorneys who have greater resources. In Figure 1, the line referred to as the State Attorney Offer Curve described the values of social and private returns for which equation (2) holds with equality. The state attorney is willing to forego all cases that are below that line because the state attorney prefers to free up time. The line slopes upward to reflect the fact that as the social value of the case increases, the state attorney is more willing to forego cases because it is more important that the case end up in the hands of the federal attorney. Many state attorneys (in discussion with one of the authors) have claimed that they are willing to let the U.S. attorneys prosecute any case that they have. If this is true, then inequality (2) holds for all cases, and that it is only the willingness of U.S. attorneys to accept cases that determines which cases become Federal. 7 Federal attorneys are willing to take a case if: Ω F π RC N F + 1 z + π θg1s z > δ F ( RC j + θg1s j ) F S Federal Cases (3) Thus, federal attorneys are likely to accept cases with higher career returns and with higher social returns. The line where this equation holds with equality is again in Figure1 and U.S. attorneys will accept all cases in the region above the Federal Attorney Acceptance Curve. The slope of this line lies between 0 and 1, and seems unlikely to be far from zero (to the extent that the difference between federal and state attorneys probability of success is small). If the magnitude of the slope is low, this means that the U.S. attorney s willingness to take cases is determined almost entirely by career concerns. As long as π > 0 and F S 6

8 δ S j + θg S j ) > δ F ( RC j + State Cases RC S ) (which both relate to the ( 1 θg Federal Cases 1 scarcity of resources in state attorney s office), then some cases will be transferred to the Federal attorney and no cases will be transferred to the state attorney. j Because the primary determinants of whether a state attorney is willing to cede a particular case that a federal attorney is willing to accept are related to career concerns, observing the cases transferred to federal attorneys suggests what types of cases yield private, career returns. As such, the model shows that by observing the cases that the federal attorneys select, we can learn about what prosecutors maximize. The framework also offers some simple comparative statics. The tendency of U.S. attorneys to take the high career return related cases (and not to care all that much about the social returns to the case) will increase when the returns to the private sector are high (i.e. when R is big relative to θ). We explore below this implication of the model. 8 This allocation system creates two types of social losses. First, Federal and state prosecutors may not be handling the right number of cases. In the presence of career concerns, the Federal attorneys may wish to keep their probabilities of success high by keeping their caseloads below those of the state attorneys. Second, holding fixed the number of cases that will be handled by the state and federal attorneys, there will be a distortion in the allocation of types of cases. For crime reduction, the high social damage cases should be given to the more effective federal attorneys. However, as we have just seen, career concerns may be the dominant force driving the distribution of returns. This key result holds in many different versions of this model, including versions where (1) the state and Federal prosecutors cooperatively divided the caseload, but the federal prosecutors have more bargaining power and where (2) the Federal prosecutors choose their cases first. The interplay between career concerns and Federalism is potentially important. It suggests that an ideal division of labor between different levels of government that is based on 7

9 social returns will not be implemented when the objective function of the prosecuting attorneys is not the objective function of society at large. In this model, the lawyers are not venal and they certainly care about crime reduction. However, because of career concerns and the differences in resources across attorneys, the federal attorney ends up taking too many easy, career-related cases, and the state attorney ends up over burdened with low career-return cases. III. Empirical Analysis We rely on two sources of data: a survey of inmates (which includes extensive questions about a variety of background characteristics) and data on admissions to prison collected from correctional agencies. We use the individual-level survey data to compare the attributes of state and federal prisoners and to examine whether federal prosecutors pursue defendants who are likely to produce contacts with private attorneys. We test several alternative hypotheses, such as whether federal inmates are more difficult to supervise or have committed offenses that lead to longer terms of incarceration. We use the aggregated data to compare the characteristics of states sending large numbers of people to federal jurisdiction to states with low rates of federalization. In particular, we try to explain variation in federalization using attributes of the criminal justice environment in the state and proxies for private returns to prosecutors. Data Sources The first source of data is the Surveys of Inmates of Correctional Facilities. The data come from two surveys of prison inmates collected by the Bureau of the Census for the Bureau of Justice Statistics in July The two surveys are essentially the same, though the sampling frames are different, with federal inmates over-sampled relative to state inmates. The surveys record a broad range of information from a sample of inmates on a given day. Questions cover individual characteristics; current offense and sentencing; criminal history; involvement with guns, drugs, and alcohol; and experiences during the current term of incarceration. 8

10 We have limited the sample in a number of important dimensions. We were interested in only those inmates with new offenses that could conceivably have been prosecuted in either federal or state court. Therefore, we included only those offenders with new drug offenses. This required us to drop those with no sentencing information, those who violated probation or parole without committing new offenses, and those who were convicted of offenses other than drug violations. Non-citizens are also omitted from the sample due to concerns about differential treatment by the federal system (Butcher and Piehl, 1999). For most analyses, we only included those who had been incarcerated for less than two years. We did this for two reasons. First, we wanted to be sure that our sample was from the "post-guidelines" era. Second, we wanted to limit the extent of survivorship bias (the excessive contribution to the estimates of those incarcerated for long terms). Unfortunately, we do not have adequate information from the prison inmate survey to actually determine the states where the federal inmates were arrested. As such, it is impossible to examine the extent to which state characteristics, such as differential punishment levels, contribute to the level of federalization among drug offenders. (Also, the number of observations per state is quite small.) However, the National Corrections Reporting Program collects information on state and federal prisoners, including information on the state of arrest, from the populations confined in correctional institutions. 10 The data cover all admissions to state and federal prisons during the calendar year Since participation by states is voluntary, we have a census, but for a subset of states. 11 Results from The Surveys of Inmates of Correctional Facilities Before turning to our results, we want to provide a little context for our discussion of drug offenders and jurisdiction. As reported in Table 1, federal inmates comprise seven percent of the flow of inmates and 7.4 percent of prison population (stock). Among those convicted of drug offenses, the federal role is much larger: among the population of inmates, 18 percent are in the federal system. As is well known, the population of inmates convicted of drug offenses has risen tremendously in recent history: an increase of over 400 percent from Since 1985, the federal share of drug prisoners has had no clear trend, though it fell substantially (to 15 percent in 1990) before rising again. 9

11 Currently, over 60 percent of the inmates in the federal prison system have been convicted of drug crimes (c.f., 23 percent of state inmates). Table 2 shows the means of our variables for our sample of drug offenders from the Surveys of Inmates. 12 We have divided the survey into inmates in federal prisons and inmates in state prisons to look for any patterns in the types of criminals who are selected into the federal and state systems. Since we are only looking with the category of drug crimes, which can almost always be prosecuted on either the federal or the state level, we believe that we will avoid most of the most obvious effects coming from the different crimes that fall into the state and federal systems. Overall 12.2 percent of our prisoners are in federal prisons. This should be compared with our 1990 figures for Table 1, which shows that 16.8 percent of drug inmates during that year were incarcerated in federal prisons. The discrepancy is accounted for by the fact that we are looking at relatively short-term prisoners (i.e., individuals who were incarcerated no more than 24 months before the date of the survey), who fall disproportionately into state systems. The first panel of Table 2 shows the basic demographic differences between state and federal inmates. Federal inmates are older, slightly more likely to be female, much more likely to be white (60.9 percent vs percent), more likely to be Hispanic 13, more likely to be currently married and less likely to be never married. In every one of the cases, the differences are strongly statistically significant. Perhaps the most striking difference are the race and marriage effects where married white men are much more likely to end up in the federal system than single non-whites. The second panel examines education, income and occupation. Federal inmates are much better educated than state inmates. For example, only 27.7 percent of federal inmates are high school dropouts where 46.3 percent of the state inmates are high school dropouts. Furthermore, federal inmates were much more likely to have worked (pre-arrest) and when they worked they were managers or technical workers rather than service workers. The annual income of the federal individuals is more than 25 percent higher. In general, the federal criminals were much less likely to admit to having illegal income, either because they actually didn't have illegal income or because of reporting reasons. 14 If we 10

12 are to believe these results, they suggest that while federal criminals may or may not have been successful in the criminal sector, they were certainly more successful in the legal sector. When we examine the attributes of the current term, there are also significant differences. The federal prisoners are more likely to be convicted of trafficking (perhaps because traffickers cross state boundaries more often) and less likely to be in on possession offenses. Of course, the current term sentence may differ from the actual crime committed. If local district attorneys are more likely to plea bargain, we wouldn't be surprised to find more traffickers in the federal system because it is the tougher sentence. The federal prisoners are less likely to have an additional offense and, despite the guidelines, have a similar average sentence. Although the sentence lengths are similar, the expected time served is very different. This is because the federal system requires inmates to serve at least 85% of their sentences, while state systems generally require much less. The average percentage of sentence served for drug traffickers released from prison in 1994 was 38% (Langan and Brown, 1997). Applying these percentages to our data, the mean expected time served is 41 months for state inmates and 88 months for federal inmates. The federal prisoners were also much less likely to plead guilty at their trial, which supports the plea bargaining hypothesis or suggests that the richer and white federal prisoners have better access to quality legal help. Since the career advantages of taking a case depend upon facing an opponent who could help spread your reputation, we would expect the federal attorneys to face criminals with many more outside attorneys and indeed they do. The proportion of federal inmates who hired their own attorney is 51.6 percent. The comparable proportion of state inmates is 28.5 percent. While this is possibly the result of federal inmates hiring an attorney as a result of going into the federal system, it seems more likely that this strong connection is the result of federal attorneys specializing in criminals with private attorneys. If hiring an attorney captures the career advantages of prosecuting a particular criminal, then an inmate s tendency to repeatedly commit crimes suggests the public gains from committing this individual to prison. Federal prison inmates are much less likely to have broken rules in prison. Federal prison inmates are also much less likely to have had a previous violent offense or a previous offense in general. The mean number of times 11

13 served (for both major and minor offenses) by inmates prior to their current term in the federal system is one third of the comparable number in the state system. The state inmates are more likely to have used drugs, have started using drugs at an early age, to have had a family member serve time, or to have been in a mental hospital. Federal inmates are more likely to have owned a gun, and gun ownership is generally associated with relatively successful people, not with destitute criminals (Glaeser and Glendon, 1998). Overall, we find these descriptive statistics quite provocative. Federal criminals are white, married, richer, better educated, more likely to hire an attorney, less likely to break the rules and less likely to have prior offenses. It appears that the U.S. attorneys may not be choosing individuals who are more likely to be repeat offenders or difficult to imprison. Instead they may be choosing the wealthier, more prestigious set of criminals who are more likely to have private lawyers. However, these means do not present us with estimates of the partial effect of each of these variables. In the next section we present multivariate estimates. For the moment, we defer consideration of alternative hypotheses to explain these stylized facts. Table 3 presents our first regression results. The basic regression format is to run a probit, where being in the federal (vs. the state) system is the dependent variable. In this way, we are examining which variables induce the probability of ending up being prosecuted by federal vs. local district attorneys. Of course, we cannot know the conviction rates of these two types of attorneys and how they differ by criminal type. So our work combines the effects of the decision to prosecute and the success probability. The first column of the table shows results for hiring a private attorney, current offense, and criminal history. Federalization is positively associated with having hired a private attorney and with having been convicted of trafficking (rather than possession). Being a federal prison inmate is negatively associated with having one or more additional offenses at the time of sentencing, with having previous violent or other offenses, and with the number of times served in the past for minor offenses. The second column includes controls for standard demographics and education. The pattern of coefficients for the variables in column (1) remains, though the size of the 12

14 coefficient on hiring an attorney is nearly halved. Whites are 6.1 percent more likely to end up in the federal system, which is a huge effect given that the federal system accounts for only 12.2 percent of our inmates. The never married are also much less likely to be federalized-- the effect is 5.1 percentage points. Additionally, schooling is also a strong determinant of the likelihood of being federalized, with college graduates being substantially more federalized than high school dropouts (11.3 percent). Finally, the third column reveals the effects of the additional controls for other past behavior. Federalization is positively associated with never having used drugs, age at first drug use and having ever owned a gun. Being a federal prison inmate is negatively associated with having been in a mental hospital. Most of the basic correlations stay strong and close to the initial estimates when we control these added controls. We are particularly interested in the effect of hiring an attorney (rather than relying on an attorney assigned by the court). We do not believe that whether or not a defendant hires an attorney is unrelated to the jurisdiction of prosecution (due to the greater consequences in the federal system, one would expect greater investment in defense). Nevertheless, the final specification in Table 3 shows that the relationship is still positive and statistically significant even after controlling for a wide range of other variables. Those with private attorneys are 3.1 percentage points (25 percent) more likely to be in the federal system, which is quite large. Due to concerns about endogeneity of hiring counsel, in Table 4 we report the results of using instrumental variables for hiring an attorney. The table reports the results of three two stage least squares specifications with sets of controls similar to those just discussed. For each specification, the first stage is the attorney variable regressed on the controls plus occupation variables. The idea here is that education is proxying for permanent income. Occupation is related to a defendant s knowledge of the legal profession and access to referrals. The first four columns present more parsimonious specifications. Our preferred specification is in columns (5) and (6). Here one sees that our instruments perform well enough the F-statistic for their inclusion is The second stage (column 6) has a full set of controls. Here, the coefficient on hiring an attorney is 0.294, substantially higher than in Table 3 and statistically significant. This is evidence that the positive coefficient in Table 3 was not the result of endogeneity and, rather than being an 13

15 overestimate of the true relationship between having a private attorney and federalization, is more likely an underestimate. The differences in other characteristics between Federal and state inmates are at least as interesting as the tendency to hire private attorneys. When Federal attorneys do not observe the actual attorney chosen in a particular case then these characteristics will help him make his decision. Alternatively, characteristics are important because they are not endogenous with respect to the practices of federal and state attorneys. Finally these characteristics might also tell us something about whether the criminal is likely to be high profile. In order to compare the whole vectors of characteristics between state and federal criminals, we form a series of indices of state and federal inmates. These indices take the form β X, where i i i β i represent the coefficients from regressions where the dependent variables are sentence length, violating prison rules, going to trial and hiring a private attorney. The indices can be interpreted as a measure of the extent to which an individual s characteristics predict that he will be prone to these outcomes. In all cases, the coefficients were estimated from regressions using the state observations alone. For individuals in the federal system, the resulting quantities represent, for example, the predicted sentence that this individual would have received if he had been part of a state system. This avoids some biases that may come about from differences between the state and federal system. By comparing these indices, we are comparing the extent to which the federal system is full of people who were predicted to get long sentences in the state system or who are predicted to hire a private attorney. First, we compare the actual sentences, the extent to which inmates violate prison rules, and the probability of going to trial (rather than settling on a sentence through a plea bargain) and hiring an attorney between the state and federal systems. As columns (1) and (2) show, federal criminals are likely to have shorter sentences (this is not significant), are less likely to be rule breakers, are more likely to have gone to trial and are much more likely to have hired an attorney. For each of these variables there are reasons to believe that simple comparisons of the variables could be biased. As discussed earlier, it is difficult to compare sentence lengths 14

16 across these jurisdictions because nominal sentence length means something different in the federal system than it does in the states. Similarly, the extent of prison rules, and the extent to which they are enforced, may also vary. Finally, for the same reasons as described for the endogeneity of hiring an attorney, the probability of going to trial is likely to be affected by jurisdiction. In columns (3)-(6) we present our indices of predicted sentence length, being a rule breaker, going to trial and hiring an attorney using various control variables. In column (3) we use only current term characteristics and criminal history to form the indices. In column (4), we also include individual demographics. In column (5), we add education and in column (6) we include other past behavior. The columns report the difference between state and federal prisoners in the various indices. The indices are starred when the differences are significant. In column (3), we show that the federal inmates have characteristics that are associated longer terms, by six months. This difference occurs because federal inmates are more likely to be traffickers and because federal inmates are less likely to have prior non-violent offenses. Surprisingly, prior non-violent offenses are negatively associated with sentence length (among inmates of state prisons). In column (4), the gap expands significantly, in part because whites have longer sentences. In column (5), the difference expands because individuals with more education generally receive longer sentences. These facts may be seen as a confirmation of the efficiency hypothesis. As prior nonviolent offenses and higher human capital are generally associated with longer sentences under the state system, these characteristics may plausibly be associated with more dangerous criminals and it is possible that it is plausibly efficient for these criminals to end up in the Federal system. Alternatively, it may be that higher human capital individuals receive longer sentences not because of efficiency reasons but because juries like imposing harsher punishments on these people (perhaps because they are harder to catch). In that case, this evidence could not be interpreted as suggesting the efficiency of the sorting of high human capital people into the Federal system. In row 2, we consider how difficult it is to incarcerate the inmates. Criminals who have characteristics associated with breaking prison rules are both more difficult to incarcerate 15

17 and, possibly, more likely to break laws after release from incarceration. The proportion who broke prison rules (reported in Table 2) was much lower among federal inmates than among state inmates, even though federal inmates had, on average, been incarcerated longer (and thus had more time to transgress). We find that federal inmates still have somewhat lower rule breaking propensities. As more control variables are added, the gap between federal and state inmates becomes larger. This means that federal criminals are not those who are particularly likely to be dangerous in prison. So, while it may be true that federal inmates are particularly dangerous socially, they are not the criminals who are difficult to handle in prison. Rows 3 and 4 help us understand the relationships among federalization, trial experience, and the association with private attorneys. We performed the same prediction exercise for went to trial and hired a private attorney. On average, federal inmates are much more likely to have gone to trial and much more likely to have hired an attorney. These differences are much attenuated once the criminal offense and history variables are taken into consideration. Nonetheless, the gaps are still appreciable and remain highly statistically significant), even with the extensive controls for current and past criminal activity. As above, as more demographic and education controls are added, the gaps between the two sets of inmates increase. In sum, inmates in the federal system are different from those in the state systems in important ways. If they had been under the jurisdiction of the states, we expect federal inmates would have had 5-15 percent longer sentences, been 5-10 percent less likely to exhibit problematic behavior in prison, been 4-14 percent more likely to have been convicted at trial, and been percent more likely to have been represented by a private attorney. These tables have suggested that federal prisoners are more likely to have hired private attorneys and that federal prisoners are the types of people who are particularly likely to have private attorneys. They are also more likely to be high human capital individuals who are likely to be high profile cases that may yield career capital. After all, the strongest image of Rudy Giuliani prior to his run for New York Mayor was handcuffing insider traders in their Wall Street offices. Of course, it may be that these more powerful individuals are able to use their greater wealth to protect themselves against the state 16

18 system, and that only the Federal system has the independence to try more successful persons. Alternatively, as Table 5 showed, these high human capital individuals may be more important criminals who are more likely to cross state lines and are naturally within the jurisdiction of Federal criminals. Results from the National Corrections Reporting Program As discussed earlier, we do not have adequate information from the prison inmate survey to actually determine the states where the federal inmates were arrested. As such, it is impossible to examine the extent to which state characteristics, such as differential punishment levels, contribute to the level of federalization among drug offenders. However, the National Corrections Reporting Program collects information on state and federal prisoners, including information on the state of arrest, from the populations confined in correctional institutions. As before, our data is on the individual level, but here we focus on state-level characteristics. 15 In particular, we try to explain variation in federalization using attributes of the criminal justice environment in the state and proxies for private returns to prosecutors. The first regression in Table 6 shows the results including both individual-level controls and state variables. The results confirm some of our prior findings on federalization and individual-level attributes. Whites and older individuals are more likely to be federalized. 16 State-level attributes contribute little to this analysis. The mandatory sentences at the state level do little to explain the level of federalization. The failure to find strong effects of state mandatory sentences 17 can either be interpreted as meaning that we have measured their effects poorly or that the federal prosecutors do not internalize the effects of their actions that reduce crime. Other controls such as state-level judicial expenditures, the number of state prosecutors, or the overall crime rate all add little explanatory power. Though the effects are generally not statistically significant, it is interesting that the number of state prosecutors per capita and the number of U.S. attorneys per capita always operate in opposite directions. The second regression adds a variable to capture the average earnings of lawyers in the state. This variable is meant to capture one aspect of the returns to private legal practice relative to public service. The assumption needed to justify this is that public returns are 17

19 only imperfectly indexed to state attorney fees. Unsurprisingly, the variable does not affect the overall level of federalization. It also has little impact on the estimates for the other variables. Our primary test of the model depends on whether the composition of admissions is affected by the extent of private returns for U.S. attorneys. We have two candidate measures. First, our theory predicts that when private salaries are higher, U.S. attorneys will be particularly focused on career concerns and facing only prisoners who are likely to hire private attorneys. Second, Eisenstein (1978) suggests that as the number of assistant U.S. attorneys increases the degree to which the office as a whole seeks private gains (relative to public service) increases. The basic argument is that as the number of attorneys rises, the chance of becoming the actual U.S. attorney decreases and therefore lawyers are more focused on jobs outside of the attorneys office. 18 Our objective is to examine whether U.S. attorneys select more strongly into prosecuting cases which are likely to have private attorneys when the returns to private careers are larger. We have only one variable that is reliable and strongly connected with hiring a private attorney in the data set: nonwhite. This race variable was strongly correlated with hiring a private attorney, where nonwhites were much less likely to hire private attorneys. To test the theory, we interact this variable at the individual level with the state average lawyers' salaries and, separately, with the number of U.S. attorneys in the state. The results of including the interactions are presented in the final three columns of Table 6. Column (3) shows the results of including the interaction of U.S. attorneys and nonwhite. As expected, there is a strong negative effect: as the number of assistant U.S. attorneys increases, the more the caseload under-represents non-white defendants (relative to the state system). Column (4) shows the effect of the salary/nonwhite interaction. Again, there is a negative effect of this interaction as the theory suggests that there should be (pvalue=0.077). In states where the private salaries are higher, federal attorneys particularly shun nonwhite criminals. In column (5) the two interactions are entered together. Here, the effect of U.S. attorneys appears to dominate. It should be noted, however, that the number of U.S. attorneys and the average lawyer s salary are highly correlated (r=0.26, p- value <0.0001). The two interactions are jointly highly significant (p-value<0.0001). 18

20 These interaction effects are large. If private lawyers salaries move from $50,000 to $100,000 annually, the connection between nonwhite and federalization more than doubles in size (column 3). This implication was a primary test of our model and it suggests that the tendency toward federalization of crimes committed by white, better educated prisoners may indeed represent concerns for private sector careers, since this tendency is stronger in the areas where private sector careers are more remunerative. Likewise, if the number of U.S. attorneys (per million population) goes from 10 to 20, the connection between nonwhite and federalization nearly doubles in size. One interpretation for this increased connection is that higher salaries increase the incentive to prosecute criminals who are more likely to have their own attorneys. As we discuss in the next section, there are also efficiency-related explanations for the tendency of Federal attorneys to face criminals with private attorneys. Validity Checks In this section we present a variety of specification checks to assess whether our findings of the importance of private attorneys to the assignment of jurisdiction stand up for various subpopulations. We redo a basic specification, regression (2) of Table 3 for a variety of subsamples. The first column just rechecks our results when we exclude all imputed data. None of our results are in any way sensitive to this change. Regression (2) shows the results for only those convicted of drug possession (in either system). This robustness check represents an attempt to abstract from the highest profile cases (e.g., prosecution of drug kingpins ). Among possession offenders, the results are qualitatively similar to those for the full sample of drug offenders: hiring a private attorney is positively associated with the federal system. Furthermore, federal prisoners have fewer private offenses, are more likely to be white, less likely to be never married and more likely to have more education. It is possible that individuals who are being imprisoned for possession are actually supercriminals who have plea bargained when facing more serious offenses. To address this possibility, in regression (3) we include only those individuals who are in prison for possession and who actually went to trial. While the sample sizes become much smaller, the results are qualitatively similar to the results in the other regressions. In fact, the effect 19

21 of hiring an attorney becomes quantitatively larger (although it becomes insignificant). Education also continues to have a positive, though insignificant, effect. Being white is still strongly associated with being in the federal system. In another attempt to abstract from the largest kingpins, the second row omits the federal inmates with the longest sentences (those greater than 120 months). (This restriction causes us to lose much less of the sample than we did by dropping all traffickers.) The results for this sample look much the same as the basic regression results. Overall, these specification checks show that our results are not particular to the definition of our sample. IV. Conclusion Drug crimes offer an important case study of the consequences of the trend in the United States toward the federalization of crime. Most drug crimes are subject to concurrent federal/state jurisdiction under which crimes can be prosecuted in either federal or state court. Although anecdotal evidence suggests that prosecutors may have substantial discretion over the extent to which crimes are prosecuted in the state versus the federal justice systems, the role of prosecutorial decision-making in the federalization of crime is not well understood. Because prosecutors may not simply act exclusively to maximize social welfare, and because jurisdictional choice is a joint decision of both federal and state officials, it is crucial to consider the incentives that each prosecutor will face in modeling the form and extent of the federalization of crime. We model theoretically the allocation of cases between state and federal prosecutors in a system in which federal prosecutors have relatively greater resources than their state counterparts. Our theoretical model illustrates that apportionment of cases between jurisdictions is in general a function of both efficiency/social welfare concerns and the private interests, or career concerns, of prosecutors. In particular, for a range of parameter values, asymmetries in resources in this context can lead federal prosecutors to try a disproportionate share of career concerns cases. Thus, observing the cases transferred to US attorneys suggests what types of cases yield private, career returns to attorneys. 20

22 Our empirical results suggest that federal prosecutors uniformly appear to prosecute high human capital individuals, who are older, more successful in their legal-sector careers, more likely to be married, more likely to be Army veterans and less likely to have a criminal record. U.S. attorneys particularly prosecute cases with private attorneys and with individuals who are likely to employ private attorneys. This is true even among criminals who are incarcerated only for drug possession. There are two natural interpretations of our results. First, these high-human-capital criminals may be particularly difficult to prosecute, particularly important, or more likely to have committed crimes that cross state boundaries; U.S. attorneys have more resources and broader jurisdictional power, and so efficiently tend to gravitate towards these more difficult cases. Second, these criminals offer the best career returns; their cases will be higher profile and lead to more skill accumulation. The attributes of Federal prisoners tend to be associated with longer sentence lengths, even within the state system, which supports the efficiency view of this sorting. On the other hand, the tendency to sort in this manner even occurs with criminals who are sentenced for possession (even after a trial). This supports the career returns view of sorting, since it is hard to believe that highhuman-capital people who are sentenced for possession are the most serious criminals. In summary, although there is evidence that supports the career concerns view, there is also evidence compatible with the view that prosecutors maximize social welfare. Future research might continue to investigate the relative importance of these factors on attorney decision-making and the federalization of crime. 21

23 References Andreoni, James Reasonable Doubt and the Optimal Magnitude of Fines: Should the Penalty Fit the Crime? Rand Journal of Economics, 22(3): Beale, Sara Sun Federalizing Crime: Assessing the Impact on the Federal Courts, in James A. Strazella, ed., The Federal Role in Criminal Law: Annals of the American Academy of Political and Social Science, vol Brickey, Kathleen F The Commerce Clause and Federalized Crime: A Tale of Two Thieves, in James A. Strazella, ed., The Federal Role in Criminal Law: Annals of the American Academy of Political and Social Science, vol Brickey, Kathleen F Criminal Mischief: The Federalization of American Criminal Law, Hastings Law Journal 46: Bureau of Justice Statistics Drug and Crime Facts, 1994, NCJ Butcher, Kristin F. and Anne Morrison Piehl The Role of Deportation in the Incarceration of Immigrants, in George Borjas, ed., Issues in the Economics of Immigration, University of Chicago Press, (forthcoming). 22

24 Curtis, Dennis E The Effect of Federalization on the Defense Function, in James A. Strazella, ed., The Federal Role in Criminal Law: Annals of the American Academy of Political and Social Science, vol Eisenstein, James Council for the United States: U.S. Attorneys in the Political and Legal Systems, Baltimore: The Johns Hopkins University Press. Flanagan, T. and K. Maguire, eds Sourcebook of Criminal Justice Statistics, 1991, U.S. Department of Justice, Washington DC: USGPO. Glaeser, E. and S. Glendon Who Owns Guns? Criminals, Victims, and the Culture of Violence, unpublished paper, Harvard University. Glaeser, Edward L., Daniel P. Kessler, and Anne Morrison Piehl What Do Prosecutors Maximize? An Analysis of Drug Offenders and Concurrent Jurisdiction, National Bureau of Economic Research working paper #6602. Gorelick, Jamie S. and Harry Litman Prosecutorial Discretion and the Federalization Debate, Hastings Law Journal 46: 967. Guerra, Sandra The Myth of Dual Sovereignty: Multijuridictional Drug Law Enforcement and Double Jeopardy, North Carolina Law Review 73: