Crime, Police, and Truth-in-Sentencing: The Impact of State Sentencing Policy on Local Communities

Transcription

1 Crime, Police, and Truth-in-Sentencing: The Impact of State Sentencing Policy on Local Communities Amanda Ross 1 Department of Economics Syracuse University Syracuse, NY alross01@maxwell.syr.edu December 3, I would like to thank Stuart Rosenthal, Badi Baltagi, Gary Engelhardt, Matt Freedman, Fariha Kamal, Jeff Kubik, and Bill Horrace, and seminar participants at the 2010 North American Regional Science Association meetings for their comments and suggestions. All remaining errors are my own.

2 Abstract This paper considers two related questions: the impact of spatial variation in crime prevention policies on the migration of criminal activity into nearby locations and the tendency for higherlevel government anti-crime policies to be offset by a scaling back of local crime deterrent efforts. A key source of identification is to draw upon variation in the timing of adoption of state-wide Truth-in-Sentencing (TIS) legislation during the 1990s. To estimate the effect of the policy on crime and police protection, a border methodology is used. The empirical design compares activity in adjacent counties on opposite sides of state boundaries in the 59 urban areas that cross state lines. There are three key results regarding the impact of TIS on local communities. First, adoption of TIS lowers the level of criminal activity in the adopting state. If both states within the urban area enact the more stringent sentencing policy, localities experience a larger reduction in violent crime. Second, adoption of the stiffer sentencing policy prompts migration of criminal activity into adjacent counties in the neighboring state. This migration effect is most pronounced in the more populated urban areas. Finally, findings indicate that after imposition of TIS by the state government, local governments reduce the level of police protection. This suggests that some of the deterrent effect of higher-level government anti-crime policy is offset by scaling back on anti-crime efforts at the local level. Key words: Economics of Crime, Truth-in-Sentencing, Police, Intergovernmental Grants, Spatial Differencing JEL Codes: H7, H72, H76

3 1. Introduction Numerous economists have used the Becker model (1968) of rational criminal behavior to estimate the success of various policies in deterring criminal activity (see Levitt (2004) and Freeman (1999) for a review). This research on the economics of crime has devoted much of its attention to deterrence, where policy is designed to increase the expected costs of crime and thus discourage criminal behavior. To raise the costs of criminal activity, policy makers typically increase police protection or implement a more stringent punishment if apprehended. Many empirical papers have examined the success of various anti-crime initiatives, such as an increase in the number of police officers (Levitt (1997), Corman and Mocan (2000), and Di Tella and Schargrodsky (2004)), the use of the death penalty (Ehrlich (1975, 1977), Dezhbakhsh, Rubin, and Shepherd (2003), Mocan and Gittings (2003), and Dezhbakhsh and Shepherd (2006), Cameron (1994)), and the implementation of Three Strikes laws (Iyengar (2008), Shepherd (2002a), and Helland and Tabarrok (2007)), in reducing criminal activity. There are two important aspects of crime prevention and deterrence that have not received as much attention in the literature, which this paper will address empirically. First, with the exception of a few papers, previous work on crime prevention has not accounted for the geographic mobility of criminal activity. When the spatial dimension is incorporated, a rational decision-maker will weigh the costs and benefits of committing a crime in alternate locations. If neighboring jurisdictions have different anti-crime policies, criminal activity may shift to areas with more lenient policies. For example, Iyengar (2008) found that adoption of Three Strikes laws in California caused criminals with one or two strikes to commit their next offense in an adjacent state. 2 Gonzalez-Navarro (2008) demonstrated that the introduction of Lojack 3 to some 2 Tabarrok and Helland (2009) also examined Three Strikes laws in California but did not find evidence of migration. 1

4 Mexican states increased the number of motor vehicle thefts in neighboring states that did not have the device. 4 The second dimension of crime prevention that has received limited attention is the interaction between the different levels of government that separately enact anti-crime policies. Crime deterrent initiatives are pursued by local, state, and federal governments. As a result, higher level government initiatives could be partially offset by a scaling back of local government expenditures on anti-crime efforts. Reducing police protection allows localities to either decrease taxes or increase expenditures on other local public services, such as schools or parks. This interaction between the different levels of government relates to two issues in public finance: the impact of grants on local government spending, as with the flypaper effect, and the offsetting of higher level government policy by the actions of lower level governments. 5 Several recent empirical papers have examined this issue in other contexts, such as Knight (2002) which looked at federal highway programs and Gordon (2003) which studied school aid. This paper uses variation across states in adoption of Truth-and-Sentencing (TIS) legislation throughout the 1990s to estimate the deterrent effect of the more stringent sentencing policy, the migration of criminal activity, and the interactions between the anti-crime initiatives of the different levels of government. In 1994, the federal government passed the Violent Crime Control and Law Enforcement Act (VCCA). One section of this law, the Violent Offender 3 Lojack is an electronic tracking device that is installed in cars by the manufacturer. If a car is reported stolen, it sends an electronic transmission which allows police to find and recover the vehicle. The device is hidden, so thieves cannot distinguish between cars that have Lojack and cars that do not. 4 Ayres and Levitt (1998) also looked at the effect of Lojack on vehicle thefts in the US. In that paper, the authors examined how Lojack creates a positive externality and found that the device is utilized at a level that is below the socially optimal quantity. 5 The flypaper effect is an anomaly in public finance. Theoretically, an unrestricted grant from a higher level of government is equivalent to an increase in income. Empirically, research finds that a dollar increase in an exogenous, unrestricted grant leads to greater public spending than an equivalent dollar increase in income. In other words, money from the exogenous grants-in-aid sticks where it hits. See Hines and Thaler (1995) and Inman (2008) for more information on the flypaper effect. 2

5 Incarceration and Truth-in-Sentencing Incentive Program, allocated funds to state governments if the state proved that violent felons served, on average, at least 85 percent of their sentence. Therefore, the policy increased the expected time served for violent crimes. By 1999, 28 states were eligible to receive the federal grant. Figure 1, Panels A, B, C, and D, show the expansion of the policy throughout the 1990s. Temporal differences in adoption of TIS provide the variation used to estimate the impact of an increase in the expected prison sentence on criminal activity and police protection. To estimate the effect of implementing TIS on local activity, I focus on the 59 urban areas that cross state lines. 6 This identification strategy has several advantages. First, adjacent counties on opposite sides of state boundaries tend to have similar unobserved local attributes. By including urban area fixed effects, common unobserved factors within that urban area are removed. Another advantage of the empirical strategy is that it utilizes a panel data set. Including time fixed effects removes even more unobserved variables that may be present. Finally, I focus on areas close to the state boundary where the migration effects are the greatest. If migration of criminal activity does not occur in these border regions, it is unlikely that there is movement across state lines, further toward the interior of the state. All regressions control for the year in which each state within an urban area enacted TIS in a manner that allows for differences in the timing of the adoption decision. In addition, the preferred specifications include an interaction between the own-state adoption indicator and the adjacent state adoption indicator. This additional term allows the impact of TIS to vary depending on whether one or both states in the urban area have implemented the policy. The empirical analysis begins by estimating the effect of TIS implementation on a county s violent crime rate. Following enactment of TIS, I find a decrease of approximately 6 The border method has been used in previous work such as Holmes (1998), Black (1999), and Pence (2006). 3

6 8.7% in the violent crime rate in counties within the adopting state. 7 In addition, results indicate that adoption of TIS by a state government causes the migration of criminal activity into the neighboring state. Specifically, enacting TIS causes an increase in the violent crime rate in counties located in the adjacent state of approximately 8.6%. The migration component is especially pronounced in the largest urban areas. Finally, if both states within the urban area enact the policy, there is a decrease in the violent crime rate of 14.9%, indicating localities experience the greatest benefit if both states enact the more stringent sentencing policy. I also consider the effect of enactment of TIS on property crime. This extension explores the relationship between property and violent crime, specifically if the two types of crime are substitutes or complements. Results indicate that TIS decreases not only violent crime rates, but property crime rates as well, suggesting there is a complementarity between the two types of crime. In addition, models are estimated stratified by the type of violent crime (murder, rape, robbery, and assault). From these regressions, I find that as the expected penalty of the crime decreases, the effect of the policy increases, which is consistent with behavior that values the present more than the future. Finally, the paper considers the impact of a harsher state-level sentencing policy on police protection. This portion of the analysis examines the effect of the federal government allocating funds to anti-crime initiatives on the spending patterns of lower levels of government. After a state implements TIS, local governments reduce the number of police officers employed by 7.2%. However, counties will increase the number of officers by 16.2% if the neighboring state enacts TIS but their own state does not implement the policy. If both states adopt the more stringent sentencing measures, there is a decrease of 4.7% in the number of police officers 7 Note that the estimates provide the total effect of adopting TIS on crime. The results could be driven by deterrence, incapacitation, or both. Deterrence is when the threat of imprisonment causes individuals to no longer engage in criminal activity. Incapacitation arises when a criminal is incarcerated and unable to commit crimes. 4

7 employed. These results suggest that local governments reduce their crime deterrent efforts as higher levels of government pursue more stringent anti-crime initiatives. In addition, this offsetting suggests that the estimates of the effect of TIS legislation on crime rates are a lower bound of the true effect of the policy on crime rates. This paper contributes to the literature on the economics of crime, as well as research in local public finance. There are three key findings. First, adoption of TIS decreases crime rates in the areas that implemented the harsher sentencing policy. This is consistent with the predictions from the standard model of rational criminal behavior. Also, differences in anticrime policy in adjacent states induce migration of criminal activity to neighboring areas with less stringent policies, especially in the largest urban areas. Finally, I find that implementation of the harsher sentencing policy by the state government decreases the amount spent on police protection by local governments. This implies that the anti-crime efforts of the state and federal governments are partially offset by the actions of the local governments. The rest of the paper proceeds as follows. Section 2 describes the 1994 Violent Crime Control and Law Enforcement Act with particular emphasis on the Violent Offender Incarceration and Truth-in-Sentencing Incentive Program. Section 3 outlines the identification strategy and Section 4 describes the data. Section 5 provides results of the impact of implementing TIS on violent crime. Extensions are presented in Section 6. The extensions include estimating the impact of TIS on property crime, stratifying the regressions by the type of violent crime, and breaking up the sample by the size of the urban area. Section 7 examines the impact of adopting the more stringent sentencing policy on police protection and Section 8 concludes. 5

8 2. Violent Crime Control and Law Enforcement Act of 1994 In 1994, Congress passed and President Clinton signed into law the Violent Crime Control and Law Enforcement Act (VCCA). This monumental legislation included many provisions that pursued a tough on crime agenda. For example, it made additional crimes eligible for the death penalty, expanded the assault weapons ban, and enacted stiffer penalties for drug trafficking. 8 It also included billions of dollars in grant programs for various crime prevention initiatives, ranging from hiring police officers to aid for battered women s shelters. One of the grant opportunities in the VCCA was the Violent Offender Incarceration and Truth-in-Sentencing (VOI/TIS) Incentive Program. This program allocated funds to any state that proved criminals convicted of any part 1 violent crime (murder, forcible rape, robbery, and aggravated assault) served, on average, at least 85 percent of their sentence. Several studies have found that the policy did increase the time served for violent criminals and was not offset by the actions of either judges or prosecutors (see Ditton and Wilson (1999), Owens (2010), and Dusek & Tsuchimoto (2009)). Therefore, TIS adoption increased the expected costs of engaging in violent criminal activity. Federal aid was allocated to the state governments to expand and build correctional facilities to ensure there were enough beds to imprison the offenders for the extended period. Between 1996 and 2001, more than $2.7 billion was allocated through the VOI/TIS grant program. 9 By 2000, 28 states and the District of Columbia had enacted legislation that made 8 For full text of the bill, see 9 See for more information about how the funds were distributed. Note that the TIS grants were formula grants. 6

9 them eligible to receive aid through the VOI/TIS grant program. Table 1a lists the states that received federal aid and the year each had legislation in place that met the TIS requirements Empirical Strategy Estimating the effect of a more stringent sentencing policy on criminal activity is difficult if crime rates are correlated with unobserved variables. Prior research has tended to circumvent this issue in one of two ways. Some use an instrumental variable approach, such as Levitt (1997) which used mayoral elections as an instrument for the number of police officers. Others use a policy change which is treated as exogenous to criminal activity, such as Di Tella and Schargrodsky (2004) which used an increase in police surveillance after a terrorist attack as an exogenous increase in police protection. This paper uses a state-level policy change, which is assumed to be exogenous to activity at the county level, to estimate the impact of adopting a more stringent sentencing policy on local crime rates and police protection. To identify the effect of the policy, I draw upon border methods developed in areas outside of the crime literature. Specifically, I examine the impact of state-specific decisions to enact TIS legislation on criminal activity and the amount of police protection in counties located in adjacent states within the same urban area. The analysis focuses on the counties within the 59 urban areas that cross state boundaries, similar to Pence (2006). To clarify how the empirical strategy controls for unobserved variables, consider the effect of various county-level characteristics on a county s crime rate, = + +, (3.1) 10 States are only considered to have adopted TIS if they were classified by the Department of Justice as a FY 2000 Truth-in-Sentencing state. Most of the remaining states also have a requirement on the percentage of their sentence violent offenders must serve, but the requirement is below 85 percent. For example, Indiana requires violent felons to serve 50 percent of the sentence. 7

10 where denotes the year, specifies the urban area to which the county belongs, and indicates which state within the urban area the county is located. The subscript indicates the counties located in the neighboring state of the urban area. is the crime rate in state of urban area in year and represents various attributes of the county. captures the characteristics of the counties within the adjacent state to control for potential spillovers effects. (as well as ) can be decomposed into three components = + +. (3.2) includes area-specific, time-invariant attributes, such as proximity to a body of water or a major highway, captures time-specific characteristics that are common across all counties, such as business cycle shocks, and represents area-specific characteristics that vary over time, such as socio-economic attributes and adoption of TIS legislation. If all possible variables that affect crime rates were observed, then I could include these control variables and estimate the model in Equation (3.1). However, this is unlikely to be true, causing these variables to become part of the error term. To address this problem of unobserved variables, I include urban area fixed effects ( ) that remove time-invariant attributes, which may be observable or unobservable, that are side specific or urban area specific. In addition, year fixed effects ( ) are included to remove any year specific characteristics that are common across the nation. This leaves me with the following equation, = (3.3) Adoption of TIS varies across urban areas and over time, making it an element of and. Equation (3.3) is therefore rewritten as, = (3.4) 8

11 where equals one after a given state implements TIS and equals one after the adjacent state adopts TIS. After removing whether and when a state adopts TIS, + contains all remaining urban area-specific, time-varying attributes. The primary challenge of the empirical work is to ensure that unobserved elements of this remaining term do not bias the estimates of the impact of the state enacting TIS on a county s crime rate and the level of police protection. In addition to the urban area fixed effects and time fixed effects, I include many socio-economic attributes of both the adopting state and the adjacent state to ensure that the estimates of and are unbiased. One concern with the model in Equation (3.4) is that it does not capture properly the effect of TIS if both states in the urban area adopt the policy. It seems likely that the effect of a state enacting the policy may differ based on if the adjacent state has implemented the same policy. To address this problem, I estimate a model which includes an interaction between the two TIS adoption variables, = (3.5) If only state has TIS, then the effect of the policy on the state s crime rate is captured by. If only the neighboring state has TIS, the effect of the sentencing policy is measured by. If both states in the urban area have the policy, then the effect of TIS is given by the sum of the coefficients, + +. After examining the impact of TIS on crime rates, I estimate the influence of TIS on the number of police officers per person in a county. If the more stringent sentencing policy causes a decrease in crime, then the local governments may devote fewer resources towards police and allocate the funds towards other local public services (e.g. schools and parks) or decrease taxes. 9

12 The same methodologies are used to estimate the effect of TIS on police protection, but with the number of police officers per capita as the dependent variable. 4. Data 4.1. Description of Data The 59 urban areas that cross state boundaries were selected based on the definitions and classifications of the US Census Bureau. 11 A list of these urban areas, as well as their populations in 2000, is presented in Table 1b. Looking across the table, we see there is a wide range in the sizes of the urban areas. Some are very large, such as New York-Newark, Chicago, and Philadelphia, which all have over 5 million people. Others are relatively small, such as Portsmouth and Lewiston, each of which has approximately 50,000 people. Information on whether and when a state implemented TIS was obtained from an Urban Institute Report (Sabal et al, 2002). Table 1a lists all the states that adopted TIS and the year each state had legislation in place that met the requirements to receive federal funding. The socio-economic variables were obtained from the 1980 and 1990 Census data. The crime statistics, based on data from 1980 to 1999, were obtained from the Federal Bureau of Investigation s (FBI) Uniform Crime Reports (UCR). 12 The UCR contains a variety of information reported to the FBI by police agencies. 13 The report submitted by each police agency includes the number of reported violent crimes (murder, rape, robbery, and assault) and 11 An urban area, defined by Census, consists of (i) core census block groups or blocks that have a population density of at least 1,000 people per square mile and (ii) surrounding census blocks that have an overall density of at least 500 people per square mile. 12 The data was obtained at the precinct level and aggregated to the county level. The UCR data is available through the National Archive of Criminal Justice Data. 13 Reporting police agencies include, but are not limited to, county sheriff departments, city police agencies, and college campus police departments. 10

13 property crimes (burglary, larceny, and motor vehicle theft). 14 The number of police officers employed by the police agency is also reported. A police officer, as defined by the FBI, is a fulltime employee of the department who has the authority to arrest individuals. Any other individual who works for the police agency is considered a civilian employee. In addition to the number of reported crimes, the UCR contains estimates of the population in the jurisdiction, which is used to calculate the crime rate and the number of police officers per capita. The data on population is provided to the FBI by the Census Bureau in census years. In non-decennial years, the population is approximated using the Current Population Survey as well as birth and death records Summary Statistics Table 1c provides summary statistics on the attributes of the counties included in the sample. These variables include the violent crime rate, the property crime rate, the number of police officers per capita, and a variety of socio-economic variables, such as the percent of the county that is Hispanic, percent of the county that is white, average age of the county, measures of the income distribution, and education measures. Summary statistics are based on 1990 data and are broken up into adopting states, Column (1), and non-adopting states, Column (2). Looking across the columns, we see that the counties in the adopting and non-adopting states are very similar. The adopting states have an average violent crime rate of approximately 40 crimes per 10,000 people in 1990, while those that did not adopt have an average violent crime rate of about 42 crimes per capita. This dissimilarity is driven by differences in the aggravated assault rate, as the rest of the violent crimes (murder, rape, and robbery) are very 14 There are several problems with the UCR data that have been documented in the literature (see Levitt (1998)). Despite its flaws, the UCR is the only data set on criminal activity available at the national level. Therefore, many academic studies use it, despite its limitations, as it is the best option available. 11

14 similar. However, while the 1990 violent crime rate in the adopting states is slightly higher than the violent crime rate in the non-adopting states, the two numbers are not significantly different from each other. In addition, the property crime rates are almost identical, and the numbers of police officers per capita are also similar. For the socio-economic attributes of the counties in the sample, there are not any significant differences in these observed variables between the states that enacted TIS legislation and those that did not. For example, both the adopting and non-adopting states are approximately 88% white, 8% black, and 4% Hispanic. The average age, percent male, measures of the income distribution, and the education measures are also not significantly different in the states that implemented TIS and those that did not. Overall, the summary statistics in Table 1c provide evidence that there is not a significant difference between the counties in the sample located in states that adopted TIS and those that did not implement TIS. 5. The Impact of TIS on Violent Crime Rates To receive federal aid through the TIS program, states had to prove that violent offenders served, on average, at least 85 percent of their sentence. Several studies, such as Ditton and Wilson (1999), Owens (2010), and Dusek & Tsuchimoto (2009), have found that adopting TIS increased the time served for violent felons and that the policy was not off-set by the actions of judges or prosecutors. Therefore, the states that enacted TIS increased the expected punishment of violent crime, raising the expected cost of violent criminal activity. Based on the simple model of rational criminal behavior, it is expected that implementation of TIS will decrease the violent crime rate within the adopting state. 12

15 I begin by estimating the effect of TIS adoption on the violent crime rate of counties within the state, as well as the effect of adoption by the adjacent state on the same county s crime rate. Results are presented in Table 2. The dependent variable is the natural logarithm of the violent crime rate. Standard errors are clustered by urban area-year. Column (1) presents estimates from a model that includes controls for if a state adopts TIS, if the neighboring state adopts TIS, and year and urban area fixed effects. The model in Column (2) includes the same variables as Column (1), as well as an interaction between the indicators for a state s adoption of TIS and the neighboring state s adoption of TIS. The interaction allows the impact of enacting TIS to differ depending on whether both states in an urban area have the policy or only one state does. Columns (3) and (4) follow the same structure as Columns (1) and (2) but also include a variety of socio-economic control variables. Note that the socio-economic attributes of the county as well as the attributes of the counties in the adjacent state are included. 15 First, compare the point estimates from Column (1) to Column (3) and Column (2) to Column (4). In these models, the magnitudes of the effects are similar, regardless of whether the socio-economic controls are included in the model. The robustness of the point estimates provides evidence that the urban area fixed effects and year fixed effects are capturing the majority of the unobserved attributes. A similar qualitative pattern emerges across all four specifications. After implementation of TIS by the state government, counties near the border of the adopting state enjoy a decline in violent crime. This decline is driven by harsher expected penalties which discourage individuals from committing crimes in that state, incarcerated individuals being imprisoned for a longer 15 Socio-economic control variables include percent male, percent Hispanic, average age, percentage with less than a high school degree, percentage that have graduated from high school, percentage with some college, and controls for the 25th, 50th, and 75th percentile of the income distribution. These same controls are included for the adjacent state as well, calculated as the average variable from all the counties in the adjacent state, in order to control for differences in the socio-economic conditions in the adjacent jurisdiction. 13

16 period of time, and the migration of criminal activity into the neighboring state. All three mechanisms reduce crime rates in the adopting state. If both states in an urban area implement TIS, there is a larger decline in the violent crime rate, suggesting that localities experience the greatest benefit if both states in an urban area enact the policy. This larger effect could be driven by the adoption of TIS in both states causing criminals to flee the urban area completely. If the neighboring state adopts TIS, there are two effects that are present which could offset each other. One is the migration effect, where criminals move to the adjacent state to avoid the more stringent sentencing policy. This migration effect would imply adoption of TIS by a state will have a positive effect on crime rate in the neighboring state. However, there is also a spillover deterrent effect which may decrease crime rates in the neighboring state. For example, when a state adopts TIS, violent offenders will be imprisoned for a longer period of time. If criminals commit crimes in both states, then the longer incarceration period will reduce the crime rate in both states. However, this deterrent effect will be smaller if only the adjacent state adopts TIS than if only the own state enacts TIS. Overall, the anticipated effect of the adjacent state adopting TIS is ambiguous. The second row of Table 2 provides some indication that the migration effect is larger. However, the estimates are noisy and significant at the 10% level in only one of the four regressions. Focusing on the estimates in Column (4), we see that after a state enacts TIS there is an 8.7% decrease in that state s violent crime rate. This model includes all the control variables, fixed effects, and TIS indicators. If a state does not adopt TIS but its neighbor does, then the counties in the non-adopting state experience an increase in the violent crime rate of 8.6%. However, this estimate is only significant at the 10% level. Finally, I find that if both states adopt the more stringent sentencing policy, there is a decrease in the county s violent crime rate 14

17 of approximately 14.9% ( = 14.9). Note that the reduction in crime is the greatest if both states in the urban area adopt TIS. The effect of adoption may be the greatest when both states adopt TIS legislation because adoption of the more stringent sentencing policy by both states may cause criminals to flee the urban area all together. 6. Extensions 6.1. Property Crime While the policy does not specifically target property crime, it is possible that enactment of TIS has an indirect impact on the property crime rate. 16 The sign of the effect depends on whether violent and property crimes are substitutes or complements. Previous research has found criminals substitute between types of crime if the penalty of one crime increases, which suggests that enactment of TIS will have a positive impact on the property crime rate (Iyengar (2008) and Yang (2008)). Shepherd (2002b) found a substitution from violent to property crime after adoption of TIS using data on all the counties within the US. Alternatively, there could be a complementarity between violent and property crime. For example, violent felons may commit property crimes as well. The increase in the time served for violent criminals may therefore cause a reduction in both the violent and property crime, due to the incarceration effect. Also, an individual who plans to commit a property crime may migrate across the state boundary to engage in criminal activity if he or she believes that the crime may become violent. This effect will be strongest in areas close to the state border. Overall, the anticipated impact of TIS on property crime is ambiguous a priori. 16 In some states, Ohio, Missouri, Delaware and Washington DC, the 85% requirement was enacted for property and some drug-related crimes as well. All models were run excluding all urban areas that included one of these states and the results are robust. 15

18 Table 3 contains estimates of the impact of TIS on a county s property crime rate. All models are estimated using the preferred specification in Column (4) of Table 2, which includes all the TIS indicators and the full set of control variables. Column (1) includes the model with the violent crime rate as the dependent variable for comparison. The dependent variable is the log of the crime rate and standard errors are reported in parentheses. Looking at Table 3, we see that adoption of TIS not only causes a decline in the violent crime rate in counties within a state, but also causes a decrease in property crime. Adoption of TIS reduces the property crime rate by approximately 7.1%. If both states in an urban area enact the new sentencing policy, there is an overall net decrease in the property crime rate of about 23.1%. If a state adopts TIS, there is not a significant effect on property crime in the counties located in the neighboring state, suggesting the migration effect is not as strong. Overall, I find evidence that violent and property crimes are complements, contrary to what previous research has found Stratified by Type of Crime The effect of TIS may vary across the types of violent crime. These differences could be driven by variance in the effect the policy will have on the expected prison sentence for the various types of crime. For example, if an individual serves 50% of his or her sentence, the average time served is 127 months (over 10 years) for murder and 36 months (3 years) for assault (Ditton and Wilson (1999)). If individuals tend to discount the future, then increasing a 10 year sentence will not have as strong of an effect as increasing a 3 year sentence. For that reason, it is expected that as the severity of the crime and the average prison sentence increase, the effect of implementing TIS will diminish. 16

19 Table 4 provides estimates of the impact of TIS on the different types of violent crime. I focus on the four violent crimes (murder, rape, robbery, and assault) because the policy only directly affected these crimes. Again, all models include the socio-economic controls of the county and the neighboring counties, urban area and time fixed effects, and the indicators for TIS adoption. Each regression uses the log of the crime rate as the dependent variable and standard errors are in parentheses. Findings show the implementation of TIS decreased the rate of murder, robbery, and aggravated assault in counties within the adopting state, though all of these coefficients are insignificant. If both states adopt TIS, I find a significant decline in the crime rate for all four types of violent crime. This indicates that TIS decreased violent crime in the urban areas, especially when both states enacted the policy. Also, note that as the severity of the crime and average time served increases, the effect of TIS on the crime rate diminishes. The second row of Table 4 shows the effect of a state s adoption of TIS on crime rates in the neighboring state. Again, there is a positive effect for murder, robbery, and assault, suggesting that there is migration of criminal activity across state boundaries in response to the policy change. However, this effect is only significant for assault. Overall, the results support the argument that the impact of the policy becomes stronger as the expected time served by the individual decreases Stratified by Size of the Urban Area Previous research has examined why crime rates tend to be higher in central cities than the surrounding suburbs (Glaeser and Sacerdote (1999)). Three explanations are provided in the literature: returns to crime are higher in cities (i.e. more attractive targets), there is a lower 17

20 probability of arrest in densely developed areas, and cities attract or create criminals. The next portion of the paper examines if this variation in crime rates is present across urban areas of different sizes. To do this, I stratify the urban areas by population and compare the 30 most populated urban areas to the 29 smallest jurisdictions. In the more populous and densely developed urban areas, such as Philadelphia and Washington, DC, there are many attractive targets. These areas may also have a relatively low probability of arrest as there are many places to hide and it is harder to recognize an individual in a large crowd. Similarly, it is expected that these large, heavily developed urban areas will have higher crime rates than the less populous urban areas, like Duluth and Binghamton. Therefore, adoption of TIS is expected to have the strongest effect in counties located in the most populated urban areas. Table 5 provides estimates of the effect of adopting TIS legislation on the violent crime rate in a given county, stratified by the population of the urban area. Again, the preferred specification from Column (4) of Table 2 is used. The dependent variable is again the log of the violent crime rate. Column (1) presents results for the 30 most populated urban areas, which have populations between 250,000 and 18 million. 17 Results indicate that implementing TIS caused a decrease in the adopting state s crime rate of 13.5%. If a state adopts TIS, the violent crime rate in the adjacent state increases by 14.4%. The magnitude of the effect of an adjacent state s adoption is similar to that of the own state s adoption. This indicates that the migration effect of the policy is large and significant in the most populated urban areas. 17 The 30 largest urban areas that cross state boundaries are New York-Newark, NY-NJ-CT; Chicago, IL-IN; Philadelphia, PA-NJ-DE-MD; Boston, MA-NH-RI; Washington, DC-VA-MD; Saint Louis, MO-IL; Portland, OR- WA; Cincinnati, OH-KY-IN; Kansas City, MO-KS; Providence, RI-MA; Memphis, TN-MS-AR; Bridgeport- Stamford, CT-NY; Louisville, KY-IN; Charlotte, NC-SC; and El Paso, TX-NM; Omaha, NE-IA; Allentown- Bethlehem, PA-NJ; Springfield, MA-CT; Toledo, OH-MI; Worcester, MA-CT; Youngstown, OH-PA; Chattanooga, TN-GA; Augusta-Richmond County, GA-SC; Spokane, WA-ID; Pensacola, FL-AL; South Bend, IN-MI; Davenport, IA-IL; Columbus, GA-AL; Round Lake Beach-McHenry-Grayslake, IL-WI; Evansville, IN-KY. 18

21 If both states enact TIS, there is an overall net decrease in a county s crime rate of approximately 3%. In these large urban areas, adoption by both states does not have a large impact on the violent crime rate. However, if there are differences in the adjacent states the effect is large and of similar magnitude. This suggests that in these large urban areas, the policy is to causes migration of criminal activity within the urban area, but does not cause a significant reduction in criminal behavior. Column (2) of Table 5 presents estimates for the 29 smallest urban areas, which have populations below 250,000 people. 18 I find that implementing TIS legislation has almost no effect on the crime rate in the smaller urban areas. In addition, if a state adopts TIS there is a slight decrease in the crime rate in the adjacent state, though this effect is small and significant at only the 10% level. This negative impact could be caused by spillover deterrent effects. For example, individuals in the adjacent state may not realize that the policy does not affect them and incorrectly choose to no longer commit crimes. Also, the more stringent penalty in one state may cause individuals to stop committing crimes in both states, causing a reduction in a given state if the adjacent state adopts TIS. When both states enact TIS legislation, however, there is a net decrease in a county s crime rate of 23.7%. 7. The Impact of TIS on Police Protection The next portion of the analysis examines how a change in the anti-crime efforts of a higher level of government affects spending on police protection by lower levels of government. 18 The 29 smallest urban areas that cross state boundaries are Nashua, NH-MA; Huntington, WV-KY-OH; Binghamton, NY-PA; Danbury, CT-NY; Fargo, ND-MN; Elkhart, IN-MI; Clarksville, TN-KY; Hagerstown, MD- WV-PA; Duluth, MN-WI; Fort Smith, AR-OK; Sioux City, IA-NE-SD; Kingsport, TN-VA; Yuma, AZ-CA; La Crosse, WI-MN; Wheeling, WV-OH; Parkersburg, WV-OH; Dover-Rochester, NH-ME; Saint Joseph, MO-KS; Weirton-Steubenville, WV-OH-PA; Texarkana, TX-AR; Michigan City, IN-MI; Dubuque, IA-IL-WI; Longview, WA-OR; Salisbury, MD-DE; Bristol, TN-VA; Grand Forks, ND-MN; Beloit, WI-IL; Portsmouth, NH-ME; Lewiston, ID-WA. 19

22 A simple political economy model, developed by Bradford and Oates (1971a, 1971b) predicts that grants from a higher level of government will be offset by the spending patterns of lower levels of government. This concept is commonly referred to as fiscal federalism or the flypaper effect. Recent empirical work has examined if this crowd out of government spending is present for various programs in the US and has found that the effect of federal grants is reduced by the spending patterns of lower levels of government (Knight (2002), Gordon (2003), and Evans and Owens (2008)). In this section, I look at adoption of TIS legislation by the state government, which resulted in the state receiving aid from the federal government, to determine if local governments respond by decreasing their spending on police protection. All levels of government, state, local, and federal, attempt to minimize crime rates, subject to a budget constraint. To decrease crime rates, the government can increase police protection or implement more stringent sentencing policies. The previous results have found that when a state enacts TIS legislation, local crime rates decline. Localities experience the benefits of the policy through lower crime rates without bearing the costs. As a result, lower levels of government may adjust their expenditures and allocate fewer resources towards police protection and either redistribute those funds towards other local public services (i.e. schools and parks) or decrease taxes. The results estimating the impact of adopting TIS on the number of police offers per 10,000 people are presented in Table 6. This table follows the same structure as Table 2 but with the number of police officers per capita in a given state as the dependent variable. 19 All specifications use the log of the number of officers per person as the dependent variable and standard errors are reported in parentheses. Column (1) estimates the model which includes an 19 Police protection is always reported as the number of officers per 10,000 people. Similarly, per capita refers to the number of police officers per 10,000 people. 20

23 indicator for if a state has TIS and an indicator for if the neighboring state has TIS. 20 Column (2) includes the same variables as the regression in Column (1), as well as an interaction that equals one if both states have implemented TIS legislation. This interaction allows for a differential effect if both states in an urban area enact the policy. Column (3) includes only the own state and adjacent state adoption controls, as well as socio-economic controls, while Column (4) includes the interaction of the TIS adoption controls as well as the additional control variables. All four models include year fixed effects and urban area fixed effects. All of the specifications follow a similar pattern. If a state implements TIS, there is a decline in the number of police officers per capita in counties near the border of that state. However, local governments increase spending on police protection if the adjacent state implements the new sentencing policy. Finally, if both states in an urban area adopt TIS, there will be a decrease in the number of police officers employed by the local government. Overall, these results indicate that the anti-crime efforts of higher levels of government are partially offset by a decrease in spending on police protection by the lower levels of government. Focusing on the model in Column (4), which includes the most complete set of control variables, we see that if a state adopts TIS legislation, the local governments decrease the number of police officers employed by approximately 7.2%. However, local governments will increase the number of officers by about 16.2% if the neighboring state adopts TIS but its own state does not. If both states in the urban area enact the more stringent sentencing policy, there will be a net decrease in the number of officers employed of 4.7%. These results of the effect of TIS on local police protection suggest that the previous estimates are a lower bound of the effect of the more stringent sentencing policy on crime rates. 20 As in the previous section, if an urban area has three states this variable equals one of any of the other states in the urban area have TIS. 21

24 Given that the local governments decrease their anti-crime efforts after the state adopts the harsher sentencing policy, the impact of TIS on crime rates is diminished. This offsetting by the lower levels of government may also explain why some of the specifications in the previous section failed to find a strong effect of the policy or found an effect that was not in the anticipated direction. 8. Conclusion This paper examines the effect of adoption of a more stringent sentencing policy on local crime rates and the level of local police protection. Focusing on areas close to the border, I find communities enjoy a decline in violent crime after implementation of Truth-in-Sentencing legislation by the state government. In addition, TIS causes a larger decrease in the violent crime rate when both states within an urban area adopt the policy. If the neighboring state adopts the policy, there is evidence of a migration effect. This is especially true in the largest urban areas, where it is relatively easy to move across state boundaries. For the most populated areas, the estimates suggest that the impact of the policy on crime is driven completely by migration of criminal activity across state boundaries. After establishing that TIS decreased violent crime rates, I estimate the impact of the new sentencing policy on police protection. Following enactment of the more stringent sentencing policy by the state government, local governments in the adopting state decrease the number of police officers per capita. If both states in the urban area adopt TIS legislation, there is still a decline in the number of police officers employed by counties in the adopting state. Therefore, the anti-crime efforts of higher levels of government are partially offset by the scaling back of 22

25 the police force by lower levels of government. Also, this finding indicates that the reduction in crime is a lower bound of the true effect of the policy. This paper has three main findings that have important policy implications. First, implementing a more stringent sentencing policy decreases criminal activity. This is consistent with the predictions from the standard model of rational criminal behavior and suggests that governments can decrease crime rates within their jurisdiction by increasing the time served if apprehended. Second, policy differences in adjacent states cause a migration of criminal activity, particularly within the largest urban areas. This implies that policy makers need to be aware of what is occurring in their community, as well as the policies that are adopted in the adjacent jurisdictions. Finally, I find the anti-crime efforts of the state government are partially offset by the actions of the local governments. State policy makers need to be aware of the behavior of the local governments as it causes the effect of their anti-crime initiatives to be diminished. 23

26 References Ayres, I., Levitt, S., Measuring Positive Externalities from Unobservable Victim Precaution: An Empirical Analysis of Lojack. The Quarterly Journal of Economics 113, Becker, G., Crime and punishment: An economic approach. Journal of Political Economy 76, Black, S., Do better schools matter? Parental valuation of elementary education. The Quarterly Journal of Economics 114, Cameron, S., A review of the econometric evidence of the effect of capital punishment. Journal of Socio-Economics 23, Corman, H., Mocan, H., A time series analysis of crime, deterrence, and drug abuse in New York City. The American Economic Review 90, Dezhbakhsh, H., Rubin, P., Shepherd, J., Does capital punishment have a deterrent effect? New evidence from postmoratorium panel data. American Law and Economics Review 5, Dezhbakhsh, H., Shepherd, J., The deterrent effect of capital punishment: Evidence from a Judicial Experiment. Economic Inquiry 44, Di Tella, R., Schargrodsky, E., Do police reduce crime? Estimates using the allocation of police forces after a terrorist attack. The American Economic Review 94, Ditton, P., Wilson, D., Truth in Sentencing in State Prisons. Bureau of Justice Statistics Special Report. Ehrlich, E., The deterrent effect of capital punishment: A question of life and death. The American Economic Review 65, Ehrlich, E., Capital punishment and deterrence: Some further thoughts and evidence. Journal of Political Economy 85, Evans, W., Owens, E., COPS and crime. Journal of Public Economics 91, Freeman, R., The economics of crime. In Handbook of Labor Economics, Ashenfelter, O., Card, D. eds. Holmes, J., The effect of state policies on the location of manufacturing: Evidence from state borders. The Journal of Political Economy 106,