THE EFFECTS OF LOCAL LAW ENFORCEMENT BLOCK GRANTS ON SERIOUS CRIME*

Transcription

1 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: 1 26-AUG-08 7:40 THE EFFECTS OF LOCAL LAW ENFORCEMENT BLOCK GRANTS ON SERIOUS CRIME* JOHN L. WORRALL The University of Texas at Dallas Keywords: Local Law Enforcement Block Grants, crime rate, Universal Hiring Program, endogenous, police levels Research Summary The Local Law Enforcement Block Grants (LLEBG) Program was second only to the Community-Oriented Policing Services (COPS) Program in its funding levels. Some $3 billion was dispensed to local jurisdictions to reduce crime and improve public safety; yet the effects of LLEBG funding on crime have been all but ignored. Accordingly, panel data from more than 5,000 cities covering a 12-year period ( ) were collected, and index crime rates were regressed on LLEBG funding and appropriate demographic controls. Additional controls for police levels and other federal grants were also introduced, proper checks for endogeneity of grants (and police levels) were performed, and the models were subjected to an array of robustness checks. A consistent message emerged: LLEBG Program funding was associated with significant reductions in serious crime. Policy Implications Although LLEBG funding seemed to reduce serious crime, the results also revealed that the decrease did not occur through the hiring of additional police officers, even though many funds were used for that purpose. Other mechanisms were thus at work, but the data did not provide insights into what these mechanisms were. In any case, every $1 in LLEBG funding per capita was associated with approximately 59 fewer index crimes per 100,000 people. When combined with the findings from recent studies of the effects of community policing grants on crime, this study suggests additional federal support for local lawenforcement agencies should be considered. * The author would like to thank Tom Jessor and the Government Accountability Office for valuable assistance in data collection and suggestions for the analysis. Direct correspondence to John L. Worall, Program in Criminology, The University of Texas at Dallas, 800 West Campbell Road, GR 31, Richardson, TX ( Worrall@utdallas.edu). CRIMINOLOGY & Public Policy Volume 7 Number 3 Copyright 2008 American Society of Criminology 325

2 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: 2 26-AUG-08 7: WORRALL Recently, researchers have expressed interest in the effects on crime of federal assistance to local criminal justice agencies. For example, Zhao, Scheider, and Thurman (2002) found that grants from the Office of Community-Oriented Policing Services (COPS Office) in the U.S. Justice Department led to significant city-level reductions in violent and property crime. 1 Others (e.g., Evans and Owens, 2007; Government Accountability Office [GAO], 2005) have found that significant reductions in serious crime could be attributed to COPS grants, particularly those under the socalled Universal Hiring Program (UHP). Whereas Zhao et al. (2002) considered the direct effects on crime of COPS grants, Evans and Owens (2007) and the GAO (2005) treated grants as instruments for police levels and thus were concerned with the indirect effects of COPS on crime. These studies have most likely focused on federal grants because of the funding amounts involved. Between 1995 and 2000 alone, some $8.8 billion in grants were awarded to local law-enforcement agencies throughout the United States (e.g., GAO, 2005). Such significant funding levels make it such that their effects on crime could be picked up in standard significance tests. COPS, however, is not the only federal program that has awarded large amounts of money to criminal justice. The Local Law Enforcement Block Grants (LLEBG) Program, which was launched in fiscal year (FY) 1996, was second only to the COPS Program in the funds it awarded to local agencies and in its costs to taxpayers. Although the program is now defunct, 2 roughly $3 billion in LLEBG funds were dispersed to police departments, local governments, and allied entities for crime control and promotion of public safety. National evaluations of the LLEBG Program have been conducted and published (Cosmos Corporation, 2005; Yin, Pate, Kim, Sheppard, and Warner, 2001), but whether the LLEBG Program served its intended purpose remains largely unknown. That is, the question of whether LLEBGs have reduced crime remains mostly unanswered. In a study of the effects of policing on crime, the GAO (2005) used LLEBG as instruments for police levels in an effort to gauge the effect of policing on crime, but direct associations between LLEBG and crime were not of particular interest. The current study attempts to fill this gap. It does so using panel data from a large sample of U.S. cities. Results suggest that LLEBG significantly reduced a variety of index offenses in cities of various sizes. 1. Worrall and Kovandzic (2007) disputed this finding. 2. A similar program has emerged in its place. It is called the Justice Assistance Grant Program.

3 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: 3 26-AUG-08 7:40 LOCAL LAW ENFORCEMENT BLOCK GRANTS 327 Federal Assistance to Local Law Enforcement The Law Enforcement Assistance Act (LEAA) of 1965 marked the first significant federal government effort to fund local criminal justice programs. The enabling legislation called for $7 million in appropriations, which is an amount that pales in comparison with funding levels of late. When the LEAA Program was discontinued, federal funding levels dwindled, but during the late 1980s, they once again became a priority. One of the largest programs launched at the time was the Edward Byrne Memorial State and Local Law Enforcement Assistance Formula Grant Program (hereinafter the Byrne program). Under this program, which was created in 1988, the Bureau of Justice Assistance awarded grants to states for use by state and local governments to improve their criminal justice system. Grants could be used for everything from enforcing laws, hiring personnel, and purchasing equipment, to providing training and technical assistance (Dunworth, Haynes, and Saiger, 1997). The next significant federal effort of note was the so-called COPS Program, which was created pursuant to the Violent Crime Control and Law Enforcement Act of 1994 (see Worrall and Kovandzic, 2007; Zhao et al., 2002). The program s mandate has been to improve community policing throughout the United States, and various programs have been implemented for this purpose. The most prominent is the so-called UHP. Other programs include Making Officer Redeployment Effect (MORE), which was a technology acquisitions program; Accelerated Hiring, Education, and Deployment; Funding Accelerated for Smaller Towns; the Youth Firearms Violence Initiative; the Anti-Gang Initiative; the Community Policing to Combat Domestic Violence Initiative; and several others. Most programs require a local match. Since the COPS Program s inception, nearly $10 billion has been awarded to local law-enforcement agencies throughout the United States. Operation Weed and Seed is another noteworthy federal program (Dunworth and Mills, 1999). It has awarded grants for the purpose of first weeding out criminals and then seeding communities with needed services, prevention programs, and neighborhood revitalization. Violence Against Women Act grants have been awarded as well. This program, which was launched in 1996, provides funding for training lawenforcement officials and prosecutors, which creates special domestic violence prosecution units and implements new procedures in the name of domestic violence prevention. More recently, federal funding has taken aim at terrorism, errors of justice, and reentry through such programs as the State and Local Terrorism Prevention Training and Technical Assistance Program, the Solving Cold Cases with DNA Program, the Paul

4 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: 4 26-AUG-08 7: WORRALL Coverdell Forensic Science Improvement Grants Program, and the Prisoner Reentry Initiative. Local Law Enforcement Block Grants The LLEBG Program was launched in 1996 to provide local government agencies (not just police departments) with funding to reduce crime and improve public safety. Grants decisions were based on the number of state and jurisdiction-level Part I violent crimes (Bauer, 2004; see below for more details). In FY 1996, the program awarded more than $400 million to local governments. Funding peaked at nearly $500 million in FY 1998, but dropped to just over $100 million in FY 2004 (Bauer, 2004). Unfortunately, the program has since been discontinued. Nevertheless, some $3 billion was awarded to local law enforcement, and at one point, the program accounted for about 20% of all federal funding for local criminal justice efforts (Yin et al., 2001). The LLEBG Program was born after an amendment to the Violent Crime Control and Law Enforcement Act of Unlike COPS grants, however, LLEBG funds could be awarded to local units of government, not just to police organizations. The program was also distinctive because it placed few restrictions on local governments, and it was less burdensome than, say, the COPS Program because it required only a 10% match at the local level. According to a Bureau of Justice Assistance publication (Gist, 2000), nearly 60% of grants went to equipment and technology purchases. The rest went, in descending order, to law-enforcement hiring and overtime (22.7%), crime prevention (10.5%), adjudication of violent offenders (3.1%), drug courts (2.7%), school security (2.3%), and multijurisdictional task forces (0.2%) (Gist, 2000). As for grant recipients, a national evaluation of the LLEBG Program found that roughly two thirds of the primary contacts on the grants were in law enforcement, and roughly three quarters of grant funds went directly to law enforcement, specifically to hiring, overtime, and equipment (Yin et al., 2001:4 6). Finally, nearly 3,000 jurisdictions proposed hiring new officers with their funds. 4 Effects of Law-Enforcement Grants on Crime An exploration of the association between LLEBG funding and crime is limited in the sense that it sacrifices detail. In other words, a measure of funding does not adequately capture the many distinct uses for the funds. 3. Public Law This value includes county-level agencies. The analyses reported in this article were city level.

5 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: 5 26-AUG-08 7:40 LOCAL LAW ENFORCEMENT BLOCK GRANTS 329 Hiring of an additional officer may not (or may) lead to the same reductions in crime that could be witnessed through implementation of a multijurisdictional task force. Likewise, funding police officers overtime may not yield the same returns as launching a drug court, the latter of which the literature supports enthusiastically (e.g., Belenko, 2001). These issues notwithstanding, it is still critical to explore, at a macro level, whether billions of dollars in taxpayer funds have been well spent. Indeed, it is all but impossible to evaluate such programs on a national scale without a focus on spending amounts. Plenty of precedents for the research are reported here. Those researchers interested in the relationship between police and crime, for instance, have sometimes used funding levels in lieu of actual officer counts (e.g., Deutsch, Simon, and Spiegel, 1990; Ehrlich, 1973; Fox, 1979; Friedman, Hakim, and Spiegel, 1989; Greenwood and Wadycki, 1973; Hakim, 1980; Jacob and Rich, 1981; Jones, 1974; Land and Felson, 1976; Liu and Bee, 1983; McPheters and Stonge, 1974; Swimmer, 1974a, 1974b; Wellford, 1974). Other researchers have focused squarely on federal grant programs, especially the COPS Program (GAO, 2003, 2005; Muhlhausen, 2001; Zhao and Thurman, 2001; Zhao et al., 2002). For example, Zhao et al. (2002:7) concluded: Our analyses suggest that COPS hiring and innovative grant programs have resulted in significant reductions in local crime rates in cities with populations greater than 10,000 for both violent and nonviolent offenses. Multivariate analysis shows that in cities with populations greater than 10,000, an increase in one dollar of hiring grant funding per resident contributed to a corresponding decline of 5.26 violent crimes and property crimes per 100,000 residents. Most recently, researchers have used federal law-enforcement grants to help identify the police-crime relationship (e.g., Evans and Owens, 2007; GAO, 2005). The problem is that although police may reduce crime, crime may lead to increases in the number of police officers, which biases any presumed relationship between both variables toward zero. No shortage of research has been published in response to this problem (e.g., Levitt, 1997, 2002), but the recent approach has been to use grants as instruments because some grants, particularly hiring grants, can be expected to affect police levels but not directly affect (or be affected by) crime. In other words, grants may, through hiring, boost police forces and thereby reduce crime. This line of research presents some fairly convincing evidence that either police levels, federal spending on local criminal justice priorities, or both, reduce crime. Moving away from policing, other researchers have explored the effects of funding levels on crime at a macro level. For example, Worrall (2004)

6 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: 6 26-AUG-08 7: WORRALL explored the effects on juvenile arrests of a delinquency prevention program in California. His main measure of the program was county-level funding allocations. Other researchers have explored associations between welfare spending and serious crime at the aggregate level (e.g., Hannon and DeFronzo, 1998a, 1998b). These efforts complement qualitative studies and site-specific evaluations, offering an assessment of whether spending for local criminal justice initiatives reduces crime, arrests, and other outcomes of interest. The research reported here continues in this vein by filling a significant void in the literature. 5 Methods This article presents the results of a macro-level assessment of the effects of LLEBG spending on serious crime. Associations between funding levels and crime rates were explored at the city level. The following sections discuss data, variables, measurement, and estimation procedures used for the analyses. Data, Variables, and Measurement Panel data from a sample of 5,199 cities, which covered the years 1990 to 2001, were gathered by the GAO and supplied to the author pursuant to a Freedom of Information Act request. Sources of the data were the Office of Justice Programs Financial Data, the Uniform Crime Reports (UCR), the Bureau of Economic Analysis in the Commerce Department, the National Center for Health Statistics, and Law Enforcement Agency Identifiers Crosswalk in the Bureau of Justice Statistics. The latter file provided geographic identification information for each unit and facilitated merging of the various data sources. The financial data from the Office of Justice Programs contained both obligations (grant awards) and draw-downs (annual amounts spent by grant recipients). This research used draw-down amounts, which avoided the need to estimate spending patterns (e.g., Evans and Owens, 2007; Zhao et al., 2002). Data on annual expenditures from several grant programs were included in the master file. These programs included, first and foremost, the LLEBG Program, followed by COPS grants (UHP, MORE, innovative COPS grants, and other COPS grants), Byrne program grants, and other federal grant programs (a catchall category for all other non-cops federal grants to local law-enforcement agencies and units of government 5. Incidentally, only one study (Cosmos Corporation, 2005) made any effort to perform site-specific evaluations of LLEBG programs, but the effects on crime were ignored.

7 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: 7 26-AUG-08 7:40 LOCAL LAW ENFORCEMENT BLOCK GRANTS 331 during the analysis period). All grants data were per capita and lagged by one year to allow for delays between receipt of funds and reductions in crime (Evans and Owens, 2007; Worrall and Kovandzic, 2007; Zhao et al., 2002; and others have taken this approach). UCR data contained the necessary crime and officer counts. These data were converted to index crimes (homicide, rape, robbery, assault, burglary, larceny, and motor vehicle theft) per 100,000 persons and officers per 10,000 persons. As discussed, roughly 25% of LLEBG Program funds did not go directly to law enforcement; however, proper controls for the size of a city s police force were necessary because 75% of LLEBG Program funds did go to police agencies. Additionally, controls for police presence were also essential because failure to do so could have resulted in spurious associations between grant spending and serious crime. Data collected from the Bureau of Economic Analysis and the National Center for Health Statistics included population counts used to construct the aforementioned rates and demographic data, which included per capita income, percent nonwhite, percent 18 to 24 years old, and percent employed. 6 These data were measured at the county level and, as such, were annual estimates. The data in the years between the decennial U.S. Census did not need to be estimated or interpolated. These four variables were included in regression models for control purposes. The need to control for racial composition finds support in the literature (e.g., Holmes, 2000; McNulty and Holloway, 2000). Income and employment controls also find support, particularly in the social disorganization literature (e.g., Bursik, 1988; Sampson, 1985; Shaw and McKay, 1972). Finally, criminalcareer research supports the controls for young males (e.g., Blumstein, Cohen, Roth, and Visher, 1986:66). Summary statistics are reported in Table 1. Estimation Procedure The author estimated a series of fixed-effects regression and instrumental variables (IV) regression models using the Generalized Method of Moments (GMM) (Wooldridge, 2001). 7 GMM is ideal, as it allows for heteroskedasticity and autocorrelation that are of unknown form. As 6. Percent employed may not be as ideal as percent unemployed, but the percent unemployed was not available in the data received from the GAO. 7. These models were implemented in Stata (StataCorp, College Station, Tex.) with the user-written command xtivreg2 (xtivreg2 can also be used with exogenous regressors). The IV regressions were estimated when one right-hand-side variable was considered endogenous. Since dedicated fixed effects commands were used, unit dummies were not added because they were conditioned out of the estimation process, thus saving thousands of degrees of freedom (an easy explanation can be found here: stata. com/support/faqs/stat/xtreg2.html). Year dummies were added to each specification.

8 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: 8 26-AUG-08 7: WORRALL Table 1. Summary Statistics Observed Mean S.D. Minimum Maximum Dependent Variables Homicides per 100,000 45, Rapes per 100,000 45, Robberies per 100,000 45, , Assaults per 100,000 45, , Burglaries per 100,000 45, , Larcenies per 100,000 45,625 2, , , MV thefts per 100,000 45, , Police Variables LLEBG a 45, Police levels b 45, UHP 45, COPS MORE 45, COPS innovative 45, COPS miscellaneous 45, Byrne program 45, Other federal 45, Other Covariates Per capita income 45,483 23, , ,479 87,098 Percent nonwhite 45, Percent years old 45, Percent employed 45, Notes: MV = motor vehicle; S.D. = standard deviation. a This and all other grant variables, besides police levels, are in (US$) draw-down amounts per capita during the calendar year. Approximately 1,600 of the 5,199 cities received no LLEBG funding during the period covered by this analysis. b Sworn officers per 10,000 people. Kovandzic, Schaffer, and Kleck (2005:18) observed, This robustness to arbitrary violations of homoskedasticity and independence is appealing to empirical researchers, not least because it means obtaining valid estimation results that does [sic] not require a researcher to model these violations explicitly or correctly. GMM is also advantageous because it nests several familiar estimators, like ordinary least squares, two-stage least squares, and instrumental variables, within a single framework. Tests for skew, heteroskedasticity, and autocorrelation in preliminary models affirmed that each problem existed in the data. Accordingly, to be safe, models also were estimated with robust standard errors (Huber, 1967; White, 1980; Williams, 2000), and they took state-level clustering into account through proper adjustments to the standard errors (see Froot, 1989; Rogers, 1993). 8 The results reported below should thus be considered robust to heteroskedasticity, autocorrelation, and arbitrary Finally, the data were stationary in levels per the augmented Dickey Fuller test and Hausman tests called for fixed effects in lieu of random effects. 8. The robust cluster option was used after xtivreg2.

9 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: 9 26-AUG-08 7:40 LOCAL LAW ENFORCEMENT BLOCK GRANTS 333 covariance across observations within each state. Lastly, all data were in levels, which was consistent with Evans and Owens (2007), Worrall and Kovandzic (2007), GAO (2005), and Zhao et al. (2002), but robustness checks adopted alternative specifications. Results The first models considered the direct effects of LLEBG on crime. Subsequent models introduced other important controls, which included police levels (officers per 10,000) and other federal grants, to the equations. Next, LLEBGs were treated as endogenous, and then indirect associations between grants and crime were explored. Finally, numerous robustness checks were performed. Direct Effects of LLEBG Funding on Crime Table 2 presents the results of models analyzing the direct effects of LLEBG funding on the seven index crimes. This approach was taken by Zhao et al. (2002), who estimated the direct effects on crime of various COPS grants. The inverse associations between LLEBG funds and crime suggest that the money led to significant reductions in all index crimes during the analysis period. 9 This interpretation is at least somewhat problematic because grants could easily be correlated with omitted variables, such as police levels and other federal grants. As a first step, however, these estimates offer some preliminary evidence that LLEBG funding may have served its intended purpose. The negative and significant associations between nonwhite and all but the homicide and assault outcomes were unexpected. Because the intraunit variation in this variable is minimal, it may have reflected the same heterogeneity the fixed-effects procedure seeks to estimate. That is, it may have been correlated with the unit effects, which introduced the signflipping (see, e.g., Plumper and Troeger, 2007). 10 Effects of LLEBG Funding on Crime, Controlling for Police Levels Zhao et al. (2002) were criticized by Worrall and Kovandzic (2007) for their failure to introduce proper controls for police levels into their models of COPS grants and crime. In the current case, it was critical to control for 9. This finding may suggest that cities that experience higher-than-usual crime rates in the past will have lower future crime rates, not that funding reduced crime. Robustness checks that account for past crime growth/decline address this issue. See Table 8 and the accompanying discussion. 10. This regressor was also less refined than separate ones for different racial categories, which also could have contributed to the unexpected results. Unfortunately, a more refined measure of race was not available.

10 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: AUG-08 7: WORRALL Table 2. Effects of LLEBG Funding on Crime Homicide Rape Robbery Assault Burglary Larceny MV Theft LLEBG (3.56)** (2.18)* (2.88)** (2.51)* (2.45)* (2.23)* (4.23)** Income (1.37) (0.37) (1.41) (0.57) (0.76) (3.42)** (1.53) Nonwhite , , , (0.62) (5.85)** (2.94)** (1.43) (3.55)** (3.99)** (2.65)* Percent years old , , , (0.18) (0.81) (0.48) (0.38) (0.81) (1.52) (1.89) Employment (1.77) (2.06)* (0.79) (1.00) (1.25) (0.09) (1.37) Observations Units Notes: Robust t statistics are in parentheses. All test statistics are robust to heteroskedasticity and clustering. MV = motor vehicle. *p <.05. **p <.01. police levels, first, because 75% of the cities sampled saw their grants go directly to municipal police departments and, second, because studies of this nature need at least some controls for other criminal justice priorities and preferences. Related to this latter point, omitted variable bias could be a problem if there is a correlation between police levels and LLEBG and police levels and crime. Accordingly, officers per 10,000 were introduced into the models reported in the previous section. The results are presented in Table 3. The author s initial expectation was that LLEBG effects would be washed out by the introduction of police controls. This expectation did not pan out. More interesting still, the only significant police coefficients were in the positive direction. This observation is likely symptomatic of the endogeneity of police levels. If, for example, the effect of crime on police is more pronounced than the effect of police on crime, a positive association may be revealed. Steps to address this endogeneity problem, and the possible endogeneity of LLEBG funding, were thus taken. The results from these models are reported below. Effects of LLEBG Funding on Crime, Controlling for Police Levels and Other Federal Grants Even with proper controls for police levels, the association between LLEBG funding and crime could be spurious because of omission of funding levels from other federal grant programs. In other words, an agency that received LLEBG may have also secured other grants, which could have been themselves inversely associated with crime. Table 4 thus presents the results of models in which controls were introduced for both police levels (per 10,000) and per-capita spending in other federal grant

11 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: AUG-08 7:40 LOCAL LAW ENFORCEMENT BLOCK GRANTS 335 Table 3. Effects of LLEBG Funding on Crime, Controlling for Police Levels Homicide Rape Robbery Assault Burglary Larceny MV Theft LLEBG (3.74)** (2.49)* (3.42)** (3.19)** (2.73)** (2.87)** (3.64)** Police levels (0.73) (4.49)** (1.96) (2.98)** (3.78)** (5.15)** (2.66)* Income (1.53) (0.30) (1.48) (0.52) (0.77) (2.99)** (1.64) Nonwhite , , , (0.38) (5.31)** (2.64)* (2.21)* (4.00)** (4.04)** (2.30)* Percent years old , , (0.09) (1.48) (0.86) (1.52) (0.21) (0.88) (2.89)** Employment (1.73) (1.91) (0.83) (0.86) (1.39) (0.18) (1.36) Observations 47,620 47,612 47,618 47,624 47,624 47,624 47,624 Units 5,054 5,054 5,054 5,054 5,054 5,054 5,054 Notes: Robust t statistics are in parentheses. All test statistics are robust to heteroskedasticity and clustering. MV = motor vehicle. *p <.05. **p <.01. programs. Spending levels from these programs were captured in six variables: UHP spending, COPS MORE spending, innovative COPS grants spending, miscellaneous COPS grant spending, Byrne program grant spending, and an additive measure of all other federal grant spending. Once again, these data were draw-downs, the actual amounts spent at the municipal level during each calendar year. The inverse associations between LLEBG spending and serious crime stood up with these controls. 11 This finding is important on two levels. First, failure to control for other federal grant programs could have led to spurious associations between LLEBG funding and crime. Second, the grant amounts could also serve as proxies for other unmeasured variables, such as agency progressiveness that would otherwise have been left out of the main models. There is no way of knowing this of course, but an agency (or city) that receives more grants than its neighbor could be more serious about crime prevention, which is something that may not have been captured by a single grant measure. The Functionally Exogenous Nature of LLEBG It is conceivable that whereas LLEBG funding could have reduced crime, crime rates may have encouraged local jurisdictions to seek LLEBG. Indeed, the LLEBG funding process required exactly this type of 11. Per a reviewer s recommendation, the author also ran models with all non- LLEBGs summed into a single variable. The results were not altered. They are available upon request.

12 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: AUG-08 7: WORRALL Table 4. Effects of LLEBG Funding on Crime, Controlling for Police Levels and Other Federal Grant Programs Homicide Rape Robbery Assault Burglary Larceny MV Theft LLEBG (3.52)** (2.06)* (3.04)** (2.49)* (2.66)* (2.30)* (5.52)** Police levels (0.65) (4.16)** (0.84) (1.22) (2.55)* (4.23)** (1.43) UHP (1.59) (1.55) (3.40)** (5.55)** (8.50)** (2.49)* (1.52) COPS MORE (0.70) (0.84) (1.07) (0.77) (3.51)** (2.33)* (2.45)* COPS innovative (1.69) (1.06) (3.85)** (2.26)* (5.22)** (3.80)** (2.94)** COPS miscellaneous (0.30) (0.66) (2.69)** (2.03)* (4.04)** (2.55)* (0.14) Byrne program (0.73) (1.27) (0.66) (0.53) (0.30) (1.11) (0.53) Other federal (1.25) (0.22) (1.34) (1.40) (1.11) (0.66) (2.02)* Income (1.34) (0.55) (1.40) (0.53) (0.70) (3.29)** (1.52) Nonwhite , , , (0.56) (5.52)** (2.90)** (1.27) (3.53)** (4.04)** (2.63)* Percent years old , , , (0.20) (0.86) (0.43) (0.38) (0.83) (1.51) (1.90) Employment (1.73) (2.13)* (0.72) (0.88) (1.16) (0.18) (1.36) Observations 38,691 38,686 38,689 38,695 38,695 38,695 38,695 Units 4,377 4,377 4,377 4,377 4,377 4,377 4,377 Notes: Robust t statistics are in parentheses. All test statistics are robust to heteroskedasticity and clustering. MV = motor vehicle. *p <.05. **p <.01. action. The formula for grant distribution followed a two-step procedure. First, states received funds based on their three-year average number of Part I violent crimes compared with the same for all other states that submitted data to the FBI (see Bauer, 2004:2). Second, individual agencies received funds following the same procedure, except allocations were based on each agency s three-year average of Part I violent crimes divided by the state total. Thus, because grant allocations were to be based on prior crime, endogeneity would seem to be a problem. Researchers rarely have the luxury of understanding the precise mechanics of a simultaneous relationship between two variables. In literature that examines police level-crime relationships, for example, researchers have generally speculated that crime boosts police levels in the same way police levels reduce (or increase) crime. Here, however, the LLEBG Program funding protocol specified, a priori, a two-way relationship with an

13 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: AUG-08 7:40 LOCAL LAW ENFORCEMENT BLOCK GRANTS 337 elaborate explanation for precisely how crime was to affect grants. This mechanism permitted easy testing to check whether, in fact, past crime was significantly associated with LLEBG. The author thus regressed LLEBG on the three-year average of each agency s Part I violent crime rate divided by its respective state s Part I violent crime rate. 12 Table 5. Effects of Past Crime on LLEBG Funding Full Model Model 2 Model 3 Model 4 LLEBG > 0 3-years past crime (1.65) (1.61) (1.89) (1.87) (0.80) Police levels (3.56)** (0.38) UHP (3.73)** (3.88)** (0.53) COPS MORE (1.57) (1.57) (0.84) COPS innovative (2.74)** (2.67)* (2.85)** COPS miscellaneous (0.33) (0.36) (0.42) Byrne program (1.71) (1.70) (0.12) Other federal (1.85) (1.86) (1.46) Income (2.64)* (2.76)** (2.30)* (0.60) Nonwhite (1.89) (1.83) (1.88) (0.69) Percent years old (0.75) (0.83) (0.55) (0.78) Employment (1.40) (1.29) (1.29) (0.43) Observations Units Notes: Robust t statistics are in parentheses. All test statistics are robust to heteroskedasticity and clustering. p <.10. *p <.05. **p <.01. The results of various models are reported in Table 5. Because past crime failed to achieve significance at conventional levels, LLEBGs were treated as functionally exogenous in the models reported throughout this 12. This estimation roughly approximated the LLEBG award process while keeping the analysis in rates.

14 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: AUG-08 7: WORRALL article. Because this finding directly contradicts the LLEBG funding formula, some explanation is in order. Recall that funding decisions were based on numbers of Part I violent crimes, not on rates. That is, the funding decisions did not take into account agency size, city population, and the like. The variables constructed for the analyses reported in this article, however, were rates. Some associations between past crime rates and LLEBG nearly achieved significance in Table 5. Had the author declared significance at the.10 level, these inverse associations may have biased the reported association between LLEBG and crime in Tables 2 4, 7, and 8. In other words, if LLEBG reduced crime, and crime in turn reduced LLEBG, then the LLEBG coefficients may have overstated the inverse association between LLEBG and crime. To guard against this possibility, the author also regressed the crime variables on endogenous LLEBG, exogenous police levels, and the controls. LLEBGs were instrumented with the second and third lags on non-llebg. The instruments passed applicable relevance and validity tests, and the inverse association between LLEBG and crime remained strong. These results are available from the author upon request. Indirect Associations between LLEBG and Crime The GAO (2005) and Evans and Owens (2007) were the first to treat federal law-enforcement spending as an instrument for police levels. Evans and Owens instrumented police levels with what they called paid officers granted (2007:188), which is a sum of grant totals from two programs: UHP and the COPS Distressed Neighborhood Program. Their logic for doing so was that hiring grants increased police levels but could not be expected to affect crime directly, except through police levels. 13 The GAO took a similar approach, but it instrumented police levels with several federal grant programs, which are the same ones also considered in this study. When police levels are instrumented with federal grants, the relationship between grants and crime effectively becomes indirect. Better stated, the federal grant variables effectively replace police levels in the main instrumental variables equations. One can estimate reduced form regressions to observe how this replacement is so. In such models, the dependent variable of interest is regressed on all instruments and exogenous regressors (see, e.g., Evans and Owens, 2007:195, Table 5). To state that an association between grants and crime is indirect may not be the technically correct description, but it is at least nearly accurate: 13. Recall that the two requirements for a good instrument are correlation with the endogenous regressor (relevance) and independence from the error term in the main equation(s) (validity).

15 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: AUG-08 7:40 LOCAL LAW ENFORCEMENT BLOCK GRANTS 339 Hiring grants may boost police forces and thereby affect crime. That is, hiring grants may not affect crime except through the police. At least this logic is offered up in previous studies (Evans and Owens, 2007; GAO, 2005). The possibility remains, of course, that hiring grants may be correlated with other unobservables, but because the police level-crime connection was of secondary concern in this article (i.e., not the key substantive concern), any instrument limitations should be taken in stride. Readers are invited to consult Evans and Owens (2007) and GAO (2005) for additional defense of using COPS grants as instruments for police levels. If, as previous research suggests, COPS hiring grants can be expected to affect crime through police levels, then it stands to reason that the same should apply in the LLEBG context. Recall that roughly 75% of LLEBG funding went to law-enforcement agencies, and several agencies used the funding to hire more officers. Thus, it is realistic to expect that LLEBG may have affected crime indirectly through additional hiring. To be sure, such an effect could have been tempered by direct linkages between LLEBG funding and crime. The alternative is that LLEBG funding did nothing to alter appreciably the ratio of sworn officers per 10,000 people. Both possibilities were explored, and the results are presented in Table 6. Table 6 contains the results of regressions of police levels on LLEBG funding and on funding from the other six federal grant programs already mentioned. 14 The individual coefficients reveal that only UHP was significantly associated with police levels. LLEBG, in contrast (along with the other federal programs), did not alter police levels. The significance of UHP has interesting implications for some results already presented. Hiring was significant in Table 5, which is not problematic because it is reasonable to expect that dedicated hiring grants increased police force sizes. The significance of hiring may also explain the significant and positive associations between police levels and crime found in Tables 3 and 4. The positive associations between police levels and crime in those tables were most certainly symptomatic of endogeneity. Accordingly, additional models were estimated in which police levels were treated as endogenous and instrumented with UHP. All other variables, which include LLEBG, were treated as exogenous. Results are presented in Table 7. Two interesting findings emerge from the results reported in Table 7. First, LLEBG funding retained its significant and inverse association with all seven index crime rates. Second, the positive associations between police levels and crime (see Tables 3 and 4) disappeared. Indeed, the coefficients became negative and significant for robbery, assault, and burglary. 14. To use instrumental or two-stage least-squares terminology, these are first stage regressions.

16 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: AUG-08 7: WORRALL Table 6. Effects of LLEBG Funding and Other Federal Grants on Police Levels LLEBG (1.56) (0.35) UHP (3.79)** (3.86)** COPS MORE (0.55) (0.11) COPS innovative (0.04) (0.80) COPS miscellaneous (0.06) (1.15) Byrne program (1.17) (1.35)* Other federal (1.54) (1.68) Income (2.33)* (2.21)* (2.33)* (2.31)* (2.31)* (2.35)* (2.39)* (2.37)* Nonwhite (0.56) (0.68) (0.52) (0.51) (0.51) (0.51) (0.49) (0.67) Percent years old (1.74) (1.72) (1.75) (1.75) (1.75) (1.77) (1.79) (1.80) Employment (0.81) (0.81) (0.80) (0.80) (0.80) (0.81) (0.80) (0.83) Observations 38,696 38,696 38,696 38,696 38,696 38,696 38,696 38,696 Units 4,377 4,377 4,377 4,377 4,377 4,377 4,377 4,377 Notes: Robust t statistics are in parentheses. All test statistics are robust to heteroskedasticity and clustering. *p <.05. **p <.01. Of all the models estimated and whose results have been reported thus far, the Table 7 models are arguably most appropriate. They accounted for endogeneity of police levels and treated LLEBG and other federal funding as exogenous. The coefficients reported in Table 7 can be interpreted as follows: For every $1 per capita of LLEBG funding, the homicide rate declined by 0.19 per 100,000. So, if a city of 100,000 people received a $200,000 grant, it could expect 0.38 fewer homicides per 100,000 people. 15 This reduction may not seem drastic, but with an average homicide rate of 4.62 per 100,000 people (see Table 1), this translates into roughly an 8% reduction. These findings are fairly consistent with previous research (see GAO, 2005, Appendix VI, Table 17). 15. Estimates near the end of Table 8 can be interpreted similarly (e.g., fewer violent crimes per 100,000 for each dollar of LLEBG spending).

17 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: AUG-08 7:40 LOCAL LAW ENFORCEMENT BLOCK GRANTS 341 Table 7. Effects of LLEBG Funding on Crime with Endogenous Police Levels (UHP Instrument) Homicide Rape Robbery Assault Burglary Larceny MV Theft LLEBG (3.60)** (2.29)* (3.63)** (2.68)** (3.40)** (2.65)* (6.06)** Police levels (1.43) (1.08) (2.28)* (2.95)** (4.03)** (1.65) (1.21) COPS MORE (0.87) (0.67) (0.33) (0.39) (1.38) (1.78) (1.72) COPS innovative (1.81) (1.11) (4.12)** (2.39)* (4.76)** (3.75)** (3.34)** COPS miscellaneous (0.13) (0.74) (2.77)** (1.95) (3.57)** (2.39)* (0.49) Byrne program (0.33) (1.11) (0.78) (0.85) (1.10) (0.72) (0.86) Other federal (1.32) (0.78) (1.42) (1.57) (1.46) (1.29) (1.47) Income (0.39) (0.17) (2.58)* (2.00) (1.66) (3.39)** (1.62) Nonwhite , , , (0.75) (6.04)** (2.35)* (1.37) (2.82)** (3.59)** (2.34)* Percent years old , , , (0.47) (0.61) (0.65) (0.34) (1.41) (1.83) (1.49) Employment (1.40) (1.69) (0.80) (1.33) (1.01) (0.15) (1.19) Observations 38,692 38,687 38,690 38,696 38,696 38,696 38,696 Units 4,377 4,377 4,377 4,377 4,377 4,377 4,377 Notes: Robust t statistics are in parentheses. All test statistics are robust to heteroskedasticity and clustering. MV = motor vehicle. *p <.05. **p <.01. Robustness Checks Table 8 presents the results of various robustness checks. LLEBG coefficients and t statistics are reported. Additionally, the base model for each row in Table 8 was the regression of the appropriate crime rate on LLEBG, endogenous police, other federal grant programs, and relevant sociodemographic controls. Researchers rarely complain when samples achieve the size of those used in the analyses reported here. But the large sample surely masked variations in the effects of LLEBG on crime that could have been attributed to jurisdictional size. In response to this concern, four additional sets of models were estimated with subsamples, which range in size from 625 cities to almost 2,500 and correspond to different population sizes (see Table 8 for the breakdown). For the most part, the LLEBG coefficients

18 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: AUG-08 7: WORRALL Table 8. Robustness Checks (LLEBG Coefficients and Associated t Statistics Reported) Homicide Rape Robbery Assault Burglary Larceny MV Theft Population 0 25,000 a (2.28)* (1.37) (2.34)* (2.03)* (1.90) (1.40) (2.90)** Population 25,001 50,000 b (1.95) (2.28)* (3.82)** (8.48)** (3.55)** (2.27)* (1.42) Population 50, ,000 c (6.18)** (4.14)** (6.46)** (5.26)** (5.82)** (2.67)* (3.22)** Population more than 100, (3.75)** (4.30)** (3.80)** (5.99)** (3.26)** (1.20) (1.57) Log-level (3.05)** (3.07)** (3.68)** (3.05)** (0.71) (1.27) (0.91) Log-log (8.62)** (1.19) (2.07)* (2.14)* (2.66)* (2.69)** (0.03) No controls (3.49)** (2.16)* (2.77)** (2.50)* (2.40)* (2.09)* (4.21)** No lags (3.28)** (2.53)* (3.92)** (3.21)** (2.91)** (3.06)** (4.65)** Two lags on LLEBG (3.02)** (2.21)* (2.88)** (2.07)* (2.49)* (2.22)* (7.30)** No clustered standard errors (5.08)** (4.22)** (5.34)** (5.43)** (4.25)** (3.76)** (5.38)** No robust standard errors (3.60)** (2.29)* (3.63)** (2.68)** (3.40)** (2.65)* (6.06)** No robust/cluster standard errors (10.32)** (8.40)** (23.70)** (22.15)** (12.15)** (7.33)** (13.86)** Heterogeneous year effects (2.97)** (1.82) (2.43)* (2.24)* (1.61) (2.55)* (2.75)** Crime growth cells (2.79)** (1.78) (2.78)** (2.21)* (1.82) (2.77)** (3.47)** Violent and property crime (2.95)** (3.69)** Aggregate crime (3.54)** Notes: Robust t statistics are in parentheses. All test statistics are robust to heteroskedasticity and clustering. Unless otherwise specified, models estimated were the same as those reported in Table 7. MV = motor vehicle. a Sample consisted of 2,485 units. b Sample consisted of 1,308 units. c Sample consisted of 695 units. *p <.05. **p <.01. remained negative and significant. The effects were minimal in the smaller cities but were pronounced in the largest cities. Next, to ensure the models were not sensitive to a linear specification, log-level and log-log models were estimated. The first of these models regressed logged crime rates on levels of the other variables. The second model regressed logs on logs. Clearly the results were not altered to a

19 \\server05\productn\c\cpp\7-3\cpp301.txt unknown Seq: AUG-08 7:40 LOCAL LAW ENFORCEMENT BLOCK GRANTS 343 significant extent. Additional models with no controls, no lags of spending, 16 two lags of spending, 17 and no adjustments for either clustering, heteroskedasticity, or both, were also estimated. The LLEBG coefficients retained their significance levels for the most part. Dummy variables for each year were included in the regressions summarized in Tables 2 7. These dummies captured year-to-year shocks that could have affected all cities simultaneously and thus could have yielded spurious associations between LLEBG funding and crime were they not included in the models. The problem is that year dummies cannot adequately capture preexisting trends in either LLEBG funding and/or police levels. If, for example, LLEBG funding was distributed during a time when police forces were expanding (clearly the case as a result of the UHP), then any association between LLEBG and crime could have also been spurious. Alternatively, if a particular jurisdiction was experiencing a significant decline in crime during the observed period, then year dummies would not have captured the trend. In response to these concerns, models with heterogeneous year effects were also estimated. These effects were calculated by (1) regressing police levels in the pre-llebg period on a linear time trend for each unit; (2) doing the same for the aggregate crime rate; (3) organizing the coefficients from each regression into quartiles; and (4) interacting the resulting cells with year dummies, for a total of 192 ( = 192) separate year effects. Similar approaches were taken in Evans and Owens (2007) and GAO (2005). These growth cells replaced the year dummies. LLEBG significance levels were not appreciably altered. Models were also estimated with only crime growth cells (4 12 = 48). The results stayed the same. Finally, models of violent, property, and total crime rates were estimated. LLEBG funding was significantly and inversely associated with each of these offense categories. All told, the results presented in Table 8, coupled with those from the earlier tables, provide some fairly convincing evidence that LLEBG funding led to reductions in serious crime throughout the United States during the analysis period. Interestingly, the effects were considerably more pronounced than those between COPS grants and crime (e.g., Zhao et al., 2002). It seems, then, that the LLEBG Program may have given more bang for its buck. The last rows of Table 8 suggest 16. These models checked whether there was a contemporaneous effect of LLEBG spending on crime. 17. These models checked whether there was a two-year delay between LLEBG spending and its effect on crime. They assumed that it took two years instead of one to put new technologies in the field, hire new officers, and otherwise spend LLEBG funds before the effects on crime could be realized.