I. INTRODUCTION

Asset forfeiture laws allow police and prosecutors to declare private assets forfeit to a government agency, with judicial approval sometimes necessary. Asset forfeiture has grown rapidly in the United States, with about $4.7 billion in deposits to federal forfeiture funds in 2012 and amounts likely in excess of a billion dollars directly seized by state and local government annually. (1) Many of the seizures are small, especially at the state and local levels, leading to increasing civil liberty concerns over the large numbers of individuals affected by forfeiture's wide reach. At the same time, proponents of forfeiture argue that it is a vital weapon in the war against crime and provides crucial funding for law enforcement.

The chief conditions for forfeiture at the federal level are that the assets have been used in, or derived from, a suspected criminal activity. Most states have adopted similar approaches. However, forfeiture is linked not to criminal activity generally, but to particular types of crime, especially drug-related crime. Further, and controversially, police agencies often can retain the assets that they seize. Given the discretion inherent in policing and prosecutorial decisions, this suggests that police arrest patterns may be influenced by the presence of asset forfeiture opportunities: police may choose to focus more on forfeiture-related arrests due to the income that they generate.

Proponents of asset forfeiture argue that it provides an effective tool in the fight against drug crime and provides vital funding for fighting a broader range of crime. Critics argue that the use of asset forfeiture creates an intrusion upon civil liberties at odds with American tradition and the understanding of the ordinary citizen. Both proponents and critics often accept without question the idea that police retention of forfeit assets achieves its goals of encouraging additional arrests for drug crimes; the dispute then centers on whether the benefits are worth the costs. However, the effectiveness of asset forfeiture is an empirical claim that must be judged against evidence. We address the empirical questions directly: does police access to forfeiture funds lead to greater policing of crimes not associated with forfeiture, an income effect argument, and does such access lead to changes in arrest patterns? We demonstrate that police retention of funds does have statistically measurable effects on crime clearances and arrest patterns, but that these effects are immaterial in economic terms.

Section II describes asset forfeiture and its institutional setting. Section III explains the relationship of our work to the existing literature. Section IV provides our modeling of police agency behavior based on the institutional nature of policing and of forfeiture. Section V discusses the data and the testing approaches, and Section VI provides the results of our empirical analysis. Section VII provides a concluding assessment.

II. ASSET FORFEITURE

Asset forfeiture is a complex set of laws and practices; here we explain the elements important to understanding our empirical research and its place in the broader controversies surrounding forfeiture. American asset forfeiture occurs at the federal, state, and local levels, and may be a criminal or a civil proceeding. Criminal asset forfeiture occurs when property used in the commission of a crime, or property gained illegally through crime, is forfeit to a government upon criminal conviction of the owner. Criminal asset forfeiture usually requires judicial due process and proof beyond reasonable doubt with respect to the underlying crime and a demonstration that the forfeit assets facilitated, or were the fruits of, that crime.

Civil asset forfeiture includes non-judicial summary and administrative actions as well as judicial civil seizures (Edgeworth 2008). Summary actions involve property that has no lawful purpose (contraband) and in general lead to the destruction of the seized material. Administrative forfeiture refers to an uncontested seizure of property by a government agency, the seizure typically initiated due to a suspected relationship of the property to a crime. Judicial civil forfeiture usually involves a party, typically the private owner of an asset, protesting a seizure, thus moving the action from a non-judicial administrative action to a judicial civil proceeding. Legally, civil forfeiture is an action against an object, such as personal or real property or financial instruments, including cash. In such in rem ("against a thing") proceedings, the property itself is the object of legal action. As the defendants are things rather than people, they are not subject to the same level of due process as in a criminal proceeding. Typically a government need show a connection of the property to an alleged criminal act through probable cause or a preponderance of the evidence, weaker standards than the "beyond a reasonable doubt" of a criminal proceeding. The underlying suspected criminal activity need not be prosecuted; an in personam ("against a person") proceeding is not a requirement for an in rem proceeding against an asset. The owner of the asset has the legal status of a third party intervener in the government's action against the property. (2)

In U.S. criminal justice, in rem proceedings were little used until 1984, when Congress passed the Comprehensive Crime Control Act (CCCA). The CCCA required that federal forfeiture proceeds be used for law enforcement and created the Assets Forfeiture Fund (AFF) within the Department of Justice to administer this. The CCCA was explicitly part of the "war on drugs" and greatly facilitated civil asset forfeiture for property associated with illegal drug activity (Benson et al. 1995). States increasingly followed the federal example in the 1980s, including the requirement that forfeiture proceeds be used for law enforcement. (3) Federal, state, and local law enforcement also increasingly cooperated in forfeiture, with forfeit assets distributed to individual agencies based in part on their contribution to the investigation. The CCCA also instituted federal "adoption," effective in 1986. State and local agencies could request that the Department of Justice "adopt" and then share the fruits of a local drug-related asset seizure even if a federal agency were not involved in the seizure. The Department of Justice would then take possession of the assets during the forfeiture process and return a percentage to the originating agencies. (4) By design, civil forfeiture is far easier for an agency to apply than criminal forfeiture and has come to dominate the value of assets seized.

During the 1990s repeated legal challenges to asset forfeiture were rebuffed by the Supreme Court; see Cassella (2013) for a legal history of that period. However, asset forfeiture, particularly civil forfeiture, has remained highly controversial. The lower standard of proof of an in rent proceeding means that an individual can be deprived of property without having been shown to have committed a criminal act. This has led to numerous abuses in which a police agency will seize assets in connection with an arrest and then retain those assets even if charges are dropped or the arrest does not otherwise lead to a conviction. (5) Nevertheless, asset forfeiture has grown remarkably in the three decades since the passage of the CCCA, in both scale and breadth. Table 1 provides one commonly cited measure, the value of total net deposits to the federal Asset Forfeiture Fund. While this must be interpreted with some caution--2012, for example, includes large financial forfeitures arising from alleged money laundering--it does give some sense of the scale and considerable growth in the programs using forfeiture. (6) The breadth of offenses subject to asset forfeiture has increased greatly at the federal level, due in part to provisions of 2000s Civil Asset Forfeiture Reform Act (CAFRA) (Cassella 2013). To some extent these federal changes have spread to state law, but state and local forfeiture remains heavily associated with drug crime. Asset forfeiture is also being increasingly adopted in other countries, for example, several Canadian provinces have passed civil asset forfeiture laws in the last 15 years (Gallant 2012).

Proponents of civil asset forfeiture make at least five sets of arguments. First, the presence of asset forfeiture for some crimes leads police to concentrate more resources on those crimes, especially if police are able to keep the proceeds of the forfeitures. This in turn is assumed to lead to more arrests and ultimately reductions in the rate of those crimes. Second, if police are allowed to keep forfeiture proceeds, asset forfeiture leads to greater policing of all crimes, assuming policing is a normal good for police forces for all crimes, including those that do not involve civil forfeiture. Third, the seizure of assets disrupts the illegal activity, particularly the drug trade. The assets used to further that trade no longer are available if forfeited, and the proceeds of that trade can no longer be reinvested if forfeited. Fourth, the potential loss of assets provides a deterrent effect. Finally, in some cases the victims of a crime may be able to receive compensation from forfeiture proceeds. (7)

Critics argue that provisions that allow police to retain seized assets can distort police incentives. This has led to a widespread view that police emphasize forfeiture-related arrests, which critics have described as "policing for profit" (Blumenson and Nilsen 1997; Williams et al. 2010; Worrall and Kovandzic 2008). (8) This may lead to policing activity at odds with civilian control, which is exercised in part through the normal budgetary processes. Critics ignore or are dismissive of the idea that forfeiture leads to greater policing of other crime (Benson 2010). Critics further object that asset forfeiture, at least in its civil forms, deprives individuals of property without due process of law and generally violates American beliefs concerning civil liberties. Finally, critics maintain that forfeiture is often at odds with the U.S. Constitution, despite U.S. Supreme Court rulings to the contrary in the 1990s (Warchol and Johnson 1996; see also the amicus curiae briefs filed in Kaley v. United States, Howe 2013).

Police retention of forfeit assets plays a central role for proponents, who not only point to police incentives to put more resources into forfeiture-related arrests, but emphasize the additional funding for police that results, to argue that drug forfeiture proceeds provide vital funding for policing generally, not just in the war on drugs. Police testimony in opposition to drafts of the CAFRA legislation repeatedly emphasized the importance of forfeiture funds to policing. (9) Worrall (2001) surveyed police agencies and found strong support for the statement that asset forfeiture is a necessary budgetary supplement. In the Kaley case (Howe 2013), the Department of Justice brief approvingly cited previous cases that concluded that "Forfeiture provisions are powerful weapons in the war on crime." and that the "interest in using the profits of crime" to fund criminal justice activities "should not be discounted" (United States Department of Justice 2013a, 41, 42).

The effect of forfeiture funds on policing should not be assumed: it is an empirical matter. The scale of forfeiture and the spread of forfeiture laws to other substantive and geographic areas only heighten the importance of understanding the effects of U.S. drug forfeiture on policing. We focus on the effects of forfeiture on crime clearances (10) and arrests because police retention of forfeit assets is the most intense area of controversy, because it directly relates to arguments for and against forfeiture, and because the public policy outcomes--patterns of policing--are of great importance. We restrict our attention to police agencies below the federal level; federal activity is certainly important, but must await separate work. (11) We focus on forfeiture associated with drug crimes, partly for data reasons, but also because this is the primary source of asset forfeiture funds at state and local levels.

III. RELATIONSHIP TO THE LITERATURE

A sizeable literature addresses asset forfeiture, but in general this has assumed the effects of forfeiture on police behavior. Much of the existing scholarship addresses legal or civil liberty concerns. Our work is empirical and so we focus on the empirical literature, which has yielded partial and somewhat contradictory results with respect to police behavior.

In work spanning two decades, Bruce Benson, David Rasmussen, and various coauthors have studied the effects of asset forfeiture on the allocation of police effort. (See Benson 2010 for a fuller account of this line of literature.) Examining Florida data, Benson et al. (1995) found that forfeiture had a statistically significant positive correlation with subsequent year police non-capital expenditures, with an implied expenditure elasticity with respect to forfeiture of .4 to .7. The authors conclude that this finding is logically consistent with the hypothesis that police might devote more resources to drug enforcement but did not test this hypothesis. Mast et al. (2000) test the hypothesis that local police responded to the incentives created by the CCCA, exercising discretion to pursue drug crimes more intensively in states that allow greater agency retention of forfeit assets. While finding a link between forfeiture regimes and drug arrests, they find inconsistent implications in the arrest rates for non-drug crimes depending on the population analyzed. Summarizing these and other studies, Benson (2010) concludes that this line of literature provides "indirect support for the contention that the upsurge in drug enforcement starting in 1984-85 is a result of the incentives created by federal legislation and DOJ decisions that alter incentives for state and local police."

Worrall (2001) develops a survey of municipal and county law enforcement that queried police attitudes toward forfeiture, finding that "dependence" on civil asset forfeiture is positively associated with revenues generated from past forfeiture activities and inversely related to fiscal expenditures. He infers a conflict of interest between "effective crime control and creative fiscal management" but does not seek to tie forfeiture to patterns of policing and arrests.

Bishopp and Worrall (2009) note that drug arrests have increased for the last 30 years while drug use, measured by surveys, has fluctuated without systematic trend. Since this period coincides with the rise of civil asset forfeiture in response to drugs, they conjecture that forfeiture may play a role in encouraging arrests but not in deterring drug use. However, their main finding is that state drug asset forfeiture laws have no discernible predictive effect on drug arrest rates. This contrasts directly with the finding of Mast et al. (2000). A secondary finding is that lagged forfeiture activity has no effect on the drug arrest rate, undercutting the idea that agencies become dependent on forfeiture proceeds.

Worrall and Kovandzic (2008) also find no clear evidence that there is more forfeiture in states that permit local agencies to receive all proceeds, but that there is clear evidence that police forces in states with restrictive policies engage in more federal adoptions. Using a broader approach to the relative restrictiveness of state forfeiture laws, while also relying on a cross-sectional approach, Holcomb et al. (2011) also find that agencies in more restrictive states receive more in federal equitable sharing proceeds, implying that state-level restrictions lead agencies to use the more generous federal adoption rules.

Baicker and Jacobson (2007) find that county and local governments often offset police seizures by making compensating reductions in police budgets, partially undermining the statutory incentives. They conclude that police respond quite strongly to the real net incentives for seizure (value of forfeiture netted for budget decreases) by increasing the drug arrest rate, but respond only weakly to the nominal incentives, the gross value of forfeitures. However, their findings are based on an allocation of the value of all forfeited assets from the federal judicial district to the county level, which in turn is compared to an aggregation to the county level of local police budgets. Consequently the implications for individual agency behavior are hard to discern.

The existing literature thus reaches inconsistent conclusions on the effects of asset forfeiture upon police activity, in particular upon arrests. Benson, Rasmussen, and various coauthors find that differences in asset forfeiture laws do affect police activity. This is supported either in two steps (forfeiture leads to increased budgets, which logically would encourage police to spend relatively more time on forfeiture-related policing), or directly through statistical testing in Mast et al. (2000). On the other hand, another set of studies (Bishopp and Worrall 2009; Holcomb et al. 2011; Worrall and Kovandzic 2008) finds that there is not a statistically significant impact of differing forfeiture regimes on drug arrests. These authors provide a possible reason for this in showing that police in states with more restrictive regimes turn more frequently to federal adoptions, thus circumventing the state-level restrictions on the seized assets. Finally, Baicker and Jacobson (2007) find a weak (but statistically significant) correlation between forfeiture regimes and drug arrests. No existing study finds a significant link between drug forfeiture and non-drug arrests.

The literature widely cites the fact that drug arrests have increased both absolutely and as a proportion of all arrests, but this of course could be due to many factors other than the rise of civil forfeiture. The finding that police agencies are cognizant of forfeiture and consider it important appears robust, being reflected not only in survey data (Worrall 2001), but in the finding that police agencies avoid local restrictions on the use of forfeiture proceeds by seeking federal adoption, and also in a host of anecdotal information. However, this finding does not in itself imply a change in police behavior due to forfeiture. Police may simply take advantage of forfeiture when available, including using federal adoption, without engaging in more forfeiture-related arrests.

Finally, the existing literature provides considerable qualitative discussion of police motives, but almost no formal modeling of the police decision making under budget constraints. The descriptions that characterize the literature make little attempt to distinguish preference changes from budget changes, or income effects from substitution effects in arrest patterns. This leads to confusion in the interpretation of results, such as the assumed relationship between greater forfeiture proceeds and lower pursuit of non-drug crimes in Benson et al. (1995).

Our work extends the literature in several ways. First, existing literature has relied heavily on instrumental variables applied to cross-sectional data, in particular the nature of state forfeiture regimes. However, as described above, recent work has concluded that police circumvent state forfeiture restrictions by turning to federal adoption, suggesting that state forfeiture regimes are an ineffective instrument in determining forfeiture's effect on behavior. Our research has revealed no other instruments that can reasonably be modeled as allowing comparison of agency behavior with and without the possibility of forfeiture. Not only do forfeiture amounts--and therefore the effects of forfeiture--appear to have little relationship to the state forfeiture instruments previously used, there is a vast degree of heterogeneity among agencies in how they approach forfeiture, which cannot be fully controlled in cross-sectional studies. Consequently, we have created a panel that incorporates forfeiture, budget, arrest, and crime data at the level of the individual agency. This allows us to condition upon unobserved agency effects, as explained in more detail in Section IV. In this, our work is similar in motivation to Cornwell and Trumbull (1994), who argue for panel data to address deficiencies in unobserved heterogeneity in prior cross-sectional work in the economic model of crime.

Second, we work from a national sample of police agencies, developing data sets that are as broadly based as possible. Much of the previous work has considered only particular states or has worked at the county level, aggregating individual agencies.

Third, we recognize the income effect argument of forfeiture's proponents and test for the effects of drug-related forfeiture on the intensity of non-drug related policing.

Fourth, we explicitly treat clearance rates and arrests as dependent variables. As noted above, some of the existing literature follows popular accounts that identify an apparent incentive to change behavior and then conclude that behavior must have changed. Our central purpose is to determine whether asset forfeiture in fact does change police behavior. We focus on clearances and arrests, rather than measures of internal agency allocations of resources, because of their direct relevance to the policy debate surrounding forfeiture.

Finally, the existing literature has reported tests of statistical significance but has provided relatively little discussion of economic significance, the impact of relationships found to be statistically significant. An important element of our work is to assess whether possible causal relationships are significant in practical terms.

IV. INSTITUTIONAL ASPECTS AND MODELING IMPLICATIONS

Police agencies face tradeoffs in the allocation of their resources, for these resources are of course subject to constraints. While discussion of forfeiture routinely speaks of the distortion of police incentives due to the ability to retain forfeited assets, this may confuse police preference sets with police budget constraints. Forfeiture can change policing not only through an alteration of underlying preferences, but by relaxing the budget constraint. Specifically, forfeiture linked to certain activities, such as drug offenses, lowers the net cost of pursuing those activities, in effect acting as a rebate. In a simple model of constrained optimization, the introduction of forfeiture linked to drug offenses will cause drug arrests to increase due both to a substitution away from other activities and to the greater income available for policing. Other activities will decrease due to substitution effects, but increase due to income effects, with the direction of the net effect being an empirical matter.

While recognizing these basic incentives, we need note several institutional aspects of policing and forfeiture that may affect forfeiture's impact on police behavior and inform our empirical modeling.

Police budgets are set by a broader civil authority such as a local or state government. Police may well have greater discretion in the use of forfeiture funds, which they may perceive as having earned, than in the use of operating budgets. Consequently the view that forfeiture funds act as a simple budget supplement should be considered an empirical matter as well: do police agencies react similarly to changes in normal budget authority and changes in forfeiture proceeds?

While the economic model of agency behavior suggests constrained optimization, the relevant constraint on policing generally, and especially on a subset of activity such as arrests, may be something other than funding, at least in the short run. For example, agencies may experience a temporary increase in funding, but not be able to alter patterns of policing significantly in response due to embedded behaviors. We have to allow for the possibility that forfeiture does not affect rational optimizing policing behavior due to structural rigidities. (12) Following the public policy debate, our concern with police behavior is in the intensity of policing, measured by crime clearances, and arrest patterns. Agencies may reallocate resources in response to forfeiture in ways that do not affect either of these outcomes.

As described above, forfeit assets may go to police agencies, but this is subject to a variety of local and state controls. Forfeiture proceeds may not benefit police agencies dollar for dollar; in the extreme case, police may not share in the proceeds at all. Further, the ability to spend the asset proceeds or to use forfeit physical assets typically lags the actual seizure considerably, since they cannot be anticipated with certainty and typically have to work through internal accounting and management procedures. The assets themselves vary widely in terms of their liquidity; by value, a majority of assets seized are cash, but they can also include personal and real property, which can be either liquidated for cash or retained for police use. All this recommends a focus on actual forfeiture receipts rather than the value at seizure of forfeited assets. We will follow the existing literature (Bishopp and Worrall 2009) and incorporate a lag between receipt of forfeit assets and their effects on arrests.

Summarizing, an economic model that treats police agencies as rational optimizers suggests that police will favor forfeiture-generating arrests through both substitution and income effects. Substitution and income effects have opposite signs, with indeterminate net effects, for other arrests. We wish to separate the analysis of forfeiture from that of other budget changes rather than treating forfeiture as simply an increase in overall budget. Any empirical test must allow for the possibility that forfeiture has no effect on patterns of police activity due to objectives not captured in this analysis or due to institutional aspects that make it difficult to adjust behavior. Finally, forfeiture's income effects are likely delayed somewhat from receipt of the assets.

V. TESTING STRATEGY AND DATA

Our objective is to determine whether the presence and extent of asset forfeiture substantially affects clearances and arrests. (13) While studies of clearances and arrests are often done at the county or state level, this can create aggregation difficulties given that most police agencies are local; two agencies with very different behaviors, facing very different circumstances, may be located in the same county. Our level of analysis is the individual local agency. To avoid the problems of much of the previous empirical work arising from the limitations of cross-sectional data, we have adopted a panel approach, developing longitudinal data sets that allow consideration of asset forfeiture, crime levels, arrest behavior, and covariates over time at the individual agency level.

We estimate unobserved (fixed) effects equations of the form:

(1) [Y.sub.i,t+1] = [rho][F.sub.it] + [mu][B.sub.it] + [beta][X.sub.i,t+1] + [[alpha].sub.i] + [[delta].sub.t] + [u.sub.it].

The dependent variable [Y.sub.it+1] is a measure of police activity, where i indexes the agency and t indexes time. [F.sub.it] is a measure of forfeiture proceeds by agency and time period, [B.sub.it] a measure of agency operating budgets by agency and time period. [X.sub.i,t+1] is a matrix of covariates that potentially affect the dependent variables, [[alpha].sub.i] represents the unobservable agency fixed effects, [[delta].sub.t] represents the year dummy variables, and [u.sub.it] represents idiosyncratic error.

We have adopted a fixed effects approach because several unobservable factors are likely to vary substantially among agencies but change only slowly for any given agency; we present the results of the formal testing against a null hypothesis of random effects in the next section. Possible unobservables include at least the following: agency attitudes toward forfeiture, perhaps conditioned by civil authorities, perhaps purely internal; policing practices that affect outcomes, such as the relative roles of patrol-and-respond and community policing; social characteristics of the community that cannot be consistently quantified; and record-keeping practices. As we regress the time-demeaned dependent variables on the time-demeaned independent variables, the unobserved agency fixed effects [[alpha].sub.i] drop out of the regression estimates.

We measured police activity through two related approaches, reflecting the available data on crimes and arrests. Both approaches rely on data gathered by the FBI; Table 2 provides the relevant crime classification codes used by the FBI for statistical purposes. Our first approach concerns offenses for which we have crime incidence data, codes 01-08 in Table 2, which we will term "serious" crime. (14) Available data allowed us to calculate a clearance rate for these offenses, the proportion of reported crimes that are cleared, either by arrest and charging or by proving unfounded upon investigation. (15) We computed clearance rates at the agency level, aggregating across the offenses (codes 01-08) for which we had both clearance and incidence data. This allowed a measure of policing intensity for these crimes and therefore a measure of whether police pursue these crimes more, or less, in response to changes in forfeiture funds received.

For other crimes, including drug crimes, agency arrest data are available but reported incidents are not. For many crimes outside codes 01-09, including drug possession, the "victim" is also often the perpetrator; for others, victims are especially reluctant to report the crime. A clearance rate generally cannot be calculated and often would not be sensible. Consequently, in our second approach we measured police activity by dividing overall arrests into three categories: arrests for codes 01-09 crimes (serious crime), arrests for code 18 (drug crime), and all other arrests. This provides a measure of the deployment of policing resources by objective.

We focus on arrests, including their use in the numerator of clearance rates, for several reasons. First, the arguments for and against forfeiture largely revolve around arrests. Proponents see forfeiture as leading to higher arrest rates in the targeted crimes, with the resulting income fueling further arrests of all kinds. Critics see it as lowering arrests for all but the targeted offenses while raising arrests for the targeted offenses at the expense of civil liberties. Second, federal and many state strictures on the use of forfeiture proceeds prohibit their application to important budgetary areas such as administration or pensions, leaving patrol and arrests as more significant areas in forfeiture-funded activity than in operating budget-funded activity. Third, arrest data are far more readily quantifiable than other measures of police activity. Fourth, the existing policing and crime literature routinely uses arrests as measures of police activity or productivity; see, e.g., Zhao et al. (2011). Finally, as one author notes, "Of the many duties performed by police officers, the arrest of suspects remains at the core" (Lyman 2011).

We have assembled a panel data set comprised of 3 years of observations for a large set of U.S. police agencies. (16) The basic panel is built from three iterations of the Law Enforcement Management and Administrative Statistics (LEMAS) survey (ICPSR Law Enforcement Management and Administrative Statistics 2011). Conducted by the Bureau of Justice Statistics, the LEMAS survey collects data from police agencies every 3-4 years. It seeks data from all agencies with over 100 sworn full-time officers (termed "self-reporting" agencies) and from a stratified sample of smaller agencies. This typically results in about 2,800 responding agencies, with response rates around 90% overall and about 95% for self-reporting agencies. We used the LEMAS survey responses for 2000, 2003, and 2007, which in general covered activities in fiscal or calendar years that incorporated June 30, 2000, June 30, 2003, and September 30, 2007, respectively; however, the questionnaires asked agencies to report forfeiture amounts for the preceding calendar years (1999,2002, and 2006). We constructed a balanced panel of police agency staffing, operating budgets, and forfeiture proceeds based upon the LEMAS self-reporting entities, meaning that our panel contains relatively large police agencies. In 2000 and 2003, LEMAS included forfeiture proceeds arising from drug crimes only. In 2007, gambling and other sources of forfeiture were added, but drug forfeitures continued to be identified separately. Throughout this study, we relied on the drug forfeiture amounts alone.

For crime and arrest data, we relied on two data sets issued under the FBI's Uniform Crime Reports (UCR) as provided by the ICPSR: the "Offenses Known and Clearances by Arrest" (the UCR Offenses Known) (ICPSR Uniform Crime Reports Offenses Known and Clearances by Arrest 2013) and the "Uniform Crime Reporting Program Data: Arrests by Age, Sex, and Race" (the UCR ASR) (ICPSR Uniform Crime Reports: Arrests by Age, Sex, and Race, Summarized Yearly 2013). The former contains incidents reported and cleared by agency for crime codes 01A through 08 (see Table 2) and was our source for clearance rates. The latter provides the number of arrests by nature of the offense, broken down by the age, gender, and race of the arrested party, and was our source for arrests by type of crime.

For each agency we calculated the clearance rates from the UCR Offenses Known data, based on all reported offenses, and also on the subset of Schedule I offenses. We merged the Offenses Known data with the LEMAS panel and developed three explanatory variables from the LEMAS surveys that were included in all of our tests: forfeiture proceeds per sworn officer, operating budget per sworn officer, and number of sworn officers per capita. Sworn officers are those employees who can carry out arrests. The forfeiture and budget variables measure the funding available per sworn officer. The sworn officers' per capita variable measures the "density" of police with general arrest powers in the population. We chose this approach, rather than simply measuring per capita forfeiture and budget amounts, to allow for the possibility that the number of officers has explanatory value independent of the level of funding. Nominal forfeiture and budget amounts were converted to constant dollars using the Bureau of Economic Analysis' GDP deflators. The natural log of the population served as a scaling variable; the UCR Offenses Known data provided the population served by each agency. We also included year dummies for the second and third years of the panel, LEMAS years 2003 and 2007, in all of our regressions, reflecting the fact that this was a period of decreasing serious crime rates generally in the United States.

For the regressions of per capita arrest rates, we included several additional variables that have been recognized as crime covariates in earlier work: the unemployment rate, the proportion of minorities in the population, and the proportion of male 15- to 24-year-olds in the population. For these covariates, we used county-level data obtained from the Census Bureau and the Bureau of Labor Statistics, which we matched to the agency data through state and county FIPS codes. We used county-level data because agency-level data is not available and because the ability of people to quickly cross agency jurisdictions suggests that a broader geographic basis is appropriate.

We analyzed local and county agencies separately. This separation was in part for consistency with much of the previous literature, in part because of very different policing patterns for the different types of agencies. County police--typically called sheriffs' offices--vary widely in their range of duties, some serving as full-service law enforcement agencies, some focusing primarily on traffic enforcement. They are much likelier to be involved in jail operations, court security, and process service than are local agencies. (17) Consequently, we report results separately for local and county agencies. We did not include state agencies; state police often do not have general arrest powers and devote a large portion of their patrol activity to traffic control.

Table 3 provides descriptive statistics based upon our local agency panel data set. Table 3 demonstrates a fact widely ignored in the academic literature: forfeiture does not form a large portion of revenues for these agencies. Arguments for its influence must rest on observation that forfeiture funds are different in important ways from general operating funds, including in the incentives that they create, the effects of occasional large forfeiture inflows for individual agencies, or the effects on smaller agencies, which are not represented in this sample. While Table 3 indicates relatively little change over time in certain variables, this disguises fluctuations at individual agencies, especially in forfeiture amounts, that are captured in the panel analysis.

VI. RESULTS

A. Policing Intensity

We test first whether forfeiture has a material effect upon clearance rates and whether any such effect differs from that of normal operating budgets. (18) Arrests for crimes coded 01A through 08 rarely lead to forfeiture, suggesting that drug forfeiture may decrease clearance rates due to a substitution toward drug policing, but may increase clearance rates due to an income effect. This leads to a pair of null hypotheses:

HI: Forfeiture receipts do not affect the clearance rate.

H2: Forfeiture's effects upon the clearance rate do not differ significantly from those of normal operating budgets.

We included the LEMAS variables (budgets, forfeiture, and number of sworn officers) for year t (2000, 2003, 2007), while measuring the offenses data at year t + 1. (19) The lag from the year of forfeiture receipt to the measurement of the clearance rate represents the delay in the realization of the income effect from forfeiture, as noted above. The effects of changes in budget or forfeiture per officer are likely to be subject to diminishing returns, so we included quadratic terms for both. Our other regressors are number of sworn officers per capita, the natural logarithm of the population served by the police agency as a scaling variable for policing responsibilities, and year dummies for the second and third periods of the panel. (20)

Results for clearance rates are shown in Table 4. The coefficients on both operating budget per officer and forfeiture receipts per officer are positive in the level of changes (increases in budget or forfeiture imply increases in clearance rates, and decreases the opposite), but with negative quadratic terms: the effects diminish with the levels of both budget and forfeiture. The forfeiture coefficients are significant at the 1% level in a two-sided test. However, inference based on statistical significance must be approached with some caution. The agencies are not a random sample of local police agencies; they are the self-reporting agencies from the LEMAS surveys, meaning that they are large local police agencies, those with 100 or more sworn officers. The high coefficients relative to the standard errors for the forfeiture variables are in this sense population descriptors for large police agencies in 2000, 2003, and 2007; the statistical significance is a statement that the correlation between the forfeiture variables and the clearance rate is not likely to be due to unobserved random effects. However, the 3 years of the LEMAS survey are a random draw from the 8-year period 2000-2007 in the sense that the years chosen resulted from administrative constraints at the Bureau of Justice Statistics; there is no reason to expect systematic bias relative to the models' variables. In this respect, the results for the 3 years of available data allow inference with respect to the years not sampled.

The number of sworn officers per capita has a positive coefficient significant at the standard 10% level. This is consistent with previous work that found positive impacts of number of officers on policing intensity within a community (Evans and Owens 2007). Interestingly, the scaling variable, the natural log of population, is highly significant with positive sign; this suggests that areas with growing populations will have increasing clearance rates, those with declining populations, the opposite. If population change is positively correlated with prosperity, or with traits of social responsibility and sense of community, the associated change in clearance rates would be unsurprising.

The year dummies are consistent with the overall averages for clearance rates reported in Table 3. The clearance rate was little changed from 2000 to 2003, and increased materially from 2003 to 2007. The regression shows that this result holds when disaggregated to the agency level.

While we can reject the first hypothesis above statistically, the materiality of the forfeiture coefficients is of key importance for understanding the effects of forfeiture changes, both absolutely and relative to operating budget changes. We can gain a sense of scale based on the mean values shown in Table 3. (21) An increase in forfeiture of $100 per officer from the mean of $723.50 per officer would increase the clearance rate by .57/1,000 against the mean clearance rate of 265.03/1,000. An increase in forfeiture of 10% from the mean would increase the clearance rate by about .16%. The results suggest that forfeiture proceeds do lead to more intensive police responses for crime codes 01A through 08, but that while statistically significant, the effect is very small. At the observed scale of forfeiture, the overall effect on clearance rates cannot be considered sufficient to support an argument that forfeiture funds have a material impact in the battle against serious crime.

A comparison of the coefficients on forfeiture and budget changes is informative. The coefficient on BUDGOFCR is only about one-twentieth the size of that on forfeiture and the difference in the coefficients is significant at the 5% level, allowing us to reject the second null hypothesis above. The negative quadratic term means that budget effects diminish with scale. (22) In all, a budget change of a given amount per officer has a much smaller effect than the already minor effect of forfeiture proceeds. (23)

The calculated clearance rate is the sum of incidents cleared for crime codes 01A through 08 divided by incidents reported for the same codes. This included two codes, manslaughter (01B) and simple assault (08), that are not part of Index I crimes as usually defined; see Table 2. To allow greater comparability with other work that focuses on Index I crimes, we ran the same specification as reported in Table 4 after removing manslaughter and simple assault from the dependent variable. The results were similar in terms of statistical significance to those for all reported incidents. The coefficient for the level of forfeiture per officer decreased from 7.74 to 6.87, remaining at an economically immaterial level.

B. Arrest Patterns

Our second set of tests compares forfeiture's effects across categories of crime. As described in Section V, offenses are not reported for crime codes other than 01A through 08, both because the idea of reported offenses makes little practical sense for many of the crimes and because the data are not gathered systematically. Consequently, to assess the impact of forfeiture on the policing of drug and other crimes, we tested whether forfeiture affects arrest patterns, dividing arrests into those related to codes 01-09 ("serious crime"), code 18 (drug crime), and all other crime. For each of the three categories, we quantified the dependent variables as the number of arrests divided by the population served by the police agency. We tested each category against the following hypotheses:

H3: Forfeiture receipts do not affect the subsequent per capita arrest rate.

H4: Forfeiture's effects upon the per capita arrest rates do not differ significantly from those of normal operating budgets.

The timing approach remains the same as for clearance rates. The test thus addresses the "dependence" argument with respect to forfeiture, that agencies become dependent upon forfeiture proceeds and thus pursue more forfeiture-related arrests, and fewer other arrests, in later periods (Bishopp and Worrall 2009; Worrall 2001) Results are shown in Table 5. Changes in forfeiture per officer did not have statistically significant effects on arrests for serious or other crime, but did increase arrests for drug crime with a coefficient significant at the 10% level. The null hypothesis of no effect of forfeiture on subsequent arrests thus cannot be rejected for the first two categories of crime, but can (weakly) for drug crime. Changes in budget per officer are positively associated with changes in per capita arrests, at a decreasing rate, for serious crimes, with statistically significant coefficients. Budget per officer did not have statistically significant effects for drug crime or other crime. Coefficients on the linear forfeiture terms were greater than those on the equivalent budget terms, implying that a $1,000 change in forfeiture has a greater impact than a $1,000 change in operating budgets; however, the differences between the forfeiture and budget coefficients are not statistically significant and we cannot reject the second hypothesis above for any group of arrests. For the number of sworn officers per capita, coefficients for all three categories of crime are positive but not statistically significant.

Before turning to the covariates, we need to consider the economic significance of the coefficients for forfeiture. In all three cases, the absolute levels of the coefficients for forfeiture are very low and one cannot view changes in forfeiture per officer as being a material determinant of changes in the arrest rates. For example, the coefficient on forfeiture is statistically significant at the 10% level with respect to drug crimes, but at the mean levels of forfeiture and drug arrests the elasticity is only .019, meaning that a 1 % increase in forfeiture implies a .019% increase in drug arrests. (24) As with the results on clearance rates, statistical significance does not imply materiality.

Population changes are negatively associated with arrest rates per capita for all categories, possibly suggesting that population growth is associated with declining crime per capita. The unemployment rate and minority proportion of population have the expected signs and statistically significant coefficients for the first category of arrests, but the opposite signs or insignificant coefficients for drug and other arrests, suggesting that police concentrate resources on more serious crimes as these covariates increase. The proportion of the population comprised of males aged 15-24 does not have a significant effect, perhaps reflecting the very small variation in this variable over the period 2001-2008 for nearly all agencies. Finally, the coefficients on the year dummies show that, relative to the base year, arrests for serious crimes decreased and arrests for drug crimes increased.

C. Further Tests and Robustness Checks

We ran the same regressions as described above on the set of 103 counties for which we were able to construct a balanced panel that included offenses known and arrests. For clearance rates, forfeiture changes had effects similar to those for local agencies: statistically significant positive coefficients with negative quadratic terms. The coefficients on the linear terms, 12.42 for all serious crime and 15.75 for Index I crime, were about twice the size of those for local agencies, but again were at economically insignificant levels. For arrest patterns, forfeiture (and budget) changes had no statistically significant impact except with respect to forfeiture's effect on "all other" arrests, which had a negative coefficient significant at the 5% level. While this may reflect a substitution away from policing of less serious crimes due to drug forfeiture, the coefficient is small economically. We provide full results in the supplementary online text (Appendix S1). Overall, the county data reinforce the outcomes on clearance rates noted for the local data while showing little pattern in the arrest data.

As described earlier, we constructed the panel so that the budget and personnel data from the LEMAS survey in year t were associated with the offenses and arrest data in year t + 1. We believe that this is the best specification given the lags inherent in applying forfeiture funds, but we also tested clearance and arrest outcomes using offenses and arrest data in year t. The results for clearance rates are similar to those in our basic tests; the coefficients on the forfeiture terms are somewhat smaller but still statistically significant at the 1% or 5% levels. The results for the budget terms also show smaller coefficients but, interestingly, much tighter standard errors and statistical significance at the 5% level. This may be due to a tighter alignment of the budget periods and arrest periods. However, in all cases the coefficients are economically very small. (25) We provide full results in the supplementary online text (Appendix S1).

As forfeiture receipts precede the dependent variables in our panel structure, we do not face the endogeneity problem that forfeiture causes drug arrests and drug arrests simultaneously cause forfeiture. The fixed effects panel structure greatly reduces the danger of endogeneity created by omitted variables, since variables that are fixed within agencies drop out of the specification. Indeed, the likelihood of such agency fixed effects motivated the use of a fixed effect panel approach. However, if changes occur in an unobserved variable during the period of analysis, the omission of that variable could bias the forfeiture coefficient if the variable is correlated with both forfeiture and the dependent variable.

One possible source of such variables would be agency characteristics that varied during the panel period. But the study's focus on large agencies meant that all of the agencies were well-established by year 2000, so there are no new-entrant behaviors that might bias the results. Forfeiture was a well-established institution by the beginning of the period of analysis and the larger agencies of our sample appear sophisticated in understanding forfeiture law. Of the 355 local agencies analyzed, only 22 had no forfeiture revenues in the first year of the panel; of the 103 counties analyzed, only 8 had no forfeiture revenues in the first year of the panel. This suggests that agencies were not moving along a learning curve with respect to forfeiture during this period. We have seen no evidence of internal agency changes that would create systematic correlations that could lead to biased estimators.

A second possible source of relevant omitted variables would be environmental factors not captured in our covariates. We followed the existing literature in the choice of the covariates and have no reason to suspect omitted variable bias. Further, we note that any such bias could well strengthen our results. Suppose, despite the covariates, increases in unobserved drug use early in the period were not identified in the regression. These could lead to increases in forfeiture receipts and, if they persisted, increases in subsequent drug arrests, causing a positive correlation between forfeiture receipts and subsequent drug arrests due to the omitted variable. This would imply that the coefficient on the linear forfeiture term in the drug arrest regression was biased upwards, exaggerating its statistical significance. However, the principal finding of our analysis is that forfeiture does not have economically significant effects. While the loss of statistical significance might be a minor tragedy in many contexts, here the decrease in the coefficient simply strengthens the conclusion that forfeiture's impact is minor. An omitted variable that instead increased the coefficients on the linear forfeiture terms would have to be strong indeed to raise those coefficients to economically meaningful levels, especially given the negative quadratic terms. (26) We stress that we have no evidence of relevant omitted variables; our purpose here is to delimit concerns about the impact of any such.

VII. CONCLUSION

Our findings show first that asset forfeiture does imply statistically significant effects on clearance rates, a measure of policing intensity, and that these differ from similar effects created by operating budget changes. Effects on arrest rates have signs consistent with a substitution effect toward drug policing. Overall, the picture is one of statistically significant income effects for serious crime clearances associated with changes in forfeiture, coupled with a substitution toward drug arrests. However, these effects, despite statistical significance, are practically very small. Further, they decrease with scale: greater forfeiture proceeds display less bang for the buck.

The lack of economic significance does not suggest that our analysis has little to say about the value of asset forfeiture. The results undercut the argument that police retention of forfeiture funds is an essential element in the fight against crime. In terms of the arguments in favor of forfeiture outlined in Section II, two of them--positive income effects for all policing and substitution effects toward drug policing--demand that the seized assets go to police agencies. Our results show that these arguments are not without a statistical basis, but that the practical effects are very small.

The analysis does not consider the deterrence and disruption arguments for asset forfeiture. Deterrence could lead to less drug crime if forfeiture changes the calculus of the rational criminal. This in turn could lead to fewer arrests than otherwise and thus bias downwards our estimate of the impact of forfeiture on arrests. Disruption assumes that loss of assets will make future drug-related activity more difficult, leading again to less drug crime. But drug traffickers in particular face far harsher penalties than forfeiture and it is questionable whether forfeiture would add to deterrence. Further, the criminology literature strongly leans to the view that deterrence has little impact on either drug trafficking or usage (Franco 2009). However, an area of further research would be to compare measures of trafficking and drug usage, drug arrests, and forfeiture, thus allowing more direct testing of the deterrence and disruption arguments.

Critics decry asset forfeiture and particularly police retention of funds for the civil liberty implications. The small changes in police behavior associated with forfeiture do not imply that forfeiture is little used; an agency may well change its policing behavior only slightly in response to forfeiture, while changing its administrative behavior to claim forfeiture whenever possible. Thus the civil liberty concerns remain: given the considerable value of assets seized, police in larger agencies appear to be seizing assets as the opportunity arises without changing their behavior much to increase those opportunities.

The civil liberties concerns also are somewhat independent of scale: a poor person who loses $400 in a cash forfeiture seizure may feel the pain as much as a wealthy person losing $400,000. Williams et al. (2010) used Freedom of Information requests to gather data from several states and to some extent recovered information on the number as well as total amounts of seizure. The median amounts seized are strikingly low; in Maine, for example, median forfeiture ranged from $ 1,820 to $2,630 per year over the course of 5 years, and in Virginia from $615 to $1,289 over the course of 12 years. The sheer number of civil forfeiture actions is a theme in much of the reporting on the topic. While this study shows that asset forfeiture has little effect on local police behavior in the senses promoted by its supporters, its civil liberty implications are little diminished by the low level of that effect.

An often-cited notion is that the widespread introduction of civil asset forfeiture for drug crimes in 1984 was a major contributor to the subsequent increase in drug arrests, absolutely and as a proportion of all arrests (Benson 2010). Our findings do not support this for the period 2000-2008, given the very modest link between drug forfeiture and drug arrests. However, we stress that the context has changed from the 1980s and 1990s to the time period covered by our study. It may well be that the rates of drug arrests had achieved such high levels by 2000 that additional incentives had little effect.

As described in Section II, asset forfeiture is a highly varied set of laws and our focus has been local agency behavior with respect to drug-related forfeiture. A possible concern is that this form of forfeiture has become identified with forfeiture generally in the eyes of the public. This could threaten the use of forfeiture in areas that many would find more palatable, such as financial fraud.

Finally, we note that forfeiture simply is not large compared with overall budgets. If it is achieving positive results, one wonders why conventional budget authority, suitably deployed, cannot do the same. Civil forfeiture raises very serious civil rights concerns and would seem a crude tool given the vastly greater funding available through conventional channels.

doi: 10.1111/ecin.12232

ABBREVIATIONS

AFF: Asset Forfeiture Fund

ASR: (Arrests by) Age, Sex, and Race

BJS: Bureau of Justice Statistics

CCCA: Comprehensive Crime Control Act

DOJ: Department of Justice

FBI: Federal Bureau of Investigation

FIPS: Federal Information Processing Standard

GDP: Gross Domestic Product

ICPSR: Interuniversity Consortium for Political and Social Research

LEMAS: Law Enforcement Management and

Administrative Statistics

UCR: Uniform Crime Reports

REFERENCES

Baicker, K., and M. Jacobson. "Finders Keepers: Forfeiture Laws, Policing Incentives, and Local Budgets." Journal of Public Economics, 91, 2007, 2113-36.

Benson, B. "The Allocation of Police," in Handbook on the Economics of Crime, edited by B. Benson and P. Zimmerman. Northampton, MA: Edward Elgar, 2010, 184-217.

Benson, B., D. Rasmussen, and D. Sollars. "Police Bureaucracies, Their Incentives, and the War on Drugs." Public Choice, 83(1-2), 1995, 21-45.

Bishopp, S., and J. Worrall. "Do State Asset Forfeiture Laws Explain the Upward Trend in Drug Arrests?" Journal of Crime and Justice, 32(2), 2009, 117-38.

Blumenson, E., and E. Nilsen. "Policing for Profit: The Drug War's Hidden Economic Agenda." University of Chicago Law Review, 65, 1997, 35-114.

Cassella, S. Asset Forfeiture Law in the United States. 2nd ed. Huntington, NY: Juris Publishing, Inc., 2013.

Cornwell, C., and W. N. Trumbull. "Estimating the Economic Model of Crime with Panel Data." Review of Economics and Statistics, 76(2), 1994, 360-6.

Edgeworth, D. Asset Forfeiture: Practice and Procedure in State and Federal Courts. 2nd ed. Chicago, IL: Criminal Justice Section of the American Bar Association, 2008. Evans, W. N., and E. G. Owens. "COPS and Crime." Journal of Public Economics, 91(1), 2007, 181-201.

Franco, C. Federal Domestic Illegal Drug Enforcement Efforts: Are They Working? Darby, PA: DIANE Publishing, 2009.

Gallant, M. "Proceeds of Crime, Forfeiture and Provincial Law." Available at SSRN 2055900 (May 10, 2012), 2012.

Hampl, E. "Message from the Director." Treasury Forfeiture Fund Accountability Report: Fiscal Year 2012, prepared by GKA, P.C. 2013. Accessed November 2013. http:// www.treasury.gov/resource-center/terrorist-illicit-finance/Asset- Forfeiture/Documents/FY%202012 %20Annual%20Report.pdf.

Holcomb, J., T. Kovandzic, and M. Wilson. "Civil Asset Forfeiture, Equitable Sharing, and Policing for Profit in the United States." Journal of Criminal Justice, 39, 2011, 273-85.

Howe, A. "Argument Preview: Court to Consider Scope of Challenges to Asset Freezes." 2013. Accessed December 11, 2013. SCOTUSblog: http://www.scotusblog .com/2013/10/argument- preview-court-to-consider-scope-of-challenges- to-asset-freezes/.

Hyde, H. Forfeiting Our Property Rights: Is Your Property Safe from Seizure? Cato Institute: Washington, DC, 1995.

ICPSR Law Enforcement Management and Administrative Statistics. Numbers 3565 (Lemas 2000), 4411 (Lemas 2003), 31161 (Lemas 2007). Accessed September 2011. http://www.icpsr.umich.edu/icpsrweb/ICPSR/ series/00092/studies?archive=ICPSR&sortBy=7.

ICPSR Uniform Crime Reports Arrests by Age, Sex, and Race, Summarized Yearly. Numbers 3729 (year 2001), 4461 (year 2004), 27642 (year 2008). Accessed August 2013. http://www.icpsr.umich.edu/icpsrweb/content/ NACJD/guides/ucr.html.

ICPSR Uniform Crime Reports Offenses Known and Clearances by Arrest. Numbers 3723 (year 2001), 4459 (year 2004), 27648 (year 2008). Accessed August 2013. http://www.icpsr.umich.edu/icpsrweb/content/NACJD/ guides/ucr.html.

Levy, L. A License to Steal: The Forfeiture of Property. Chapel Hill, NC: University of North Carolina Press, 1996.

Lyman, M. D. "Arrest Practices," in Police and Law Enforcement, edited by W. J. Chambliss. Los Angeles: Sage Publications, 2011.

Martinez, M. "City of Berkeley Fights U.S. Civil Action against Marijuana Dispensary." CNN, 2013. Accessed December 9, 2013. http://www.cnn.com/2013/07/03/ us/california-medical-marijuana-lawsuit.

Mast, B., B. Benson, and D. Rasmussen. "Entrepreneurial Police and Drug Enforcement Policy." Public Choice, 104, 2000, 285-308.

Noble, A. "D.C. Attorney General: Civil Asset Forfeiture Bill Could Endanger Profits." Washington Times, 2013. Accessed December 2013. http://www .washingtontimes.eom/news/2013/jul/11/dc-attorney-general-civil- asset-forfeiture-bill-co/?page=all.

Stillman, S. "Taken." The New Yorker, August 12, 2013. Accessed December 2013. http://www.newyorker.com/ reporting/2013/08/12/130812fa_fact_stillman.

The Economist. "Fighting Crime through Superior Steak." Nov. 2, 2013, 35.

United States Department of Justice. "Guide to Equitable Sharing for State and Local Law Enforcement Agencies." 2009. Accessed November 2013. http://www .justice.gov/usao/ri/projects/esguidelines.pdf.

--Crime in the United States. Washington, DC: Office of the Attorney General, 2011.

--Kaley v. United States of America: Brief for the United States. Washington, DC: Office of the Solicitor General, 2013a.

--. "Asset Forfeiture Fund Report to Congress. 2000-2013." Accessed November 2013b. http://www .justice.gov/jmd/afp/02fundreport

--. "Attorney General Prohibits Federal Agency Adoptions of Assets Seized by State and Local Law Enforcement Agencies Except Where Needed to Protect Public Safety." 2015. Accessed February 2015. http:// www.justice.gov/opa/pr/attorney-general-prohibits-federal-agency- adoptions-assets-seized-state-and-local-law

United States Senate. Oversight of Federal Asset Forfeiture: Its Role in Fighting Crime, Senate Hearing 106-673. Washington DC: U.S. Government Printing Office, 1999.

Warchol, G., and B. Johnson. "Guilty Property: A Quantitative Analysis of Civil Asset Forfeiture." American Journal of Criminal Justice, 21, 1996, 61-81.

Williams, M., J. Holcomb, T. Kovandzic, and S. Bullock. Policing for Profit: The Abuse of Civil Asset Forfeiture. Institute for Justice: Washington, DC, 2010.

Worrall, J. "Addicted to the Drug War: The Role of Civil Asset Forfeiture as Budgetary Necessity in Contemporary Law Enforcement." Journal of Criminal Justice, 29, 2001, 171-87.

Worrall, J., and T. Kovandzic. "Is Policing For Profit? Answers from Asset Forfeiture." Criminology and Public Policy, 7, 2008, 219-44.

Zhao, J. S., Y. Zhang, and Q. Thurman. "Can Additional Resources Lead to Higher Levels of Productivity (Arrests) in Police Agencies?" Criminal Justice Review, 36(2), 2011, 165-82.

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version of this article:

APPENDIX S1. Data Notes

(1.) See Table 1 and note 6, below.

(2.) For a fuller presentation, see Edgeworth (2008), Chapters 1 and 2. Forfeiture has numerous variations, but these do not affect our empirical analysis. For example, a few states rely on civil in personam actions to accomplish goals similar to those of civil in rem proceedings, albeit with different legal procedures. Also, the distinction between administrative and judicial civil forfeiture in practice is often narrow and indeed generates controversy. Typically law enforcement takes control of an asset and gives notice of the intent to declare it forfeit. If no challenge is filed, the forfeiture is administrative. If a challenge is filed, the forfeiture then is subject to a civil judicial hearing and is judicial. Mandated minimum response times are often short and the distinction between administrative and civil forfeiture may depend simply on whether the property owner had the knowledge and ability to respond to the notification within the legal time period.

(3.) Edgeworth (2008) notes that 47 states have civil in rem forfeiture statutes, most of them drug forfeiture statutes based on the federal example. State laws vary widely in the due process associated with civil asset forfeiture; see Edgeworth (2008) and Williams et al. (2010) for extensive reviews of state forfeiture laws.

(4.) State and local governments sometimes restrict the proportion of seized assets that police are allowed to keep. However, during our study period a local police agency that faced the loss of seized assets to general operating funds could apply for federal adoption rather than using state forfeiture laws. In such cases, the assets returned by the federal government to the agency--typically 80% of the asset value--had to be used for law enforcement (United States Department of Justice 2009). On January 16,2015, the Department of Justice announced the end of the adoption program for both the Justice and Treasury forfeiture funds except in cases involving public safety (United States Department of Justice 2015).

(5.) Asset forfeiture's uses and abuses have attracted considerable press coverage and popular accounts over the years. In the 1990s, books by Congressman Henry Hyde (1995) and by Leonard Levy (1996) emphasized the cost to civil liberties of asset forfeiture, providing many examples. Recent press coverage includes the controversy surrounding the D.C. police seizure of vehicles (Noble 2013), threatened U.S. seizure of commercial buildings in California that rented space to marijuana dispensaries (Martinez 2013), forfeiture being used as a funding tool in Texas (Stillman 2013), and "Fighting crime through superior steak" in The Economist (Nov. 2, 2013, 35). Legally, the 2013-2014 Supreme Court docket included Kaley v. The United States (Howe 2013), which concerned the loss of funds for legal defense due to their seizure (Howe 2013). Further examples are provided in Williams et al. (2010).

(6.) The DOJ Asset Forfeiture Fund (AFF) does not in general include another set of federal asset forfeitures, those going through a Department of Treasury fund, many of which arise from border control. The Treasury Forfeiture Fund had inflows of $868.1 million in fiscal year 2011 and $516.6 million in fiscal year 2012, with a small degree of overlap with the AFF funds (Hampl 2013). More importantly for our purposes, the federal figures do not include state and local seizures that are not processed through the AFF, which are heavily drug-related. No accurate figures exist for these, but we can make a rough inference for certain years from the ICPSR Law Enforcement Management and Administrative Statistics (LEMAS) reports. We estimate drug-related forfeiture income in year 2000 of $665 million, in year 2003 of $538 million, and in year 2007 of $934 million at the state and local levels, a minority of which overlaps with the federal figures of Table 1. These figures substantially understate the amounts because the LEMAS surveys did not sample from all types of agencies, particularly in 2003 and 2007. The figures also do not include forfeiture proceeds flowing to other elements in the criminal justice system, including prosecutors' offices.

(7.) The compensation rationale applies primarily to financial fraud cases rather than drug-related forfeiture and is much more prominent in federal than state and local practice. See Cassella (2013).

(8.) "The most troubling aspect of modern civil forfeiture laws is the profit incentive at their core" (Williams et al. 2010).

(9.) "Furthermore, police departments across this nation already have severely restricted budgets and by lessening income potential from asset forfeiture through this bill, the federal government would be drastically handicapping law enforcement capabilities in seizing illegal property. The ability of law enforcement to seize property is an important tool in this nation's 'war on drugs'." Letter from the National Association of Police Organizations, July 15, 1999, in United States Senate (1999), 9.

(10.) Clearances refer to the ratio of crimes cleared over crime reported. Section V and the supplementary online text (Appendix S1) provide further detail. The large majority of clearances are due to arrests.

(11.) Further, local police agencies are numerous and allow statistical reasoning concerning forfeiture's effects. Relatively few federal agencies are involved and intensive case-by-case analysis may be the best approach for these.

(12.) Of course, even without a binding budget constraint, police may wish to pursue forfeiture-related arrests due to greater freedom in applying forfeit assets than normal budgets. Indeed, a widespread criticism of forfeiture is that it leads to police confiscating or purchasing luxury or prestige items that have no impact on policing. See, for example, the incidents cited in Williams et al. (2010).

(13.) Anecdotal evidence makes clear that there are individual instances in which police behavior has been affected by asset forfeiture opportunities; see Stillman (2013). The goal here is to determine whether this reaches a statistically and economically meaningful level over a broad selection of police agencies.

(14.) "Schedule I" (that is, Part 1) crime is the subject of most criminological studies and many of the public crime reports. As indicated in Table 2, Schedule I crimes comprise codes 01 through 09 less codes 01B and 08. Compared to the data with which we work, this includes arson but excludes manslaughter by negligence (code 01B) and simple assault (code 08). We exclude arson (code 09) simply for data reasons; fortunately, it is a small category, and many existing studies exclude it as well. While code 01B also is a small category, code 08 contains numerous incidents, leading to a significant difference between the incidents underlying our measure and Schedule I. Consequently we separately tested all crime with reported offenses (codes 01-08) and Index I crimes (offenses known less codes 01B and 08); we present results for both tests in Table 4.

(15.) An incident is cleared by arrest when three conditions are met: at least one individual has been arrested, charged with the offense, and turned over to the court for prosecution. Clearances can differ from simple arrests due to the second and third of these criteria. Moreover, several arrests for a single crime counts as a single clearance, and one arrest that clears several incidents counts as multiple clearances. See United States Department of Justice (2011) for further discussion.

(16.) The supplementary online text (Appendix SI) provides details concerning the data sources and processing.

(17.) For example, in 2007, self-reporting local agencies averaged 157 sworn officers compared to 131 for self-reporting counties, but total employees averaged 206 for local agencies and 263 for counties. The average number of personnel involved in jail operations (2.7 local, 72.5 county), court security (1.0 local, 36.6 county), and process service (0.6 local, 9.5 county) varied widely between the two sets of agencies. Source: authors' compilations from LEMAS 2007 survey.

(18.) All of the tests were conducted using a fixed effects model with time dummies. A null hypothesis of random effects is rejected in all cases; F values are provided in the tables. Robust errors were used throughout.

(19.) As described above, the LEMAS questionnaire requests data based on budget periods incorporating June 30 for 2000 and 2003 and September 30 for 2007. This means that budget periods may partially overlap with the offenses data from 2001, 2004, and 2008. We do not view this as critical, since budgets of course affect present as well as future policing. The LEMAS questionnaires ask for forfeiture amounts for the preceding years, which mean that there should be no overlap between the forfeiture and crime periods. This avoids the potential endogeneity problem of forfeiture funds affecting policing in the same period that policing is itself the providing those forfeiture proceeds.

(20.) The paucity of explanatory variables may raise concerns, in particular the lack of covariates with underlying crime. The clearance rate is a ratio, resolved incidents divided by reported incidents. The traditional demographic and sociological correlates of Index I crime, such as poverty or unemployment, have no readily apparent relationship to the proportion of clearances to crimes reported.

(21.) Because of the negative coefficients on the quadratic terms for forfeiture and budget, the marginal impact of changes in those variables decreases with their levels. The coefficients on the forfeiture variables imply that clearance rates will change with respect to forfeiture at the rate of 7.74-1.13 * FORFOFCR, where FORFOFCR is measured in thousands of dollars per sworn officer. A small change in forfeiture [DELTA]FORFOFCR from a base of zero would increase clearance by 7.74 x [DELTA]FORFOFCR, a small change from a base of $1,000 per sworn officer would increase clearance by 6.61 x [DELTA]FORFOFCR, and so forth. The marginal rate of change at the forfeiture mean of $723.50 is 6.92 x [DELTA]FORFOFCR and the elasticity (proportional change in the clearance rate for a proportional change in forfeiture per officer) at that level is .0189. The net effect of the two terms turns negative at forfeiture/officer of $6,850; only nine of the 1,165 individual observations have forfeiture/officer amounts above this level.

(22.) Interestingly, the net effect of the two terms is slightly negative at the mean: 0.374892 - .0037 * 102.46 = -0.004, or $4. As the coefficients are not statistically significant, not too much should be made of this, but we note that there are several possible reasons that clearance rates could fall at higher budgetary levels. For example, budget increases at those levels may serve to lower crime rates for crimes that are relatively easy to solve or budget increases may increase bureaucracy at the cost of efficiency.

(23.) The interpretation of the coefficient for sworn officers (SWRNPC) is that an increase of one officer per thousand individuals in the agency's service area will increase the clearance rate by 22.7 per thousand incidents. The mean number of officers per thousand residents is 2.09 and the mean clearance rate is 265, so at the mean the implied elasticity (proportional change in the clearance rate for a proportional change in sworn officers per capita) is (22.7) * (2.09/265) = 0.179.

(24.) The coefficients on the forfeiture variables for drug arrests imply that arrest rates will change with respect to forfeiture at the rate of 0.187 - .023 * FORFOFCR, where FORFOFCR is measured in thousands of dollars per sworn officer. Substituting in average forfeiture per officer of $723.50, the net marginal effect at the average is 0.17 additional drug arrests per thousand residents, against a mean of 6.63 drug arrests per thousand. This yields the elasticity at the mean (0.17) * (0.7235/6.63) = 0.019.

(25.) The same tests applied to clearances measured in the year of receipt show no significant correlation between forfeiture receipts and clearances, supporting the observation that it takes some time to realize the benefits of forfeiture proceeds.

(26.) Also, the upward trend in drug arrests during the period suggests that decreases in unobserved drug use, not captured by the covariates, are relatively unlikely.

BRIAN D. KELLY and MAUREEN KOLE *

* We are grateful to the editor and two anonymous referees for very helpful comments, to participants in the WEAI's Crime and Punishment session at its 88th Annual Conference in June 2013, to participants in Albers Scholarship Seminar Series in November 2013, and to Yantao Wang. Marc Cohen, Claus Portner, and Vladimir Bejan of Seattle University. We also acknowledge the excellent research assistance of Scott Harp and Colin Kelly.

Kelly: Department of Economics, Seattle University, Seattle, WA 98122-1090. Phone 206-296-5711, E-mail [email protected]

Kole: Unaffiliated. E-mail [email protected]

TABLE 1
Net Deposits to the U.S. Department of Justice
Asset Forfeiture Fund

Fiscal Year       Amount (Thousands $)

2000                     507,033
2001                     439,930
2002                     453,133
2003                     466,968
2004                     537,113
2005                     578,804
2006                   1,143,308
2007                   1,583,389
2008                   1,327,605
2009                   1,404,823
2010                   1,600,371
2011                   1,684,810
2012                   4,221,910

Notes: Source: Department of Justice, Asset Forfeiture
Fund Reports to Congress, FY 2000 through FY 2012 (United
States Department of Justice 2013b). Amounts do not include
state and local forfeitures unless these go through the Asset
Forfeiture Fund.

TABLE 2
Codes Used in the UCR Crime Reports

01A Murder and non-negligent manslaughter *
02 Forcible rape *
03 Robbery *
04 Aggravated assault *
05 Burglary--breaking or entering *
06 Larceny--theft (not motor vehicles) *
07 Motor vehicle theft *
09 Arson *
01B Manslaughter by negligence
08 Other assaults
10 Forgery and counterfeiting
11 Fraud
12 Embezzlement
13 Stolen property--buy, receive, poses
14 Vandalism
15 Weapons--carry, possess, etc.
16 Prostitution and commercialized vice
17 Sex offenses (not rape or prostitution)
18 Total drug abuse violations **
180 Sale/manufacture (subtotal) **
185 Possession (subtotal) **
18A Sale/mfg--opium, coke, and their derivatives **
18B Sale/mfg--marijuana
18C Sale/mfg--truly addicting synthetic narcotics **
18D Sale/mfg--other dangerous non-narc drugs **
18E Possession--opium, coke, and their derivatives **
18F Possession--marijuana **
18G Possession--truly addicting synthetic narcotics **
18H Possession---other dangerous non-narc drugs **
19 Gambling (total)
19A Bookmaking (horse and sports)
19B Number and lottery
19C All other gambling
20 Offenses against family and children
21 Driving under the influence
22 Liquor laws
23 Drunkenness
24 Disorderly conduct
25 Vagrancy
26 All other non-traffic offenses
27 Suspicion
28 Curfew and loitering violations
29 Runaways
998 (M) Not applicable

Notes: The UCR Offenses Known ("Return A") report
number of offenses for each of the eight Part 1 crimes as well
as codes 01B and 08. The UCR Arrests by Age, Sex, and Race
report arrest data for all crime codes. Codes other than Part 1
are referred to as Part II.

LEGEND

* PART I CRIMES:

Violent: 01A, 02, 03, 04
Property: 05, 06, 07, 09

** DRUG CRIMES

TABLE 3
Descriptive Statistics for the Local Agency Panel Data

                                         2000             2003

Number of agencies                       355              355
Population served                     71.571.944       73.193,948
Sworn officers                         159.531          161,893
Officers/agency                          449              456
Officers/1,000 residents                 2.10             2.08
Budget                              13,988.946.104   15,402,779.417
Budget/agency                         39,405,482       43,388,111
Budget/sworn officer                    91.661          101.699
Forfeiture                           118,222,848      117,231,545
Forfeiture/agency                      333,022          330,230
Forfeiture/sworn officer                 761              634
Average unemployment rate (%)            4.7              5.6
Minority % of population                 15.2             15.4
Male, age 15-24, % of population         7.3              7.5
Clearance rate                          258.36           258.74
Arrests. Index I/1,000 residents        15.72            15.45
Drug arrests/1,000 residents             6.42             6.84
Arrests, Index II/1,000 residents       31.85            32.91

                                         2007           Average

Number of agencies                       355              355
Population served                     75,371,136       73,379,009
Sworn officers                         164,132          161,852
Officers/agency                          462              456
Officers/1,000 residents                 2.08             2.09
Budget                              17,746,973,077   15,712,899,532
Budget/agency                         49,991,473       44,261,689
Budget/sworn officer                   114,018          102,459
Forfeiture                           107,338,688      114,264,360
Forfeiture/agency                      302,363          321,871
Forfeiture/sworn officer                 775              724
Average unemployment rate (%)            5.9              5.4
Minority % of population                 15.7             15.4
Male, age 15-24, % of population         7.5              7.4
Clearance rate                          277.98           265.03
Arrests. Index I/1,000 residents        15.82            15.67
Drug arrests/1,000 residents             6.62             6.63
Arrests, Index II/1,000 residents       31.01            31.92

Notes: Values are in constant 2000 dollars. The clearance rate (1,000
* offenses cleared-offenses known) is for crime codes 01 through 08.
"Arrests, Index I" includes Index I crimes plus manslaughter by
negligence (code 01B), simple assault (code 08), and arson (code 09).
"Arrests, Index II" includes all crimes other than codes 01 through
09 or drug crimes. Data for the unemployment rate, minority, and age
15-24 percentages, and the clearance and arrest data are for the year
following the three LEMAS years: 2001, 2004, and 2008.

TABLE 4
Panel Regression of Clearance Rates for Reported Offenses

                                     Clearance Rate,   Clearance Rate,
                                      All Offenses      Index I Crime
                                     Known Estimate     Only Estimate
Variable                              and Standard      and Standard
Name              Definition              Error             Error

Forfofcr      Forfeiture proceeds      7.74308 ***       6.87467 ***
                  per officer           (2.97860)         (2.42060)
               ($ 1,000/officer)
Sqforfofcr    Square of Forfofcr       -.56502 ***       -.49068* **
                                        (.20770)          (.16470)
Budgofcr      Budget per officer         .37489            .37740
               ($ 1,000/officer)        (.31050)          (.28950)
Sqbudgofcr    Square of Budgofcr       -.00185 **         -.00177**
                                        (.00087)          (.00080)
Ofcrpc        Sworn officers per        22.74470*        27.07877 **
                   thousand            (11.93490)        (11.24770)
              (1,000 x officers/
                 pop. served)
LnPOP           Natural log of        109.78580 ***     82.21983 ***
               population served       (32.44530)        (24.82980)
year2003      Dummy for year 2003        -.54060           2.95251
                                        (3.38010)         (2.78060)
year2007      Dummy for year 2007     17.54237 ***      19.45937 ***
                                        (4.51640)         (3.85330)
[R.sup.2]                                 .7564             .7228
N            3 years, 355 agencies        1065              1065
F value        Null of no fixed           6.42              5.44
                    effects

Notes: The table shows the results of a fixed effects panel
regression of clearance rates, measured as 1,000 x offenses cleared/
offenses reported, for the aggregate of all offenses included in the
UCR Offenses Known data, and the same regression for the aggregate of
all offenses known for Index I crimes. Clearance and population
served data are derived from the UCR Offenses Known reports, budget,
and personnel information from the LEMAS surveys. Data are for local
police agencies from the LEMAS survey and so do not include state or
county agencies.

Statistical significance at the 1%, 5%, and 10% levels in two-tailed
tests is indicated by ***, **, and *, respectively.

TABLE 5
Panel Regression of Arrest Rates per Thousand People Served

Variable                           Arrests,     Arrests,     Arrests,
Name            Definition       Codes 01-09     Code 18    All Other

Forfofcr    Forfeiture proceeds     .03022      .18747 *      .49940
                per officer        (.18060)     (.1 1380)    (.57290)
             ($ 1,000/officer)
SqForfofcr  Square of Forfofcr     -.00304       -.01172     -.02894
                                   (.01400)     (.01310)     (.04080)
Budgofcr    Budget per officer     .01917 *      .01512      -.03233
             ($l,000/officer)      (.01040)     (.00975)     (.04990)
SqBudgofcr  Square of Budgofcr    -.00006 **     -.00004      .00020
                                   (.00003)     (.00003)     (.00015)
Ofcrpc      Sworn officers per      .46193       1.02777     15.17595
                 thousand         (1.39030)     (.68140)    (19.30310)
            (1,000 x officers/
               pop. served)
LnPOP         Natural log of     -5.62199 ***  -2.48483 **  -12.16840
             population served    (1.68360)     (1.04350)   (11.19800)
Unempr       Unemployment rate    .47363 **    -.25021 **    -.89142
                                   (.20590)     (.10400)     (.78310)
Minority      Percentage of       .90906 ***     .15696       .79613
             population black      (.24614)     (.12484)    (1.57500)
Age 1524      Percentage of        -.25573       .01354      -2.97234
              population ages      (.46522)     (.25739)    (1.97500)
                   15-24
Year2003      Year 2003 dummy    -.73403 ***   .65402 ***    2.53072
                                   (.28320)     (.16990)    (1.88720)
Year2007      Year 2007 dummy     -.68418 *     .46136 **     .57272
                                   (.37160)     (.21630)    (1.47180)
[R.sup.2]                           .8699         .8800       .7437
N            Three years, 355        1065         1065         1065
                 agencies
F value      Null of no fixed       11.08         10.31        4.50
                  effects

Notes: The table shows the results of a fixed effects panel
regressions of arrest rates, measured as 1,000 x number of arrests-
population served. Arrest data are derived from the UCR ASR reports,
budget, and personnel information from the LEMAS surveys. Data are
for self-reporting local police agencies from the LEMAS survey. The
unemployment rate is from the Bureau of Labor Statistics, and
Minority and Age 1524 percentages from the Census Bureau. These three
variables are at the county level, the remaining variables are agency
level.

Statistical significance at the 1%, 5%, and 10% levels in two-tailed
tests is indicated by ***, **, and *, respectively.