Less Cash, Less Crime: Evidence from the Electronic Benefit Transfer Program Richard Wright Georgia State University Erdal Tekin American University Volkan Topalli Georgia State University Chandler McClellan American University Timothy Dickinson University of Texas at El Paso Richard Rosenfeld University of Missouri–St. Louis Abstract It has been long recognized that cash plays a critical role in fueling street crime because of its liquidity and transactional anonymity. In this paper, we investigate whether the reduction in the circulation of cash on the streets associated with electronic benefit transfer (EBT) program implementation had an effect on crime. To address this question, we exploit the variation in the timing of EBT implementation across Missouri counties and counties in the states bordering Missouri. According to our results, the EBT program had a negative and significant effect on the overall crime rate and specifically for burglary, assault, and larceny. The point estimates indicate that the overall crime rate decreased by 9.2 percent in response to the EBT program. Interestingly, the significant drop in crime in the United States over several decades coincided with a period of steady decline in the proportion of financial transactions involving cash. 1. Introduction Criminologists have long known that most predatory street criminals are motivated by the acquisition of cash—as opposed to alternative forms of monetary transfer such as debit or credit cards—because its liquidity and transactional anonymity are critical to the functioning of the underground economy (Varjavand 2011). Economists have also emphasized the role that cash plays in fueling street We thank seminar participants at the University of Arizona, George Mason University, and the Institute for the Study of Labor and conference participants at the 2014 annual meetings of the European Society for Population Economics, the European Association of Law and Economics, and the American Society of Criminology for comments and suggestions. [ Journal of Law and Economics, vol. 60 (May 2017)] © 2017 by The University of Chicago. All rights reserved. 0022-2186/2017/6002-0012$10.00 361 This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). 362 The Journal of LAW & ECONOMICS crime, recognizing that neighborhood drug dealers, prostitutes, and pawn brokers are not inclined to accept noncash forms of payment for their services (see, for example, Foley 2011; Armey, Lipow, and Webb 2014). If that is so, then any reduction in the amount of cash in circulation should produce concomitant reductions in acquisitive street crimes. In poor neighborhoods where street offenses are concentrated, a major source of circulating cash used to stem from public assistance or welfare payments. Prior to the late 1990s, the welfare assistance program, now referred to as Temporary Assistance for Needy Families (TANF), issued payments in the form of paper checks, which required a significant proportion of unbanked or underbanked recipients to cash them at independent check-cashing establishments (see Leyshon and Thrift 1994, 1995). However, that changed over the last 2 decades, with paper checks being replaced by the electronic benefit transfer (EBT) program, a digital, debit-cardbased system. Mandated by the federal government but enacted at the state level, the changeover from paper to EBT was implemented variably in states, most often on a county-by-county basis. In this paper, we hypothesize that the introduction of such a system will reduce the amount of cash circulating on the streets, especially in low-income neighborhoods, which will result in a reduction in rates of acquisitive street crimes. To examine this hypothesis, we assemble monthly data on various types of crimes from all of the counties in Missouri and the bordering counties in its neighbor states between 1990 and 2011. We then exploit the variation in the timing of the implementation of the EBT program across counties to examine the impact on various crimes of reduced circulation of cash caused by the EBT program.1 To the extent that there is no other plausible channel though which EBT implementation can produce an independent effect on crime, any association between EBT implementation and crime can then be attributed to the removal of cash from the streets. Indeed, our results indicate that the EBT program’s implementation is associated with a significant decrease in the overall crime rate and the rates of burglary, assault, and larceny. Our analysis provides suggestive evidence that the EBT program also reduced robbery rates. Finally, we find suggestive evidence of concomitant reductions in arrests in response to the EBT program’s implementation, which is consistent with the hypothesis of a negative effect of EBT on crime. 1 Note that the electronic benefit transfer (EBT) program was implemented in phases in many other states as well. Although it would be interesting to conduct a nationwide analysis, data on implementation dates at the county level are not currently available for all states. In an attempt to identify such EBT implementation dates, we contacted an official who served as an evaluator of the system-wide EBT implementation in several states. He acknowledged that no one involved in the process recognized the potential for tracking county implementation dates for the purposes of research (John Kirlin, economist at the US Department of Agriculture’s Economic Research Service, e-mail correspondence with the authors, April 29, 2013). We thank him for providing that information. This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Electronic Benefit Transfer Program 363 2. Background Beginning in the early 1980s, the federal government gradually shifted toward the use of EBT as a means for disbursing government benefits to recipients. A number of EBT demonstration programs were implemented during the decade, which culminated in congressional passage of the Hunger Prevention Act of 1988 and the Mickey Leland Memorial Domestic Hunger Relief Act of 1990 (US Department of Agriculture 2014b). During the 1990s, political support for EBT increased as a result of an ideological shift reflected in the conference report on the Omnibus Budget Reconciliation Act of 1993 and the 1993 National Performance Review, which urged states to establish these systems (see Personal Responsibility and Work Opportunity Reconciliation Act, 64 Fed. Reg. 28,763 [May 27, 1999]; Office of the Vice President 1993). Welfare-reform legislation signed in 1996 required every state to develop systems to issue food-stamp program benefits electronically by 2002. Missouri lawmakers responded by enacting a statute (Mo. Rev. Stat., sec. 208.182) in 1994 that stipulated the establishment of EBT pilot programs in Missouri counties with a population of 600,000 or more (thus including the city of St. Louis2 and Jackson County, which contains Kansas City). These pilot programs were initiated in mid-1997, shortly after the federal government restructured welfare with the Personal Responsibility and Work Opportunity Reconciliation Act (Pub. L. No. 104-193, 110 Stat. 2105). During the testing of EBT, Missouri recipients of food stamps and temporary assistance residing in pilot locations received EBT cards in place of traditional paper checks. Funds were added to the cards according to recipients’ month of birth and the first letter of their surname, a practice continuing to the present day.3 The switch to EBT was received positively by both retailers and recipients.4 It increased the speed and efficiency of transactions, and the resemblance of the EBT cards to credit and debit cards reduced the stigma experienced when making a purchase with distinctive checks.5 (Impor2 St. Louis is an independent city but was included in the statute mandating the rollout of EBT statewide as follows: “The family support division shall establish pilot projects in St. Louis City and in any county with a population of six hundred thousand or more” (Mo. Rev. Stat., sec. 208.182) 3 Kay Martellaro, EBT/Food Distribution Unit Manager, Family Support Division, Missouri Department of Social Services, e-mail correspondence with the authors, May 15, 2013. 4 Missouri recipients of EBT are given detailed instructions about the use of their cards, including how and where they may be used, when monthly benefits are disbursed, and the fees and surcharges associated with withdrawing cash from automated teller machines and point-of-sale terminals (Missouri Department of Social Services, Family Support Division, Benefit Availability Dates [https:// mydss.mo.gov/food-assistance/ebt/ebt-availability-dates]; Missouri Department of Social Services, Family Support Division, EBT Cardholder Information [https://mydss.mo.gov/food-assistance /ebt]). Although recipients of temporary assistance in Missouri were strongly urged by the Department of Social Services to establish bank accounts during the initial years of the restructured program, the majority of individuals continue to receive temporary assistance via EBT cards (Missouri Department of Social Services 2013a). This is consistent with national trends indicating that a significant proportion of the poor are “unbanked” or “underbanked” (Federal Deposit Insurance Corporation 2012; see also Rhine and Greene 2013). 5 Missouri Department of Social Services, Family Support Division, EBT Cardholder Information [https://mydss.mo.gov/food-assistance/ebt]). This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). 364 The Journal of LAW & ECONOMICS tantly, the EBT system obviated the need for recipients to cash their checks and carry paper money around with them. Note that TANF benefits were the most likely source of circulating cash on the streets in low-income communities prior to EBT because they were the predominant form of benefits distributed via checks. But this does not mean that other types of benefits did not also contribute cash to the streets. Food-assistance benefits, now referred to as Supplemental Nutrition Assistance Program (SNAP) benefits, were previously distributed as food stamps, which were neither cash nor checks but a token-based closed transactional medium. The argument could be made that the bulk of the decrease in crime attributable to the EBT implementation was due to the removal of TANF-based welfare, as that mechanism created a platform for conversion to cash via checks. Meanwhile, SNAP benefits bypassed this process by moving from a token-based system to an electronic system, with no legal mechanism for conversion to cash. On the one hand, this could be construed to provide further support for our hypothesis, wherein the key mechanism driving the crime reduction is not the existence of electronic systems per se but instead the removal of cash from the market resulting from the operation of those systems. On the other hand, the role of SNAP in reducing crime may not be so simple. While it is true that SNAP benefits were never connected to a formal mechanism designed for the conversion of value to cash, it is also true that SNAP food stamps were often illegally converted to cash by recipients (and likely by those stealing them) who wished to access services and commodities not permitted by food stamps—everything from cigarettes to lottery tickets to drugs—thereby contributing to the total amount of cash circulating on the streets (see, for example, Pulliam 1997; Edin and Lein 1997; Macaluso 2000; Whitmore 2002). It is noteworthy that a key reason why the federal government wanted to facilitate the EBT process was to reduce this kind of fraud (see Exec. Order No. 13681, 79 Fed. Reg. 63,489 [October 23, 2014]; Macaluso 2003; US Department of Agriculture 2014a). It is not unreasonable to assume, therefore, that the integration of SNAP benefits into the EBT system also contributed to the decrease in crime uncovered in our analysis. We also make use of data from the counties that border Missouri in the eight surrounding states. These include 57 counties in Arkansas, Illinois, Iowa, Kansas, Kentucky, Nebraska, Oklahoma, and Tennessee. Adding these counties clearly enlarges our sample. More important, it increases variation in the EBT implementation dates across counties and over time that is key to identification. The EBT implementation dates for the border counties are April 1998 for Arkansas, November 1997 for Illinois, October 2003 for Iowa, March 1997 for Kansas, November 1999 for Kentucky, September 2002 for Nebraska, January 1998 for Oklahoma, and August 1999 for Tennessee. The EBT program was implemented in eight phases in Missouri localities between June 1997 and May 1998.6 Figure 1 illustrates the EBT program’s implementation for Missouri and the bordering counties. 6 The numbers of counties in the eight phases (including 114 Missouri counties and the city of St. Louis) are eight, 17, 19, 12, seven, two, six, and 44 in order of implementation. This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Electronic Benefit Transfer Program 365 Figure 1. Implementation of the electronic benefit transfer program in Missouri and surrounding counties. 3. Crime Data The crime data are from the Uniform Crime Reporting (UCR) program of the Federal Bureau of Investigation (FBI). The UCR program represents “a nationwide, cooperative statistical effort of nearly 18,000 city, university and college, county, state, tribal, and federal law enforcement agencies voluntarily reporting data on crimes brought to their attention.”7 Under the UCR system, law enforcement agencies submit crime data either through a state UCR program or directly to the FBI’s UCR program on a monthly basis. These consist mainly of crimes reported to the police by the general public but may also include offenses that police officers discover or learn about through other sources. 7 US Department of Justice, About the Uniform Crime Reporting Program (https://www.bjs.gov /ucrdata/abouttheucr.cfm). This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). 366 The Journal of LAW & ECONOMICS Table 1 Descriptive Statistics: Mean Crime Rates per 100,000 Persons Full Sample Total crime rate Robbery rate Assault rate Burglary rate Larceny rate Motor-vehicle theft rate SNAP caseload TANF caseload Unemployment rate County population Rate N With EBT Without EBT 395.68 (310.79) 8.24 (14.74) 104.44 (83.33) 61.23 (50.69) 192.46 (146.15) 18.16 (27.64) 22,739.34 (32,630.35) 6,235.26 (11,311.60) 5.92 (2.52) 32,985.51 (72,812.18) 33,252 382.27 (282.14) 6.59 (10.73) 105.04 (74.81) 56.46 (43.63) 187.47 (140.84) 15.74 (22.45) 22,805.14 (33,378.55) 4,073.18 (7,283.49) 6.04 (2.47) 34,191.84 (75,501.28) 424.68 (363.44) 11.82 (20.49) 103.13 (99.28) 71.53 (62.07) 203.25 (156.47) 23.38 (35.86) 22,596.50 (30,945.45) 10,931.58 (16,076.94) 5.65 (2.59) 30,647.98 (67,237.98) 33,247 33,252 33,251 33,251 33,239 23,616 23,628 33,252 33,252 Note. Standard deviations are in parentheses. Monthly data for Supplemental Nutrition Assistance Program (SNAP) and Temporary Assistance for Needy Families (TANF) are available only for Missouri counties. EBT = electronic benefit transfer. For this investigation, we assembled county-level monthly data for total crime, assault, burglary, robbery, larceny, and motor-vehicle theft. The mean crime rates for our sample variables weighted by county population in counties in Missouri and the eight surrounding states are presented in Table 1. The rate of monthly total crime is about 396 per 100,000 persons per county. Larceny constitutes a large share of overall crime, followed by assault and burglary. Robbery and motor- vehicle theft are the least prevalent crimes in our data, with averages of 8 and 18 per 100,000 persons per month, respectively, and are concentrated in urban areas. When we compare the average crime rates for EBT and non-EBT observations, we see that the rate is lower in observations with EBT programs for all types of crime, which suggests that there may be a reduction in crime associated with EBT implementation. The only exception to this pattern is assault, for which the difference between the two categories is statistically insignificant. We illustrate the monthly proportion of each crime reported for Missouri and its border counties in Table 2. The patterns for each type of crime are as expected, with peaks in summer months and lowest reported in February because of fewer days. This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Electronic Benefit Transfer Program 367 Table 2 Monthly Distribution of Crime Rates Missouri: January February March April May June July August September October November December Bordering counties: January February March April May June July August September October November December Total Crime Robbery Assault Burglary Larceny MotorVehicle Theft 7.58 6.82 7.96 8.01 8.55 8.67 9.26 9.34 8.87 8.80 8.05 8.09 8.20 6.82 7.59 7.46 7.94 8.03 8.73 9.02 9.01 9.22 8.74 9.25 7.29 6.92 8.21 8.53 9.19 8.90 9.26 9.07 9.02 8.68 7.54 7.38 7.94 6.72 7.70 7.59 8.20 8.27 9.10 9.28 9.06 8.98 8.56 8.61 7.46 6.75 7.97 8.01 8.50 8.79 9.33 9.51 8.75 8.79 8.02 8.12 8.36 7.08 7.70 7.35 7.71 8.22 9.25 9.37 8.79 8.79 8.59 8.79 6.78 6.15 9.36 7.33 8.04 10.41 8.77 8.66 10.23 7.95 7.07 9.25 8.48 6.85 8.89 6.90 7.52 8.98 7.47 8.36 9.31 7.82 8.65 10.77 6.75 6.32 9.79 7.69 8.37 10.52 8.50 8.46 10.27 7.83 6.71 8.79 7.51 6.34 8.51 7.33 8.02 9.32 9.04 8.95 9.56 8.46 7.82 9.14 6.49 5.96 9.35 7.21 7.92 10.76 8.84 8.65 10.38 7.93 7.01 9.51 7.61 6.94 9.73 6.54 7.42 9.70 8.88 9.07 9.99 7.65 7.20 9.26 Note. Values are proportions of crime reported. In Figures 2 and 3, we illustrate the patterns in total crime rate and rates of part I crimes, respectively, for Missouri and the United States.8 Crime rates fell sharply in both Missouri and the United States during the period of analysis. Note that this pattern is present for both the overall crime rate and each of the part I offenses. In general, the trends in crimes in Missouri appear to be quite similar to those of the US averages.9 8 The Uniform Crime Reporting program indexes two main categories of crime. Part I crimes, our focus in this study, include two categories: Violent crimes include forcible rape, aggravated assault, murder, and robbery. Property crimes include arson, burglary, larceny-theft, and motor-vehicle theft. Part II crimes are less serious and involve such offenses as loitering, embezzlement, forgery and counterfeiting, disorderly conduct, prostitution, vandalism, vagrancy, and weapons offenses. Note that the patterns are similar for the eight neighboring states. 9 The corresponding patterns for the total crime rate and rates of part I crimes for Missouri and surrounding states compared with the Unites States are shown in Figures A1 and A2. These patterns are very similar to those in Figures 2 and 3. This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Figure 2. Trends in total crime rates for Missouri and the United States Figure 3. Trends in part I crime rates for Missouri and the United States This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Electronic Benefit Transfer Program 369 4. Estimation Method Our goal is to estimate the change in the rate of crime caused by the implementation of the EBT program. One key empirical challenge to accomplishing this stems from the possibility that factors leading people to carry cash may also be correlated with crime. For example, if people decide not to carry a lot of cash with them as a reaction to increased crime, then any observed negative relationship between the crime rate and EBT would be overstated. Our approach to guarding against this problem is to exploit a policy change that has no direct association with crime but that leads to a reduction in the circulation of cash on the streets. Then any reduction in crime associated with implementation of the policy can be attributed to the reduction in the circulation of cash. The validity of this approach hinges on whether any of the factors driving the change in policy are correlated with street crime. There is no evidence to suggest that crime reduction was ever mentioned as a reason or justification for implementing the EBT program. Rather, EBT policy was instituted to reduce program fraud (unrelated to street crime), ensure ease of use of food benefits by the program’s participants, and reduce the stigma associated with using food stamps (see, for example, US General Accounting Office 1994). Therefore, any variation in the circulation of cash generated by the EBT implementation should be exogenous to crime. In addition, the EBT program was implemented in a staggered fashion across various jurisdictions. It is this variation across jurisdictions that we exploit in our empirical analysis, using a difference-in-differences method in which we estimate the difference in average crime rates in the jurisdictions with an EBT program before and after the implementation net of the difference in rates in jurisdictions without an EBT program. The key assumption in the method is that in the absence of the implementation of EBT policy, crime rates would have trended similarly between treatment and nontreatment counties. We believe this is a plausible assumption since EBT implementation is likely to be exogenous to crime for reasons explained above. While this assumption is not directly testable, valuable insights can be gained by comparing the crime trends for each set of treatment jurisdictions prior to any EBT program implementation. Note that we do not have a set of control counties that were never treated since all counties eventually implemented an EBT program. In different months between 1997 and 2003, however, each set of treatment counties served as a control for the other counties. To motivate our empirical strategy further, we provide additional visual evidence of the effect of EBT implementation on crime rates in Missouri. In Figure 4, we illustrate the trends in the average rates of our crime variables up to 24 months before and after EBT implementation, weighted by county population. The vertical line indicates when the EBT program became effective in each of the treatment counties. Since the program went into effect at different points in time, the graph is centered on the month and year of implementation (time 0) and tracks crime rates up to 24 months before and after that time. As shown in Figure 4, there is a reversal in the trend for the rate of total crime at the time of EBT im- This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). 370 The Journal of LAW & ECONOMICS Figure 4. Trends in Missouri’s crime rates before and after implementation plementation. With the exception of larceny, individual crimes exhibit a similar pattern.10 While the patterns presented in Figure 4 are indicative of a causal relationship between the EBT program’s implementation and crime rates, a stronger test of our hypothesis would take advantage of both within- and between-county differences in the crime rates between counties with and without EBT implementation in each month of each year. We also allow the crime rates to trend differently across counties by accounting for county-specific linear trends in our empirical analysis. One key characteristic of our crime data is that it is a nonnegative count with 10 Figure A3 shows data for Missouri and bordering counties. These patterns are, although somewhat weaker, not inconsistent with those in Figure 4. This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Electronic Benefit Transfer Program 371 a large number of zeros. Therefore, we employ a count model for estimating the effect of the EBT program’s implementation on crime.11 In particular, we conduct our empirical analyses using a negative binomial regression model.12 To estimate the negative binomial model in our context, it is assumed that the count for each of our crime outcomes in a particular county c in a particular month m and calendar year y, Crimecmy, follows a negative binomial distribution with the parameters νcmy and ϕcmy, where E(Crimecmy) = ϕcmy and var(Crimecmy) = ϕcmy(1 + ϕcmy /νcmy).13 Following Greene (2008), covariates incorporated into this model can be represented in a regression framework by the following expression: Crimecmy = kcy exp EBTcmy + β c +lmy +c t + ecmy , (1) where EBTcmy is our key treatment variable that equals one if county c has an EBT policy in effect in month m in calendar year y and zero otherwise. The term βc is a vector of county fixed effects that serve to account for any permanent differences across counties that may affect crime. The term λmy is a set of month-byyear fixed effects that serve to control for seasonality in crime and any changes in crimes that are common to all counties. The county-specific linear time trends ηct capture the influence of difficult-to-measure factors at the county level that trend linearly. In equation (1), κcy represents the exposure variable, county population. Errors εcmy are clustered at the county level. The unit of observation in equation (1) is at the county-year-month group level. The coefficient of interest is α, the impact of EBT policy on crime rates. As noted earlier, our identification strategy does not require the levels in crime rates between treatment and control counties to be equal prior to implementation of the EBT program. Rather, it assumes that, in the absence of the EBT program, the rates of crime could have trended similarly in the two types of counties. Although we cannot test this assumption directly, we further assess the credibility of our research design by examining whether there were any systematic changes in crime rates prior to the EBT program’s implementation. One way to do this is to conduct an event study, which would allow us to trace out the trends in our outcomes month by month for the periods leading up to and following the implementation of the EBT program. In practice, we implement this by 11 Estimating our models using linear probability largely produced results that are similar to those presented in this paper, although the point estimates were less precise. This is not surprising given the large number of zero values in our outcomes. Another alternative count-data method is Poisson regression. A potential limitation of the Poisson regression is that it does not allow for overdispersion in the outcome variable. Despite this limitation, the estimates presented here remained very similar when we estimated our models using Poisson regression instead of negative binomial regression. 12 The negative binomial model is widely used in studying questions concerning crime (see, for example, Grogger 1990; Osgood 2000; Jacob and Lefgren 2003; Lang 2016; Anderson, Crost, and Rees, forthcoming). 13 The negative binomial distribution has the following probability density function: Pr(Crimecmy ) = [G(Crimecmy + n cmy )/ G(Crimecmy + 1)G(n cmy )] n Crimecmy ´ [n cmy /(n cmy + fcmy )] cmy [fcmy /(n cmy + fcmy )] , where Γ(.) is the gamma function, νcmy > 0, and ϕcmy > 0. This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). 372 The Journal of LAW & ECONOMICS Figure 5. Event-study estimates for Missouri augmenting equation (1) to allow the coefficients on EBT implementation to vary with time for up to 18 months, where month 0 is the month that the EBT program status changes from zero to one. Figure 5 plots these coefficients and their 95 percent confidence intervals from the event-study analysis for Missouri counties. The patterns in Figure 5 support the validity of the identification strategy, showing no evidence of systematic changes in crime rates in the months prior to implementing the EBT program and indicating that our results do not simply reflect the continuation of long-run preexisting trends in crime rates. Nevertheless, we also allow county-specific unobservable factors to trend differently by including county-specific trends in our specifications. The patterns in Figure 5 also suggest that the changes in EBT program status were followed This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Electronic Benefit Transfer Program 373 Table 3 Effect of Implementation on the Logarithm of Crime Rates Total Crime SE N Robbery SE N Assault SE N Burglary SE N Larceny SE N Motor-vehicle theft SE N Missouri and Border Counties All Excluding Kansas City and St. Louis Missouri Counties −.092* .042 37,308 −.003 37,297 .077 −.107** .041 37,308 −.104** .037 37,307 −.105* .043 37,307 −.058 .080 37,296 −.132** .051 30,816 −.161 30,809 .108 −.103 .065 30,816 −.121* .052 30,815 −.173** .036 30,815 −.081 .103 37,296 −.098** .036 26,268 .028 26,259 .040 −.125** .041 26,268 −.079+ .042 26,267 −.096* .045 26,267 .028 .039 26,257 Note. Robust standard errors are clustered at the county level. + P < .10. * P < .05. ** P < .01. by a downward trend in crime rates. Note that the patterns in Figure A4, which displays the event-study estimates for Missouri and bordering counties, are very similar to those in Figure 5. 5. Empirical Results We present the results from the estimation of equation (1) in Table 3.14 Values are the size of the effect of the EBT program’s implementation on each of the six crime outcomes. The estimates for Missouri and the border counties indicate that the EBT implementation had a negative effect on all of the crime rates analyzed. Furthermore, the estimates are statistically significant for four of the six outcomes. According to point estimates, the EBT program is associated with a 9.2 percent reduction in total crime rate per 100,000 persons. The magnitudes of the effects for other individual crimes are also sizeable, suggesting declines of 10.7 percent for assault, 10.4 percent for burglary, and 10.5 percent for larceny. The estimates for robbery and motor-vehicle theft are both negative but not statistically significant. We also examine the extent to which our estimates are driven 14 We also estimated our models using different end points for our analysis sample. Our estimates in the most comprehensive specification are robust to periods of 1990–2006, 1991–2005, 1992–2004, and 1993–2003. These results are available from the authors on request. This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). 374 The Journal of LAW & ECONOMICS by urban population centers like St. Louis and Kansas City. The estimates remain qualitatively similar but generally become larger in terms of magnitude. The implementation of EBT is associated with a 13.2 percent decline in total crime according to these estimates. Interestingly, the EBT estimate on the robbery model is negative, sizeable, and almost statistically significant with a P-value of .136. The estimates for Missouri counties are again mostly consistent with those in the first two columns, which suggests that the EBT program’s implementation had a negative effect on the total crime rate and rates of assault, burglary, and larceny. To put the significant results into perspective, we next conduct back-of-the- envelope calculations for the number of crimes that are prevented as a result of the EBT program’s implementation in Missouri. Given that the total crime rate per 100,000 persons was 482.62 in the counties in the month before implementation of the EBT program, an effect of 9.8 percent implies that approximately 47 fewer crimes per 100,000 persons would be committed per county per month as a result of the program. Since the average population in these counties was 62,870, the reduction in the number of crimes would be about 30. The corresponding figures for burglary, assault, and larceny would be 3.7, 5.9, and 14.2 per month, respectively. If the EBT program’s implementation had a negative impact on the number of crimes committed, then a related question is whether there was an associated decrease in the number of arrests that coincided with the decrease in crime. To provide insights into this question, we assembled data on various types of arrest records for Missouri counties at the monthly level between 1990 and 2011, which are available from the UCR program. In particular, we compiled data on the breakdown of arrests for non-drug-related offenses and then arrests for possession of any illicit drug, possession of cocaine, possession of marijuana, possession of synthetic narcotics (for example, Demerol, methadone, and so on), arrests for driving under the influence (DUI) of liquor or narcotic drugs, and arrests for a violation of liquor laws. Then we separately estimated our models using these variables as outcomes for our three samples. The estimates on the EBT program indicator from our most comprehensive specification are shown in Table 4. The estimates for the sample of Missouri and border-state counties indicate that there is a statistically significant and negative relationship between the EBT program’s implementation and arrests for possession of cocaine and nondrug arrests. The estimates for other drug-arrest outcomes are not estimated with precision. Consistent with Table 3, the estimates are more precisely estimated when we exclude Kansas City and St. Louis. In particular, the impact of EBT is negative and statistically significant for arrests for possession of any drug and possession of marijuana and for DUI violations and nondrug arrests. Finally, the estimates derived from the Missouri counties reveal a negative impact for arrests for possession of cocaine and nondrug arrests. The negative sign of the estimates that are statistically significant in Table 4 may be interpreted as suggestive evidence of a negative causal relationship between the EBT program’s implementation and arrest rates. That being said, the pattern in This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Electronic Benefit Transfer Program 375 Table 4 Effect of Implementation on the Logarithm of Arrest Rates Possession of any drug SE N Possession of cocaine SE N Possession of marijuana SE N Possession of synthetic narcotics SE N DUI violation SE N Violation of liquor laws SE N Nondrug arrest SE N Missouri and Border Counties All Excluding Kansas City and St. Louis Missouri Counties −.058 .077 26,039 −.317* .141 7,927 −.064 .076 24,306 .012 .121 9,523 −.049 .070 31,058 −.117 .074 21,825 −.071+ .037 34,551 −.179+ .098 14,875 −.073 .102 3,657 −.219** .043 13,884 .101 .280 5,485 −.123+ .062 17,424 −.0092 .061 11,763 −.127** .034 20,009 −.014 .108 19,028 −.300+ .165 6,000 −.010 .091 17,928 −.033 .170 7,746 −.125 .093 21,856 −.078 .050 15,607 −.092+ .103 24,646 Note. Robust standard errors are clustered at the county level. DUI = driving under the influence. + P < .10. * P < .05. ** P < .01. statistical significance across the columns lacks consistency. Therefore, these estimates should be viewed with caution. One possible challenge to the results presented thus far is that a reduction in the circulation of cash on the streets might induce some criminals to travel to neighboring counties without an EBT program to conduct illegal acts. This would suggest that crime is displaced from a treatment county to a neighboring control county and results in no net change in overall crime. While such rational behavior on the part of criminals is theoretically plausible, available evidence from the criminological literature suggests that most offenders tend to operate within their own geographical activity or awareness spheres, which usually do not extend beyond several blocks (Brantingham and Brantingham 1981, 1984). Nevertheless, we conduct three robustness analyses to address this concern.15 In the first analysis, we estimated our models excluding counties that are located on either side of the borderline for EBT implementation. Therefore, any 15 Results from these analyses were presented in an earlier version of this paper but are excluded here to economize on space. They are available from the authors on request. This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). 376 The Journal of LAW & ECONOMICS crime committed by criminals who travel to a non-EBT county is excluded from our analysis. Under the assumption that criminals might be able to travel to a neighboring county but no farther than that, this analysis should produce the effect of EBT implementation on crime net of any cross-border crimes. This analysis produced results that are consistent with those presented in Table 3. Next we estimated all of our models controlling for an indicator variable for whether any of the neighboring counties has an EBT program effective in each of our county-year-month observations. The estimate on the binary indicator is negative in four of the six models and never statistically significant. Therefore, it is not surprising that controlling for this variable does not cause an appreciable change to our findings presented in Table 3. In our third robustness analysis, we redefined our treatment indicator such that it equals one if any of the neighboring counties has an EBT program and zero otherwise. The idea is that if there is a displacement effect associated with EBT implementation, then we should obtain a positive association between this indicator and crime. Again, we do not find any evidence supporting this argument. There is no particular pattern in the sign of the indicator for EBT implementation in a neighboring county. Furthermore, the effects are very small, and none of them are estimated with statistical significance. The timing of EBT implementation in Missouri coincides with a period of decline in the unemployment rate and declines in the caseloads for SNAP and TANF. Note that participants in these two welfare programs together constitute the overall client base for the EBT program. It could then be argued that the decline in crime associated with EBT implementation might be attributable to the fact that the period of interest was associated with a decreasing number of EBT card recipients, who are also the potential victims of street crime. While this may first appear to be a plausible story, it is important to remember that the identification of an EBT program effect does not come from a before-and-after comparison of the program’s implementation. Rather, we compare the change in crime associated with EBT between treatment and control counties. If caseloads of SNAP and TANF exhibited similar trends across counties, then our results should not be affected by this development. Furthermore, the variation in the declining unemployment and caseloads for the two programs should, at least partially, be captured by county-specific linear time trends. Even so, we estimated our models controlling for monthly caseloads of SNAP and TANF between 1990 and 2011. Controlling for these variables does not cause any appreciable change to our main estimates presented in Table 3. Unemployment has been shown to be an important determinant of criminal behavior (see, for example, Raphael and Winter-Ebmer 2001). The second half of the 1990s was a particularly favorable period for labor-market conditions in most of the United States, and Missouri was no exception. To the extent that county unemployment rate is uncorrelated with the timing of the EBT program’s implementation, our results should not be sensitive to accounting for the unemployment rate. Given the timing of the EBT rollout, it is difficult to conceive of a This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Electronic Benefit Transfer Program 377 plausible scenario wherein such a correlation should exist.16 Nevertheless, we estimated our models controlling for county unemployment rate obtained from the Local Area Unemployment Statistics program of the Bureau of Labor Statistics. Again, our results do not change in any meaningful way in response. 6. Conclusions In this paper, we show that the transition from check-based welfare payments to EBT is associated with a decrease in street crime, using data from counties in Missouri and counties in bordering states. Our results are particularly strong for burglary, larceny, and assault, with some indication of a nonzero effect for robbery. In robustness analyses, we also document that our results are not driven by displacement of crime between counties that implemented EBT and those that had not yet done so. Furthermore, the results from auxiliary analyses reveal suggestive evidence that EBT implementation might also have been associated with a decline in arrests. Taken together, these findings indicate that the reduction of cash on the streets associated with EBT implementation had a negative impact on crime. For the removal of cash from circulation to result in a concomitant decrease in crime, it must be assumed that a desire to acquire cash is a primary factor motivating street criminals. To test this assumption, we extracted data from the National Crime Victimization Survey for the years overlapping with our analysis period, 1990–2011, and constructed a measure for the proportion of cash stolen as a fraction of the value of all items stolen in completed burglaries, robberies, and thefts that resulted in a loss of at least $1. As shown in Figure 6, loss of cash appears to account for a nontrivial fraction of the total dollar value of all items stolen in these three types of crime. Therefore, the most likely explanation for our overall findings is that moving from a check-based system to EBT effectively reduced the amount of cash on the streets available to be stolen or used for illegal purposes. 16 Note that the federal welfare-reform legislation that was signed into law in July 1996 introduced major changes to the way eligibility for welfare programs and the conditions for maintaining eligibility were defined. However, the associated policy changes in Missouri were not significant because Missouri had already begun implementing its own reform measures under the Beyond Welfare initiative in 1993, and no state legislation was passed in the wake of federal welfare law (Eaton 1998; Carrington, Mueser, and Troske 2002). One exception was the 5-year time limit on receipt of Temporary Assistance for Needy Families, which was introduced in Missouri in July 1997. However, the time limit was a statewide requirement, so there is no possibility that it could be correlated with the EBT implementation. Furthermore, no family had exceeded its time limit before July 2002. This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Figure 6. Ratio of cash to total value lost 378 This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Appendix Crime Patterns for Missouri and Surrounding Counties Figure A1. Trends in total crime rates for Missouri and surrounding counties and the United States. Figure A2. Trends in part I crime rates for Missouri and surrounding counties and the United States. 379 This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Figure A3. Trends in crime rates for Missouri and surrounding counties before and after implementation. 380 This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Figure A4. Event-study estimates for Missouri and surrounding counties before and after implementation. 381 This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). 382 The Journal of LAW & ECONOMICS References Anderson, D. Mark, Benjamin Crost, and Daniel Rees. Forthcoming. Wet Laws, Drinking Establishments, and Violent Crime. Economic Journal. Armey, Laura E., Jonathan Lipow, and Natalie J. Webb. 2014. The Impact of Electronic Financial Payments on Crime. Information Economics and Policy 29:46–57. Brantingham, Paul J., and Patricia L. Brantingham, eds. 1981. Environmental Criminology. Beverly Hills, CA: Sage. ———. 1984. Patterns in Crime. New York: Macmillan. Carrington, William J., Peter R. Mueser, and Kenneth R. Troske. 2002. The Impact of Welfare Reform on Leaver Characteristics, Employment, and Recidivism. Discussion Paper No. 561. Institute for the Study of Labor, Bonn. Eaton, Peter. 1998. Welfare Reform in Missouri. Missouri State Census Data Center Economic Report Series No. 9803. Kansas City: University of Missouri, Center for Economic Information. Edin, Kathryn, and Laura Lein. 1997. Making Ends Meet: How Single Mothers Survive Welfare and Low-Wage Work. New York: Russell Sage Foundation. Federal Deposit Insurance Corporation. 2012. National Survey of Unbanked and Underbanked Households: Executive Summary. Washington, DC: Federal Deposit Insurance Corporation. http://www.fdic.gov/householdsurvey/2012_unbankedreport_execsumm .pdf. Foley, C. Fritz. 2011. Welfare Payments and Crime. Review of Economics and Statistics 93:97–112. Greene, William. 2008. Functional Forms for the Negative Binomial Model for Count Data. Economics Letters 99:585–90. Grogger, Jeffrey. 1990. The Deterrent Effect of Capital Punishment: An Analysis of Daily Homicide Counts. Journal of the American Statistical Association 85:295–303. Jacob, Brian A., and Lars Lefgren. 2003. Are Idle Hands the Devil’s Workshop? Incapacitation, Concentration, and Juvenile Crime. American Economic Review 93:1560–77. Lang, Mathew. 2016. State Firearm Sales and Criminal Activity: Evidence from Firearm Background Checks. Southern Economic Journal 83:45–68. Leyshon, Andrew, and Nigel Thrift. 1994. Access to Financial Services and Financial Infrastructure Withdrawal: Problems and Policies. Area 26:268–75. ———. 1995. Geographies of Financial Exclusion: Financial Abandonment in Britain and the United States. Transactions of the Institute of British Geographers, n.s., 20:312–41. Macaluso, Theodore F. 2000. The Extent of Trafficking in the Food Stamp Program: An Update. Washington, DC: US Department of Agriculture, Food and Nutrition Service. ———. 2003. The Extent of Trafficking in the Food Stamp Program: 1999–2002. Report No. FSP-03-TRAF. Washington, DC: US Department of Agriculture, Food and Nutrition Service. http://www.fns.usda.gov/sites/default/files/Trafficking1999-2002.pdf. Office of the Vice President. 1993. Reeingineering through Information Technology: Accompanying Report of the National Performance Review. Washington, DC: Office of the Vice President. http://govinfo.library.unt.edu/npr/library/reports/it.html. Osgood, D. Wayne. 2000. Poisson-Based Regression Analysis of Aggregate Crime Rates. Journal of Quantitative Criminology 16:21–43. Pulliam, Kathryn O’Neill. 1997. Muddying the Water: Electronic Benefits Transfer after the Welfare Reform Act of 1996. Georgia State University Law Review 14:515–44. Raphael, Stephen, and Raphael Winter‐Ebmer. 2001. Identifying the Effect of Unemploy- This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c). Electronic Benefit Transfer Program 383 ment on Crime. Journal of Law and Economics 44:259–83. Rhine, Sherrie L. W., and William H. Greene. 2013. Factors That Contribute to Becoming Unbanked. Journal of Consumer Affairs 47:27–45. US Department of Agriculture. Food and Nutrition Service. 2014a. SNAP Integrity Efforts Reduce Fraud. Press Release No. FNS 0001.14. Washington, DC, March 25. http://www .fns.usda.gov/pressrelease/2014/fns-000114. ———. 2014b. Supplemental Nutrition Assistance Program (SNAP): A Short History of SNAP. http://www.fns.usda.gov/snap/rules/Legislation/about.htm. US General Accounting Office. 1994. Food Assistance: Reducing Fraud and Abuse in the Food Stamp Program with Electronic Benefit Transfer Technologies. Hearing before the Senate Committee on Agriculture, Nutrition, and Forestry (testimony of Robert A. Robinson). 103d Cong., 2d Sess., February 2. http://www.gao.gov/products/T-RCED -94-125. Varjavand, Reza. 2011. Growing Underground Economy: The Evidence, the Measures, and the Consequences. Journal of International Management Studies 11:133–42. Whitmore, Diane. 2002. What Are Food Stamps Worth? Working Paper No. 468. Princeton University, Industrial Relations Section, Princeton, NJ. This content downloaded from 130.064.011.153 on October 28, 2017 12:26:08 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).