Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
5 An Empirical Assessment of the Contemporary Crime Trends Puzzle: A Modest Step Toward a More Comprehensive Research Agenda Eric P. Baumer The main purpose of this chapter is to report findings from an origi- nal analysis that aims to add a heretofore missing element to the extant crime trends literature: a comprehensive assessment that includes most of the major factors that have been identified as potential keys to resolving the recent crime trends puzzle. I begin with an overview of the state of the existing research and then outline the ways in which the present study goes beyond previous efforts. I then describe the sample and data used in the empirical analysis, summarize the key empirical findings and assess them in comparison to conclusions drawn in other recent studies, and close with a call for additional research that can build on the analysis to help establish the kind of research agenda needed to make significant progress in develop- ing a more definitive portrait of the determinants of recent crime trends. THE CRIME TRENDS PUZZLE The basic portrait of U.S. crime trends during the past three decades is now well known. There were steep increases in rates of robbery, motor vehicle theft, and overall homicide from the mid- to late 1980s through the early 1990s. The patterns for homicide during the 1980s varied by age and method, with youth firearm homicide rates following the trend shown for overall homicide, but adult homicide rates and nongun homicide rates fall- ing modestly throughout the 1980s, much like the observed trends in bur- glary (e.g., Blumstein and Rosenfeld, 1998; Cook and Laub, 1998). Crime patterns were much more consistent across crime types in the 1990s, as all forms of crime declined considerably, a trend that showed signs of slowing 127
128 UNDERSTANDING CRIME TRENDS only during the early years of the present decade (Blumstein and Wallman, 2006a; Zimring, 2006). In the past few years, attention has turned to increases in violence observed in some cities across the United States (e.g., Police Executive Research Forum, 2006). What explains these recent shifts in crime rates? Are they the result primarily of modifications to the quantity and quality of policing and incarceration? Were shifts in abortion laws, demographics (e.g., age struc- ture and immigration), or the economy (e.g., unemployment and wages) important? Have changes in illicit drug (e.g., crack cocaine) involvement or alcohol consumption played a role? Do these or other factors help explain the substantial degree of variability in crime trends observed across places? And, ultimately, which factor or set of factors has contributed most to shaping recent crime trends? These are the questions of primary interest in this volume and the ones that I have been asked to examine in this chapter. Addressing these questions is difficult but vitally important for shaping perceptions of public safety among citizens, informing public policy debates about how best to respond to crime, and identifying conditions that are most apt to produce or prevent major shifts in crime. THE STATE OF KNOWLEDGE Policy makers, the media, and other citizens have rightly pressed for answers to the puzzling changes in U.S. crime rates over the past three decades. However, the existing empirical research on recent crime trends is in the early stages of development and is not at the point of sufficient breadth or depth to provide definitive evidence on which factors mattered a lot, which mattered relatively little, and, importantly, which mattered the most. The research community appears to have been reluctant to admit this apparent fact, concluding instead that the available evidence supports either the conclusion that virtually all of the dozen or so factors implicated in the theoretical literature played some role in shaping recent crime trends or the conclusion that one or more specific factors were very important and others mattered little. In my view, neither of these conclusions is supported strongly by the available evidence, which consists primarily of inconclusive circumstantial patterns and empirical evidence based on limited data and models that, as elaborated below, simply do not permit strong conclusions one way or another. Despite bold claims about which factors mattered and which did not (e.g., Levitt, 2004), as Travis and Waul (2002, p. iii) summed up after a national forum on the subject, although a good deal has been learned from prior research, there are no âdefinitive answers to the ques- tions raised by the recent crime [trends]âthat would require more research, new data, and a sustained effort to reconcile every competing claimâ (see also Blumstein and Wallman, 2006b; Rosenfeld, 2004).
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 129 The uncertainty associated with identifying more clearly the primary sources of recent crime trends is not due to a scarcity of ideas about why crime probably changed in the manner it has or to the absence of sophis- ticated empirical investigations. Within just the past decade, four books (Blumstein and Wallman, 2006a; Conklin, 2003; LaFree, 1998; Zimring, 2006), a National Institute of Justice (NIJ) symposium published in the Journal of Criminal Law & Criminology (Travis, 1998), a major NIJ intramural research project (Lattimore et al., 1997), and several articles and conference panels have been devoted to explaining the crime trends observed in the United States during the 1980s and 1990s. As illustrated in Figure 5-1, this attention has generated a rich and creative set of plausible hypotheses for recent crime trends. Surely, one or more of the factors iden- tified in the figure was highly influential in shaping recent crime trends in the United States. But if the hypothesized causes have been identified, why Illicit Drug Use Alcohol Police Force and Market Consumption Firearm Size Activity Prevalence Public Order and + + + Weapons Enforcement _ _ Incarceration _ Volume of + Crime Offender Reentry Rates + _ Unemployment _ _ Wages for + Legal Work Percent Age Immigration Percent Age 45 and Older 15 to 24 Levels _ _ High-Risk Births Levels of and Births to High-Risk Fertility Control, t-15-24 Women, t-15-24 FIGURE 5-1â Heuristic model of hypothesized main effects on recent crime trends. Figure 5-1, fully editable
130 UNDERSTANDING CRIME TRENDS does one not know something more definitive empirically about the sources of recent crime trends, and specifically about which factors mattered or were most influential? When criminology and criminal justice scholars are asked why crime rates have taken their observed path since the early 1980s, why is the modal response something along the lines of âit is unclear, but probably several things worked together to bring crime downâ (see, e.g., Tierney, 2007)? In my view, there are two major impediments to providing definitive responses to key theoretical and policy questions associated with recent crime trends: There simply has not been a sufficient amount of empiri- cal research on the matter, and the existing body of research is largely unsystematic along a variety of dimensions. In a vibrant research agenda, a diversity of data, measures, and methods brought to bear on a problem often are welcomed, because a high volume of research makes it is possible to see meaningful patterns emerge from a collective effort that helps to establish an overall sense of the issue at hand, net of the different ways in which the issue has been studied. But the volume of empirical research in this area does not seem sufficient at present to generate a landscape ripe for producing enough information to glean emergent patterns across the diverse approaches currently taken among those studying crime trends. What is more, the relatively barren landscape of crime trends research has provided fertile ground for advocates of the importance of particular fac- tors (e.g., policing, incarceration, crack, abortion, immigration) to draw relatively strong conclusions in an empirical vacuum in which the various factors that also may have mattered (and possibly mattered more) are rarely considered simultaneously. There are several excellent empirical papers on recent crime trends that provide persuasive evidence that a given factor was critical to shap- ing recent trends, yet most of the factors identified in Figure 5-1 have not been evaluated empirically very often or in systematic ways and, even when they are examined, the overall body of relevant empirical research on crime trends suffers from two major limitations that have served as impedi- ments to establishing a more definitive body of knowledge on why crime rates have taken their observed path since the early 1980s: (1) a narrow research focus on only a few of the factors believed to be important for shaping recent crime trends and, consequently, a high degree of empirical misspecification and (2) substantial variability in the analytical methods applied across studies, including the explanatory variables considered, units of analysis employed, and the types of models estimated. Regarding the first limitation, although most observers seem to agree that several factors probably coalesced to shape recent crime trends (e.g., Blumstein and Wallman, 2006a; Rosenfeld, 2004; Travis and Waul, 2002; Zimring, 2006), the empirical literature generally has focused narrowly on
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 131 a small subset of the potentially relevant factors, most typically police force size, drug arrest rates, and incarceration. City- and county-level studies of recent crime trends (e.g., Baumer et al., 1998; Gallup-Black, 2005; Levitt, 1997; Lott, 1998; Ousey and Lee, 2002; Phillips, 2006) have focused on the first two of these factors, but they typically exclude time-varying indicators of nearly all of the other factors shown in Figure 5-1 and, perhaps most importantly, they rarely incorporate any indicators of incarceration or jail confinement rates, which have emerged as central in many other studies. Similarly, although state-level studies of recent crime trends routinely include indicators of overall incarceration rates (e.g., Levitt, 1996; Liedka, Piehl, and Useem, 2006; Marvell and Moody, 1994, 1997), they have not considered age- or crime-specific incarceration rates, average sentence length, time served, or prisoner release rates, even though these factors have been highlighted as potentially important in theoretical discussions and are available as data elements in the public domain. It is also noteworthy that none of the existing studies of recent crime trends has included indicators of recent immigration flows and, surprisingly, only a few have incorpo- rated direct indicators of changes in the nature of policing across multiple geographic units (MacDonald, 2002; Messner et al., 2007; Rosenfeld, Fornango, and Rengifo, 2007). A corresponding story emerges for many of the economic and demographic conditions emphasized as potentially important in the theoretical literature on recent crime trends. Although the occasional study examines wages (Gould, Weinberg, and Mustard, 2002), levels of âdomesticityâ (Dugan, Nagin, and Rosenfeld, 1999), and indica- tors relevant to assessing the role of abortion law changes (Donohue and Levitt, 2001, 2006), most studies do not consider these factors, despite evidence suggesting that they may be very important. In short, a common theme in the extant research on recent crime trends is that most studies have a limited scope, focusing on a few select factors and ignoring many other potentially important ones. The high level of empirical misspecification makes the findings reported in much of the prior research on crime trends open to charges of spuriousness, including studies often cited as the primary regression-based evidence of a significant link between recent crime trends and such factors as crack cocaine involvement and incarceration (e.g., Baumer et al., 1998; Ousey and Lee, 2002; Stemen, 2007). Some studies minimize the serious omitted variables bias concern by incorporating âfixed effects,â but this strategy does not say anything about the specific role of the omitted factors for which the fixed portions of the model are serving as a surrogate. Overall, despite much speculation about the explanatory potential of several fac- tors, the limited scope of existing research and differences across studies in the variables considered make it difficult to draw definitive conclusions
132 UNDERSTANDING CRIME TRENDS about the role of given factors, much less to make concrete determinations about the relative impact of specific factors. A second major impediment to establishing from existing research a more definitive body of empirical evidence and knowledge about the fac- tors that have shaped recent crime trends is the substantial variability in analytical methods applied across studies (see also Spelman, 2008). Beyond the significant differences in model specification already noted, the extant research also is hard to pin down systematically because of (a) the use of different units of analysis across studies and (b) the reliance on different types of statistical methods to estimate key parameters. With respect to the former issue, research on recent crime trends has been conducted across multiple units of analysis, most often states, counties, cities, neighborhoods, and police precincts. It is not necessarily important that a particular unit of analysis be identified, a priori, as superior for studying crime trends, for the reality is that each of these units has conceptual merit, and there are important trade-offs in the choice of unit. Nonetheless, it would be use- ful to know the empirical implications of using different units of analysis, something that cannot be deciphered easily from existing research. There also appears to be little consensus on the type of methodological approach that is best suited for studying crime trends, especially in subÂ national (states, counties, cities, neighborhoods) studies, and consequently there is a tremendous degree of variation across studies in the Â methods applied. Some studies have used methods geared toward identifying classes of crime âtrajectoriesâ (e.g., Weisburd et al., 2004), while most have applied different versions of two suitable analytical strategies: pooled time- series cross-sectional panel models (e.g., Donohue and Levitt, 2001; Gould, Weinberg, and Mustard, 2002; Phillips, 2006) and multilevel growth curve models (e.g., ÂBaumer et al., 1998; Kubrin and Herting, 2003; Ousey and Lee, 2004; Rosenfeld, Fornango, and Rengifo, 2007). The specific choice between the latter two approaches is not very important, for they can be made to be equivalent with proper modifications, but it is critical to rec- ognize that the two approaches typically are implemented in ways that are likely to generate different findings and conclusions even when applied to the same data (see, e.g., Phillips and Greenberg, 2008). It is also important to note that there are some important differ- ences in how each of the two most common approaches is implemented. For instance, some studies of recent crime trends employ econometric panel models appropriately by testing for stationarity in the variables (e.g., P Â hillips, 2006), but most others do not do this (e.g., Donohue and Levitt, 2001; Gould, Weinberg, and Mustard, 2002). Some estimate models in both levels and differences (Moody and Marvell, 2005), while most focus on levels only (for a review, see Moody, 2007). And some include unit-specific trends (e.g., Raphael and Winter-Ebmer, 2001), while most studies do not.
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 133 In many instances there are legitimate disagreements about which of the many possible specifications is most appropriate under particular circum- stances, but it is important to recognize that the different specifications used are likely to generate different results and conclusions even when applied to the same data (Spelman, 2008). Recognizing this is, of course, key to compiling a systematic body of knowledge that may or may not point to particular answers to questions about crime trends. RESEARCH NEEDS AND THE PRESENT STUDY Recent crime trends represent a major social phenomenon and a fun- damental research issue for which criminological researchers should pro- vide concrete answers. We should and can do better than âa lot of things mattered,â but doing so will require a more vibrant research agenda that focuses on modeling recent crime trends in a much more comprehensive and systematic way. Ideally, this research agenda would incorporate a comprehensive set of measures that mirror as closely as possible the fac- tors emphasized in theoretical and policy discussions of crime trends. It would also apply appropriate methodological techniques uniformly across multiple units of analysis (or a common unit) while attending to impor- tant econometric issues (e.g., stationarity, spatial dependence, endogeneity) that are critical for the inferences drawn from temporal data. As Spelman (2008) has recently demonstrated, attending to these issues even for a single crime rate covariateâincarceration ratesâin a bivariate context is a highly complex enterprise, so extrapolating the effort to a larger scale will not be easy. Nonetheless, this kind of systematic effort is feasible and would help to better pinpoint empirical patterns in available data and minimize the degree to which conclusions are confounded by differences across studies in specification, unit of analysis, or method. Developing the research agenda just described will take a concerted effort by a community of scholars who share data and ideas to build a knowledge base on crime trends incrementally and systematically. The goal here is to make a modest contribution to this effort by significantly broaden- ing the scope of empirical research to incorporate not only the factors often considered (e.g., police force size, drug use and market Âactivity, age struc- ture, incarceration) in the literature, but also various others that have been highlighted in theory but only occasionally considered in empirical studies (e.g., immigration, wages, alcohol consumption, levels of Âdomesticity, and youthful cohort âqualityâ). The main contribution, then, lies in an expan- sion of the empirical specification typically applied in research on recent crime trendsâit is the first study of which I am aware that examines simul- taneously each of the major factors shown in Figure 5-1 that have been emphasized in theoretical and policy discussions of recent crime trends.
134 UNDERSTANDING CRIME TRENDS This strikes me as a logical starting point for establishing a more definitive knowledge base about recent crime trends. I must emphasize that the analysis described below is relatively Âmodest and only an initial step in moving toward a more definitive knowledge base from which to draw answers to the important questions that have emerged about recent crime trends. As mentioned above and elaborated below, there are also several unresolved and complicated methodological issues that should be tackled in a comprehensive research agenda on crime trends. In an exhaustive analysis, for instance, the econometric properties of each of the variables considered should be evaluated (e.g., Are the variables station- ary or nonstationary?) and appropriate transformations should be made to the data and estimation techniques (e.g., Should the variables be differ- enced? Are the variables cointegrated?). Furthermore, the many instances of possible endogeneity in models of crime trends should be addressed, spatial dependence should be assessed and, if necessary, modeled appropriately, and there should be a systematic approach to model selection so that one can make informed choices about the factors that emerge as most relevant. These matters are not merely statistical exercises; they have important implications for the inferences drawn about crime trends. Although there is important work being done on these issues both in criminology and other disciplines, collectively they represent a challenging set of issues that is more complex than often portrayed in the literature and which cannot be addressed satisfactorily here given the space constraints of this volume. Instead, although the present work expands on recent crime trends research by considering a much larger set of the factors most commonly emphasized in theoretical discussions, it applies the same econo- metric tools used in many existing subnational studies. Consequently, the conclusions that can be drawn must be viewed as tentative; the important methodological issues noted earlier will need to be addressed to assess the validity of the results reported below. â ational-level N studies of crime and other social phenomena have addressed these issues, and some panel studies have entertained these issues as well, but in general they have not been dealt with effectively in the criminological literature. Among other things, most panel stud- ies that employ panel unit root tests have used so-called second-Âgeneration tests that do not account for spatial dependence, which is likely to be present in most of the subnational data used to study crime trends and could bias results. Also, there are many questions about how to handle heterogeneous panels, some of which contain unit roots and others that do not, that have not been adequately resolved. Differencing is a common solution when nonstationarity is found in panel studies, but, without testing for cointegration, it is not clear whether this is an appropriate solution or one that produces valid results. Finally, assessments and corrections for endogeneity vary wildly in the literature, so its impact remains unclear, and there is little guidance from the literature about how best to approach such issues in practice.
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 135 BROADENING THE SCOPE OF CRIME TRENDS RESEARCH This study expands the scope of research on recent crime trends by taking a more comprehensive approach to measuring and modeling effects of factors that have been well represented in prior research (e.g., such com- monly considered factors as policing, incarceration, and illicit drug Âactivity), by incorporating measures of factors that are irregularly included in the extant research (e.g., largely neglected factors, such as alcohol consump- tion, legal wages, levels of domesticity, immigration, and cohort Â quality indicators) and, more generally, by estimating models in which all of these factors are considered simultaneously. Many of these factors have been discussed extensively elsewhere, including prior chapters in this volume, so they do not need to be reviewed in detail here. But I elaborate on some relatively neglected issues and outline the ways in which I go beyond past work. Reconsidering Commonly Addressed Factors Two factors that have received a particularly high level of attention in public discourse on recent crime trends, and especially the 1990s crime decline, are changes in policing and incarceration. With respect to the for- mer, Eck and Maguire (2006) provide an excellent treatment of the various changes in the quantity and quality of policing that have occurred in the past several decades, focusing especially on (1) increases in the number of police officers devoted to helping address crime problems and (2) enhance- ments to the manner by which police agencies have approached their work, particularly a move toward a targeted policing focus on behaviors thought to facilitate crime, such as levels of public disorder and the prevalence of weapon carrying. Several studies examining the link between crime trends and police size have generated inconsistent results (see Eck and Maguire, 2006, for an exhaustive review). A relatively large body of research also has examined the effects of different policing approaches on levels of crime (e.g., Sampson and Cohen, 1988; for a review, see MacDonald, 2002), but only a small handful of studies have assessed explicitly the role of recent changes in the nature of policing on contemporary crime trends. Overall, the existing research suggests that policing efforts that targeted public order violations and weapon carrying may have had a modest effect on crime trends during the 1990s (Braga et al., 2001; Kennedy et al., 2001; Messner et al., 2007; Piehl et al., 2003; Rosenfeld, Fornango, and Rengifo, 2007), but the limited attention to this issue precludes more definitive conclusions being drawn. I build on existing work by examining whether a changing police focus on public order crimes and weapons offensesâas measured by the number of arrests per 100,000 residents for weapons
136 UNDERSTANDING CRIME TRENDS violations, vandalism, prostitution, gambling, liquor laws, drunkenness, disorderly conduct, vagrancy, curfew violations, loitering, and suspicion (for similar measures, see Messner et al., 2007; Rosenfeld, Fornango, and Rengifo, 2007)âis associated with recent crime trends across a relatively large sample of cities. The other major criminal justice factor emphasized in the literature on recent crime trends is the well-documented substantial increase in incarcera- tion rates, which have more than tripled in the United States since the early 1970s (Zimring, 2006). There is a long history of linking incarceration to crime rates through incapacitation and/or deterrent processes (e.g., Zimring and Hawkins, 1973) and a relatively large and growing empirical literature. Yet, despite the substantial attention devoted in prior research to the role of overall incarceration rates and what appears to be some consensus on the overall impact of incarceration on crime trends (Stemen, 2007; but see Spelman, 2008), two aspects of the link between incarceration and crime have been neglected in prior work and warrant additional consideration: (1) the analysis of both âstockâ (i.e., the overall number of persons per capita in prison at a given point in time) and âflowâ (i.e., the number of persons per capita admitted to prison and released from prison in a given year) measures of incarceration and (2) the analysis of temporal variability and scale effects for incarceration. Discussions of incarceration effects tend to emphasize the crime reduc- tion that may result from relatively immediate sentencing actions, such as the recent removal from the street of each additional offender. Most studies of incarceration effects estimate how stock incarceration rates for a given year affect crime in that year or the next. However, during periods of sentence enhancements, such as the 1980s and 1990s, the stock incar- ceration rate may not be a very good indicator of how many offenders were removed from the streets and placed in prison in a particular year (it will reflect admissions in that year and many before it). Thus, modeling its contemporaneous or one-year lagged effect on crime rates may yield a misleading picture of the effect of recent incarceration practices on crime rates. Annual flow indicators of the number of persons admitted to prison (less the number of persons released), of course, are well suited for gauging such effects. The present study therefore estimates models of crime for both stock and flow measures of incarceration to evaluate in a more comprehensive manner the role of incarceration in shaping recent crime trends. Consid- ering the prison flow measures not only provides a more precise look at incarceration effects than relying solely on the stock incarceration rate, but it also permits an independent assessment of prison releases on recent crime trends. There has been a lot of attention recently to the consequences and challenges of a large volume of prisoners moving from prison back to
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 137 their home communities (e.g., Travis and Visher, 2005) yet very little direct empirical investigation of whether and how trends in prison releases may have affected crime trends during the past few decades. Only one of the regression-based studies of recent crime trends has considered the quan- tity or quality of prison releases (Kovandzic et al., 2004). No significant impact of prison release rates was observed in that study, but the focus was exclusively on homicide, a relatively rare crime, and it is uncertain whether similar findings would emerge for more common offences. Recent research on incarceration rates and crime trends also challenges the assumptions shared in most previous studies of linear and time-invariant incarceration effects. Some have argued that the elasticity of incarceration has changed over time, although there is disagreement about the direc- tion of this change (see Liedka, Piehl, and Useem, 2006; Spelman, 2000). Spelmanâs (2000) research suggests that the effectiveness of prisons may have increased over time because of growth in the scale of imprisonment, the proportion of crime committed by adults, and the selectivity of law enforcement efforts (i.e., the degree to which serious offenders are impris- oned). More recently, Liedka, Piehl, and Useem (2006) suggest that the crime reduction benefits of incarceration are likely to be reduced as the scale of incarceration reaches very high levels and may even reverse, such that very high levels of incarceration may actually increase crime. Their state- level panel analysis of data of 1972-2000 reveals evidence consistent with this claim. Furthermore, although they do not examine the issue directly, they suggest that in contrast to Spelmanâs argument about the increasing effectiveness of incarceration over time, one implication of their findings may be that the elasticity of incarceration has probably declined (i.e., become less negative) as incarceration rates in many states have approached and surpassed an effective deterrent or incapacitation level. These recent studies point to the need for additional refined analyses of incarceration effects that move beyond estimating elasticities under the assumption of temporal invariance and that are attentive to potential vari- ability in elasticity by changes in scale and the composition of the prison- bound population. I explore some of these issues by examining not only the main effect of the overall prison admissions rate, but also the possibility of a nonlinear response for this variable and whether the estimated effects have changed over time. The role of illicit drug use and market activity, especially with respect to crack cocaine, also has received a good deal of attention in the theoretical and empirical literature on recent crime trends. Although there are some doubters (e.g., Zimring, 2006), there seems to be a fairly strong consensus that the rise and fall of crack use and crack markets are important pieces of the crime trends puzzle over the past 25 years (Blumstein and Wallman, 2006a; Johnson, Golub, and Dunlap, 2006; Levitt, 2004). The central
138 UNDERSTANDING CRIME TRENDS arguments provided for the link between crack and violence are logical and persuasive (Baumer et al., 1998; Blumstein and Rosenfeld, 1998), and the demographic features of recent homicide trends certainly fit well with the idea that crack use and crack markets were an important facilitator (Blumstein, 1995). However, the systematic empirical evidence in support of this hypothesis is not as abundant or definitive as one might suspect, and it is not clear precisely how important changes in drug market activity and drug use were during the rise in violence observed in the 1980s and the decline observed in the 1990s and beyond. The relatively few studies that have examined the issue generally show that cities with higher levels and greater increases in crack use and market activity experienced larger increases in violence during the 1980s (e.g., Baumer et al., 1998; Cork, 1999; Fryer et al., 2006; Grogger and Willis, 2000). However, these studies may overstate the magnitude of the link between crack and violence during the 1980s because they omit many other potentially important factors that changed over time and are thought to be connected to recent crime trends. The evidence is also mixed with respect to the role of changes in crack use and crack markets for the 1990s crime decline (see Corman and Mocan, 2000; Fryer et al., 2007; Messner et al., 2007; Rosenfeld, Fornango, and Baumer, 2005; Rosenfeld, Fornango, and Rengifo, 2007). In short, while the rise and fall of the crack epidemic probably played an important role in recent crime trends, especially youth violence, there are several unresolved issues. Some of the evidence in support of the hypoth- esized connection can be criticized for being generated from empirical models that omit many of the other factors thought to be relevant to recent crime trends. Also, although some scholars have emphasized the possible interactive effects of drug markets and the legitimate economy on recent crime trends, aside from assessments of the contingent role of drug markets in areas with different levels of economic and social disadvantage (e.g., Ousey and Lee, 2004), this has rarely been examined systematically. Finally, one potentially important aspect of Blumsteinâs argument about the link between crack markets and violence has not been examined in previous work. Specifically, Blumstein (1995) suggests that a distinctive feature that made crack markets particularly violent in the late 1980s and early 1990s was their age composition, in particular the fact that they were dominated by younger people. As the demand for crack grew and the adult sellers who dominated markets were arrested and imprisoned, Blumstein notes that crack markets became staffed largely by young and inexperienced street sellers who, compared with their older counterparts, were more reckless and irresponsible. They lacked the necessary maturity and skills to resolve conflicts in nonphysical ways, stimulating them to use guns with little restraint (Blumstein, 1995, pp. 29-31). This suggests that recent crime trends may be linked to the degree to which drug markets are âstaffedâ by
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 139 younger people, an issue that has not been examined directly. The present study contributes to the literature by examining multiple measures of crack cocaine involvement, including overall arrest rates for cocaine and heroin as well as an indicator of the age structure of crack markets and indicators of drug-related mortality, in a comprehensive empirical model that also incorporates other factors that might be relevant. Attending to Some Neglected Factors In addition to evaluating some expanded measures of commonly consid- ered causes of recent crime shifts, I also examine several factors highlighted in theoretical and policy discussions but rarely examined systematically in the empirical literature. These include changes in alcohol consumption, legal wages, levels of domesticity, immigration, and birth cohort quality. Alcohol has long been linked to violence, perhaps even more so than other drugs (Fagan, 1990; Parker and Rebhun, 1995), and national-level trends in alcohol consumption during the past several decades yield patterns similar to trends in homicide rates, especially the observed trends in adult homicide rates since 1980 (Parker and Cartmill, 1998). Although the avail- able evidence suggests that alcohol consumption may play a significant role in shaping violence levels and trends at the national level and across states, metropolitan areas, cities, and neighborhoods (e.g., Fagan, 1990), most studies of recent crime trends have not considered this possibility. In fact, to my knowledge, there is not a single published study that has considered the link between trends in alcohol consumption and city or county crime trends using annual panel data. This is likely to be the case because data on alcohol consumption are not readily available for counties or cities. In light of this, I evaluate the role of alcohol consumption on recent crime trends using a proxy measureâthe percentage of fatal traffic acci- dents involving alcoholâthat exhibits a strong temporal correspondence with alcohol consumption at the national level and that is available for cities and counties annually from the late 1970s to the present. Incorporat- ing this measure into the analysis presented below not only provides a way to assess the independent effect of alcohol consumption on recent crime trends, but also may enhance the identification of other effects, including unemployment (see, e.g., Cook and Zarkin, 1985; Raphael and Winter- Ebmer, 2001). A wide range of economic conditions have been mentioned in the literature on recent crime trends (see Rosenfeld and Fornango, 2007), but unemployment rates and wages have been the economic factors most often implicated as shaping the likelihood of offending directly as well as medi- ating the effects on crime of related variables, such as educational attain- ment (e.g., Blumstein and Wallman, 2006a). Wages are hypothesized to be
140 UNDERSTANDING CRIME TRENDS negatively associated with crime rates, while unemployment is hypothesized to be positively related to crime. There is a voluminous empirical literature on the link between unemployment rates and crime rates using state- and national-level data; overall the evidence from this work reveals little, if any, effect of unemployment on violence rates but that property crime rates tend to decrease by about 1 to 5 percent with each percentage point reduction in unemployment rates (Levitt, 1996, 1997, 2001; Raphael and Winter-Ebmer, 2001). Much less empirical attention has been devoted to assessing the pos- sibility that changes in legal wages are associated with recent crime trends. Gould, Weinberg, and Mustard (2002) have conducted the only aggregate- level study of which I am aware that examines the relationship between wages and crime rates. Overall, their results suggest that changes in real wages for unskilled men account for more than one-third of the increase in crime rates observed in U.S. counties during the late 1980s, but less than 5 percent of the decline in crime rates between 1993 and 1997. The present study builds on recent research on the effects of unemployÂ ment and wages on recent crime trends in two ways. First, the key economic indicators are measured at a more local level in the present work, capturing economic conditions and crime at the county and city levels of analysis rather than the state or national level. This is potentially important given the high degree of local variability in both economic conditions and crime (Levitt, 2001), yet none of the existing city-level studies of crime trends includes a time- varying indicator of unemployment rates, and only a few county-level studies have done so (e.g., Phillips, 2006). Second, the present research explores interactions between unemployment and wage indicators and measures of the magnitude of the crack cocaine markets that characterize the large U.S. cities selected by the Committee on Law and Justice for Âexaminationâreferred to here as âNRC cities.â As noted above, several scholars have alluded to the possibility of this type of interaction (e.g., ÂBlumstein and Rosenfeld, 1998; Fagan and Freeman, 1999; ÂGrogger, 2006; Zimring and Hawkins, 1997), but it has rarely been examined systematically. Although a variety of demographic features have been alluded to in the literature as potentially relevant to recent crime trends, the existing empirical literature has taken a relatively narrow approach to measuring the role of demography. This work adds to the literature by explicitly con- sidering three demographic factors highlighted in theoretical and policy discussions of recent crime trends but that have received meager attention in the empirical literature: levels of domesticity (Rosenfeld, 2006), levels of immigration (Sampson, 2006), and the conditions under which the contemporary youth population was born (Donohue and Levitt, 2001; Sampson and Wilson, 1995). Rosenfeld (1997) directs attention to a potentially important social change witnessed during the past few decadesâthe substantial retreat from
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 141 marital unions (Amato et al., 2007)âand hypothesizes that the associated âdeclining domesticityâ may be key to the observed reductions in adult homicide since the early 1980s and intimate partner homicide in particular. The rationale underlying this link is simple: when the fraction of the popu- lation that is married (or gets married) falls, so too does the overall number of reoccurring opportunities for lethal violence between intimates. Detailed analysis of adult spousal homicide trends (Rosenfeld, 1997; Blumstein and Rosenfeld, 1998) across groups that differ substantially on marriage and divorce propensities reveals evidence consistent with patterns one would expect if declining domesticity were an important contributor to the decline in intimate partner homicide. Dugan, Nagin, and Rosenfeld (1999) also provide support for the notion that declining domesticity is important for understanding declines in intimate partner homicide. The present study, building on this earlier work, evaluates whether changes in levels of domesticity during the past 25 years are associated with changes in city crime rates. Like earlier work, the analysis examines the role of the decline in marital unions in shaping adult homicide trends. But I also extend prior work by considering whether recent changes in the prevalence of both marital unions and nonmarital cohabitation may help to explain changes in violence and changes in burglary rates since the early 1980s. Rates of nonmarital cohabitation have increased considerably dur- ing the past three decades (e.g., Amato et al., 2007; Casper and Cohen, 2000) and have substantially offset the decline in marriage rates. Given that cohabiting relationships have been shown to yield more violence than other types of relationship statuses (Brownridge, 2004; Shackelford, 2005), it is important to consider trends in cohabitation along with trends in mar- tial unions, both because the former are interesting in their own right and because not doing so may bias estimates of the latter. It is also plausible that domesticity effects are relevant for burglary rates. From a routine activities perspective, for example, domesticity should be inversely associated with changes in burglary rates. I evaluate these predictions using annual state- level data on household composition from the Current Population Survey and testing whether cities located in states that exhibited greater changes in the proportion married or cohabitating experienced more substantial changes in rates of adult homicide, burglary, and other crimes. Another demographic feature that has been linked to recent crime trends in the United States is the level of immigration. In a 2006 New York Times op-ed contribution, Sampson extrapolated from the findings revealed in recent individual- and multilevel studies of the role of immigrant status in shaping involvement in crime and violence (e.g., Butcher and Piehl, 1998; Sampson, Morenoff, and Raudenbush, 2005; for reviews, see Hagan and Palloni, 1998; Martinez and Lee, 2000) to suggest that recent increases in levels of immigration may be a major factor in the decline in crime during
142 UNDERSTANDING CRIME TRENDS the 1990s in the United States as well as the leveling off of crime in the early part of the 2000s (see also, Sampson, 2008). The logic of his argument relies heavily on the relatively lower rates of offending exhibited by immi- grants and the fact that, if additions to the population due to immigration are primarily nonoffenders, the crime rate will by definition drop. Increased immigration can affect aggregate crime rates in a variety of ways, however, beyond the criminal offending rates of new arrivals. Sampson has alluded to the possibility that the influx of immigrants in the 1990s may have reduced crime because immigration increased collective efficacy (Press, 2006). In addition, an influx of immigrants may bolster local economies and thus reduce pressures to engage in illicit conduct, or a large pool of immigrants may bring with them a value system that eschews violence as a means of settling interpersonal disputes (Reid et al., 2005; Sampson, 2008). Sampsonâs speculation about an inverse association between changes in immigration and changes in crime rates is plausible, but what does the empirical evidence say about this possibility? Have changes in levels of immigration been relevant to recent crime trends, and, if so, have changes in immigration been associated with increases or decreases in crime? The circumstantial evidence is persuasive. Many U.S. border cities consistently exhibit relatively low crime rates (Martinez and Lee, 2000), and a cursory look at the cities that experienced the largest declines in crime during the 1990s (see, e.g., Zimring, 2006) reveals that many are places that routinely experience high levels of immigration. Also, at the national level, immigra- tion grew substantially during the 1990s, and this growth accelerated right around the time (about 1994) when the crime decline accelerated (Simanski, 2005). Yet there is relatively little systematic empirical evidence on the link between immigration and crime at the aggregate level, and only three s Â tudies of which I am aware consider empirically the role of immigration on recent crime trends (Butcher and Piehl, 1998; Rosenfeld, Fornango, and Rengifo, 2007; Sykes, Hangartner, and Hathaway, 2007). The weight of the evidence from aggregate cross-sectional research on immigration and crime appears to be that the relationship is either nonexistent or negative (ÂMartinez and Lee, 2000; Reid et al., 2005), but findings are d Â ecidedly mixed in other types of studies, with the effects positive for some crimes or contexts and negative for others (e.g., Hagan and Palloni, 1998). Butcher and Piehl (1998) found no significant association between changes in crime rates and changes in the stock of foreign-born or the flow of new immigrants during the 1980s. Rosenfeld, Fornango, and Rengifoâs (2007) analysis of 1990s crime trends for New York City police precincts shows that areas with a higher percentage of foreign-born residents experienced significantly greater declines in robbery rates, but they found no significant relationship between changes in percentage foreign-born and changes in
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 143 crime. In contrast, a state-level panel analysis conducted by Sykes, Hangart- ner, and Hathaway (2007) reveals a significant negative association between changes in the percentage foreign-born and changes in rates of property and violent crime, which is consistent with Sampsonâs argument. More research is needed to assess the merits of the idea that immigra- tion flows are associated (negatively or positively) with recent crime trends. Admittedly, estimating the number of people who immigrate to the United States with precision in any given year is very difficult (Hagan and Palloni, 1998), but the present study builds on recent panel analyses of this issue by drawing on data from the Immigration and Naturalization Service to estimate annual figures on legal immigrants who reported an intended residence in the metropolitan areas in which the NRC cities are located. These data were used to estimate the annual number of newly admitted legal immigrants (per 100,000 current residents) intending to reside in the metropolitan statistical areas in which the NRC cities are located. Perhaps the most controversial demographic argument that has emerged in discussions of recent crime trends is Donohue and Levittâs âabortion divi- dendâ thesis. Although the details of this argument are somewhat complex, in essence Donohue and Levitt (2001) invoke a classic cohort theory argu- ment (e.g., OâBrien, Stockard, and Isaacson, 1999) in suggesting that the legalization of abortion in the early 1970s in the United States served to reduce crime substantially during the 1990s because it resulted in smaller cohorts of teenagers and young adults (e.g., ages 15-24) in this period and, more importantly, because a smaller proportion of this age group had high- risk birth attributes and/or a smaller proportion were born to high-risk mothers, as indicated, for example, by maternal age, marital status, and educational and economic status at birth. Much of the research attention on and discussion about the abor- tion dividend argument understandably has focused exclusively on the direct link between 1970s abortion law changes and recent crime trends. Although some observers have expressed skepticism about this link because the timing of abortion law changes and the beginning of observed declines in the 1990s do not coincide neatly (e.g., Blumstein and Wallman, 2006a; Fox, 2006; Rosenfeld, 2004), and others have raised concerns about ques- tionable assumptions and empirical specifications associated with some of the initial findings presented by Donohue and Levitt (2001) (e.g., Foote and Goetz, 2005; Joyce, 2004; Sykes, ÂHangartner, and ÂHathaway, 2007; Z Â imring, 2006), the empirical evidence on the association between abor- tion law changes and contemporary crime trends presented by Donohue and Levitt (2001, 2006) has so far withstood the challenges. Moreover, others have reported results that affirm their core findings (Berk et al., 2003; Sorenson, Wiebe, and Berk, 2002), and the magnitude of the effects of abortion law changes implied in their work is substantial, account-
144 UNDERSTANDING CRIME TRENDS ing for perhaps half of the crime decline in the United States during the 1990s. Nevertheless, the legalization of abortion is but one of many social changes that could yield the differential fertility outcomes posited by D Â onohue and Levitt (2001, 2006) to link abortion to contemporary crime trends. In my view, the key criminological and more proximate causal questions that emerge from their work are not whether abortion laws are associated with crime trends, but rather whether recent crime trends were significantly shaped by (1) the percentage of persons in high-crime-rate age groups (e.g., 15-24); (2) the relative number of persons in this age group who were born in high-risk family situations (e.g., unmarried mothers, teenage mothers, mothers with relatively little education or low economic status); and (3) the relative number of persons in this age group who expe- rienced high-risk or suboptimal birth conditions (e.g., low birth weight, prenatal exposure to drugs and alcohol, neurological birth complications). The first of these issues has a long history in aggregate-level studies of crime, but the latter two rarely have been directly addressed in the crime trends literature. Little is known about the possible role of high-risk births and births to high-risk women in the 1970s on shaping contemporary crime trends. The types of birth quality indicators most relevant to Donohue and Levittâs (2001) argumentâparental age, marital and economic status, prenatal health, and such attributes as birth weightâhave been linked to a height- ened likelihood of involvement in delinquency and crime, but few studies have assessed their potential impact on aggregate crime trends. Research by OâBrien, Stockard, and Isaacson (1999) reveals strong evidence that a lagged measure of the percentage of births to unmarried mothers is signifi- cantly associated with contemporary trends in age-specific homicide arrest rates, which is consistent with the logic of Donohue and Levittâs (2001) argument (see also OâBrien and Stockard, 2002). I build on this work by more squarely evaluating whether recent city-level crime trends are associ- ated with various features of the cohorts born 15 to 19 years earlier in the metropolitan area in which these cities are located, including the percentage born to unmarried mothers and teenage mothers and the percentage clas- sified as low-weight births. Summary of Prior Work and the Scope of the Present Study There are many rich ideas about the factors that probably were respon- sible for the crime trends observed in the United States since 1980. These include changes in the quantity and quality of policing, incarceration, drug and alcohol use, drug markets, unemployment rates and real wages, the prevalence of firearms, domesticity, age structure, lagged birth cohort
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 145 features, and immigration. Although the extant empirical research has contributed much to the understanding of what happened with respect to crime trends during the past 25 years, as well as why it happened, there are numerous questions that either have not been addressed or have not been addressed sufficiently. One of the significant omissions from the exist- ing empirical literature is a simultaneous assessment of the many factors hypothesized to shape recent crime trends. The attempt here is to fill this gap by considering the effects on crime trends of each of the major factors emphasized in the literature and discussed above. DATA AND METHODS Units of Analysis and Sample As explained in the first chapter of the volume, the Committee on Law and Justice of the National Research Council (NRC) selected the units of analysis and defined the sampling universe for this study. The units of analy- sis chosen were large U.S. cities, and the original database included 240 cities with populations of 100,000 or more based on the 2000 census. In theory, the use of subnational units, like cities, provides a better chance of solving the contemporary crime trends puzzle than national- level studies. As Levitt (2001) points out, such an approach permits the estimation of panel time-series models, which can address a wider array of research questions and are much better suited for controlling for confound- ing temporal and spatial factors than national-level approaches. Cities also are a particularly sensible choice for studying crime trends given that most law enforcement agencies, and hence data on the volume of crime, are organized at the level of cities. And although decisions about crime policy often are influenced by state and county developments and politics, they are typically implemented by city personnel. Although cities are a sensible unit of analysis for studying crime trends, there are two major drawbacks associated with this choice. First, very little of the requisite data one might use to measure directly key explana- tory variables is collected for cities, or at least not on a regular basis. The original NRC city-level database supplied included only two time-varying city-level indicators as explanatory variables: police officers per capita and drug arrest rates. This was not an oversight. It is merely the reality of the current data infrastructure; many of the relevant time-varying indicators one might want for studying crime trends simply are not readily available for U.S. cities for most of the period under review. Given the importance of incorporating time-varying indicators in a study of crime trends and the desire to take a comprehensive approach, I therefore drew from a variety of sources to construct annual estimates that might be reasonably allocated
146 UNDERSTANDING CRIME TRENDS to describe city conditions. In some instances, potentially relevant mea- sures were available only for the counties, metropolitan areas, or states in which the NRC cities are located. It is arguably better to use county or Âmetropolitan-area data to estimate annual trends for cities than to use n Â othing or to use city-level data from decennial censuses and simply assume a linear trend between decennial periods to estimate these conditions, which could introduce an artificial temporal relationship with crime rate trends. Nevertheless, the unit discrepancy between the NRC primary sample units (cities) and the lowest level of geography for which many potentially key indicators are available (counties) is a drawback of the present work, and future studies should evaluate its implications. Second, one trade-off in attempting an inclusive assessment of contem- porary crime trends is that the requisite data needed to do so could not be located or were very incomplete for many of the cities included in the origi- nal NRC sampling frame. Consequently, several of the cities were excluded from the analysis. Some of the cities became incorporated government units only in the late 1970s, which limits some of the data elements that can be gathered for them, and many of these cities and other areas did not report crime or arrest data consistently during the study period, especially during the early 1980s. Overall, I was able to locate complete data for 151 of the 240 cities in the original NRC sample frame. About one-third of the sample attrition arose because of the inclusion of disaggregated homicide rates in the study, which were available on a consistent basis across the study period for only 205 of the 240 original cities. Much of the remaining attrition was due to the inclusion of key explanatory variables, especially indicators of drug market activity and drug use, which contain a substantial amount of missing data. I estimated models of contemporary crime trends for the full sample of 151 cities for which I could locate complete data, but the analysis reported below is based on a subsample of 114 of the cities with a population of 100,000 or more in 1980. As noted above, the NRC sample universe was defined as all large cities with populations of 100,000 or more based on the 2000 census. I modified this universe to use instead the 1980 census population counts to define large cities, retaining the 100,000 person mini- mum value. Doing so seemed sensible, given that the study focuses on crime trends from 1980 forward, since most of the extant city-level research on contemporary crime trends has used 1980 population counts to define large cities, and because using the 2000 census to define large cities would have resulted in a sample skewed toward newly emerging urban areas, especially in the western and southern regions of the United States (i.e., California and Texas), where recent population growth has been concentrated. Thus, the results presented below are based on the 114 cities in 1980 that had
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 147 populations of 100,000 or more and for which data on all variables could be located. Data and Measures Variable definitions, sources, unit of measurement, and summary statis- tics for all variables included in the study are listed in Table 5-1. Subscripts in the table identify variables measured in lagged versus contemporaneous form. The original NRC database included annual rates of Uniform Crime Reports (UCR) homicide, robbery, burglary, and motor vehicle theft for large cities as well as annual indicators of city drug arrest rates and police force size, and state-level data on annual levels of incarceration. In addition, various demographic and economic indicators (age and racial composition, population size, family structure, poverty, unemployment, inequality, etc.) from the decennial censuses of 1980, 1990, and 2000 were included. To facilitate a more comprehensive analysis of contemporary crime trends, I modified the NRC database in three ways. First, given that prior research has shown that changes in youth homicide rates and gun homicide rates were distinct from other homicide trends during the 1980s and 1990s (e.g., Blumstein and Rosenfeld, 1998), I added data on these forms of lethal violence from the Supplementary Homicide Reports. Second, because the census data on social and economic attributes included in the NRC data- base were available for only 3 of the 25 time points examined in the study, I drew from a variety of additional sources to construct annual indicators of these conditions. Finally, some of the conditions emphasized in the litera- ture as potentially important for understanding recent crime trends, such as the prevalence of firearms, levels of immigration, domesticity, alcohol con- sumption, birth cohort conditions, and wages received for legal work were not included in the NRC data, so I added measures of them to the data. As noted above, some of the factors emphasized in the literature as potentially relevant to shaping recent crime trends are not available for most cities, so in some cases the annual indicators used in the study describe conditions in the counties, metropolitan areas, or states in which the cities are located. â he wage data were deflated using the region-specific consumer price index for urban T consumers published by the Bureau of Labor Statistics. Because annual data on age and race composition are currently not available for U.S. cities, annual county population estimates were used to compute year-to-year changes in the number of blacks and the number of persons between ages 15-24 and persons 45 and older in the counties in which the NRC cities fall. These county growth rates were applied to the available decennial (1980, 1990, and 2000) city-level estimates of population by race and age to compute city-level intercensal (1981-1989, 1991-1999, 2001-2004) values for percentage black, percentage 15-24, and percentage 45 and older for the NRC cities.
TABLE 5-1â Description of Variables Included in Analysis of Recent Crime Trends (N = 114) 148 Overall Within-City Variable Variable Definition and Source Mean SD SD Dependent Variables â Homicide rate Homicides per 100,000 residents (UCR) 12.69 10.03 4.70 ââ Gun homicide rate Gun homicides per 100,000 residents (SHR) 7.78 7.29 3.63 ââ Nongun homicide rate Nongun homicides per 100,000 residents (SHR) 4.41 3.27 2.07 ââ Youth homicide rate Homicides involving a person ages 15-24 per 100,000 residents 22.23 20.71 14.04 ages 15-24 (SHR) ââ Adult homicide rate Homicides involving a person ages 25-44 per 100,000 residents 11.25 9.54 6.27 ages 25-44 (SHR) â Robbery rate Robberies per 100,000 residents (UCR) 395.57 301.66 138.98 â Burglary rate Burglaries per 100,000 housing units (UCR and Census Bureau) 4,144.75 1,951.18 1,410.78 â Motor vehicle theft rate Motor vehicle thefts per 100,000 motor vehicles (UCR and 807.70 1,056.05 471.88 Census Bureau) Explanatory Variables â Criminal justice factors â State stock incarceration ratet-1 Persons incarcerated in state prison per 100,0000 residents (BJS) 308.59 157.79 134.85 â State prison admission ratet-1 Persons admitted to state prison per 100,000 residents (BJS) 174.96 97.06 74.84 â State prison release ratet-1 Persons released from state prison per 100,000 residents (BJS) 158.46 94.35 74.17 â City police force sizet-1 Police officers per 100,000 residents (UCR) 207.67 76.50 23.62 â City public order and Number of arrests per 100,000 residents for weapons violations, 282.92 343.79 276.35 weapons arrest ratet-1 vandalism, prostitution, gambling, liquor laws, drunkenness, disorderly conduct, vagrancy, curfew violations, loitering, and suspicion (UCR) â City serious crime arrest Ratio of arrests for murder, robbery, burglary, and motor vehicle 14.32 6.39 3.38 certaintyt-1 theft to the number of murders, robberies, burglaries, and motor vehicle thefts known to the police (UCR) â Economic conditions â City job availability Ratio of jobs in city commuting zone to city residents ages 14.53 25.19 12.09 16-64 (BEA and Census Bureau)
â City unemployment rate % of civilian labor force unemployed (BLS) 6.33 2.55 1.58 â County average real wages Mean real annual wage across all industries (BEA) 19,358 3,313.59 1,790.81 â Guns, drugs and alcohol â County firearm prevalence % of suicides committed with a firearm (NCHS) 53.35 14.23 7.71 â City cocaine/heroin arrest rate Number of arrests per 100,000 residents for sale or possession 112.35 184.09 123.40 of cocaine/heroin (UCR) â City % cocaine arrests < 18 % of cocaine/heroin arrests attributed to persons under 18 (UCR) 7.72 11.78 10.67 â County cocaine mortality rate Cocaine-related deaths, per 100,000 county residents (NCHS) 1.04 1.76 1.57 â City alcohol-related traffic % of traffic fatalities that involved a drunk driver (NHTSA FARS) 35.93 16.26 14.53 fatalities â Demographic characteristics â State % married couple % of state households with a married couple (CPS) 38.96 3.33 2.09 households â State % cohabiting couple % of state households with two unmarried adults of opposite 3.37 1.20 .95 households sex who share living quarters (CPS) â City % ages 15-24 % of population ages 15-24 (Census Bureau and SEER) 16.96 3.82 2.19 â City % ages 45+ % of population ages 45 and older (Census Bureau and SEER) 29.47 4.46 2.33 â City population size Total population (Census Bureau and SEER) 39,2342 78,6873 58,192 â City % black % of population who are black (Census Bureau and SEER) 19.12 16.31 2.27 â MSA immigration rate New immigrants intending to reside in MSA per 100,000 MSA 171.51 173.23 58.50 residents (INS and Census Bureau) â MSA % of births to teenage % of births in MSA to women ages 15-19, lagged 15-19 years 15.85 3.69 1.75 woment-15-19 years (NCHS Natality Files) â MSA % of births to % of births in MSA to unmarried women, lagged 15-19 years 13.49 6.55 5.36 unmarried (NCHS Natality Files) woment-15-19 years â MSA % of births < 2500 % of babies born in MSA who weighed < 2,500 grams, lagged 7.34 1.09 .496 gramst-15-19 years 15-19 years (NCHS Natality Files) NOTE: BEA = Bureau of Economic Analysis; BJS = Bureau of Justice Statistics; BLS = Bureau of ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ Labor Statistics; FARS = Federal Accident Reporting System; INS = Immigration and Naturalization Service; MSA = metropolitan statistical areas; Nï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ CHS = National Center for Health Statistics; NHTSA = National Highway and Traffic Safety Administration; SEER = Surveillance Epidemiology and End Results; SHR = Supplementary Homicide Reports; ï¿½ ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ï¿½ 149 UCR = Uniform Crime Reporting Program.
150 UNDERSTANDING CRIME TRENDS Methods Given the research issues addressed here and various features of the data used to do so, the present study applies econometric panel modeling techniques to evaluate the effects of the factors outlined earlier on recent crime trends. A series of two-way fixed-effects panel models of crime rates are estimated and reported below. This specification includes fixed effects that control for stable unmeasured city attributes and temporal shocks that are shared across cities. Also, linear and quadratic trend variables are included in the models to account for unit-specific temporal shocks (Raphael and Winter-Ebmer, 2001; Worrall and Pratt, 2004). The annual crime data examined in the study exhibit significant serial autoÂcorrelation, so the models include as an explanatory variable lagged forms of the dependent variable (Beck and Katz, 1995). Finally, preliminary analyses of the data used here indicate the presence of substantial cross-sectional cor- relation in disturbances across cities. Failing to account for these features of the data can lead to invalid inferences, so in the models shown below I report panel-corrected standard errors, which allow the disturbances to be heteroskedastic and contemporaneously correlated across panels (Wilson and Butler, 2007). Before presenting results, it is important to highlight two issues that war- rant careful consideration while proceeding in a panel estimation of crime trends. First, although the issue of stationarity has been studied extensively with national-level crime data (e.g., Greenberg, 2001), it has largely been ignored in the literature on recent crime trends. In fact, only a few of the studies commonly included in overviews of the literature on recent crime trends even mentions this issue (see also Moody, 2007; Spelman, 2008). The typical subnational study of recent crime trends assumes stationarity in the variables and proceeds by estimating panel regression models in levels (for some exceptions, see McDowall, Loftin, and Wiersema, 2000; Moody and Marvell, 2005). If crime rates and the explanatory variables are stationary, then this is an appropriate way to proceed. However, if crime rates and/or the explanatory variables are nonstationary, the specter of spurious regression emerges and traditional panel model estimation strategies in levels may be inappropriate (see also Bushway and McDowall, 2006). One approach often applied when nonstationarity is suspected or found is to difference the variables, which can induce stationarity, and to estimate panel models on the transformed variables. Although easy enough to implement, this seem- â he T results presented are robust to alternative specifications. Models were also estimated without the unit-specific trends and with autocorrelation modeled as a nuisance parameter in lieu of the lagged dependent variable (e.g., Beck and Katz, 1995; Wolfers, 2006). The results of these supplementary analyses were substantively identical to those reported below.
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 151 ingly easy and quick fix may not necessarily be a wise decision or produce more valid results, because differencing masks the stable long-run levels relationships that may exist between variables (i.e., cointegration), which raises a whole host of additional modeling issues that can be critical for the inferences drawn (see Baumer and Rapach, 2007, for a review). Over- all, the preferred strategy would be to evaluate stationarity formally with appropriate tests (e.g., panel unit root tests that account for cross-sectional dependence), assess the implications of some of the imprecision that is likely to result from such tests (e.g., some units and some variables may exhibit stationarity while others do not), test for cointegration if necessary, and proceed accordingly with panel regression estimation based on the results of such tests. Spelmanâs (2008) recent analysis of state-level data on incarcera- tion and crime exemplifies the kind of analytical approach needed, though the fact that his scope in doing so in a full paper was limited to a single relationship with few explanatory variables underscores the complexity of the issues that need to be addressed in subsequent research. A second issue that deserves mention is simultaneity. Several of the factors thought to be instrumental in raising or lowering crime during the past two decades are also likely to be affected by crime rates. For example, increases in incarceration, police force size, and arrest rates may be impor- tant for determining subsequent crime levels, but probably also are to some large extent the consequent of rising crime rates (Greenberg, Kessler, and Logan, 1979; Levitt, 1996, 2002). Economic conditions (e.g., wages and unemployment rates), firearm prevalence, and levels of drug use also have been suspected of exhibiting a simultaneous relationship with crime rates (e.g., Fagan, 1990; Raphael and Winter-Ebmer, 2001; Rosenfeld, Baumer, and Messner, 2007). In short, many of the factors routinely mentioned as possible causes of recent crime trends also might plausibly be consequences of crime. If so, and if this possibility is not formally considered in the analytical approach to studying crime trends, the inferences drawn may be misleading. There are a variety of ways to address endogeneity concerns, including Granger causality and instrumental variables analysis. None of the available strategies is ideal or fully satisfactory under typical conditions, but address- ing this issue is critical for drawing more definitive conclusions about the factors associated with recent crime trends. Granger tests are useful for assessing whether simultaneous relationships are present, but it appears that the greatest strides in tackling the endogeneity problem in crime trends research will come in the form of identifying and incorporating valid and relevant instrumental variables that can help separate simultaneous effects (see also Spelman, 2008). This will not be an easy task, but a growing lit- erature has documented potentially useful instrumental variables that yield better estimates of the effects on crime rates of firearms (Kleck, Kovandzic,
152 UNDERSTANDING CRIME TRENDS and Schaffer, 2005; Rosenfeld, Baumer, and Messner, 2007), incarceration rates (e.g., Spelman, 2005), unemployment rates (Raphael and Winter- Ebmer, 2001), and police strength (e.g., Levitt, 2002). Additional efforts to locate alternative instruments for these and other crime predictors would be useful, and estimating models that attempt to account for endogeneity in key predictors should become standard practice in a comprehensive and systematic research agenda on crime trends. Given the space constraints of the present volume and my chosen focus of expanding the typical set of factors considered in crime trends, it is not possible to deal satisfactorily with these two important methodological issues. Examining stationarity and cointegration in a panel setting raises a series of complex issues that warrant detailed attention, discussion, and analysis of issues that have yet to be fully resolved in the literature. Some of the appropriate panel unit root tests (e.g., those that account for cross- sectional dependence) have only recently been developed in econometrics, and it is unclear how they should be applied and interpreted in common crime panel data settings, which are likely to contain some variables and some units that are stationary and others that are nonstationary. In short, addressing the issue of nonstationarity is not merely a Âmatter of differencing the data and seeing what happens. Although a research agenda on dealing with these issues in a panel setting is developing and beginning to sort through some of the relevant issues (Baumer and Rapach, 2007; Moody, 2007; Spelman, 2008), this work is still in early stages, and incorporating the necessary procedures in the present study would require a substantial expansion of scope. Addressing the prospects of cointegration is even more complex, especially in the context of relatively large multivariate models, such as those developed here. I therefore proceed âas usualâ for subnational crime studies and assume stationarity in the variables and esti- mate panel models in levels. Also, although there may be several instances of simultaneity in the models presented below, the general strategy adopted here is to follow the typical approach used in crime trends research and use lagged explanatory variables to minimize endogeneity concerns for attributes that theoretically are expected to have a delayed effect on crime. Although this is a common strategy, extensions of this research should assess its validity by incorporating instrumental variables as noted above. In summary, the analysis presented below adopts methods that have been used in most of the extant research on recent crime trends, yet it extends that work by incorporating, to a much greater degree, indicators of each of the major factors emphasized in theoretical and policy discussions. These results represent an assessment of the role that these factors played in shaping city crime trends since the early 1980s, as well as their relative contributions, and can be compared meaningfully with much of the extant research on recent crime trends. Nonetheless, like all research findings, the
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 153 validity of these results rest on the validity of the underlying assumptions. As noted above, drawing definitive conclusions from the findings reported below should await a more rigorous assessment of the assumptions of stationarity, lack of cointegration, and exogeneity, in particular. I therefore do not focus on illustrating specific results in detail, but rather emphasize the general implications of results based on an analysis that expands the typical specification employed and on comparing these results with existing research that is based on less comprehensive approaches. RESULTS Given the qualifications just noted, the analysis proceeds as follows. Results are first presented for two sets of regression models: the first set covers the full time frame covered in the NRC database (1980-2004) and includes all variables described above except immigration and lagged birth cohort features, which are not available for the full period; the second set adds these indicators and covers the shorter period (1984-2000) for which these variables are available. Results for a parallel set of models that con- sider alternative measures (i.e., incarceration flow measures), functional forms, and temporal interactions for state-level incarceration rates for the full period are also discussed. After providing an overview of the regression models, their overall implications for recent crime trends are compared with recent overviews based on previous research. Finally, the implications of the results for predicting subsequent crime rates are outlined. A Comprehensive Model of Recent Crime Trends Table 5-2 displays estimates for two-way fixed-effects panel regression models for the 114 cities included in the analysis for the full study period (1980-2004). Table 5-3 presents results for a set of models for a slightly shorter period (1984-2000) that are identical, save for the addition of the potentially important measures of immigration and lagged birth cohort attributes (unavailable for the full period). In each case, results are pre- sented in tabular form both for the four crime types included in the NRC database (total homicide, robbery, burglary, and motor vehicle theft) and for four disaggregated homicide measures (gun homicide, nongun homicide, youth homicide, and adult homicide). With some notable exceptions, the results for common variables in the two sets of models (i.e., the analyses for 1980-2004 presented in Table 5-2 and the expanded specification for 1984- 2000 presented in Table 5-3) are very similar, so I summarize the general conclusions that emerge across these analyses and highlight noteworthy differences when relevant rather than describing each table in detail sequen-
154 UNDERSTANDING CRIME TRENDS TABLE 5-2â Two-Way Fixed-Effects Models of Crime Rates, 1980-2004 (N = 114) Logged Logged Logged Logged Explanatory Variable Homicide Robbery Burglary MV Theft Once-lagged crime rate (logged) .065 .722* .775* .856* (.057) (.031) (.031) (.026) State % households with cohabiting .036 .021* .012* .006 couple (.022) (.007) (.005) (.006) State % households with married .007 .001 â.001 â.0001 couple (.007) (.002) (.002) (.003) City % ages 15-24 .026* â.0001 .002 .002 (.006) (.002) (.002) (.002) City % ages 45+ â.021* .002 .001 â.0002 (.007) (.002) (.001) (.002) City population size (logged) .118 .140* .024 .180* (.125) (.044) (.034) (.042) City % black .031* .007* .004* .004* (.005) (.001) (.001) (.002) County firearm prevalence .003* .001 .0001 â.0001 (.001) (.001) (.0003) (.0004) City cocaine/heroin arrest rate (logged) .026* .007* â.001 .003 (.010) (.003) (.002) (.003) County cocaine mortality rate (logged) .012 .007* .005* .005 (.010) (.003) (.002) (.004) City % cocaine/heroin arrests < 18 .003* .001* .002 .001* (.001) (.0003) (.002) (.0002) City % crashes with a drunk driver â.001 â.0001 â.0002 â.0002 (.001) (.002) (.0002) (.0002) City job availability .001 .001* â.0001 .0001 (.0006) (.0002) (.0002) (.0001) City unemployment rate .001 â.001 .002 â.011* (.007) (.003) (.003) (.003) County average real wages (logged) â.504* â.052 â.001 â.041 (.237) (.064) (.052) (.057) State stock incarceration ratetâ1 (logged) â.347* â.112* â.054* â.089* (.076) (.027) (.026) (.029) City police force sizetâ1 (logged) â.095 â.069* â.005 â.044 (.109) (.029) (.021) (.029) City public order and weapons arrest .102* .009 â.004 â.010 ratetâ1 (logged) (.035) (.010) (.006) (.009) City serious crime arrest certaintytâ1 â.099 â.047* â.014 .005 (logged) (.058) (.014) (.009) (.013) R-Squared .768 .965 .957 .950 continued
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 155 TABLE 5-2â Continued Logged Logged Logged Logged Gun Nongun Youth Adult Explanatory Variable Homicide Homicide Homicide Homicide Once-lagged crime rate .113* .085 .011 .026 (.053) (.051) (.050) (.052) State % households with cohabiting .067* .038 .072 .070 couple (.031) (.025) (.046) (.037) State % households with married .002 â.002 â.007 .005 couple (.009) (.009) (.016) (.012) City % ages 15-24 .027* .009 â.025 .025* (.008) (.008) (.014) (.012) City % ages 45+ â.030* â.009 â.016 â.010 (.008) (.007) (.012) (.009) City population size (logged) â.008 .139 .106 .332 (.149) (.173) (.256) (.213) City % black .027* .039* .033* .029* (.006) (.005) (.010) (.008) County firearm prevalence .006* â.0003 â.001 .002 (.002) (.002) (.003) (.002) City cocaine/heroin arrest rate (logged) .015 .037* .041* .025 (.012) (.014) (.024) (.020) County cocaine mortality rate (logged) .006 .007 .041 .006 (.015) (.015) (.024) (.020) City % cocaine arrests < 18 .003* .003* .008* â.001 (.001) (.001) (.003) (.002) City % crashes with a drunk driver â.002 .0002 â.002 â.002 (.001) (.001) (.002) (.002) City job availability .002 â.0002 .002 â.001 (.001) (.001) (.001) (.001) City unemployment rate .028* â.027* .012 .015 (.010) (.010) (.018) (.014) County average real wages (logged) .220 â1.06* â.789 â.515 (.289) (.268) (.481) (.407) State stock incarceration ratetâ1 (logged) â.348* â.270* â.512* â.420* (.104) (.094) (.157) (.145) City police force sizetâ1 (logged) â.239 â.129 â.319 â.330 (.151) (.147) (.298) (.216) City public order and weapons arrest .151* .050 .254* .098 ratetâ1 (logged) (.046) (.041) (.068) (.057) City serious crime arrest certaintytâ1 â.085 â.095 â.131 â.065 (logged) (.063) (.064) (.099) (.073) R-Squared .716 .614 .545 .578 *p < .05
156 UNDERSTANDING CRIME TRENDS TABLE 5-3â Two-Way Fixed-Effects Models of Crime Rates, 1984-2000 (N = 114) Logged Logged Logged Logged Explanatory Variable Homicide Robbery Burglary MV Theft MSA immigration rate â.055 .011 .038* .021 (.080) (.025) (.017) (.021) MSA % teenage birthstâ15-19 years .023* .001 â.003 .008 (.013) (.005) (.004) (.005) MSA % nonmarital birthstâ15-19 years â.002 â.001 â.001 â.002* (.003) (.001) (.001) (.001) MSA % low birth weight â.083* .019 .014 .016 birthstâ15-19 years (.040) (.015) (.013) (.018) Once-lagged crime rate (logged) .021 .661* .706* .755* (.079) (.049) (.051) (.037) State stock incarceration ratetâ1 (logged) â.373* â.154* â.099* â.151* (.116) (.040) (.041) (.039) City police force sizetâ1 (logged) â.073 â.114* â.056* â.122* (.133) (.035) (.028) (.038) City public order and weapons arrest .130* .018 â.006 â.024 ratetâ1 (logged) (.047) (.015) (.029) (.013) City serious crime arrest certaintytâ1 â.152* â.065* â.036* .005 (logged) (.071) (.017) (.012) (.017) City job availability .0002 .003 .0002 â.0001 (.0007) (.002) (.0002) (.0002) City unemployment rate .004 â.006 .003 â.007 (.010) (.004) (.003) (.004) County average real wages (logged) â.724* â.050 .057 .004 (.363) (.111) (.077) (.087) County firearm prevalence .003 .0001 â.0001 .0002 (.002) (.001) (.0004) (.0006) City cocaine/heroin arrest rate (logged) .030* .010* .002 .009* (.012) (.004) (.003) (.003) County cocaine mortality rate (logged) .010 .004 .003 .003 (.012) (.004) (.003) (.004) City % cocaine/heroin arrests < 18 .004* .001* .0002 .001 (.001) (.0004) (.0002) (.0002) City % crashes with a drunk driver â.001 .002 â.0002 â.0001 (.001) (.003) (.0002) (.0002) State % households with cohabiting .046* .022* .017* .007 couple (.023) (.009) (.007) (.008) State % households with married .008 â.001 â.001 .003 couple (.008) (.002) (.002) (.003) City % ages 15-24 .026* â.001 .0003 .002 (.007) (.001) (.002) (.003) City % ages 45+ â.027* .001 .001 â.002 (.007) (.002) (.001) (.002) City population size (logged) .128 .106 â.065 .725* (.228) (.081) (.055) (.069) City % black .041* .008* .007* .018* (.007) (.002) (.002) (.002) R-Squared .781 .966 .951 .951 continued
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 157 TABLE 5-3â Continued Logged Logged Logged Logged Gun Nongun Youth Adult Explanatory Variable Homicide Homicide Homicide Homicide MSA immigration rate â.053 â.035 .191 â.065 (.115) (.098) (.129) (.139) MSA % teenage birthstâ15-19 years .039* â.012 .090* â.004 (.018) (.022) (.034) (.026) MSA % nonmarital birthstâ15-19 years .002 â.001 .008 â.012* (.004) (.003) (.006) (.006) MSA % low birth weight â.149* â.026 â.127 â.089 birthstâ15-19 years (.061) (.070) (.104) (.091) Once-lagged crime rate .110* .055 â.011 â.023 (.071) (.070) (.066) (.071) State stock incarceration ratetâ1 (logged) â.562* â.197 â.668* â.694* (.150) (.120) (.261) (.187) City police force sizetâ1 (logged) â.205 â.250 â.539 â.376 (.195) (.177) (.349) (.291) City public order and weapons arrest .192* .036 .289* .069 ratetâ1 (logged) (.065) (.054) (.091) (.078) City serious crime arrest certaintytâ1 â.093 â.213* â.182 .019 (logged) (.088) (.091) (.144) (.108) City job availability .001 â.001 .001 â.003 (.001) (.001) (.001) (.002) City unemployment rate .036* â.036* .012 .010 (.013) (.015) (.018) (.017) County average real wages (logged) .214 â1.80* â1.63* â.983 (.444) (.347) (.784) (.651) County firearm prevalence .004 .001 â.006 .003 (.002) (.002) (.004) (.003) City cocaine/heroin arrest rate (logged) .017 .056* .064* .015 (.016) (.017) (.026) (.024) County cocaine mortality rate (logged) .016 .005 .035 .012 (.016) (.019) (.030) (.026) City % cocaine arrests < 18 .005* .002 .012* â.0004 (.002) (.002) (.003) (.002) City % crashes with a drunk driver â.002 â.0003 â.003 â.001 (.001) (.002) (.002) (.002) State % households with cohabiting .085* .029 .060 .079 couple (.034) (.032) (.051) (.042) State % households with married .005 â.008 â.024 â.009 couple (.010) (.011) (.017) (.014) City % ages 15-24 .026* .008 â.041* .034* (.010) (.008) (.018) (.016) City % ages 45+ â.019* â.009 â.009 â.026* (.009) (.010) (.017) (.010) City population size (logged) .071 â.287 â.084 .394 (.285) (.338) (.421) (.412) City % black .030* .037* .011* .047* (.008) (.009) (.014) (.012) R-Squared .744 .616 .596 .589 *p < .05
158 UNDERSTANDING CRIME TRENDS tially. Overall, there are five noteworthy patterns revealed in the findings displayed in these two tables. First, the coefficient on the once-lagged measure of stock incarceration is significant and negative in all but one of the crime models estimated (nongun homicide, 1984-2000). Consistent with other recent studies, the estimated elasticities for incarceration rates are higher for violence than property crimes and range from â.05 percent (burglary, Table 5-2) to â.67 (youth homicide, Table 5-3) across the crime types considered (see, e.g., Stemen, 2007). These results add to a large and growing body of evidence that reveals significant effects for incarceration during the period under consideration. Indeed, like other studies, incarceration rates emerge as particularly important here. Nevertheless, as noted earlier, there are several questions about these effects that have not received much attention in the literature, including whether stock and flow measures yield different effects and whether these patterns vary by scale or across time. To explore these issues in a preliminary way, I estimated two sets of additional models, building from the models shown in Table 5-2: (1) I substituted once-lagged indicators of rates of prison admissions and prison releases for the stock incarceration measure and (2) I assessed whether the effects of both the stock (i.e., overall incarceration rates) and flow (i.e., prison admission and release rates) measures of incarceration varied according to incarceration scale and over time by adding the relevant product terms (e.g., incarceration rate X incarceration rate; incarceration rate X year). The supplementary analysis results (not shown in tabular form due to space constraints) indicated that despite a very strong correlation between the two flow measures (r > .95), increases in state prison committals per 100,000 residents tend to reduce crime in the following year, while increases in the number of persons released from state prisons per 100,000 residents tend to increase crime in the next year. The estimated coefficients were not consistently significant across crime types, but the pair of coefficients for these two indicators of prison flow (i.e., admissions and releases) is statis- tically significant at conventional levels for four of the eight crime types considered (overall homicide, gun homicide, robbery, and burglary) and significant using a one-tailed test for two of the others (youth homicide and adult homicide). The supplementary analyses also revealed some significant variability in the estimated incarceration effects across time and at different levels of scale (results not shown), although the details varied across crime types. More specifically, the data examined here suggest that incarceration effects for lethal violence increased in magnitude during the 1980s and 1990s, while for the property crimes considered the evidence suggests significant dimin- ishing returns for incarceration over time (for robbery) or no significant change (for burglary and motor vehicle theft). A parallel story emerged for
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 159 the analysis of scale effects, with generally increasing elasticities for homi- cide as the scale of incarceration increases, declining elasticities for robbery as incarceration rates reach very high levels, and no significant scale effects for burglary and motor vehicle theft. Thus, depending on the crime type under investigation, there is evidence both for the notion that the crime reduction effects of growth in levels of incarceration have increased over time and with the scale of imprisonment, which is consistent with Spelmanâs (2006) research, and with âdiminishing returnsâ arguments over time and with enhanced scale, which is consistent with recent state-level research by Liedka, Piehl, and Useem (2006). Overall, these results suggest that rely- ing solely on the commonly used stock incarceration rate and assuming linearity and time invariance mask important information about the role of incarceration. Perhaps even the factor most often examined in crime trends researchâincarcerationâcalls for a more systematic and comprehensive research agenda that can sort out these details (see also Spelman, 2008). A second noteworthy finding that emerges from the regression analysis is that the policing variables yield inconsistent findings. There is no evidence in the data that cities in which the police focused more heavily on public order and weapons offenses (as measured by arrest rates) exhibited signifi- cantly lower crime rates. This finding is perhaps not surprising in light of recent evidence that such approaches had relatively small effects on crime trends in New York City, where they have evidently been implemented on a particularly grand scale (Messner et al., 2007; Rosenfeld, Fornango, and Baumer, 2005; Rosenfeld, Fornango, and Rengifo, 2007). However, the other two policing variables consideredâpolice force size and arrest cer- tainty for serious crimesâdo yield significant effects in the expected direc- tion, especially in the analysis restricted from 1984 to 2000 (Table 5-3). For this time frame, cities that increased their police forces experienced significantly greater declines in logged robbery, burglary, and motor vehicle theft. And areas in which the arrest certainty for serious crimes (i.e., the ratio of arrests for serious crimes to the number of serious crimes known to the police) was higher exhibited lower logged levels of homicide, robbery, and burglary. Third, overall the results point to a relatively limited role for changes in the economy. The indicator of job availability is not significantly associated with trends in any of the crimes considered, and unemployment and wage effects appear to be limited to lethal violence. Specifically, unemployment rates during this period were positively associated with gun homicide rates, and wages were negatively associated with nongun homicide. As elaborated below, these significant effects were fairly substantial in magnitude. But none of the economic variables considered here exerted significant effects on the crimes one would most expect to see such effects influence: property crimes. Perhaps indicators better able to capture economic changes in the
160 UNDERSTANDING CRIME TRENDS low-skilled sectors would fare better (e.g., Gould, Weinberg, and Mustard, 2002), but such data are not presently available over time for a large sample of cities. Fourth, the evidence for the effects of guns, drug activity, and alcohol consumption on recent crime trends is mixed. Trends in alcohol consump- tion do not appear to play a significant role in shaping recent crime trends, firearm prevalence is significantly associated with overall homicide but not other crimes, and the indicators of change in crack cocaine use and market activity exert significant and meaningful effects on recent crime trends, albeit in somewhat inconsistent ways across measures and crime types. The indicator of alcohol consumption employed is insignificant in all of the esti- mated models. Perhaps a more direct measure of alcohol consumption, or even an age-specific version of the one used in this study, would point to a different conclusion, but as measured and modeled in this study it appears that trends in alcohol consumption did not play a significant role in recent crime trends, net of other factors. A similar story can be told for firearm prevalence. Although the indicator used hereâthe fraction of suicides com- mitted with a firearmâhas been used extensively in prior research and is considered by many to be the gold standard for gauging geographic varia- tion in household gun ownership, its validity and reliability for tracking gun ownership trends has been critiqued (e.g., Kleck, 2004), and it is unclear how well it measures the stock of firearms available to would-be offenders. The indicators of illicit drug use and drug market conditions yield a more intuitive and substantively meaningful pattern of effects. The overall drug arrest rate for cocaine/heroin and the drug market age structure measure (as indicated by the percentage of persons arrested for possession or sale of cocaine/heroin who are under 18) yield the most consistently significant effects and are strongest for youth homicide, as expected on the basis of the underlying theoretical arguments. Finally, some of the demographic variables exert significant and inter- esting effects on crime trends. The indicator of the relative size of the black population exhibits a significant positive effect across all of the crime models. This reinforces a persistent finding in the literature that crime rates tend to be highest in cities with a high percentage of black residents (see, e.g., Land, McCall, and Cohen, 1990). There is much speculation about the reasons behind this association, but little convincing empirical evidence on the matter in aggregate crime studies. The age structure effects appear â Additional analyses (not shown in tabular form) revealed no significant evidence of an interaction between the indicators of the legitimate economy and indicators of drug use and drug market measures. In practice, it is very difficult to assess this argument with the avail- able aggregate-level data. But it is noteworthy that the observed effects of unemployment and wages show no evidence of being significantly conditioned by the drug indicators used in the study, or vice versa.
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 161 to be limited to homicide and in this instance are consistent with the idea that relatively larger cohorts of young persons are positively related and relatively larger cohorts of older persons are negatively related to vio- lence. The findings for the indicators of domesticity suggest no significant effects of marriage rates but significant effects of cohabitation rates on logged robbery, burglary, and homicide. Rising levels of cohabitation yield similar patterns in the models of adult homicide, although in this instance the Â coefficient does not quite reach conventional levels of statistical sig- nificance. The link between cohabitation and homicide is consistent with Blumstein and Rosenfeldâs (1998) domesticity argument. However, the significant association between trends in cohabitation and trends in rob- bery and burglary and the lack of a significant association in the models for adult homicide suggest that the cohabitation effects may reflect more general lifestyle patterns that raise the risk of victimization and offending, rather than increasing domesticity per se. It would be interesting to explore this further, especially in models of intimate partner homicide. Levels of immigration and lagged birth cohort conditions were available only from 1984 to 2000 (Table 5-3). The results displayed in Table 5-3 show no support for the idea that increasing flows of immigra- tion were significantly or inversely associated with aggregate-level crime rates between 1984 and 2000 (see also Butcher and Piehl, 1998). Even after trimming economic conditions from the model, which represent one of the pathways through which immigration has been posited to affect crime trends, no evidence of a significant negative immigration effect on crime trends was detected (not shown in tabular form). Perhaps addi- tional analyses that incorporate data on immigration through the early years of the 21st century (currently not available publicly) and also that adjust for the stock of foreign-born would yield different findings, but the conclusion supported here is that immigration had negligible effects on recent crime trends. The indicators of lagged nonmartial births and the lagged prevalence of low-birth-weight babies are not associated with contemporary crime trends in the expected positive direction; in fact, these variables exhibit significant negative effects in a few cases. However, the findings do show that recent city-level homicide trends are significantly influenced by the percentage of â ccording A to these results, cities situated in metropolitan areas with greater increases in immigration actually experienced significantly elevated rates of burglary during the period. However, a supplementary analysis (not shown) on a slightly shorter time frame (1984-1997) and with a measure of city-level rates of immigration (i.e., the number of immigrants intend- ing to live in the sampled citiesânot merely the metropolitan statistical areas in which they are locatedâper 100,000 city residents) does not yield such a pattern and, more importantly, affirms that the most consistent pattern is that immigration flows are not significantly associ- ated with recent crime trends.
162 UNDERSTANDING CRIME TRENDS the contemporary youthful cohort (i.e., persons ages 15-19) estimated to have been born to teenage mothers. This finding emerges as statistically significant for overall homicide and youth homicide, and it is strongest for the latter, as would be expected if the lagged teen birth prevalence indicator gauges differences in birth and childrearing conditions that yield conse- quences specific to the contemporary cohort defined by such conditions. It is important to acknowledge, however, that this finding also could reflect more contemporary family structure effects or other types of lagged social and economic conditions that are not considered here. In general, aside from cohort size, aggregate-level crime research has paid little attention to the possible role of the conditions under which contemporary populations were born or grew up, and the results shown in Table 5-3 suggest that this could be an important oversight. The Relative Contribution of the Factors Overall, what do these findings tell us about which factors contributed most to contemporary crime trends? As noted above, a much more rigorous analysis should be applied to the data before one can draw precise conclu- sions about the bigger picture, but to address this issue in a preliminary way and make general comparisons with recent overviews of the research, I used the results shown in Table 5-3, coupled with information about observed changes in crime rates and the explanatory variables, to compute the esti- mated percentage of the overall change in crime rates that can be attributed to each factor considered. I used these procedures to compute the relative contributions of the factors separately for the two major crime trend eras of the past two decades, defined here as 1984-1992 and 1993-2000. Thus, I first estimated the mean change in each of the crime variables and explana- tory variables across the 114 cities between 1984-1992 and 1993-2000, respectively. Using the coefficients shown in Table 5-3, which are based on models that also incorporate city and year fixed effects and city-specific time trends, I then calculated the expected or predicted change in the crime variables given the amount of observed change in each explanatory vari- able, and then divided it by the observed change in the crime variables to generate the fraction of the observed change that can be attributed to each factor. The end result is an estimate of the relative impact of each explana- tory variable on the observed change in each of the eight crime types for the two periods under consideration. In keeping with the scope of the present â he findings are very similar, and even somewhat stronger, if the youth homicide rate is T defined to match more precisely the lagged birth cohort measures (i.e., homicide rates for persons ages 15-19).
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 163 investigation, I emphasize here the general conclusions that emerge from this exercise, beginning with the 1980s. Although there has been a good deal of attention devoted to 1980s crime trends, to my knowledge there has not been a systematic assessment that breaks down given factors on the basis of their relative contributions, at least not in the same way seen in the literature on the 1990s crime decline (e.g., Levitt, 2004; Zimring, 2006). Most observers attribute the rise in youth gun violence, robbery, and some forms of auto theft during the 1980s largely to the emergence and proliferation of crack cocaine, and the results reported in Table 5-3 support that conclusion (e.g., Blumstein and ÂRosenfeld, 1998; Blumstein and Wallman, 2006a). Although the role of the three indicators of drug market activity and drug use vary across crime types, the results suggest that together they account for between 20 and 40 percent of the observed increases in overall homicide, gun homicide, and youth homicide and about 10 percent of the observed increase in Ârobbery rates. The results also point to the relevance of some factors that have not been given much weight in most discussions of crime trends during this period, however, such as the rise in cohabitation and changes in the preva- lence of births to teenage mothers in an earlier period. The results suggest that the rise in cohabitation levels across the 114 cities accounts for roughly 15-25 percent of the observed increase in lethal violence between 1984 and 1992. Also, during the 1980s, the percentage of young persons estimated to have been born to teenage mothers increased slightly, and the results show that this trend accounted for about 5-10 percent of the overall increase in homicide, especially youth homicide. According to the data used in this research, there were slight declines in the availability of jobs and increases in unemployment during the 1980s, but, aside from gun homicide, for which the rise in unemployment contributed to an estimated 10 percent of the increase, the economy appears to have had relatively little direct impact on 1980s crime trends, at least based on the measures and models employed in this study. Finally, consistent with other reports (e.g., Levitt, 2004), the analysis indicates that lethal violence would have increased even more had it not been for a substantial increase in levels of incarcera- tion and a considerable decline in the relative size of the youth population (i.e., the percentage ages 15-24). Incarceration also emerged as a primary contributor to the decline in burglary and adult homicide, accounting for more than half of the observed declines in both of these crimes (see also Rosenfeld, 1998). What about the widespread crime decline of the 1990s? Here, there is a clearer record of claims that have been made in the extant literature with respect to what mattered and what mattered most. In particular, Levitt (2004) has boldly outlined the four factors that mattered and the six that did not, and Zimring (2006) also has drawn fairly precise conclusions about
164 UNDERSTANDING CRIME TRENDS TABLE 5-4â Conclusions About Factors Associated with the 1990s Crime Decline A. Levitt (2004) Factors That Probably Mattered Quite A Bit Factors That Probably Did Not Matter Much Increases in incarceration rates Improving economic conditions Increases in police per capita Changes in policing focus Decline in crack Smaller youth cohorts 1970s abortion legalization B. Zimring (2006) Factors That Probably Mattered Quite A Bit Factors That Probably Did Not Matter Much Increases in incarceration rates 1970s abortion legalization Improving economic conditions Decline in crack (except youth violence) Smaller youth cohorts Increases in police per capita (except NYC) Regional cyclical factors Changes in policing focus (except NYC) C. The Present Study Factors That Probably Mattered Quite A Bit Factors That Probably Did Not Matter Much Increases in incarceration rates Decline in crack (10-35%) Changes in policing focus Improving economic conditions Smaller youth cohorts (10-30%) Changes in domesticity Decline in âlaggedâ teen births (10-35%) Larger adult cohorts (4-8%) Increases in police per capita (3-7%) NOTE: In Panel C, the percentages in parentheses represent a range across crime types of the estimated contribution of each factor to the observed crime declines. what did and did not matter. I summarize their conclusions in Panels A and B of Table 5-4. The conclusions displayed in the upper two panels are not derived from formal meta-analysesâsomething that does not seem pos- sible given the current shape of the literatureâbut rather in both cases the authors have culled from existing research the most pertinent evidence and provided educated overviews of what it says. I have argued in this chapter that taking a more comprehensive approach to measuring and modeling the factors thought to be associated with recent crime trends could yield different conclusions. In fact, this does appear to be the case. Panel C of Table 5-4 shows the conclusions supported by the present research, which incorporates a broader set of factors compared with previous studies. Overall, the results concur with the conclusions drawn by others about the likely importance of incarceration for the 1990s crime decline. The models reported in Table 5-3 suggest that the continued rise in incarceration during the 1990s accounted for 10 to 35 percent of the decline in crime
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 165 rates across crime types, with property crimes (i.e., robbery, burglary, and motor vehicle theft) defining the lower end of this range and lethal violence defining the higher end. The results also indicate, however, that the improving economy may have been more important during the 1990s than suggested by Levitt and Zimring. Although the effects are not uniform across crime types and, curiously, are evident only for lethal violence (and not property crime), the results indicate that the drop in unemployment during the 1990s can account for 10-15 percent of the decline in overall homicide and gun homicide. This coupled with the rise in real wages during the period explains as much as 30 percent of the observed decline in youth and nongun homicide rates. Contrary to Zimringâs review, the decline in the lagged prevalence of births to teenage mothers also seems to have made a substantively important difference, accounting for about 10 percent of the decline in overall homicide and approximately one-third of the decline in youth violence during the 1990s. This is less than the estimated 50 percent attributed to the rise in abortion by Donohue and Levitt (2001), but it is still significant and also more clearly specifies one of the mechanisms that might link a rise in abortion (or other actions that control fertility) to lower subsequent crime many years later. The findings reported here diverge somewhat from others with respect to age structure. Consistent with Levitt and contrary to Zimring, I found that changes in the relative size of youth cohorts do not appear to have made a big impact on the crime decline of the 1990s. However, the rise in the fraction of the population ages 45 and older emerges as a notable fac- tor for the observed declines in lethal violence, accounting for between 4 and 8 percent of the observed declines in homicide subtypes considered in the study. The role of older cohorts in shaping crime trends has rarely been explored in prior work, which is somewhat surprising given the dramatic changes in the relative size of this low-offending-rate group. Zimring (2006) does not give much weight to suggestions that increases in police force size were very consequential to the 1990s crime decline, except for perhaps New York City, where such increases were especially dramatic. Levitt (2004) also does not see enhancements to police force size as a major factor but does estimate that it probably accounted for about 5-6 percent of the decline. The results in Table 5-3 yield very similar esti- mates, indicating that about 3-7 percent of the observed decline in crime during the 1990s can be attributed directly to increased police forces, with property crimes defining the lower end and adult homicide and nongun homicide defining the upper end. The other factors examined do not appear to have played a major role in the 1990s crime decline, at least as I have measured and modeled them. One of theseâthe prevalence of crack cocaine use and market activityâdid reveal significant effects on many of the crime types considered, but the
166 UNDERSTANDING CRIME TRENDS indicators used did not decline substantially during the 1990s, rendering their ability to account for the observed changes in crime to be minimal, in the neighborhood of 1-2 percent. Some of the other factors, like changes in policing focus (e.g., arrest rates for public order and weapons offenses), levels of domesticity, immigration rates, alcohol consumption, and firearm prevalence may emerge as more relevant in analyses that better attend to issues of measurement error, unit measure mismatch, stationarity, and simultaneity, but for now I would conclude that their contributions to the 1990s crime decline do not appear to be substantial. Out-of-Sample Predictions Thus far I have focused on attempting to explain what happened with respect to recent crime trends. Another objective of the NRC crime trends workshop was to assess the capacity for existing models to provide fore- casts or predictions of subsequent crime rates beyond the period covered in the study (i.e., post-2004). Although it is unclear whether and how decent crime forecasts would be put to use, the idea of knowing what is coming is enticing. For instance, if it was known sooner in the mid-1980s what may have been on the horizon, perhaps law enforcement and other public policy agents could have responded in ways that could have reduced the dramatic rise in violence that occurred. Furthermore, in a time of tight local budgets, high levels of anxiety among the public, and claims makers who tend to sensationalize highly visible violent incidents, it might be useful to have some sense of whether crime is likely to continue the descent seen for most of the 1990s and beyond, whether it is likely to rise significantly, or whether it might maintain a steady state. Despite some courageous and sophisticated previous efforts, crime forecasting is highly undeveloped at the present time and, in part because of this, I chose to focus most of my efforts in this chapter on developing explanatory models of recent crime trends rather than forecasting crime beyond the study period. Nevertheless, part of the value in assessing the factors that contributed to recent crime trends lies in what these models say about the likely direction of crime beyond the period under study. So what do the models outlined above, which focus on trying to explain crime trends between 1980-2004, say about the path of crime in subsequent years? To address this question, I used the coefficients from Table 5-2, which includes city and year fixed effects and city-specific time trends, to esti- mate crime levels (rates of homicide, robbery, burglary, and auto theft) for 2005. I first obtained, or in some cases estimated, city-specific values for each explanatory variable included in the models shown in Table 5-2. Since the focus was on estimating levels of crime in 2005, for the one-year lagged variables in the model (i.e., police force size, incarceration rates,
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 167 public order and weapons arrest rates, and serious arrest certainty) the city- specific values reflect 2004 conditions, and for all other variables, which were measured contemporaneously, the city-specific values reflect 2005 conditions derived from additional data collection. I then multiplied these city-specific covariate observed values by their corresponding Â coefficients from the Â models shown in Table 5-2 to generate a predicted crime level for 2005. This process yielded predicted 2005 crime levels for each of the cities included in the study. I next obtained from the UCR the observed crime levels for the 50 largest of these cities and used the observed values to compute the percentage change between 2004 and 2005 in observed crime levels and the predicted percentage change between 2004 and 2005 using the predicted values for 2005 derived from the procedures just outlined. Table 5-5 summarizes the results of the exercise. The UCR data reveal that the 50 largest cities included in the study experienced an average (median) increase of about 3.79 percent in logged homicide rates between 2004 (the last year of the study period) and 2005. This figure represents the observed change in homicide over the one-year period. Applying the procedures outlined above, my analysis predicted an average (median) decrease of 2.74 percent over this one-year period based on the results of the panel crime regression model displayed in Table 5-2 (including city and year fixed effects and city time trends) and the city- s Â pecific values on the explanatory variables. For robbery, the observed change was a 1 percent increase, and the prediction was for a .40 percent decline. As Table 5-5 shows, the predicted and observed change values are closer for burglary and especially motor vehicle theft; in the latter case, the model predicted a slight decline (â.27 percent) that was very close to the observed change (â.34 percent). The gaps here are not large in absolute terms, but at least for homicide, robbery, and burglary they probably are unacceptable, given the high stakes associated with crime prediction. In each of these three cases, for instance, one would have predicted decreases TABLE 5-5â Predictions of 2005 Crime Levels for 50 Largest Cities from Regressions of Recent Crime Trends, 1980-2004 Homicide Robbery Burglary Auto Theft % Change 2004-2005 â Predicted â2.74 â.40 â.39 â.27 â Observed 3.79 1.02 .01 â.34 % Cities predicted in right direction 44 40 46 48 NOTE: Figures for predicted and observed change represent median values for the 50 largest cities in the sample for which this information could be computed.
168 UNDERSTANDING CRIME TRENDS in crime when, in fact, there were increases. Indeed, another way to look at forecasting is to ask more simply about predictions in the direction of crime changes. In other words, is crime going to increase or decrease next year? The bottom row of Table 5-5 presents results relevant to this question. The procedures outlined above yielded an accurate prediction compared to the observed change in 44 percent of the cities in the case of homicide, 40 per- cent for robbery, 46 percent for burglary, and 48 percent for motor vehicle theft. Overall, the model based predictions that emerge from my study do not appear to be very goodâone could do better in predicting the direction of changes in these crimes by flipping a coin. It is possible that 2005 was an aberrant year. When I replicated Table 5-5 predicting changes in crime for 2003-2004, the predicted and observed robbery and burglary rates were closer and the record for predicting the direction of crime changes improved dramatically, with the right direction predicted in two-thirds to three-quarters of the cities (results not shown). However, the gap between observed and predicted homicide rates in this supplementary analyses were no better, and when I repeated the process yet again for predicting changes in crime for 2002-2003, the results were mixed (not shown), with some estimates outperforming those shown in Table 5-5 and others doing less well. Maybe 2005 was aberrant in the degree of unexplained change, maybe one or more of the other years was, maybe the types of model-based predictions summarized here are not well suited for predicting crime, or maybe crime trends are not highly predictable, save for periods of major shifts, such as the 1990s, when a predicted decline in crime would have been a good bet (as it turns out) for many years. Overall, like crime trends research in general, the issue of forecasting crime is in the early stages of development and more work needs to be done to better understand the nuances of making future crime predictions, to outline the best approaches to take given the reality of existing data, and to define acceptable parameters of prediction performance. There are several reasons why the regression models used to predict crime in this study probably did not fare better, including measurement mismatches between the city-level outcome variables and, in some cases, state-level explanatory variables, other sources of measurement error, methodological limitations of the models used to generate the estimates (e.g., no attention to Â stationarity and little formal attention to endogeneity), and simpler things, like the assumption of linearity and temporal invariance in the estimated effects of the explanatory variables. With respect to the latter, as noted earlier, one of the more important factors in explaining recent crime trendsâincarcerationâappeared in this study to exhibit some nonlinear and temporally variable effects, and these were not captured in the predic- tion exercise summarized above (the pooled estimates were used instead). Other studies, too, have shown that incarceration effects could vary by
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 169 location (e.g., DeFina and Arvanites, 2002), another issue not explored here that could have introduced some error into the effort of predicting crime from the models estimated. Overall, the main story that emerges from the prediction exercise is that better predictions will probably require a more comprehensive modeling strategy that attends to the fundamental method- ological issues and various analytical nuances mentioned above. CONCLUSION In this chapter I have attempted a more comprehensive approach to the measurement and modeling of contemporary crime trends. Most of the extant literature has focused on a small number of potentially relevant factors, even though the theoretical literature highlights numerous other f Â actors that may have been important for shaping recent crime trends. Tak- ing a more comprehensive measurement approach required, in some cases, the use of state-, metropolitan area-, and county-level explanatory variables to explain city-level crime rates (cities were the unit of analysis chosen by the NRC). Future research should assess the implications of this type of unit mismatch by replicating the models developed here with crime rates measured at levels that match the explanatory variables. However, given the measures and methods used in this research, one can conclude that the findings generated from a more comprehensive approach affirm some of the results reported elsewhere with respect to incarceration, drug market conditions and drug use, lagged birth cohort conditions, and the economy, but also point to some additional factors that show relevance in shaping recent crime trends, including changes in levels of cohabitation, prisoner release rates, and the drug market age structure. As summarized in Table 5-5, the overall conclusions diverge somewhat from two widely cited reviews of prior work (Levitt, 2004; Zimring, 2006), a result that under- scores the importance of simultaneously considering the various factors emphasized in the theoretical literature rather than focusing on a select few factors. In essence, it seems important to take a comprehensive measure- ment approach to studying crime trends, especially if the goal is to assess which of the various factors hypothesized to shape recent trends actually matter (and if so, how much). I close by reiterating that, in some respects, it is highly premature to draw definitive conclusions from this or most previous work about the fac- tors that were mostly responsible for shaping recent crime trends. Although the public and the media are anxious to know what happened and why with respect to contemporary crime trends, the reality is that empirical literature on crime trends is in the early stages of development. Much more research is needed to develop answers in which there can be a high degree of con- fidence. I have argued that increasing the breadth of empirical studies to
170 UNDERSTANDING CRIME TRENDS incorporate measures of each of the major factors emphasized in theoretical discussion and policy debates is important, but there are other fundamental issues that require more attention as well before strong inference can be drawn. For one, the magnitude of the effects of criminal justice factors and other variables may be misestimated in most studies, including the pres- ent research, because of possible simultaneous relationships between these indicators and crime rates. One of the trade-offs of doing a comprehensive study is that it is very difficult to deal with these issues adequately, but doing so is an important next step. A second, perhaps even more fundamental issue, which needs to be addressed more systematically in research on contemporary crime trends before definitive conclusions can be drawn, concerns the time-series proper- ties of crime rates and the factors thought to be important for shaping crime trends. As noted at the outset, most of the extant research on contemporary crime trends and the present study assume stationarity in the variables and proceed by estimating panel regression models in levels. Although this may be an appropriate approach, if crime rates or the explanatory variables (or both) are nonstationary, the results that emerge could be spurious, which obviously has important implications about the most important factors in shaping recent crime trends. A variety of methodological issues need to be sorted out to satisfactorily address the time-series properties of variables considered, many of which currently are or have been explored (Baumer and Rapach, 2007; Moody, 2007; Spelman, 2008) but are not yet resolved. Pursuing such research more vigorously should better clarify the methods most appropriate for drawing valid inferences from panel studies of crime trends. In conclusion, this work takes some small but necessary steps toward addressing the two questions around which the NRC crime trends workshop has been organized: (1) Which factors were most important for explaining city-level crime trends observed between 1980-2004? (2) What might one reasonably expect for city crime levels in the years following this period? Tentative answers to these questions have been generated from a database supplied by the NRC, to which I added several measures. A more definitive resolution to these issues would be valuable, but achieving that goal will require a much larger and elaborated effort that retains the comprehensive approach to measurement and modeling outlined here, but also attends to fundamental methodological issues with respect to time-series estimation, simultaneity, model selection, spatial dependence, forecasting, and other issues. Although the NRC workshop on crime trends is a good start, much more research is needed to gain a full understanding of the past and future path of crime in the United States.
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 171 ACKNOWLEDGMENTS Thanks to Richard Rosenfeld for his insights on many of the issues discussed in this paper and to David Rapach for advice on many of the econometric issues that arise in a study of recent crime trends. Thanks also to Robert Fornango, Kim Martin, Michele Stacey, and Jessica Johnston for research assistance. Direct correspondence to Eric Baumer, College of Criminology and Criminal Justice, Florida State University, Tallahassee, FL 32306. REFERENCES Amato, Paul R., Alan Booth, David R. Johnson, and Stacy J. Rogers. (2007). Alone together: How marriage in America is changing. Cambridge, MA: Harvard University Press. Baumer, Eric P., and David Rapach. (2007). Spurious relationships in panel crime regressions. Unpublished manuscript. University of Missouri-St. Louis. Baumer, Eric P., Janet L. Lauritsen, Richard Rosenfeld, and Richard Wright. (1998). The influ- ence of crack cocaine on robbery, burglary, and homicide rates: A cross-city, longitudinal analysis. Journal of Research in Crime and Delinquency, 35, 316-340. Beck, Nathaniel, and Jonathan N. Katz. (1995). What to do (and not to do) with time-series cross-section data. American Political Science Review, 89, 634-647. Berk, Richard, Susan Sorenson, Douglas Wiebe, and Dawn Upchurch. (2003). The legalization of abortion and subsequent youth homicide: A time series analysis. Analyses of Social Issues and Public Policy, 3, 45-64. Blumstein, Alfred. (1995). Youth violence, guns, and the illicit-drug industry. Journal of Crimi- nal Law and Criminology, 86, 10-36. Blumstein, Alfred, and Richard Rosenfeld. (1998). Explaining recent trends in U.S. homicide rates. Journal of Criminal Law and Criminology, 88, 1175-1216. Blumstein, Alfred, and Joel Wallman. (2006a). The crime drop in America (revised ed.). New York: Cambridge University Press. Blumstein, Alfred, and Joel Wallman. (2006b). The crime drop and beyond. Annual Review of Law and Social Science, 2, 125-146. Braga, Anthony, David M. Kennedy, Elin J. Waring, and Anne M. Piehl. (2001). Problem- oriented policing, youth violence, and deterrence: An evaluation of Bostonâs Operation Ceasefire. Journal of Research in Crime and Delinquency, 38, 195-225. Brownridge, Douglas A. (2004). Understanding womenâs heightened risk of violence in com- mon-law unions. Violence Against Women, 10, 626-651. Bushway, Shawn D., and David McDowall. (2006). Here we go againâCan we learn anything from aggregate-level studies of policy interventions? Criminology & Public Policy, 5, 461-470. Butcher, Kristin F., and Anne M. Piehl. (1998). Cross-city evidence on the relationship between immigration and crime. Journal of Policy Analysis and Management, 17, 457-493. Casper, Lynne M., and Phillip N. Cohen. (2000). How does POSSLQ measure up? New his- torical estimates of cohabitation. Demography, 37, 237-245. Conklin, John E. (2003). Why crime rates fell. Boston: Allyn and Bacon. Cook, Philip J., and John H. Laub. (1998). The unprecedented epidemic in youth violence. Crime and Justice, 24, 27-64. Cook, Philip J., and Gary A. Zarkin. (1985). Crime and the business cycle. Journal of Legal Studies, 14, 115-128.
172 UNDERSTANDING CRIME TRENDS Cork, Daniel. (1999). Examining space-time interaction in city-level homicide data: Crack markets and the diffusion of guns among youth. Journal of Quantitative Criminology, 15, 379-406. Corman, Hope, and H. Naci Mocan. (2000). A time-series analysis of crime, deterrence, and drug abuse in New York City. American Economic Review, 90, 584-604. Donohue, John, and Steven Levitt. (2001). The impact of legalized abortion on crime. Quar- terly Journal of Economics, 116, 379-420. Donohue, John, and Steven Levitt. (2006). Measurement error, legalized abortion, and the decline in crime: A response to Foote and Goetz. Unpublished manuscript, Yale Univer- sity Law School. Dugan, Laura, Daniel Nagin, and Richard Rosenfeld. (1999). Explaining the decline in intimate partner homicide: The effects of changing domesticity, womenâs status, and domestic violence resources. Homicide Studies, 3, 187-214. Eck, John E., and Edward R. Maguire. (2006). Have changes in policing reduced violent crime? An assessment of the evidence. In Alfred Blumstein and Joel Wallman (Eds.), The crime drop in America (revised ed.). New York: Cambridge University Press. Fagan, Jeffrey. (1990). Intoxication and aggression. Crime and Justice, 13, 241-320. Fagan, Jeffrey, and Richard B. Freeman. (1999). Crime and work. Crime and Justice, 25, 225-290. Foote, Christopher, and Christopher Goetz. (2005). Testing economic hypotheses with state- level data: A comment on Donohue and Levitt. Federal Reserve Bank of Boston Working Paper No. 05-15, 1-18. Fox, James Allen. (2006). Demographics and U.S. homicide. In Alfred Blumstein and Joel Wallman (Eds.), The crime drop in America (revised ed.). New York: Cambridge Uni- versity Press. Fryer, Roland, Paul Heaton, Steven Levitt, and Kevin Murphy. (2006). Measuring the impact of crack cocaine. (NBER Working Paper #11318.) Cambridge, MA: National Bureau of Economic Research. Gallup-Black, Adria. (2005). Twenty years of rural and urban trends in family and intimate partner homicide: Does place matter? Homicide Studies, 9, 149-173. Gould, Eric D., Bruce. A. Weinberg, and David B. Mustard. (2002). Crime rates and local labor market opportunities in the United States, 1979-1997. Review of Economics and Statistics, 84, 45-61. Greenberg, David. (2001). Time series analysis of crime rates. Journal of Quantitative Crimi- nology, 17, 291-327. Greenberg, David F., Ronald C. Kessler, and Charles H. Logan. (1979). A panel model of crime rates and arrest rates. American Sociological Review, 44, 843-850. Grogger, Jeffrey. (1998). Market wages and youth crime. Journal of Labor Economics, 16, 756-792. Grogger, Jeffrey. (2006). An economic model of recent trends in violence. In Alfred Blumstein and Joel Wallman (Eds.), The crime drop in America (revised ed.). New York: Cambridge University Press. Grogger, Jeffrey, and Michael Willis. (2000). The emergence of crack cocaine and the rise in urban crime rates. Review of Economics and Statistics, 82, 519-529. Hagan, John, and Alberto Palloni. (1998). Immigration and crime in the United States. In N Â ational Research Council, The immigration debate. James P. Smith and Barry ÂEdmonston (Eds.). Panel on the Demographic and Economic Impacts of Immigration. Committee on Population and Committee on National Statistics, Commission on ÂBehavioral and Social Sciences and Education. Washington, DC: National Academy Press.
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 173 Johnson, Bruce D., Andrew Golub, and Eloise Dunlap. (2006). The rise and decline of hard drugs, drug markets, and violence in inner-city New York. In Alfred Blumstein and Joel Wallman (Eds.), The crime drop in America (revised ed.). New York: Cambridge University Press. Joyce, Ted. (2004). Further tests of abortion and crime. (NBER Working Paper #10564.) Cambridge, MA: National Bureau of Economic Research. Kennedy, David M., Anthony A. Braga, Anne M. Piehl, and Elin J. Waring. (2001). Reducing gun violence: The Boston Gun Projectâs Operation Ceasefire. National Institute of Justice Research Report. Washington, DC: U.S. Department of Justice. Kleck, Gary. (2004). Measures of gun ownership levels for macro-level crime and violence research. Journal of Research in Crime and Delinquency, 41, 3-36. Kleck, Gary, Tomislav Kovandzic, and Mark E. Schaffer. (2005). Gun prevalence, homicide rates and causality: A GMM approach to endogeneity bias. (CEPR Discussion Paper #5357.) London: Center for Economic Policy Research. Available: http://www.cepr. org/pubs/new-dps/dplist.asp?dpno=5357 [accessed August 2008]. Kovandzic, Tomislav V., Thomas B. Marvell, Lynn M. Vieraitis, and Carlisle E. Moody. (2004). When prisoners get out: The impact of prison releases on homicide rates, 1975- 1999. Criminal Justice Policy Review, 15, 212-228. Kubrin, Charis E., and Jerald Herting. (2003). Neighborhood correlates of homicide trends: An analysis using growth-curve modeling. Sociological Quarterly, 44, 329-350. LaFree, Gary. (1998). Losing legitimacy: Street crime and the decline of institutions in ÂAmerica. Boulder, CO: Westview. Land, Kenneth C., Patricia L. McCall, and Lawrence E. Cohen. (1990). Structural covariates of homicide rates: Are there any invariances across time and social space? American Journal of Sociology, 95, 922-963. Lattimore, Pamela K., K. Jack Riley, James Trudeau, Jordan Leiter, and Steven Edwards. (1997). A study of homicide in eight U.S. cities: An NIJ intramural research project. (National Institute of Justice Research in Brief, November.) Washington, DC: U.S. De- partment of Justice. Levitt, Steven D. (1996). The effect of prison population size on crime rates: Evidence from prison overcrowding litigation. Quarterly Journal of Economics, 111, 319-352. Levitt, Steven D. (1997). Using electoral cycles in police hiring to estimate the effect of police on crime. American Economic Review, 8, 280-290. Levitt, Steven D. (2001). Alternative strategies for identifying the link between unemployment and crime. Journal of Quantitative Criminology, 17, 377-390. Levitt, Steven D. (2002). Using electoral cycles in police hiring to estimate the effects of police on crime: Reply. American Economic Review, 92, 1244-1250. Levitt, Steven D. (2004). Understanding why crime fell in the 1990s: Four factors that explain the decline and six that do not. Journal of Economic Perspectives, 18, 163-190. Liedka, Raymond V., Anne Morrison Piehl, and Bert Useem. (2006). The crime control effect of incarceration: Does scale matter? Criminology and Public Policy, 5, 245-276. Lott, John. (1998). More guns, less crime: Understanding crime and gun-control laws. Chi- cago: University of Chicago Press. MacDonald, John M. (2002). The effectiveness of community policing in reducing urban violence. Crime and Delinquency, 48, 592-618. Martinez, Ramiro, and Matthew T. Lee. (2000). On immigration and crime. In Gary LaFree (Ed.), The nature of crime: Continuity and change (vol. 1, pp. 485-524). ÂWashington, DC: National Institute of Justice. Available: http://www.nicic.org/Library/016702 [ Â accessed August 2008]. Marvell, Thomas B., and Carlisle E. Moody. (1994). Prison population growth and crime reduction. Journal of Quantitative Criminology, 10, 109-140.
174 UNDERSTANDING CRIME TRENDS Marvell, Thomas B., and Carlisle E. Moody. (1996). Specification problems, police levels, and crime rates. Criminology, 34, 609-646. Marvell, Thomas B., and Carlisle E. Moody. (1997). The impact of prison growth on homi- cide. Homicide Studies, 1, 215-233. McDowall, David, Colin Loftin, and Brian Wiersema. (2000). The impact of youth curfew laws on juvenile crime rates. Crime and Delinquency,Â 46, 76-91. Messner, Steven F., Glenn D. Deane, Luc Anselin, and Benjamin Pearson-Nelson. (2005). Locating the vanguard in rising and falling homicide rates across U.S. cities. Criminol- ogy, 43, 661-696. Messner, Steven F., Sandro Galea, Kenneth J. Tardiff, Melissa Tracy, Angela Bucciarelli, Tinka Markham Piper, Victoria Frye, and David Vlahov. (2007). Policing, drugs, and the homi- cide decline in the 1990s. Criminology, 45, 385-414. Moody, Carlisle E. (2007). Is crime a random walk? Unpublished manuscript, College of William and Mary. Moody, Carlisle E., and Thomas B. Marvell. (2005). Guns and crime. Southern Economic Journal, 71, 720-736. OâBrien, Robert M., and Jean Stockard. (2005). Variations in age-specific homicide death rates: A cohort explanation for changes in the age distribution of homicide deaths. Social Science Research, 31, 124-150. OâBrien, Robert M., Jean Stockard, and Lynne Isaacson. (1999). The enduring effects of cohort characteristics on age-specific homicide rates, 1960-1995. American Journal of Sociology, 104, 1061-1095. Ousey, Graham C., and Matthew R. Lee. (2002). Examining the conditional nature of the illicit drug market-homicide relationship: A partial test of the theory of contingent causa- tion. Criminology, 40, 73-102. Ousey, Graham C., and Matthew R. Lee. (2004). Investigating the connections between race, illicit drug markets, and lethal violence, 1984-1997. Journal of Research in Crime and Delinquency, 41, 352-383. Parker, Robert Nash, and Randi S. Cartmill. (1998). Alcohol and homicide in the United States 1934-1995: Or one reason why U.S. rates of violence may be going down. Journal of Criminal Law and Criminology, 88, 1369-1398. Parker, Robert Nash, and Linda-Anne Rebhun. (1995). Alcohol and homicide: A deadly com- bination of two American traditions. Albany: State University of New York Press. Phillips, Julie A. (2006). The relationship between age structure and homicide rates in the United States, 1970-1999. Journal of Research in Crime and Delinquency, 43, 230-260. Phillips, Julie A., and David F. Greenberg. (2008). A comparison of methods for analyzing criminological panel data. Journal of Quantitative Criminology, 24, 51-72. Piehl, Anne M., Suzanne J. Cooper, Anthony A. Braga, and David M. Kennedy. (2003). Testing for structural breaks in the evaluation of programs. Review of Economics and Statistics, 85, 550-558. Police Executive Research Forum. (2006). Chief concerns: A gathering stormâviolent crime in America. Washington, DC: Police Executive Research Forum. Press, Eyal. (2006, December 3). Do immigrants make us safer? The New York Times. Raphael, Stephen, and Rudolf Winter-Ebmer. (2001). Identifying the effect of unemployment on crime. Journal of Law and Economics, 44, 259-283. Reid, Lesley W., Harald E. Weiss, Robert M. Adelman, and Charles Jaret. (2005). The i Âmmigration-crime relationship: Evidence across U.S. metropolitan areas. Social Science Research, 34, 757-780. Rosenfeld, Richard. (1997). Changing relationships between men and women: A note on the decline in intimate partner homicide. Homicide Studies, 1, 72-83.
EMPIRICAL ASSESSMENT OF THE CONTEMPORARY CRIME TRENDS PUZZLE 175 Rosenfeld, Richard. (2004, February). The case of the unsolved crime decline. Scientific American, 82-89. Rosenfeld, Richard. (2006). Patterns in adult homicide: 1980-1995. In Alfred Blumstein and Joel Wallman (Eds.), The crime drop in America (revised ed.). New York: Cambridge University Press. Rosenfeld, Richard, and Robert Fornango. (2007). The impact of economic conditions on robbery and property crime: The role of consumer sentiment. Criminology, 45(4), 735-769. Rosenfeld, Richard, Robert Fornango, and Eric Baumer. (2005). Did Ceasefire, Compstat, and Exile reduce homicide? Criminology and Public Policy, 4, 419-450. Rosenfeld, Richard, Joel Wallman, and Robert Fornango. (2005). The contribution of ex-prisoners to crime rates. In Jeremy Travis and Christy Visher (Eds.), Prisoner reentry and crime in America (pp. 80-104). New York: Cambridge University Press. Rosenfeld, Richard, Eric Baumer, and Steven F. Messner. (2007). Social trust, firearm preva- lence, and homicide. Annals of Epidemiology, 17, 119-125. Rosenfeld, Richard, Robert Fornango, and Andres Rengifo. (2007). The impact of order- m Â aintenance policing on New York City robbery and homicide rates: 1988-2001. Crimi- nology, 45, 355-383. Sampson, Robert J. (2006, March 11). Open doors donât invite criminals. The New York Times. Sampson, Robert J. (2008). Rethinking crime and immigration. Contexts, (Winter), 28-33. Sampson, Robert J., and Jacqueline Cohen. (1988). Deterrent effects of the police on crime: A replication and theoretical extension. Law and Society Review, 22, 163-189. Sampson, Robert J., and John H. Laub. (1993). Crime in the making: Pathways and turning points through life. Cambridge, MA: Harvard University Press. Sampson, Robert J., and William Julius Wilson. (1995). Toward a theory of race, crime and urban inequality. In John Hagan and Ruth D. Peterson (Eds.), Crime and inequality (pp. 37-54). Stanford, CA: Stanford University Press. Sampson, Robert J., Jeffrey D. Morenoff, and Stephen Raudenbush. (2005). Social anatomy of racial and ethnic disparities in violence. American Journal of Public Health, 95, 224-232. Shackelford, Todd K. (2005). Partner killing by men in cohabiting and marital relationships. Journal of Interpersonal Violence, 20, 1310-1324. Simanski, John. (2005). Mapping trends in U.S. legal immigration: 1980-2003. Working paper, Homeland Security, Office of Immigration Statistics. Sorenson, Susan, Douglas Wiebe, and Richard Berk. (2002). Legalized abortion and the homicide of young children: An empirical investigation. Analyses of Social Issues and Public Policy, 2, 239-256. Spelman, W. (2000). The limited importance of prison expansion. In Alfred Blumstein and Joel Wallman (Eds.), The crime drop in America. Cambridge: Cambridge University Press. Spelman, William. (2005). Jobs or jails? The crime drop in Texas. Journal of Policy Analysis and Management, 24, 133-165. Spelman, William. (2006). The limited importance of prison expansion. In Alfred Blumstein and Joel Wallman (Eds.), The crime drop in America (revised ed.). New York: Cambridge University Press. Spelman, William. (2008). Specifying the relationship between crime and prisons. Journal of Quantitative Criminology, 24, 149-178. Stemen, Donald. (2007). Reconsidering incarceration: New directions for reducing crime. New York: Vera Institute.
176 UNDERSTANDING CRIME TRENDS Sykes, Bryan L., Dominik Hangartner, and Earl Hathaway. (2007). Fertility and the Âabortion- crime debate. Unpublished manuscript, Department of Sociology, University of ÂCalifornia, Berkeley. Tierney, John. (2007). The crime mystery. The New York Times, February 16. Travis, Jeremy. (1998). Symposium: Why is crime decreasing? Foreword. The Journal of Criminal Law and Criminology, 88(4), 1173-1174. Travis, Jeremy, and Christy Visher. (2005). Prisoner reentry and crime in America. New York: Cambridge University Press. Travis, Jeremy, and Michelle Waul. (2002). Reflections on the crime decline: Lessons for the future? Washington, DC: Urban Institute Justice Policy Center. Wilson, Sven E., and Daniel M. Butler. (2007). A lot more to do: The sensitivity of time- series cross-section analyses to simple alternative specifications. Political Analysis, 15, 101-123. Wolfers, Justin. (2006). Did unilateral divorce laws raise divorce rates? A reconciliation and new results. American Economic Review, 96, 1802-1820. Worrall, John, and Travis Pratt. (2004). Estimation issues associated with time-series-cross- section analysis in criminology. Western Criminology Review, 5, 35-49. Zimring, Franklin E. (2006). The great American crime decline. New York: Oxford University Press. Zimring, Franklin E., and Gordon Hawkins. (1973). Deterrence: The legal threat in crime control. Chicago: University of Chicago Press. Zimring, Franklin, and Gordon Hawkins. (1997). Crime is not the problem: Lethal violence in America. New York: Oxford University Press.