Download The Impact of Economic Adversity, Inflation, and Incarceration on Crime Rates and more Study notes Literature in PDF only on Docsity!
The author(s) shown below used Federal funds provided by the U.S.
Department of Justice and prepared the following final report:
Document Title: Expanding the Scope of Research on Recent
Crime Trends
Author: Eric P. Baumer, Richard Rosenfeld, Kevin T.
Wolff
Document No.: 240204
Date Received: November 2012
Award Number: 2008-IJ-CX-
This report has not been published by the U.S. Department of Justice.
To provide better customer service, NCJRS has made this Federally-
funded grant final report available electronically in addition to
traditional paper copies.
Opinions or points of view expressed are those
of the author(s) and do not necessarily reflect
the official position or policies of the U.S.
Department of Justice.
Report Title: Expanding the Scope of Research on Recent Crime Trends
Award Number: 2008-IJ-CX-
Authors: Eric P. Baumer, Richard Rosenfeld, Kevin T. Wolff
Abstract
Statement of Purpose While there is a burgeoning research literature on crime trends, much of the extant research has adopted a relatively narrow approach, efforts across studies are highly variable, and the overall conclusions that can be drawn are ambiguous. In our judgment, one reason for this state of affairs is that the current data infrastructure that supports crime trends research is incomplete and scattered, yielding redundant efforts and highly inconsistent approaches. The primary purpose of this project was to enhance the data infrastructure by compiling in a centralized location the most commonly referenced datasets and measures. An ancillary objective was to illustrate the utility of the resulting data archive. We do so by considering three substantive research issues: (1) a uniform set of analyses across states, counties, and cities; (2) an assessment of the conditional effects of economic conditions on recent crime trends; and (3) an expanded analysis of the effects of key criminal justice attributes (e.g., the nature of policing, age- and crime-specific imprisonment rates) on recent crime trends that have not been considered extensively in prior research.
Methods The specific samples, time frames, and measures employed vary somewhat across the three substantive issues addressed, but our general analytical strategy in addressing these issues is to construct when possible from the Crime Trends Data Archive (CTDA) produced in the project a panel database with requisite measures centered on the following time points: 1980, 1985, 1990, 1995, 2000, 2005, and 2010. This approach marshals the strength of a pooled cross-sectional design, while also avoiding the significant data imputation that is needed to support panel analyses of annual time periods for sub-national geographic units. The three sets of empirical analyses reported in the project include models of overall homicide, non-lethal violence (robbery and aggravated assault), and non-violent property crime (burglary, motor vehicle theft, and larceny). We estimate a series of two- way fixed-effects panel models of crime rates that include fixed effects that control for stable unmeasured city attributes and temporal shocks that are shared across cities.
Results Our uniform empirical analysis across units of analysis revealed that minimalist specifications can yield misleading conclusions. It also revealed that age structure and divorce rates are robust predictors of crime rates, with higher crime in areas with a larger percentage of persons aged 15- and where divorce rates are higher. Additionally, the results reaffirm findings shown in other work on non-violent property crime by showing that incarceration rates tend to yield lower crime rates, but at a diminishing rate as incarceration reaches very high levels. Our analysis of conditional economic effects pointed to a tendency of “objective criminal justice risk” to lessen the criminogenic consequences of elevated unemployment rates and depressed wages. Another intriguing pattern that emerged is that the estimated adverse effects of unemployment and wages on non-lethal crime (both violent and property) are weaker in the face of elevated levels of income maintenance payments (i.e., SSI, Snap, family assistance). Finally, our expanded analysis of criminal justice factors showed that, at least for non-lethal violence and non-violent property crime, arrest certainty for these crimes is a robust predictor of lower crime rates. We also found that non-violent
This document is a research report submitted to the U.S. Department of Justice. This report has not^1 been published by the Department. Opinions or points of view expressed are those of the author(s)
Table of Contents
Executive Summary pp. 4-
Technical Report Statement of Problem and Relation to Existing Literature pp. 10- Rationale “Enhancing the Data Infrastructure” pp. 12-
“Clarifying and Expanding the Scope of Existing Research” pp. 18- A uniform empirical specification across states, counties, and cities. An assessment of conditional economic effects on recent crime trends Expanding the typical set of criminal justice variables considered Methods pp. 26-
Results A uniform empirical specification across states, counties, and cities. pp. 35-
An assessment of conditional economic effects on recent crime trends pp. 47-
Expanding the typical set of criminal justice variables considered pp. 54-
Conclusions pp. 60-
References pp. 62-
Dissemination of Research Findings pp. 69-
Appendix A. The Crime Trends Data Archive (CTDA) pp. 70-
This document is a research report submitted to the U.S. Department of Justice. This report has not^3 been published by the Department. Opinions or points of view expressed are those of the author(s)
Executive Summary
Statement of Purpose This project focused on addressing two general issues, as outlined in the original proposal: (a) enhancing the data infrastructure available to study recent American crime trends; and (b) clarifying and expanding the scope of empirical analysis directed at describing and explaining recent crime trends. Both objectives are motivated by the relatively strong and growing interest among policy makers, the media, criminal justice practitioners, and the general public in the properties and predictors of crime trends across America.
Rationale Enhancing the Data Infrastructure Notwithstanding the extraordinarily useful features of the NACJD, which contains many pertinent data elements that will be of interest to researchers who wish to study recent crime trends, the existing data infrastructure is limited in two notable ways: (1) it is incomplete and/or somewhat decentralized; and (2) it contains pieces of the puzzle but little shared work-product about how those pieces might be and often are fitted together. Before elaborating on the first issue, it is important to acknowledge that the NACJD and other data archives bring together an enormous volume of data, much of which is pertinent to research on recent crime trends. Indeed, without the NACJD it would be nearly impossible to access in short-order the various components needed to sufficiently examine crime trends across multiple units, such as states, counties, or cities. Nevertheless, there are several data resources that are relevant to studying crime trends that do not fall within the normal confines of the NACJD. This includes data on social, economic, and demographic attributes from the decennial census and the annual American Community Survey (ACS), data on annual unemployment rates from the Bureau of Labor Statistics (BLS), data on wages and other economic indicators from the Bureau of Economic Analysis (BEA). Other pertinent data sources (e.g., the National Corrections Reporting Program [NCRP]) are included in NACJS, but not in forms that are easily modified to support their use in studies of crime trends. Finally, some of the available resources housed at the NACJD possess notable limitations for studying contemporary crime trends. For example, though the NACJD produced county-level crime and arrest data avoid many of the issues that emerge when researchers estimate county data from NACJD (or FBI) agency-level data (see Maltz and Targonski, 2002, 2004), the current county-level holdings at NACJD apply divergent imputation procedures from 1994 onward, introducing an important break from prior years that can be problematic for studies of crime trends that span before and after this period. While there is a need to enhance the data infrastructure for studying crime trends by broadening the data that are available in a centralized location, an even more pressing matter is to develop an archive of the substantial data processing and manipulation required to put the available data to work. Studying crime trends is somewhat different than many other research endeavors. Rather than drawing on a single survey or a few major data sources that encompass one or perhaps a small handful of data “waves,” a typical study of recent crime trends entails the combination of a very large number of distinct data sources, and often in each case ten, twenty, or perhaps even more “temporal installments” of those sources. The nature of the task at hand means that simply providing an archive of raw data with machine readable “setup” files will help, but this represents a relatively small portion of the overall effort involved in generating the data needed to study crime trends. The point being made here is not that the remainder of the work needed to do a crime trends study should fall on the NACJD, but rather that a highly useful archive would not only house a wide array of data but also would document how such data are combined and analyzed to generate meaningful information about crime trends. This is the general direction in which data archiving has
This document is a research report submitted to the U.S. Department of Justice. This report has not^4 been published by the Department. Opinions or points of view expressed are those of the author(s)
In summary, while there are no strong reasons at the present time to preference a priori a particular unit of analysis or analytical strategy for studying crime trends, it would be useful to know the empirical implications of using different units of analysis and different approaches, something that cannot be deciphered easily from existing research. Building on comparable approaches to cross-sectional crime research (e.g., Land, McCall, and Cohen, 1990), one of the key sets of analysis presented below is directed at advancing the literature by applying a uniform set of empirical specifications and procedures across multiple units of analysis. We specifically address whether identical empirical specifications applied to data for a given time period yields comparable results when applied to American states, counties, and cities. The objective of this portion of our work is modest—does the same specification yield comparable findings across cities, counties, and states?— but we consider an important step in clarifying the existing research on crime trends.
Expanding analyses of economic factors. The project also highlights a few more specific topics, including the potential conditional effects of economic factors. Specifically, we examine whether two of the most common indicators of economic adversity in crime trends research–high unemployment rates and depressed wages -- exhibit effects on crime rates that are contingent on other factors, including prevailing levels of inflation, the extent of unemployment insurance and other income supportive benefits dispensed, police force size, and incarceration rates. Adverse economic conditions have been linked to elevated crime rates through a variety of theoretical frameworks, encompassing arguments about cost-benefit assessments, heightened stress and anxiety, and shifting routine activities among others. Drawing on this literature, the accumulated body of evidence on the relationship between crime rates and adverse economic conditions (e.g., rising unemployment and falling wages) suggests that sometimes significant downturns in the economy yield an increase in crime and sometimes they do not (e.g., Bushway, Cook, and Phillips, 2010; Cook and Zarkin, 1985; Chiricos and DeLone, 1992; Smith, Devine and Sheley, 1992). The mixed or ambiguous empirical literature on macro-economic conditions and crime rates sometimes is written-off as a function of empirical misspecification (Greenberg, 2001; Raphael and Winter-Ebmer, 2001). However, another possibility is that adverse economic circumstances yield increases in crime (and good times yield decreases in crime) only under certain conditions. In essence, the idea that increasing economic adversity should simply yield a linear increase in crime rates is unlikely to capture the full range of behavioral realities that may be observed. Most prior research has focused on the effects of adverse economic conditions on crime rates, irrespective of the broader context in which those conditions arise or play out. Using the data generated for this NIJ DRP project, we expand prior efforts by evaluating the influence on crime rates of two common indicators of economic adversity – rising unemployment and declining wages. More specifically, we examine both the overall influence of these conditions on crime and the degree to which their effects on crime may differ depending on other factors. As elaborated in the full report, we specifically evaluate: whether inflation levels moderate the effects on crime of unemployment rates and wages; whether the effects on crime rates of unemployment rates and wages are moderated by levels of spending on unemployment insurance and income maintenance; and whether the effects on crime rates of unemployment rates and wages are conditioned by changes in police force size and incarceration rates.
Expanding the typical set of criminal justice variables considered. The two factors that perhaps have received the most attention in public discourse on recent crime trends, and especially the 1990s crime decline, are changes in policing and incarceration. Each of these factors, and criminal justice actions more generally, has been linked to crime trends mainly through their theoretical capacity to serve an incapacitation or deterrent function. Prior research on crime trends often includes some indicator of criminal justice activity, but the specific factor(s) included vary across studies. More important from our standpoint, though, is that the extant research tends to approach criminal justice factors from a relatively narrow vantage point. In the context of policing, this narrowness manifests most frequently in a focus on the quantity of policing and a parallel neglect in the quality of policing. With
This document is a research report submitted to the U.S. Department of Justice. This report has not^6 been published by the Department. Opinions or points of view expressed are those of the author(s)
respect to incarceration, as already noted city- and county-level studies often ignore incarceration rates altogether. But a more general limitation of the extant research is its reliance on overall incarceration rates. As elaborated below, a broader consideration of post-arrest criminal justice indicators may prove useful for expanding our understanding of how criminal justice responses affect crime trends. Accordingly, we extend the typical empirical specifications employed by considering additional criminal justice variables that may be relevant to crime trends, including proactive policing, and age- and crime-specific imprisonment rates per capita. As Eck and Maguire (2006) note, most of the studies of crime trends that emphasize the quantity of policing do so in a context that ignores the quality of policing. A handful of studies have shown that various changes in the nature of policing that occurred in the 1980s and 1990s, and especially policing efforts that target particular types of behaviors thought to facilitate crime, such as levels of public disorder and the prevalence of weapon carrying, can have important implications for crime levels. Early studies of jurisdictional differences in “arrest certainty” (e.g., Yu and Liska, 1993) come to mind here, as does New York City’s highly lauded organizational shifts and orchestrated “order maintenance” approach to policing and “reclaiming public spaces” during the first half of the 1990s. This research motivates the need to further assess the role of police force size, but also directs attention to a wider array of policing indicators, including shifts over time and places in “arrest certainty” and the types of activities to which police resources are allocated. The latter has been referred to in the literature in a variety of ways, including “order maintenance policing” (Rosenfeld et al., 2005) and “proactive policing” (e.g., Sampson and Cohen, 1988; Kubrin et al., 2010). The present study builds on existing work by examining whether proactive policing (see also Messner et al., 2007; Rosenfeld et al., 2007), is associated with recent crime trends. We also expand recent investigations of the role of incarceration in shaping crime trends. Several studies have focused particularly on estimating the effects on crime trends of shifts in levels of incarceration (see Stemen, 2007, for a review). Despite the substantial attention devoted in prior research to the role of incarceration and the possibility of an emerging consensus about the magnitude of its effects on recent crime trends (e.g., Goldberger and Rosenfeld, 2008), several issues warrant additional consideration and are examined in the present study. Specifically, the present study goes beyond prior research by estimating the effect on recent crime trends of age- and crime- specific measures of incarceration rates and by evaluating the role of an alternative indicator, which we label imprisonment-arrest ratio. Though the standard approach of including overall incarceration rates is a useful beginning point, both of the primary arguments that have been used to link shifts in incarceration to crime trends—incapacitation and deterrence—imply that a more specific focus may be meaningful. In particular, we explore whether age- and crime-specific incarceration rates may yield differential effects on age- and crime-specific offending trends. Analyzing age-specific data seems quite important given the emphasis placed on incapacitation effects in the extant literature. Suffice it to say that analyzing the association between age-specific indicators of imprisonment and crime provide a more direct assessment of such effects than non-disaggregated data. Additionally, though offenders do not tend to specialize in most instances, since a large fraction of the mass imprisonment era has been led by shifts in imprisonment among drug offenders, analyzing overall imprisonment and crime might mask potentially important deterrent and/or incapacitation effects of imprisonment. Crime-specific analyses will enable a closer examination of this possibility.
Methods The specific samples, time frames, and measures employed vary somewhat across the three substantive issues addressed, but our general analytical strategy in addressing these issues is to construct when possible from the CTDA a panel database with requisite measures centered on the following time points: 1980, 1985, 1990, 1995, 2000, 2005, and 2010. This approach marshals the strength of a pooled cross-sectional design, while also avoiding the significant data imputation that is
This document is a research report submitted to the U.S. Department of Justice. This report has not^7 been published by the Department. Opinions or points of view expressed are those of the author(s)
types and units. Most notably, these models suggest that age structure and divorce rates are robust predictors of crime rates, with higher crime in areas with a larger percentage of persons aged 15- and where divorce rates are higher. Additionally, the results reaffirm findings shown in other work on non-violent property crime by showing that incarceration rates tend to yield lower crime rates, but at a diminishing rate as incarceration reaches very high levels. An assessment of conditional economic effects on recent crime trends. The results point to a tendency of “objective criminal justice risk” to lessen the criminogenic consequences of elevated unemployment rates and depressed wages. Another intriguing pattern that emerges is that the estimated adverse effects of unemployment and wages on non-lethal crime (both violent and property) are weaker in the face of elevated levels of income maintenance payments (i.e., SSI, Snap, family assistance). Finally, contrary to expectations, inflation levels do not moderate the effects of wages and unemployment rates in our analysis. Overall, we find moderately strong evidence that the assumed main effects of wages and unemployment rates in most previous studies is questionable. The influence of these economic conditions on contemporary crime trends is contingent on other conditions, and this may be one reason why past research yields highly inconsistent empirical patterns for these attributes. Expanding the typical set of criminal justice variables considered. Our expanded analysis of criminal justice factors shows that homicide rates appear to be insensitive to the quantity and quality of policing or levels of incarceration. A consistent finding that emerged was that, at least for non-lethal violence and non-violent property crime, arrest certainty for these crimes is a robust predictor of lower crime rates. This implies that prior county-level studies, which typically have not included policing measures, may be misspecified, and it also highlights an important dimension of policing for shaping recent crime trends. In contrast, police size and proactive policing do not yield the anticipated negative associations with crime rates and, in fact, exhibit positive signs in several of the models. With respect to prison measures, the findings for non-lethal violence are not consistent with expectations. County variation in robbery and aggravated assault is generally not responsive to county differences in imprisonment risk. However, non-violent property crime rates are lower in counties with higher imprisonment rates and in counties situated within states that have higher imprisonment rates. The results for the age-specific county imprisonment variables indicate that both yield significant negative associations with non-violent property crime, but imprisonment rates of younger persons (i.e., ages 18-34) is stronger. Finally, consistent with expectations, rates of non- violent property crime are not affected by imprisonment rates for homicide or non-lethal violence, but they are influenced by imprisonment rates for non-violent property crime.
Conclusions The primary purpose of this project was to enhance the data infrastructure by compiling in a centralized location the most commonly referenced datasets and measures. The key product of the grant -- the Crime Trends Data Archive (CTDA) -- should prove valuable to those involved in or considering the study of contemporary American crime trends (the CTDA is described in greater detail in Appendix A). Thus, while some of the empirical results presented can be read in terms of policy implications, the most important contribution of the present work to policy and practice is in generating a research infrastructure that can facilitate timely and informative research on crime trends and policy issues moving forward. To fully take advantage of the products of this grant, we offer two recommendations. First, it would be a good investment for the appropriate government agency to continue development of the CTDA, including updates as new data become available. Second, the CTDA will be useful to the extent that it is made widely available to other scholars. While the project uses many resources that are already archived, and thus could be retained as separate data collections, the utility of the CTDA lies in its integration of the various components needed to generate meaningful assessments of crime trends in a centralized space. We recommend
This document is a research report submitted to the U.S. Department of Justice. This report has not^9 been published by the Department. Opinions or points of view expressed are those of the author(s)
that the NACJD use the CTDA as the basis of establishing a permanent, distinct archive for studying crime trends. Technical Report
I. Introduction: Statement of the problem and relation to existing literature This project focused on addressing two general issues, as outlined in the original proposal: (a) enhancing the data infrastructure available to study recent American crime trends; and (b) clarifying and expanding the scope of empirical analysis directed at describing and explaining recent crime trends. The rationale for tackling each objective is elaborated below. We begin with a general statement of the problem that motivates these efforts and then elaborate on the specific objectives pursued in the project.
Ia. Enhancing the infrastructure. The substantial shifts in crime observed in the United
States since the early 1980s have stimulated a strong and growing interest among policy makers, the media, criminal justice practitioners, and the general public in the properties and predictors of crime trends. Yet, the extant empirical research on recent crime trends has taken an overly narrow empirical approach, yielding significant ambiguity in the conclusions that can be drawn. One apparent reason for this state of affairs is that the underlying data infrastructure that supports current work is incomplete and highly inconsistent across studies. As elaborated below, the literature on crime trends is relatively modest in size; though it has grown in recent years, the work that has been done varies significantly in terms of the measures employed, temporal coverage, and units of analysis. One of the likely major reasons for this is that the data needed to study crime trends tend to be highly decentralized. Even as electronic access to data has increased substantially and highly germane topical archives such as the National Archive of Criminal Justice Data (NACJD) have evolved into invaluable resources for researchers, some of the data elements frequently used to study crime trends are missing or are not easily attainable in available archives, and others are provided in forms that require extensive processing to be ready for analysis. Several scholars have taken the needed steps to assemble the requisite data, but the substantial work product from such efforts typically has not been archived, yielding a very inefficient process in which each crime trends study essentially must start from scratch. Accordingly, one of the primary objectives of the proposed project was to contribute to a comprehensive data infrastructure that would support an ongoing and more systematic research agenda on recent crime trends. Though the resulting, centralized data archive does not contain all of the pertinent pieces that might be relevant to resolving existing crime trends “puzzles,” our hope is that it will serve as a launching point to stimulate others to expand and add to the effort on an ongoing basis.
Ib. Clarifying and expanding the scope of empirical analysis. Empirical studies of
crime trends tend to be highly variable and relatively narrow in their application. Both of these issues—inconsistency and limited scope—have impeded the accumulation of knowledge on the factors most germane to explaining recent crime trends. These issues are interrelated, and tend to manifest in empirical studies in at least three ways: (a) differences in the variables included as covariates; (b) the use of different units of analysis; and (c) the application of different statistical methods to estimate key parameters. We discuss first the highly variable and somewhat limited empirical specifications employed in the literature on recent crime trends, which also illuminates the disparate approaches often taken in studies that rely on different units of analysis. We then highlight the variety of statistical approaches taken in such studies, many of which are defensible for the task at hand but which nonetheless add a layer of complexity for extracting emergent patterns from the existing research. Though there are plenty of sophisticated and useful empirical studies of crime trends, many are focused on a small subset of potential factors. Often, studies focus on a single factor, such as
This document is a research report submitted to the U.S. Department of Justice. This report has not^10 been published by the Department. Opinions or points of view expressed are those of the author(s)
data (Phillips and Greenberg, 2008). Overall, though we have learned much from existing studies of crime trends, the relatively low degree of uniformity in model specification and statistical methods within and across units of analysis makes it difficult to detect general emergent patterns from the extant research and to draw definitive conclusions about the relevance of given factors. In light of the noted ambiguities of extant research on recent crime trends, in addition to the data infrastructure component of the project, three sets of analysis were conducted in the project with an aim toward illuminating the utility of the data product generated in the project and for clarifying and expanding the scope of research on recent crime trends: (1) the estimation of a parallel series of “baseline” empirical models of recent crime trends using different units of analysis, focusing on state, counties, and cities^1 ; (2) an expanded analysis of the effects of key criminal justice attributes (e.g., the nature of policing, age- and crime-specific imprisonment rates) on recent crime trends that have not been considered extensively in prior research; and (3) an assessment of the conditional effects of economic conditions on recent crime trends. The first of these was explicitly described in the original proposal. The second two were implied but not detailed in the proposal, in large measure because the proposal focused on data infrastructure enhancement and was relatively vague in terms of the specific substantive analyses to be considered. Nonetheless, during the course of the project these issues emerged as logical and particularly fruitful substantive applications of the overarching focal point of the study (i.e., to clarify and expand the scope of crime trends research). Expanding the criminal justice attributes typically considered seems useful both because it focuses on central policy variables and because it draws on several existing NACJD sources, which is the essence of the NIJ Data Resources Program (DRP) under which the project falls. Further, the detailed assessment of conditional economic effects on crime trends is timely in light of the major recession that hit the nation during the latter part of the 2000s, which has occurred largely without notable increases in crime. Not surprisingly, this has stimulated renewed calls for research to explore the conditions under which adverse economic circumstances are or are not likely to yield significant rises in crime rates.
II. Rationale for the research The main objectives of this project were to enhance the data infrastructure for studying crime trends , and to clarify and expand the scope of existing research on crime trends. As noted above, the latter effort was focused on three sets of analyses (i.e., analysis of a parallel baseline model across multiple units of analysis; an expanded analysis of criminal justice attributes; and an assessment of conditional economic effects on crime trends). The rationale for considering these issues is described in more detail below in this section, and the data, methods, and results related to them are discussed in subsequent sections. Before turning to these substantive issues, however, we discuss some pertinent issues regarding the backdrop for pursuing the first stated objective. Enhancing the data infrastructure for studying crime trends entailed a major effort directed toward centralizing the somewhat scattered data sources that often are used in studies of crime trends, adding data that are not routinely considered, and producing an integrated set of data files and data manipulation files
(^1) As noted in the original research proposal, we also considered the possibility of using Metropolitan Statistical Areas (MSAs) as a distinct geographic unit in this project. Though we concur with others that MSAs can serve as a useful level of geographic and social aggregation (Stowell et al., 2009), we steered away from generating distinct data sources because of its overlap with the state- and county-level components of our research, and because of two unique complications that arise when studying MSA crime trends. First, MSAs represent county-group areas that have changed considerably in composition over time, and though some MSA crime data are routinely published by the FBI, it is not possible to impose from published sources a comparable set of MSA definitions across the three decades considered in our study. Second, while MSA crime estimates could be derived in a consistent manner across time by aggregating county-level crime data with careful attention to shifting MSA boundaries, doing so compounds the general problems that have been documented about aggregating crime data (Maltz and Targonski, 2004).
This document is a research report submitted to the U.S. Department of Justice. This report has not^12 been published by the Department. Opinions or points of view expressed are those of the author(s)
that could facilitate in a more efficient manner research in this area of inquiry. We next describe the data resource produced in the project, which also serves as the “results” of this effort.
IIa. Enhancing the data infrastructure
Notwithstanding the extraordinarily useful features of the NACJD, which contains many pertinent data elements that will be of interest to researchers who wish to study recent crime trends, the existing data infrastructure is limited in two notable ways: (1) it is incomplete and/or somewhat decentralized; and (2) it contains pieces of the puzzle but little shared work-product about how those pieces might be and often are fitted together. Before elaborating on the first issue, it is important to acknowledge that the NACJD and other data archives bring together an enormous volume of data, much of which is pertinent to research on recent crime trends. Indeed, without the NACJD it would be nearly impossible to access in short-order the various components needed to sufficiently examine crime trends across multiple units, such as states, counties, or cities. Nevertheless, there are several data resources that are relevant to studying crime trends that do not fall within the normal confines of the NACJD. This includes data on social, economic, and demographic attributes from the decennial census and the annual American Community Survey (ACS), data on annual unemployment rates from the Bureau of Labor Statistics (BLS), data on wages and other economic indicators from the Bureau of Economic Analysis (BEA). Other pertinent data sources (e.g., the National Corrections Reporting Program [NCRP]) are included in NACJS, but not in forms that are easily modified to support their use in studies of crime trends. Finally, some of the available resources housed at the NACJD possess notable limitations for studying contemporary crime trends. For example, though the NACJD produced county-level crime and arrest data avoid many of the issues that emerge when researchers estimate county data from NACJD (or FBI) agency-level data (see Maltz and Targonski, 2002, 2004), the current county-level holdings at NACJD apply divergent imputation procedures from 1994 onward, introducing an important break from prior years that can be problematic for studies of crime trends that span before and after this period. While there is a need to enhance the data infrastructure for studying crime trends by broadening the data that are available in a centralized location, an even more pressing matter is to develop an archive of the substantial data processing and manipulation required to put the available data to work. Studying crime trends is somewhat different than many other research endeavors. Rather than drawing on a single survey or a few major data sources that encompass one or perhaps a small handful of data “waves,” a typical study of recent crime trends entails the combination of a very large number of distinct data sources, and often in each case ten, twenty, or perhaps even more “temporal installments” of those sources. The nature of the task at hand means that simply providing an archive of raw data with machine readable “setup” files will help, but this represents a relatively small portion of the overall effort involved in generating the data needed to study crime trends. The point being made here is not that the remainder of the work needed to do a crime trends study should fall on the NACJD, but rather that a highly useful archive would not only house a wide array of data but also would document how such data are combined and analyzed to generate meaningful information about crime trends. This is the general direction in which data archiving has been heading for the past several years within the NACJD, and in the area of crime trends research it should be especially welcomed. With these points as a backdrop, this project focused in part on assembling in a centralized location the various data resources that might be used in a standard crime trends study to ensure a reasonable, theoretically informed, baseline empirical specification. Many of these sources already are contained within the NACJD, but others were culled from other sources. Importantly, in both of these instances, we include in a centralized directory the raw data, machine readable setup code, and programming syntax used to extract pertinent data and integrate with other sources to generate analysis-ready datasets for states, counties, and cities. This is not to suggest that we have resolved all
This document is a research report submitted to the U.S. Department of Justice. This report has not^13 been published by the Department. Opinions or points of view expressed are those of the author(s)
pertinent issues of data integration and estimation that are relevant to many common approaches to studying crime trends, including the strategy adopted in the analyses reported below.
Crime Data. The CTDA provides UCR city, county, and state crime estimates that are, in
our judgment, preferable in many respects to existing resources. There are several possible sources from which one might obtain estimated UCR crime counts, including annual agency-level and county-level estimates housed in the NACJD. Research on crime trends have drawn from a wide variety of these sources, and sometimes it is not clear where exactly the data used in a given study were obtained. The agency-level NACJD data are useful for estimating city crime totals, but for reasons outlined in detail by Maltz and Targonski (2004), they pose problems for generating valid county-level estimates. Among the more problematic issues are that the available agency-level files—even when linked to geographic cross-walk files that identify the counties in which agencies fall--do not provide a straightforward means by which to apportion data from agencies that serve multiple counties and are prone to yielding double counting of both crime counts and population counts of the areas represented. Additionally, the agency-level data for a given year include only the jurisdictions that report crimes to the FBI during the period, which may be limiting if researchers are interested in estimating total county crime volume that adjusts for non-reporting agencies. In part as a response to the limitations of the standard agency-level UCR files just noted, the NACJD also produces an annual county-level crime dataset. As Maltz and Targonski (2002) note, these NACJD county-level files differ from the county-level data one can generate independently from publicly available agency-level files in two very important ways. One is that NACJD use as their starting point internally produced FBI files commonly referred to as “Crime by County” files. These are agency-level (i.e., ORI) files that are organized by counties as defined by the FBI, and they are superior to standard agency-level files one can access from NACJD because they explicitly apportion agency-level data to the counties in which they fall (up to a maximum of three) and they properly identify “zero-population agencies” that report crime but do not represent populations that are distinct from other agencies in a county. Another important feature of the NACJD county-level files is that they yield estimates that explicitly account for underreporting and non-reporting at the agency level. Unfortunately, though, the procedures used by NACJD to impute missing or incomplete data have changed significantly over time, yielding in particular a major a break in the available county series in 1994 (see Maltz and Targonski, 2002, Table III). On its face, the imputation procedures used by NACJD since 1994 are an improvement over the method used before that point, but to our knowledge the validity and reliability of the different procedures have not been assessed systematically. Further, the series break imposed by the shift in imputation procedures in 1994 makes the NACJD county crime data quite limited for purposes of studying trends during a period that is often of significant interest to crime trends scholars (e.g., the 1980s and 1990s). Two nice features of the CTDA are that it includes the “raw” agency-level FBI “Crime by County” files annually from 1980-2010, and that it includes agency-level “population” files obtained from the FBI for 1980, 1990, 2000, and 2010 that define all law enforcement agencies that fall within each U.S. county and the population served by these agencies. By integrating these files, it is possible to generate a full series of annual county-level crime estimates using a set of procedures that are comparable to those currently used by the NACJD, but without the break in series in 1994. This component of the CTDA should prove useful for researchers who wish to study longer-term county crime trends and, importantly, it can facilitate detailed evaluations of different imputation methods for generating county crime estimates from agency-level data. For the purposes of the county-level crime estimation applied for the analyses reported below, we followed these steps. First, we begin with a full listing of local law enforcement agencies as reported in the “Population Files” provided to us by the FBI for 1980, 1990, 2000, and 2010, a file that includes agency ORI identifiers, population counts, and internal FBI county codes. Using
This document is a research report submitted to the U.S. Department of Justice. This report has not^15 been published by the Department. Opinions or points of view expressed are those of the author(s)
these four files as bookmarks, we then used linear interpolation to estimate population for agencies between the periods. We subsequently linked to this file annual agency-level UCR crime data files obtained from the NACJD using a geographic crosswalk that provides a translation table between ORI identifiers, FBI internal county codes, and FIPS county codes (The geographic cross-walk file was obtained from NACJD as well). An important step taken in appending these files was to identify ORIs that spanned multiple counties (up to three, as is the practice used by the FBI) and to estimate the proportion of the population covered by such agencies that falls within each specified county. As Maltz and Targonski (2002, 2004) note, a considerable number of law enforcement agencies either report less than 12 months of data to the FBI or do not report at all in a given year. Following the practice of NACJD (from 1994 onward), we adjust crime counts for the full period of our data coverage upward for agencies that report between 3-11 months of data by multiplying the reported crime count by the proportion of months reported (i.e., 12/n, where n is the number of months reported). We deviate somewhat from the approach adopted by NACJD to impute data for agencies that report less than 2 months of data in a given year. Specifically, from 1994 onward, the NACJD replaces crime counts for agencies reporting less than two months of data with a value equal to Cs * P (^) a /P (^) s , where C (^) s is the crime count and P (^) s is the population of agencies of similar size within the same state , and P (^) a is the agency population. While this strikes us as a reasonable approach, it ignores the significant heterogeneity within states in crime levels, making the questionable assumption that agencies of the same size in one part of a specified state are comparable to those quite far away (e.g., rural and urban agencies of similar size). Given this, for agencies that report 0- months of crime data, we replace the observed value with an estimate that is equivalent to the average crime count of agencies of similar size within the same county. 2 When a comparable agency is not available as defined here, we allocate a value for these agencies that is equivalent to the average crime count of similarly sized agencies within the same state. The end result is that we are able to generate crime estimates for the vast majority of agencies that fall within U.S. counties, using a consistent set of procedures that parallel in important respects those applied by the NACJD but with an added layer of drawing information about non-reporting agencies and “low reporting agencies” (i.e., those that report less than 3 months of data) from comparable agencies within the same county. This permits us to aggregate the agency-level data to produce county- and state-level UCR estimates using a consistent set of procedures. Note that we also include in the CTDA state- level crime estimates drawn from the Bureau of Justice Statistics (BJS) UCR Datatool (http://bjs.ojp.usdoj.gov/ucrdata/Search/Crime/State/StateCrime.cfm). These state estimates are generated by BJS by aggregating crime data only from law enforcement agencies that report 12 months of data within a given year. Given the relatively high rates of non-reporting and under- reporting among U.S. law enforcement agencies, this leaves a significant amount of data out of the state-estimates. It is unclear how these estimates compare to those generated from those obtained in other ways (e.g., the imputation procedures currently used by the NACJD), but this is an important issue worthy of exploration. As Appendix A shows, the CTDA also includes other crime data besides city and county UCR offenses known. For instance, the archive includes agency-level SHR data from 1976-2009, which can readily be used to support city-level analyses of detailed homicide rates. Though we do not do so for the analysis reported below, using the adjustment and imputation procedures described above, it would be relatively straightforward to generate county- and state-level SHR estimates. To facilitate such estimations, we provide all of the needed components in the CTDA, and also the programming code used to generate adjusted agency-level data as described and aggregate to counties.
(^2) We used the following population groups for these computations: (a) under 5,000; (b) 5,000 – 15,000; (c) 15,000 – 45,000; and 45,000+.
This document is a research report submitted to the U.S. Department of Justice. This report has not^16 been published by the Department. Opinions or points of view expressed are those of the author(s)
structure (i.e., population size and density), divorce rates, immigrant concentration (e.g., % Latino and % foreign born), and resource deprivation (e.g., poverty rates, % non-Latino black, % female headed families, median family income). Though census data for states, counties, and cities are available from a variety of sources, including public repositories such as the Inter-University Consortium for Political and Social Research (ICPSR), of which NACJD is a component, in our experience acquiring the various files that are needed can be quite cumbersome in a standard crime trends study that spans several decades. Additionally, some of the most recent census data—most notably the aggregate ACS data—are not yet widely available in an analysis ready format. The CTDA provides extensive data from the ACS and its predecessor (i.e., the decennial census STF files).
Economic Data. The CTDA contains a relatively rich array of data sources on economic
conditions. The census-based sources just described include indicators of unemployment and poverty rates, but supplemented such data with annual estimated unemployment rates from the Bureau of Labor Statistics (BLS) for cities, counties, and states from the early 1980s through 2010. Additionally, the CTDA includes from the Bureau of Economic Analysis (BEA) state and county estimates of average wages, levels of unemployment insurance and income maintenance, and state estimates of GDP and GSP. Finally, we incorporate regional indicators of the consumer price index (which enable us to measure inflation across regions and over time, and also adjust wages for inflation) and the Index of Consumer Sentiment (ICS). The latter has been chronicled in several recent papers on trends in violence and property crime (e.g., Rosenfeld and Fornango, 2007; Rosenfeld, 2009).
IIb. Clarifying and expanding the scope of empirical analysis
While the data resource just described was a major focus of the project, the overarching motivation for producing this resource was to facilitate meaningful analyses of recent crime trends. The original proposal outlined several possibilities along these lines, with an emphasis on clarifying observed empirical patterns that are often generated from studies that apply different empirical specifications to different units of analysis and on expanding the typical set of criminal justice variables considered. The rationale for these two substantive issues and a third--an assessment of conditional economic effects on recent crime trends—are described in the paragraphs that follow. The data, methods, and results of analyses directed at these issues are presented in subsequent sections of the report.
IIb.1. Uniform empirical specification across states, counties, and cities. An important initial set of analyses conducted in the study encompasses the estimation of parallel empirical models of crime trends across multiple units of analysis, focusing on the most commonly used geographic boundaries employed in the extant research: states, counties, and cities. The objective is to “hold constant” differences in model specification to discern potential uniformities in empirical results that may emerge across levels of analysis (or, alternatively, to identify meaningful differences in patterns across different geographies). Though there are many examples of highly sophisticated studies of crime trends, the cumulative body of literature in this area tends to be highly variable and relatively narrow in its application. As Baumer (2008) elaborates, prior research on crime trends often focuses on somewhat different time periods and outcome measures, but it is particularly inconsistent in three ways: (a) the variables included as covariates; (b) the units of analysis used; and (c) the application of different statistical methods to estimate key parameters. While a diversity of approaches can be a healthy feature of scientific research, it can generate ambiguity if it is not accompanied by some systematic assessments that hold constant key factors that tend to vary across studies. For instance, the conclusions one draws about the role of many factors (e.g., incarceration, police size, and age
This document is a research report submitted to the U.S. Department of Justice. This report has not^18 been published by the Department. Opinions or points of view expressed are those of the author(s)
structure) appear to be highly contingent on the unit of analysis used, and the estimated role of these and other factors (e.g., unemployment) seems highly sensitive to model specification and/or analytical procedure (e.g., Liedka et al., 2006; Defina and Arvanites, 2002; Eck and Maguire, 2006; Spelman, 2005). It could be that crime trend patterns and predictors vary meaningfully across cities, counties, and states. But the current research literature cannot tell us this, because the typical empirical specification adopted is highly variable both within and across studies based on different units of analysis. The lack of systematic research on crime trends across units of analysis may contribute to gaps in knowledge of the factors thought to be most pertinent to shaping contemporary crime trends. For example, shifts in illicit drug use and market activity, and especially crack-cocaine involvement, have received significant attention as explanations for recent changes in crime levels (Blumstein and Wallman, 2006). The arguments presented are compelling, but the empirical research is not as convincing. Some of the studies that represent the primary regression-based evidence for the role of crack cocaine activity on recent crime trends (e.g., Baumer et al., 1998; Ousey and Lee, 2002, 2004) do not include any other time-varying indicators other than temporal variability in crack use and market activity, which obviously can contribute to misleading results. Other recent sub-national studies have adopted estimation procedures that control for unmeasured heterogeneity and shared temporal trends across units, which strengthens the faith we can have in estimates obtained for the few factors that are directly measured. However, this strategy divulges little about the role of the unmeasured factors hypothesized to have played a potentially significant part in shaping recent crime trends. We also see a high degree of inconsistency in model specification for economic factors. State-, and regional-level studies suggest that economic indicators such as GDP, consumer sentiment, and wages are significant predictors of crime trends, but very few city- and county-level studies incorporate time-varying estimates of these factors. Indeed, city-level studies also routinely omit time-varying measures of another economic factor–the unemployment rate—that has been shown to be relevant in other research using larger aggregates (see Levitt, 2001 for a review). Along the same lines, though there is a long history of linking incarceration to crime rates through incapacitation and/or deterrent processes, and there also is a voluminous empirical literature that focuses on estimating the relationship between rates of incarceration and crime (see Spelman, 2006), most city- and county-level studies of crime trends have either omitted incarceration rates altogether or applied state-level incarceration rates uniformly to all places within the same state. Neither approach is ideal. On the one hand, the literature shows fairly consistently that incarceration rates are an important part of the puzzle of recent crime trends, and therefore omitting them may yield biased parameters for the variables that are measured and undermine the validity of results reported in county- and city-level analyses. On the other hand, applying state-level incarceration rates to all places within a state may introduce significant measurement error. There is great variability within states in prison admission rates and, if would-be offenders are affected more by local than by state realities, the estimation of state-level incarceration rates on county or city crime rates may yield highly imprecise estimates. Finally, studies of crime trends often apply different methods both within and across units of analysis to estimate key parameters. Although some studies have used methods geared toward identifying classes of crime “trajectories” (e.g., Weisburd et al., 2004), perhaps because a central interest in this area of research is in explaining recent crime trends, most studies have applied different versions of two suitable analytical strategies: pooled time-series and multilevel growth curve models. Briefly, most applications of growth curve models in the study of crime trends, including prior work by the principal investigator, impose deterministic trend parameters and usually do not account for unmeasured stable characteristics of the spatial units (see, e.g., Baumer et al., 1998; Ousey and Lee, 2002), whereas the typical econometric panel models estimated to study crime trends do not make a priori assumptions about how the unit intercepts shift over time and almost
This document is a research report submitted to the U.S. Department of Justice. This report has not^19 been published by the Department. Opinions or points of view expressed are those of the author(s)
