Abstract
This research examines the association between economic insecurity and imprisonment rates in the United States. Building on Garland’s thesis about punishment and late modernity, it is hypothesized that rising economic insecurity in a population is associated with an increase in the imprisonment rate. This hypothesis is tested with state-level data for the years 1986–2013. Results indicate a robust association between changes in economic insecurity, measured as the percentage of households in a state losing a quarter or more of their income in a single year, and changes in imprisonment rates. This finding suggests that economic insecurity is not only relevant for explaining large-scale shifts in penal philosophy and practice, as prior sociological theory has argued. It also explains some of the year-to-year variation in imprisonment rates and points to another way in which inequality is associated with punishment.
Introduction
The U.S. incarceration rate increased fivefold between 1972 and 2010 (National Research Council 2014). Even with recent declines in the prison population during the last decade, the rate at which people are incarcerated in the United States remains far above the nation’s historical average and is incomparable to the rates in other western democracies. This growth in the number of people behind bars, combined with the marked differences across states and countries, has provided social scientists with the raw material for testing extant sociological theories of punishment and for developing new explanations for the emergence and persistence of mass incarceration (Garland 2001; Raphael and Stoll 2013; Spelman 2009; Sutton 2000). This body of scholarship generally agrees that crime rates, alone, do not fully explain temporal and spatial variation in incarceration rates, and that other explanatory factors must also be considered. For instance, research suggests that incarceration rates are influenced by changes in racial demographics and the degree of Republican strength in government (Jacobs and Carmichael 2001), anxieties associated with cultural change (Garland 2001), shifts in public punitiveness (Enns 2014), and worsening economies (Chiricos and Delone 1992).
The latter of these potential explanatory factors—economic conditions—is of particular interest in the current study. A prominent line of scholarship, born out of the pioneering work of Rusche and Kirchheimer ([1939] 2003), suggests that prison populations are responsive to the level of surplus labor in a society. According to this thesis, motivation to commit crimes and the incentive for the state to punish offenders are higher when unemployment increases (Greenberg 1977; Jankovic 1977; Quinney 1977; Spitzer 1975). As such, the imprisonment rate is expected to increase when unemployment swells, a supposition that has been supported in several empirical studies, particularly those conducted prior to the steep escalation of imprisonment rates in the 1990s. For instance, Chiricos and Delone’s (1992) review of dozens of studies of unemployment and incarceration rates concluded that the correlation between the two variables was statistically significant, substantively meaningful, and frequently replicated.
Yet in the decades since Chiricos and Delone’s (1992) careful appraisal of research on labor markets and incarceration, scholars have increasingly questioned whether Rusche and Kirchheimer’s thesis remains a viable explanation of imprisonment rates (D’Alessio and Stolzenberg 1995; Pfaff 2008; Sutton 2004). Empirically, studies that include data from portions of the mass incarceration era find no association between unemployment rates and imprisonment rates (Beckett and Western 2001; Jacobs and Carmichael 2001; Jacobs and Helms 1996). Theoretically, Sutton (2004) notes that unemployment is merely one facet of a complex economy, and scholars must consider related factors and mechanisms. As he states, “business cycles are multidimensional phenomena, comprising several empirical trends that are often independent of each other and that have different implications for class-based social division” (Sutton 2004:171).
In the present study, we build on the premise that economic conditions are relevant for understanding imprisonment rates, yet we also pay heed to the critiques of Sutton (2004) and Garland (1990), who demonstrate that unemployment is but one facet of a complex political economy. To this end, we direct attention to the related concept of economic insecurity, a condition that affects both the unemployed and many gainfully employed workers who fear that they are at risk of substantial economic loss in the future. Drawing on work in political science (Hacker 2008) and sociology (Kalleberg 2009; Western et al. 2012), we conceptualize economic insecurity as the risk of a substantial loss of personal or household income due to job loss or unexpected costs, such as a major health care expense, which can create unease about one’s future financial security. By this definition, economic insecurity has increased since approximately the late 1970s, as a larger swath of American workers suffered significant income losses and increasingly expressed fears of being laid off (Hacker 2008). Recent work on the topic indicates that such financial strains have social and psychological consequences. Concerns about debt and the ability to “make ends meet” take an emotional toll on families (Cooper 2014), increase support for state intervention in the economy (Hacker, Rehm, and Schlesinger 2013), and adversely affect personal health (Catalano 1991).
Economic insecurity may also be consequential for punishment. Garland (2001) makes this case in his account of the rise and persistence of mass incarceration in the United States and Great Britain. According to Garland (2001:79), from World War II (WWII) through to the 1970s the American working class had “an unprecedented level of economic security in their lives,” and the economic philosophy that emphasized an interventionist state and a safety net for workers mapped on to a set of criminal justice policies guided by a preference for rehabilitation, judicial discretion at sentencing, and modest use of imprisonment. Thereafter, Garland argues, policy changes and deunionization stripped away some of the economic security that previously characterized the working and middle classes, which had psychological as well as financial consequences. Declining economic security, combined with higher crime rates from the 1970s to the early 1990s, increased public anxiety about personal and financial security, which led to an anxious and more punitive population that was increasingly receptive to punitive policies and harsher punishments (Garland 2001:156; see also Wacquant 2010:204). It follows from this line of theory that state imprisonment rates would be associated with macro-level changes in economic insecurity.
To date, the connection between insecurity and punishment has largely been tested at the level of individuals’ punitive attitudes, which yields a mixed set of findings (Costelloe, Chiricos, and Gertz 2009; Johnson 2001; Lehmann and Pickett 2017). Macro-level studies have investigated this thesis only insofar as unemployment or wages are assumed to be fair proxies for the level of economic insecurity, which is indeed a reasonable assumption. However, research shows that gainfully employed workers and even middle-class households can be at risk of sudden financial loss and economic hardship (Hacker 2008). We thus make use of an alternative measure of insecurity that captures sudden financial loss, and we test the hypothesis that economic insecurity at the macro-level is associated with imprisonment rates.
We proceed with a brief review of the existing research on economic conditions and punishment, positioning our case as an extension of that line of work. We then test our hypothesis about insecurity and imprisonment rates with a pooled time-series analysis of state imprisonment rates from 1986 through 2013 (years for which we have data on the measure of economic insecurity). We close with a discussion of the implications for understanding temporal variation in imprisonment rates.
Theory and Prior Research
Economic Conditions and Punishment
Perhaps the most frequently tested theory that connects macroeconomic conditions and punishment is the Marxian thesis that the criminal justice system is used, in part, to regulate and manage surplus labor. This idea, often associated with the work of Rusche and Kirchheimer ([1939] 2003), suggests that the predominant type of punishment found in a society is partly determined by the economic mode of production. Furthermore, within a capitalist state, imprisonments and deportations are expected to be more frequent during periods of high unemployment (King, Massoglia, and Uggen 2012). Research as of the early 1990s generally supported this hypothesis. As mentioned above, Chiricos and Delone (1992) reviewed 44 studies on the topic and concluded that “labor surplus is consistently and significantly related to prison populations” (p. 431). Cross-national research comparing common law countries also found that imprisonment rates were higher when unemployment increased (Sutton 2000).
However, not all research supports this argument, and prior work identifies at least three reasons for questioning whether the association between unemployment and incarceration is robust. First, the effect of unemployment is contingent on characteristics of the political economy. For instance, Michalowski and Carlson (1999) show that the correlation between unemployment and imprisonment varies over time, depending on the degree to which the state intervenes in the economy. The unemployment coefficient is weaker when the state provides a more robust safety net to lessen the pains of unemployment (e.g., the 1930s and 1940s). Second, unemployment and its consequences are to some extent determined by policy making and the structure of the economy. To this end, Sutton’s (2004) comparative study shows that the correlation between unemployment and imprisonment rates disappears when controlling for measures of partisan politics and structural features of the economy. Third, and related to the previous points, the empirical evidence that surplus labor is associated with imprisonment is equivocal. This statement would not be true three decades ago; however, several studies since the time of Chiricos and Delone’s (1992) appraisal of research find no significant association between unemployment and incarceration rates (e.g., Campbell, Vogel, and Williams 2015; Jacobs and Carmichael 2001; Jacobs and Kleban 2003).
Our point in highlighting these critiques is not to imply that we should abandon the idea that economic conditions are relevant for understanding the style and intensity of punishment. Rather, we merely suggest that unemployment is only one facet of an economy; one that arguably has outsize influence on theories of punishment. As a complement to the labor market thesis, we draw on a line of theory that emphasizes the related concept of economic insecurity.
Economic Insecurity and Imprisonment Rates
The clearest theoretical statement about insecurity and punishment is found in Garland’s (2001) The Culture of Control, in which he describes and explains a major shift in American criminal justice policy and practice that unfolded during and after the 1970s. According to Garland, crime control in the decades prior to the 1970s was shaped by a belief that rehabilitation should be the guiding philosophy behind criminal justice policy and decision-making. To facilitate this correctional approach, sentencing policies permitted substantial discretion to judges so that punishments could be tailored to the offenders more than the offenses, and policy was to some extent guided by social science research on the root causes of crime. This approach to corrections and punishment coincided with stable imprisonment rates, which for many decades held steady at about 110 per 100,000 people, a rate that was comparable to other western democratic nations.
This “penal welfarist” approach, as Garland (2001:3) labeled the era, gave way to a different set of crime control principles and strategies that became entrenched after the 1970s. Determinate sentencing policies, such as mandatory minimums and three-strikes laws, replaced indeterminate sentencing regimes, decision-makers focused less on the root causes of crime in favor of focusing on situational factors and control mechanisms, and rehabilitation gave way to the philosophies of deterrence and incapacitation. Essential to these new objectives was a greater reliance on prisons, which contributed to the U.S. imprisonment rate increasing during the last quarter of the twentieth century (Garland 2001).
Garland argues that this sea change in punishment coincides with several changes in American politics and society that fomented support for punitive public policies and tolerance for exceptionally high imprisonment rates. For Garland, of particular importance was the growing sense of anxiety and insecurity in middle-class households that emerged after the 1970s. Middle-class insecurity during this period had multiple sources, two of which are relevant for the present argument. The first was the rising levels of crime and violence. The violent crime rate doubled during the 1960s and then doubled again between 1970 and 1995. The victims of crime throughout this period were most often the poor and racial minorities, but the objectively high crime rates, combined with sensationalized media coverage, made fear of crime “an established part of daily existence” for the middle class as well (Garland 2001:152).
Second, and most relevant here, anxiety about the crime problem was accompanied by a growing sense of economic insecurity. According to Garland, the decline of the welfare state and the rise of market-based approaches that place greater responsibility on individual workers to provide for their retirement were among the most significant changes in American society during the late twentieth century. The proportion of jobs with pensions declined and retirement savings became tethered to the market, for instance through 401(k) and similar programs. In addition, union membership declined significantly (Western and Rosenfeld 2011) and job security grew more precarious. This aspect of what Garland refers to as “late modern society” is precisely what Hacker (2008) labels “the great risk shift,” a period characterized by the reapportionment of risk from the state to the individual.
Welfare state retrenchment and the growing precariousness of work (Kalleberg 2009), particularly when combined with heightened anxieties about crime, was consequential for criminal justice policy and practice. To quote Garland (2001) at length: [T]his new element of precariousness and insecurity is built into the fabric of everyday life . . . . Little surprise then, that the felt need to establish control over risks and uncertainties, and the desire to stave off insecurity, become ever-more urgent aspects of middle-class psychology and culture. Little surprise too that people increasingly demand to know about the risks to which they are exposed by the criminal justice system and are increasingly impatient when that system fails to control “dangerous” individuals who are within its reach. . . Anxieties of this kind are often mixed with anger and resentment and, when experienced en masse, can supply the emotional basis for retaliatory laws and expressive punishments. (pp. 155–56)
In short, a central pillar in Garland’s argument is that the changing nature of employment contributed to a sense of anxiety and frustration that, in turn, helped usher in an era of pessimism, fear, intolerance, and harsh sanctions.
While Garland set out to explain a seismic shift in crime control philosophy, policy, and strategy, our inquiry is focused on short-term fluctuations in the use of imprisonment. Specifically, we investigate whether a measure of economic insecurity explains some of the temporal variation in state imprisonment rates during the era of mass incarceration. In line with Garland (2001) and related work by Melossi (1985, 1993) we theorize that economic anxieties result in a harsher public discourse about the crime problem, regardless of the actual crime rate (Melossi 2000:298), and that these anxieties and fears can influence criminal justice decision-making.
The notion that insecurity is associated with fear, intolerance, and punitive sentiment has some empirical support, although it is by no means axiomatic. For instance, one study finds that fear of crime is higher among those with pessimistic views about their futures (Vieno, Roccato, and Russo 2013), which is consequential for imprisonment rates because criminal sentencing is more punitive where fear of crime is higher (Baumer and Martin 2013). Research also finds that feelings of uncertainty and insecurity are associated with greater intolerance toward others and more support for extremist ideas (McGregor et al. 2008). Pessimism about one’s own and the nation’s economic future is also associated with higher levels of racial prejudice (Burns and Gimpel 2000) and support for immigration restrictions (Citrin et al. 1997). With respect to punitive attitudes, a study of Florida residents finds that individuals reporting a sense of economic insecurity in their personal lives—measured by an expectation of being financially worse off next year—scored higher on an index of punitive attitudes (Costelloe et al. 2009; see also King and Maruna 2009).
However, not all studies agree on this point. An analysis of General Social Survey data finds no association between whites’ punitive beliefs and economic insecurity, which was measured by responses to questions about satisfaction with the respondent’s present financial situation and whether this situation has gotten better or worse in recent years (Johnson 2001). Another study of insecurity and punitiveness, using data from a single county in Florida, finds mixed results. Measuring economic insecurity with questions about future financial expectations, results showed no association for white males but a significant and positive association for women and non-whites (Hogan, Chiricos, and Gertz 2005). Finally, using a national sample of registered voters collected during the Great Recession, Lehmann and Pickett (2017) conclude that support for the death penalty is actually lower among respondents who expressed more pessimism about their future financial situation.
Taken together, theory suggests that incarceration rates are responsive to aggregate levels of economic insecurity (Garland 2001; Melossi 2000; Wacquant 2010). The key mechanisms connecting these macro-level phenomena are public and elite sentiment, including working-class and middle-class anxiety, elevated fear, punitiveness, and, following Melossi (1985, 1993), a harsher discourse about crime among the public and elites. We are unable to test each mechanism, and we acknowledge that prior research on economic insecurity and punitiveness yields mixed results (although some research on related outcomes, such as fear and intolerance, finds the expected statistical association; see Burns and Gimpel 2000; Vieno et al. 2013). The theoretical connection between a fearful or punitive public and punitive sanctions also has some support in earlier studies on sentencing (Baumer and Martin 2013) and national imprisonment rates (Enns 2014). Mindful of variability in conclusions from prior research, in our assessment there is a sound theoretical rationale for hypothesizing that imprisonment rates increase following a rise in macro-level economic insecurity, above and beyond the influence of crime and unemployment. We test this proposition while giving attention to other factors that could also be associated with imprisonment and insecurity. These include other aspects of the economy, political partisanship, and demographic change.
Other Determinants of Imprisonment Rates
Although economic insecurity is our primary focus, it is necessary to assess its influence while taking into account several other factors that could be associated with insecurity and imprisonment. First, we must be attentive to other economic indicators, such as the level of income inequality. According to one line of theory, a larger gap between the “haves” and the “have nots” yields higher imprisonment rates because, as Chambliss and Seidman (1980) claim, the criminal law becomes an instrument through which the wealthy can “enforce through coercion the norms of conduct that guarantee their supremacy” (p. 33). However, research does not always support this argument. Measures of inequality, such as the Gini coefficient, do not strongly correlate with imprisonment rates in studies of U.S. states (Campbell et al. 2015; Jacobs and Carmichael 2001) or in national time-series analyses (Enns 2014). Still, it is plausible that inequality influences imprisonment rates and that economic insecurity could correlate with the unequal distribution of income, and hence we must assess the effect of economic insecurity while controlling for a measure of inequality.
We also consider characteristics of the welfare state. For instance, it is plausible that social welfare programs such as unemployment insurance suppress crime rates during times of economic hardship. Low crime rates, in turn, mitigate the need for imprisonment. Sutton (2000:360) mentions this connection in his comparative work on imprisonment in common law countries, where he finds an inverse relationship between imprisonment rates and country-level expenditures on unemployment compensation, work injury benefits, and public assistance. This finding aligns with other research on the welfare state and imprisonment. For instance, Beckett and Western (2001) examine changes in U.S. state imprisonment rates between 1975 and 1995 and identify a correlation between welfare state retrenchment and the expansion of state prison populations, particularly in the 1990s. Given this evidence of a correlation between welfare state generosity and imprisonment, along with the argument that welfare state retrenchment is a major cause of economic insecurity (Hacker 2008), we assess the effect of insecurity while accounting for relevant measures of welfare generosity.
The level of insecurity may also be associated with Republican Party control (Garland 2001; Hacker 2008). The rise of mass incarceration coincides with a rightward shift in American politics (Beckett 1999; Garland 2001) and some studies find that imprisonment rates are on average higher when and where conservatives are in power (Jacobs and Carmichael 2001, 2002; Jacobs and Helms 1996; Sutton 2000). In the United States, it has been argued that Republican politicians advocated for punitive responses to crime as an electoral strategy to woo white, working-class voters (Scheingold 1992). Furthermore, the conservative movement’s suspicion of a social safety net is cited as a primary reason for the rise of economic insecurity (Hacker 2008), thus making it necessary to account for partisanship in an analysis of imprisonment and economic insecurity.
Finally, racial demographics may influence imprisonment. Macro-level work in the tradition of Blalock’s (1967) theory of intergroup relations suggests that punishment is to some extent an expression of racial prejudice. Akin to Blalock’s claim that discrimination against blacks in the post-WWII era was closely connected to the relative black population size, the criminological variant of Blalock’s thesis understands crime control institutions as channeling the prejudices of the population. Accordingly, imprisonment rates are expected to be higher when and where the percentage of blacks is larger (Greenberg and West 2001; Jacobs, Carmichael, and Kent 2005). Not all empirical work supports this thesis (Spelman 2009), and coefficients on measures of minority group size are generally small (Campbell et al. 2015). Still, we must consider the racial composition of the population when attempting to isolate the influence of economic insecurity.
Methods
Data and Measurement
We test our hypothesis by analyzing state-level data for the period of approximately 1986 to 2013 (specific years may shift for some analyses). As we clarify below, the selection of years and states was largely determined by the years for which we had information on the focal independent variable, economic insecurity.
Dependent variable
The dependent variable is the number of sentenced prisoners under state jurisdiction per 100,000 population in a given state-year. We use this measure of the imprisonment rate, as opposed to the rate of prison admissions, as our primary outcome variable for two reasons. First, this variable is most often used in the extant research on the economy and punishment, which allows us to maintain continuity with prior state-level studies. Second, economic conditions could affect both the likelihood of entering prison and the timing of release, each of which determines the overall imprisonment rate. Nonetheless, in the interest of completeness, we also present the results of an analysis that substitutes prison admission rates for total imprisonment rates. The prison admission rate is measured as the number of sentenced prisoners admitted to state prison per 100,000 population in a given state-year. State prison admissions and prison population data, collected as part of the National Prisoner Statistics program, were obtained from the Bureau of Justice Statistics. Consistently measured data for the dependent variables are available after 1977; however, the first and last years for which our measure of economic insecurity is available are 1986 and 2012, respectively (more on this below).
Independent variables
Economic insecurity can exist at the micro- and macro-level. To date, the concept has most frequently been used in surveys of individuals, particularly studies in which survey respondents are queried about their financial outlook and their ability to cover a sudden financial hardship. 1 Research on this dimension of economic insecurity is based on individual approximations and perceptions of future volatility as opposed to economic hardships actually experienced by an individual in the present or past. In recognition of this gap in the policy and related literature, Hacker and colleagues (2014) developed the Economic Security Index (ESI) as a standardized measure of exposure to economic insecurity at the macro-level (state and national).
The ESI is measured as the percentage of all individuals in a state, including children, that experienced at least a 25 percent decline in their household income during the previous year and who lack the financial safety net to replace the lost income. 2 Thus, the ESI measures insecurity, with higher values signaling more people in a state of economic insecurity. Importantly, the ESI accounts for income losses that reduce available household income (adjusted for inflation, household size, debt payments, retirement assets, and retirement status) as well as total nondiscretionary, out-of-pocket medical expenses (premiums included). The ESI thus incorporates several relevant predictors of economic insecurity in a straightforward manner, is more robust than less sophisticated indices or individual measures that are unable to capture the relationship between resource availability and financial loss, and avoids the complexities and sensitivities that accompany weighted indices generated from multiple indicators (Hacker et al. 2014). The ESI data used in this study were acquired from the ESI (2016) and constructed by Hacker et al. using matched raw data available from the Current Population Surveys dating back to 1986. 3 Given that we lag the measure of economic insecurity in the analyses, our time span ranges from 1987 to 2013 for the contiguous 48 states, resulting in a total of 1,296 possible observations. 4
As expected, the ESI is correlated with the unemployment rate, but the correlation is not particularly strong. The average Pearson correlation within each state (i.e., calculating the within-state correlation over time, and then averaging across states) is +.25. We control for the unemployment rate using data obtained from the Bureau of Labor Statistics, but foresee no issue with identifying the unique contribution of the ESI in the models. Additionally, we capture the degree of state income inequality with the Gini coefficient for each state-year (Frank 2014). The Gini coefficient is a measure of the distribution of personal income for a given state-year and ranges from 0 to 1, with higher values indicating more income inequality. To account for the expansiveness of public welfare programs in each state, we utilize two separate indicators. First, we include a measure of the available support for the unemployed. Specifically, we draw from reports available from the U.S. Department of Labor, Employment and Training Administration, Office of Unemployment Insurance (2017) to calculate the maximum weekly unemployment benefits as well as the maximum duration of these benefits. Following prior work that utilizes these data (Cylus, Glymour, and Avendano 2014), we use the product of the inflation-adjusted amount (in 2013 dollars) and duration to calculate unemployment benefit generosity. Second, we include a measure of the combined monthly maximum of aid to families with dependent children/temporary assistance for needy families (AFDC/TANF) and supplemental nutrition assistance program/food stamp (SNAP/FS) benefits for a four-person family. This measure was obtained from the University of Kentucky Center for Poverty Research’s (2018) National Welfare Data. Finally, we control for the labor force participation rate for each state and year using data from the U.S. Bureau of Labor Statistics.
We control for six additional factors that may be associated with state imprisonment rates. First, Hacker (2008) suggests that insecurity is partly a result of neo-conservative success in American politics, and research cited above connects conservative politics with imprisonment rates. We measure gubernatorial partisanship with a dummy variable coded 1 if the state has a Republican governor and coded 0 otherwise (Klarner 2013). Like other research, we focus on the governor because that office has some control over prison construction, corrections budgets, appointment of parole board members, and setting state law enforcement priorities (Greenberg and West 2001:625). Second, we include Berry et al.’s (1998) updated measure of citizen ideology to account for voter policy preferences. This measure represents the mean location of the state electorate on a continuum that ranges from 0 (most conservative) to 100 (most liberal). Third, we control for the crime rate using data from the Federal Bureau of Investigation’s Uniform Crime Reports. We include a measure for the total crime rate per 100,000 residents, with the assumption that higher crime rates will increase imprisonment rates (Spelman 2009). Total crime rates include murder, rape, robbery, aggravated assault, burglary, larceny-theft, and motor vehicle theft. Fourth, an inflation-adjusted median household income variable derived from Current Population Survey (CPS) data is included in each model. Prior research suggests a positive association between median household income and imprisonment rates because higher incomes increase the tax base, and imprisonment tends to increase when there is money in the state’s coffers to finance it (Jacobs, Malone, and Iles 2012; see Spelman 2009 for related argument about spending and imprisonment). Fifth, we control for the percentage of state residents living in urban areas. This variable is derived from urban and rural population data available from the Integrated Public Use Microdata Series (Manson et al. 2018). Residents in urban areas tend to be less punitive than residents in rural areas (Jacobs et al. 2012). Sixth, we control for the percentage of non-Hispanic black residents in a state using detailed sex, race, and Hispanic origin population data from the U.S. Census. Descriptive statistics are shown in Table 1.
Descriptive Statistics.
Note. “L.” indicates lagged. “Δ” indicates first-differenced. “L.Δ” indicates lagged first-difference. AFDC = Aid to families with dependent children; TANF = Temporary assistance for needy families; SNAP = Supplemental nutrition assistance program; FS = Food stamp.
Estimation
The data are organized by state and year, and the time period of analysis is 1987–2013 after incorporating the lags on the independent variables. Given the nesting of time points within geographic units and the potential correlation between time and other variables, we are attentive to several potential patterns in the data that would make a simple linear regression model problematic. For example, a common issue in time-series analysis is autocorrelation, a condition in which the errors from different points on a time series are serially correlated. If positive autocorrelation exists, then the standard errors are likely underestimated and the risk of type I error increases. Trends in data used to construct some variables of interest (e.g., percentage black and the imprisonment rate) also raise the possibility of nonstationarity, a condition in which the statistical properties of a variable change over time. Some stochastic variation over time is expected and acceptable, so long as the data generally revert back to the central tendency and the variance does not change appreciably. Trending data can also lead to heteroscedasticity, which occurs when the variance of the error term is uneven across values of the independent variables in a model, thus affecting standard errors. Finally, cross-sectional dependence between states can bias results. While the use of year dummy variables in a model effectively controls for economic, policy, and other common factors that similarly affect all states in a particular year, a formal test for cross-sectional dependence is still needed to mitigate concerns about shocks and other factors that influence only a few states or that produce heterogeneous affects across all states, thus violating assumptions of independence and potentially biasing results.
We initially used a two-way fixed effects (i.e., state and year) estimator to assess the association between imprisonment rates and the lagged explanatory variables (see Tables A1 and A2 in the Supplemental Appendix for descriptive statistics and results, respectively). This model allows us to control for time-constant omitted variables that may be correlated with the outcome, such as regional or historical factors that have been shown to have a long-run influence on punishment (e.g., past lynchings; see Jacobs et al. 2005; Jacobs et al. 2012). 5 A test proposed by Wooldridge (2002; see also Drukker 2003), however, detected autocorrelation in the residuals of this model. Furthermore, an inspection of a plot of the residuals over the predicted values revealed some heteroscedasticity, and a postestimation test for cross-sectional dependence proposed by Pesaran (2004) indicated that our observations were correlated across panels. Unit-root tests also showed that our panels were not stationary (Pesaran 2003, 2007; see also Lewandowski 2007; Sangíacomo 2014).
In response to these issues and in keeping with prior time-series scholarship on imprisonment rates (Jacobs and Helms 1996; Sutton 2000), we first-differenced the variables on each side of the equation (save the year dummies). Subsequent diagnostic and postestimation tests indicated that the conversion of these variables from levels to first-differences effectively addresses the issues of serial correlation, cross-sectional dependence, and nonstationarity. An inspection of a residuals-versus-fitted plot generated after the first-difference regression model yielded a proper random cloud of points, alleviating concerns about heteroscedasticity. And despite the moderate degree of correlation between some of our untransformed variables, an inspection of the variance inflation factor (VIF) values for the transformed predictors revealed an acceptable average value of 2.21. Economic insecurity and unemployment rates, two variables of particular interest in this study, had respective VIF values of 1.73 and 3.69. Consequently, we are confident that multicollinearity is not an issue in our first-differenced models (the Stata code and results of these tests are available from the authors on request).
The outcome variable in our first-differenced models is the year-over-year change in the imprisonment rate per 100,000 people (Y t –Yt-1). The independent variables are lagged first-differences (Xt-1–Xt-2) because we continue to assume that changes in the predictors influence the outcome with a short time lag. For example, we expect that changes in economic insecurity between 2010 and 2011 will affect imprisonment rates during the 2011 to 2012 interval. By taking the lagged first-differences of the predictors, we lost data from the first year of our series and the analytic sample for this model effectively became 1988 through 2013 for the contiguous 48 states.
Finally, we estimate one additional model in which we substitute state prison admission rates for the state imprisonment rates. Prison admissions are likely more responsive to contemporary social, political, and economic changes than imprisonment rates because the latter reflects past practices and policy decisions (Pfaff 2008:614). Consequently, results from this model should indicate that the relationship between economic insecurity and prison admission rates is comparable to, or perhaps stronger than, the association between economic insecurity and total imprisonment rates located in prior model results.
Results
We begin with a baseline first-difference model of the imprisonment rate regressed on economic insecurity and year dummy variables (see Table 2, Model 1). As hypothesized, we observe a positive and significant correlation between economic insecurity and state imprisonment rates (b = 1.364). In Model 2, we introduce three additional variables that are theoretically associated with imprisonment rates: crime rate, percent black, and percent residing in urban areas. The crime rate is not significantly associated with changes in imprisonment, which is consistent with other work that analyzes changes in imprisonment (Enns 2014). This does not imply that the crime rate is irrelevant for understanding imprisonment rates, but the coefficient suggests that an increase in crime has a negligible effect on the change in imprisonment rates the following year. The percent black and urbanicity, however, are significant and in the anticipated direction (Model 2). That is, imprisonment rates are positively correlated with the percent black and negatively associated with urbanicity. The coefficient for economic insecurity remains significant when controlling for these variables. As reported in Model 3, we find no evidence that imprisonment rates are associated with our two measures of the political environment, and no meaningful change in the economic insecurity coefficient when adding these measures to the model.
First-Difference Regression Models: Change in Imprisonment Rates per 100,000 Population on Predictor Variables.
Note. “L.Δ” indicates lagged first-difference. All models include year dummy variables (coefficients not shown).
p < .05. **p < .01. ***p < .001 (all tests two-tailed).
Finally, we add indicators of economic conditions and welfare generosity in Model 4. Three findings are particularly noteworthy. First, the coefficient on economic insecurity remains consistent across models and does not appear to be confounded by unemployment or other economic indicators. Each percentage increase in insecurity is associated with an increase of around 1.5 imprisonments per 100,000 state population (b = 1.459 in Model 4). This effect size is modest in magnitude but robust across models. Second, of the five covariates added to Model 4, only welfare generosity is significantly associated with imprisonment rates. Specifically, for each hundred-dollar increase in the monthly maximum for AFDC/TANF and SNAP/Food Stamp benefits, there is a corresponding drop in state imprisonments of 6.2 per 100,000 population. Third, the percent black and percent urban remain associated with imprisonment. Each 0.1 percentage increase in the black population of a state (approximately the standard deviation; see Table 1) is associated with an increase of about 1.2 imprisonments per 100,000 state population. In addition, a percentage increase in urban residents corresponds to reduction in imprisonments of around 5 prisoners per 100,000 state population, although rarely does this variable change by a full percent in a year.
As a further test, we also estimated a model using the change in prison admission rates, which is a common alternative measure of the dependent variable. Prison admissions reflect the flow of sentenced offenders into prison throughout the year rather than the total stock of prisoners in custody at year end. As expected, the economic insecurity index remains positively and significantly associated with prison admission rates (b = 3.385, Table 3), such that a percentage increase in insecurity is associated with almost 3.4 new admissions for every 100,000 state residents. In a state the size of Michigan (10 million), this translates into about 340 additional admissions. The only other variable that is significantly associated with admission rates in Table 3 is median household income. Consistent with prior research (Spelman 2009), an increase in median income is associated with increases in prison admissions per capita (b = .861).
First-Difference Regression Models: Change in Prison Admission Rates per 100,000 Population on Predictor Variables.
Note. “L.Δ” indicates lagged first-difference. All models include year dummy variables (coefficients not shown).
p < .05. **p < .01. ***p < .001 (all tests two-tailed).
Finally, we report the results of a placebo regression model used to test for spuriousness (Wildeman 2010; see Table A3 of the Supplemental Appendix). In this model, the outcome at time t is regressed on the predictor variable at t + 1, with the assumption that a significant correlation would indicate possible spuriousness (because Xt + 1 cannot logically cause Yt). A null association when using a “lead variable” (as opposed to a lagged variable) on the right side of the equation would be consistent with, although not absolute proof of, a causal relationship. In supplementary analyses, we regressed the change in the imprisonment rate on the change in economic insecurity the following year, and as shown in Table A3, the coefficient on the lead variable is no longer statistically significant and the direction is negative. This pattern of results is consistent with what we would expect if economic insecurity is driving some of the change in imprisonment rates.
Discussion and Conclusion
Each state in the country experienced two trends between the mid-1980s and the early part of the last decade: higher imprisonment rates and a larger proportion of persons who felt financially insecure. Garland (2001) made a cogent argument as to why these trends may be related. As he theorized, economic insecurity fosters a sense of anxiety and uncertainty in a population, which in turn makes high imprisonment rates tolerable, if not desired. We built on Garland’s thesis by using a state-level measure of economic insecurity developed by Hacker and colleagues (2014) that is designed to measure risk of loss rather than sheer unemployment. As both Hacker (2008) and Garland (2001) have noted, insecurity is felt by a significant proportion of the employed, and even a large segment of the middle class.
Results of time-series analyses support the hypothesis that increases in economic insecurity are associated with increases in state imprisonment rates. Findings show that imprisonment rates increase following an uptick in our measure of economic insecurity, even when controlling for the crime rate, unemployment, and other economic conditions. The magnitude of the correlation is modest, and we do not argue that economic insecurity was the sole driver of year-to-year changes in imprisonment rates. However, there is a sound theoretical and empirical basis for stating that changes in economic insecurity in a population are consequential for punishment.
We are careful to point out some differences between Garland’s theoretical argument and our empirical case. Garland, like Wacquant (2010), views insecurity and its associated anxieties as staples of late-modern society. However, neither Garland nor Wacquant argue that year-to-year fluctuations in insecurity will push and pull on the prison population. In this sense, our work supports, but also extends, Garland’s thesis to explain short-term fluctuations in imprisonment rates.
In addition to insecurity, we find some evidence that increases in AFDC/TANF benefits are associated with decreases in imprisonment rates, consistent with prior work (Beckett and Western 2001). Likewise, changes in racial demographics are associated with changes in imprisonment rates, as predicted by the criminological adaptation of Blalock’s racial threat theory. We interpret these correlations with caution because they were not robust across model specifications. For instance, AFDC/TANF benefits were not significant in the analysis of prison admissions (Table 3), and percent black was only significant when analyzing changes in imprisonment rates (Table 2). The measure of economic insecurity, however, was the only variable that was significantly associated with imprisonment and prison admissions.
We also draw attention to variables that were not consistently correlated with imprisonment, at least not as predicted by theory. For instance, the coefficient on unemployment was inconsistent with the oft-cited Rusche and Kirchheimer thesis. Sutton (2004) provides several reasons as to why labor market conditions are inadequate for explaining cross-national variation in imprisonment rates, and our work finds no support for this thesis when explaining within-state change. We are careful not to overreach and suggest that labor markets are irrelevant for understanding imprisonment rates, but we suggest that the study of punishment would benefit from considering facets of the economy other than the standby measures of unemployment and inequality (i.e., the Gini coefficient). We suggest that economic insecurity is a viable complement to the more traditional measures of economic conditions.
We are mindful of some limitations in the present research. For one, data availability limited our time series to a period that misses the early years of the mass incarceration era. We would have preferred data for the decades prior to 1986 to assess whether the relationship persists when analyzing a period in which imprisonment rates were not trending upward. Our method of detrending the time series data gives us some confidence that we are not simply capturing coincidental trends, but the inability to capture the 1970s and early 1980s is nonetheless a concern. In addition, we cannot empirically pin down the precise mechanism that accounts for this relationship. We follow Garland’s lead and theorize that insecurity causes a sense of unease, precariousness, anger, and stress in the population (Garland 2001:154–56), which provides the emotional basis for a tough-on-criminals approach to criminal justice. As discussed earlier, some prior work supports the premise that insecurity is associated with punitive attitudes, but related work using different samples and measures of insecurity yields different results (Costelloe et al. 2009; Johnson 2001; Lehmann and Pickett 2017).
In sum, the findings suggest that economic insecurity—a defining feature of late modern society—exerts some influence on imprisonment rates. If replicated in future research, this finding may also have implications for decarceration. Pundits, politicians, and think tank experts of all political stripes acknowledge that change is needed in the American criminal justice system, particularly for imprisonment rates. A plausible implication of our analysis is that remedial punishments may be more palatable to the public when they have a sense of economic security in their lives. On the flipside, it may be difficult to reduce imprisonment rates when large segments of a population feel angst about making financial ends meet.
Supplemental Material
Appendix – Supplemental material for Stressed to the Punishing Point: Economic Insecurity and State Imprisonment Rates
Supplemental material, Appendix for Stressed to the Punishing Point: Economic Insecurity and State Imprisonment Rates by Chad A. Malone and Ryan D. King in Social Currents
Footnotes
Acknowledgements
We are particularly grateful to Peter Enns, David Garland, Joshua Guetzkow, Marianne Ulriksen, and the anonymous reviewers for their feedback.
Authors’ Note
An earlier version of this paper was presented at the Societies under Stress Workshop at The Ohio State University in December of 2018. The authors benefited from many comments and suggestions at this workshop.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research received support from the Department of Sociology at The Ohio State University.
Supplemental Material
Supplemental material for this article is available online.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
