A macro-level perspective on prison inmate deviance

Abstract

A macro-level perspective for understanding differences in levels of inmate assaults and nonviolent offenses across US prisons is presented which emphasizes the greater relevance of population composition as opposed to environmental and managerial controls over inmates. We argue that population effects are more relevant due to a heavy reliance on custodial risk assessments which place high concentrations of the most dangerous offenders within particular facilities or units within facilities, rendering these environments less effective for controlling inmate misconduct regardless of tighter security and greater use of administrative controls relative to lower security environments. A partial test of our model was conducted with a national US sample of 247 state prisons for men and women. Results indicate that while both facility and population factors are relevant predictors of offense levels, population effects are stronger and more prevalent. The implications of specific findings are discussed.

Keywords

inmate assaults prison environment;prison inmates

Introduction

Prison scholars have contributed a variety of perspectives on inmate deviance in efforts to explain individual-level differences in the odds of engaging in crimes and other rule infractions during incarceration. Yet, there is a distinction between understanding differences in the odds of individual deviance versus understanding differences in levels of deviance across prisons. For example, programming might occupy inmates' time while also offering them opportunities to invest in more conventional goals (Colvin, 1992; Huebner, 2003), but the assumption that such activities can effectively suppress levels of inmate deviance remains an empirical question given limited program resources for often large inmate populations.

Here we argue that the composition of inmate populations in the USA is more relevant than managerial and environmental controls for understanding differences in levels of assaults and of nonviolent rule infractions across state prisons. The question of whether population composition matters more than organizational context for shaping the behaviors of its members has been debated for decades among sociologists interested in other social institutions (e.g. Hauser’s (1969) and Hauser et al.’s (1974) studies of whether school environments mattered more than the composition of student bodies for shaping educational outcomes, or Petersen’s (1972) study of whether the context of Danish naval bases mattered more for shaping naval criminality relative to the composition of the seamen assigned to those bases). This question is also applicable to an understanding of inmate misconduct in prison organizations. The idea that inmate population composition might matter more than the prison context is realistic given the heavy emphasis on custodial risk assessments in the USA which serve to place high concentrations of the most dangerous offenders within particular facilities or units within facilities. Perhaps ironically, effective risk assessment may hinder the ability of more secure prisons to effectively control levels of inmate misconduct regardless of the available resources for tighter security and a greater use of administrative controls relative to less secure contexts. In other words, once facilities are populated with the types of offenders for which they are designed, certain levels of offending may be foregone conclusions based solely on the types of offenders housed.

Inmate population composition versus the prison context

Managing large criminal populations via security classification has been a cornerstone of US institutional corrections since the first part of the 20th century. However, risk assessment began to evolve in the 1970s and has since become more efficient at distinguishing between groups of offenders based on their danger to others and to themselves, their risk of recidivism, and their amenability to treatment (Bonta, 1996; see also Andrews and Bonta, 2010, for a review of the evolution of risk-need assessments and the importance of incorporating “dynamic” predictors into actuarial assessments in the 1980s). During this era of dramatic improvements in risk assessment, the US prison population rose from roughly 300,000 in 1980 to 1.1 million by the mid-1990s (Sentencing Project, 2012). Thus, an improved ability to “correctly” classify more dangerous offenders coupled with a nearly exponential increase in prison populations generated larger concentrations of more dangerous offenders within more secure facilities. Of course, these same processes also generated larger concentrations of less dangerous offenders in less secure facilities.

Placing higher risk offenders in more secure facilities in order to more effectively reduce threats to prison and public safety is premised in the assumption that prison environments can be constructed in order to effectively achieve this mission. Although these prison environments are by no means chaotic (which they could be in the absence of any effective prison management), it remains to be seen whether more secure environments generate lower levels of inmate misconduct (or at least comparable levels) compared to less secure environments. In other words, whether the context of the prison environment is more relevant than the composition of the inmate population for shaping the level of crime within a facility.

A sociological perspective on the subject would necessarily favor the argument that organizational environment/context matters more for shaping the behaviors of individuals relative to the composition of the individuals who make up the organization. However, the empirical literature testing this perspective for various social entities (e.g. schools, neighborhoods) includes a mix of findings depending on the entity as well as on the sample. For example, Hauser (1969) uncovered evidence favoring the idea that the composition of student bodies in grades 7–12 (particularly students’ IQs and fathers’ education levels) was far more relevant for predicting students’ scores on standardized math and reading tests compared to the actual schools attended. As another example, Galster and Hesser’s (1981) study of residents’ satisfaction with their neighborhoods revealed only a few significant neighborhood contextual effects versus stronger effects of household population characteristics on average levels of “satisfaction”. A review of the empirical literature on contextual versus compositional effects on health risk prior to 2000 (Pickett and Pearl, 2001) revealed modest neighborhood effects and much stronger compositional effects (e.g. population SES) across studies. Similarly, Veenstra (2005) found stronger compositional effects on health outcomes (e.g. income, levels of trust in government) in British Columbia.

On the other hand, and more specific to crime, Petersen’s (1972) study of criminality among Danish seamen placed at different naval bases across the country produced evidence of contextual effects in that crime rates varied significantly across naval bases despite similar populations of seamen, although he recognized that some of these differences might have relied on interactions between population composition and environmental context. Liska and Chamlin (1984) found significant differences in arrest rates across US cities that corresponded fairly closely with the racial/economic composition of residential populations. Miethe and McDowall (1993) found significant neighborhood (contextual) effects on victimization risk in Seattle although these effects paled in comparison to compositional effects based on the crime prevention behaviors of residents. Closer to the topic of how criminal justice organizations can influence crime, Sampson (1986) found significant contextual effects of jail incarceration rates, imprisonment rates, and public order crime arrest rates on robbery and/or homicide rates, controlling for compositional differences in populations based on age, race, and sex. These types of studies are important for assessing the feasibility of controlling outcomes by manipulating social environments, and this is a particularly salient issue for crime control within prison organizations.

Misconduct levels as artifacts of population composition

A compositional perspective challenges the assumption that inmate misconduct levels can be controlled by prison officials. Although it seems intuitive that a greater focus on management and control might reduce misconduct levels if inmates are kept on a “shorter leash” during confinement (DiIulio, 1987), there is no empirical evidence that decreasing rates of officially recorded assaults in US prisons since the 1980s are due to more effective prison policies, despite what some descriptive analyses suggest (e.g. Association of State Corrections Administrators, 2010; Useem and Piehl, 2006). For instance, this trend could have resulted from changes in how officers use their discretion in more crowded facilities, or from increases in the numbers of less serious offenders sent to prison (increasing the population base used in the calculation of these rates).

Improvements in custodial risk assessment coupled with the “get tough” movement beginning in the 1970s (which reduced further the already modest emphasis on rehabilitation in US prisons) may have only served to shift portions of misconduct into particular facilities or units within facilities. Irwin (2005) described how the state of California constructed large warehouse prisons for “low risk” inmates and supermax prisons for “high risk” inmates. He observed that violence is relatively rare in the warehouse facilities while violence rates are three to four times higher in the supermax facilities (Irwin, 2005). This idea is the gist of our argument that inmate population composition is more relevant than formal controls for understanding between-prison differences in offending levels in the USA.

A parallel can be drawn between perspectives on neighborhoods and crime and the process of creating more dangerous prisons by shaping inmate populations. Higher concentrations of high risk inmates found in particular types of prisons are analogous to larger concentrations of individuals who are higher risk to offend that reside in particular types of neighborhoods. The convergence of large numbers of socially and economically disadvantaged residents in certain neighborhoods might contribute to cognitive landscapes where mainstream values have become attenuated which, in turn, further isolates those groups from conventional society and contributes to crime in those locales (Krivo and Peterson, 1996; Sampson and Bean, 2006; Sampson and Wilson, 1995; Wacquant, 2001; Warner, 2003). Informal controls are less effective for reducing crime when the strength of conventional values is weaker (Kornhauser, 1978; Sampson and Bean, 2006; Warner, 2003). Formal controls are also heavier in more crime-ridden communities (e.g. heavier police patrols) as well as in prisons for higher risk offenders (e.g. higher officer to inmate ratios). In other words, both situations characterize more crime-prone environments where larger numbers of individuals with deficits in social and human capital, whose mainstream values may have been weakened, are in relatively close physical proximity to one another. Moreover, the huge numbers of minorities now held in US prisons, constituting 60 percent of the inmate population (Sentencing Project, 2012), have created a migratory loop for high risk offenders between the most economically disadvantaged urban neighborhoods, where these individuals most often reside, and maximum security units/prisons, where these same individuals are more often sent upon conviction. The highest levels of crime exist within these most “severe” environments, perhaps a foregone conclusion of parallel lifestyles inside and outside of prison (see also Wacquant, 2001).

Administrative controls and misconduct levels

Aspects of administrative control theory (DiIulio, 1987) are relevant to the perspective that a greater focus on management and control should reduce misconduct levels. These aspects include the roles of supervision, segregation, and program resources in prison. Administrative control theory predicts that inmate deviance results from poor facility management (DiIulio, 1987; Useem and Kimball, 1989). Facilities with higher levels of supervision, segregation, and so on might have lower misconduct levels because these factors help to promote “good order” within a facility (Bottoms, 1999; DiIulio, 1987; Sparks et al., 1996; Steiner, 2008). Aside from formal controls that are more coercive in nature (e.g. segregation), Colvin (1992) discussed the relevance of remunerative controls (e.g. program participation and paid jobs) and advocated for their greater use because they offer incentives for inmates to comply with facility rules without damaging their dignity and without feeding their cynicism toward authority (see also Colvin, 2007). Remunerative controls tend to vary inversely with security level but are typically available to some degree at each level except death row.

From this perspective, misconduct levels might be influenced by environmental controls (security, supervision, expenditures, and punishment philosophies), coercive controls (disciplinary and administrative segregation), and remunerative controls (programs and jobs). We separate environmental controls from managerial (coercive and remunerative) controls because the former are likely to influence the latter and, therefore, may have both direct and indirect effects on levels of inmate misconduct.¹

An empirical model of violent and nonviolent misconduct levels

Our perspective is that inmate misconduct levels in the USA are influenced primarily by the composition of the inmate population (demographics, social demographics, and criminal histories) rather than by facility controls. Environmental and managerial controls might still be relevant for shaping misconduct levels, although these effects might be substantively weaker than population effects. As such, population composition might influence inmate misconduct levels both directly and indirectly via prison context (i.e. population risk influences the level and types of control exerted over the population which, in turn, may influence misconduct levels). Central to this argument is the assumption that causality moves from population to facility characteristics. Offenders are placed into certain facilities based on their risk evaluation, so facility features are dictated by the types of inmates housed. More secure facilities “draw” more violent offenders, for example. Placing high risk offenders into maximum security prisons, however, does not mean that security level necessarily becomes the primary effect on misconduct.²

A partial test

Findings from a partial test of our perspective are described below in order to encourage further discussion and research. This was a “partial” test because the data examined were not compiled specifically for this purpose and more rigorous measures are required for a more reliable test.

Population factors relevant to our perspective might include sex composition, age, race and ethnicity, education, marital status, employment prior to incarceration, histories of sexual and other forms of physical abuse, drug abuse, and violent versus nonviolent offenders. Scholars have demonstrated significant links between crime rates and residential populations with more youthful age structures and higher proportions of minorities, non-married residents, unemployed adults, and adults with less education (e.g. Land et al., 1990; Messner and Sampson, 1991; Morenoff et al., 2001; Peterson and Krivo, 2010; Sampson et al., 1997). Rule violations may be more common among more disadvantaged inmates if they bring their ecologically structured beliefs regarding legal authority and deviance into prison (Harer and Steffensmeier, 1996; Irwin, 1980; Irwin and Cressey, 1962). Even if these beliefs become latent once offenders are pulled from their environment, they may eventually influence inmates’ behaviors if these beliefs are relevant to solving problems posed by prison environments (Steiner and Wooldredge, 2009).

The corrections literature also underscores consideration of substance abuse, violent offending, and prior violent victimizations as influences on offenders’ future behaviors (e.g. Miller and Miller, 2010; Mosher and Phillips, 2006; Ostermann, 2009). A greater prevalence of these traits among prison inmates might reflect more volatile populations with lower levels of self-control.

Factors relevant to an administrative control perspective include “environmental controls” (a facility’s physical security, levels of supervision over inmates, expenditures, and state sentencing policies), “coercive controls” (levels of segregation), and “remunerative controls” (paid work assignments, jobs outside the facility, and program availability including academic education, vocational training, and counseling). Regarding outside jobs, incentives to follow rules might be stronger when inmates work apart from the more institutional setting. Conversely, inmates without these jobs might grow weary of the more authoritative environment inside prison.

Individual-level studies have uncovered lower odds of misconduct among inmates in education programs (Lahm, 2009; cf. Gendreau et al., 1985), who completed drug or alcohol treatment in prison (Innes, 1997; Langan and Pelissier, 2001; cf. Welsh et al., 2007), and among inmates in treatment-oriented facilities (Dietz et al., 2003; Prendergast et al., 2001). Again, such individual-level effects may not have aggregate-level equivalents. On the other hand, evidence that a greater use of remunerative controls is significantly linked to misconduct levels has been found by Camp et al. (2003), Huebner (2003), Steiner (2009), and Useem and Reisig (1999). Anecdotal evidence on the role of supervision for reducing misconduct rates has been offered by Bottoms (1999) and DiIulio (1987).

The severity of state sentencing policies may be relevant if tougher policies reflect a more punitive ideology shared by state government officials, with greater emphases on deterrence and incapacitation as opposed to rehabilitation. From an administrative control perspective, more punitive ideologies might promote tighter controls over inmates, in turn reducing levels of offending during confinement. On the other hand, more punitive ideologies at the state level might be a response to higher crime rates and higher risk offender populations, suggesting that population risk might ultimately influence facility misconduct levels if greater punitiveness does not compensate for higher risk populations. The specific policies examined are described in the next section.

Based on the above discussion, we predicted that misconduct levels would be driven primarily by the direct effects of population composition and, to a much lesser extent, by the direct effects of environmental and managerial controls as well as the indirect effects of population composition via these types of controls. Important to recognize, however, is that between-facility differences in offending levels may be smaller in states with facilities housing all security levels on the same property.

Methods

A national sample of 247 state-operated prisons for adults across the USA was examined. Data were obtained from the 1997 Survey of Inmates in State and Federal Correctional Facilities (ICPSR 2598) and the 1995 and 2000 Censuses of State and Federal Adult Correctional Facilities (ICPSR 4021 and ICPSR 6953, respectively). All data sets were compiled by the United States Bureau of the Census on behalf of the Bureau of Justice Statistics. The Survey data were self-report data collected from random samples of inmates (as units of analysis) whereas the Census data were official data compiled from prison administrators (with facilities as units). We aggregated the Survey data across prisons and merged it with the Census data to create the macro-level data set examined here. (Some of the Survey data are “restricted” and were obtained with permission from ICPSR.) Our reliance on the Survey data meant that the sample of prisons examined here included only the prisons from which inmates were sampled for the 1997 Survey of Inmates.

The sampling design for the Survey involved a stratified two-stage selection, with facilities selected at the first stage followed by inmates selected randomly from those facilities. The sampling frame for the first stage included all correctional facilities in the USA (N = 1409 state and federal facilities housing men and/or women, including residential treatment centers). The 13 largest confinement facilities for men were selected with certainty while the remaining facilities were separated into strata defined by region of the country and facility type, and were ordered by security level and inmate population size within each stratum. The sample of facilities drawn by the US Bureau of the Census included 271 prisons relevant to our focus on state-operated confinement facilities (223 state-operated confinement facilities for men and 48 such facilities for women). We excluded all federal facilities in addition to boot camps, work release centers, residential treatment facilities, and drug dependency facilities.³ Due to our interest in aggregating the inmate survey data to the prison level, we wanted facilities with at least 40 sampled inmates in order to generate reliable estimates of the prison factors derived from the Survey data. The total number of inmates originally selected for interviews in each facility was based on the size of the inmate population, and a random sample of inmates was chosen from each facility based on a list of all inmates in the facility who occupied a bed the previous night. Twenty-four of our 271 facilities of interest included samples smaller than 40 inmates based on sampling proportionate to the size of the inmate population, thereby reducing the sample to 247 facilities (203 for men and 44 for women, which reflects the same ratio of facilities for men versus women in the sampling frame). Survey data compiled for over 12,000 inmates were used to construct aggregate measures for these 247 facilities. These facilities were spread across the 40 states with the largest inmate populations. The sample included five supermax prisons, and roughly 38 percent of the facilities overall were classified as maximum/close/high security.

Based on the complex sampling design, the sample of prisons examined here (reflecting the facilities included in the 1997 Inmate Survey) was weighted for the analysis. Each prison was weighted inversely to its probability of selection for the 1997 Survey. Prison weights were not available in the original survey data but were created using the detailed information provided in the codebook on how facilities were selected based on size, region of the country, and facility type. Weights were normalized so that sample size would not differ from the pool of 247 facilities.

Measures

The observed measures and corresponding data sources are described in Table 1. Figure 1 displays our theoretical model including both latent and observed variables. Most observed variables were transformed either because of heavily skewed distributions or because the scales were bounded between 0 and 1 (Fox, 2008). These transformations are described in the Appendix.

Figure 1.

Proposed macro-level model of inmate misconduct.

Table 1.

Description of variables (n₁ = 247 state correctional facilities; n₂ = 40 states)

Variables	Label	Mean	SD
Offense levels
prop. inmates assaulted other inmates^a	ASSAULTS	.15	.12
prop. inmates committed nonviolent offenses^a	NONVIOLENT	.45	.18
Environmental controls
Less punitive orientation^c	< PUNITIVE
sentencing policy “discretion”	DISCRETION	23.52	13.51
annual costs (dollars) per inmate	COSTS	31,270	13,022
Security	SECURITY
prop. of pop. housed in maximum security/custody^b	MAXSEC	.20	.29
ratio correctional officers to inmates^b	CO2INM	.23	.12
prop. of pop. in single, one-person cells^b	SINGLCELL	.16	.27
Prison capacity
design capacity – inverse^b	LOWERCAP	.002	.002
Administrative controls
Remunerative-1	REMUN1
# education and treatment programs^b	PROGS	7.91	2.12
prop. inmates in paid work^a	PAIDJOBS	.40	.32
Remunerative-2	REMUN2
prop. inmates with outside work assignments^a	OUTJOBS	7.91	2.12
prop. population in any job^b	ANYJOBS	.40	.32
Coercive
prop. pop. in administrative and punitive segregation^b	TOTALSEG	.05	.08
Inmate population composition
Lower risk-1	LOWER RISK-1
prop. male^b	MALE	.82	.38
prop. w/o histories of sexual/other physical abuse^a	NOABUSE	.79	.19
Lower risk-2	LOWER RISK-2
prop. incarcerated for non-violent crimes^a	INCNONVIOL	.46	.19
prop. no drug use month before arrest^a	NODRUGS	.44	.09
Lower risk-3	LOWER RISK-3
mean age^a	AGE	33.54	3.24
prop. married^a	MARRIED	.17	.08
prop. employed before prison^a	EMPLOY	.65	.12
prop. with high school degree^a	HS	.58	.16
prop. white^b	WHITE	.37	.17
Control variable
mean time (# months) served^a	TIMESERVED	48.54	23.08

Notes: ^aprison-level measure derived from inmate survey data; ^bprison-level measure derived from facility census data; ^cstate-level measures with data sources cited in text.

The observed endogenous variables included the proportion of each facility’s inmates who engaged in physical assaults on other inmates, and the proportion of inmates who engaged in nonviolent offenses aside from drugs, both aggregated from the inmate survey data. Nonviolent offenses included “crimes” (e.g. theft, property damage) as well as noncriminal infractions (e.g. refusing to obey staff).⁴ These proportions were transformed into logits due to their positively skewed distributions (Fox, 2008). The transformed distributions were approximately normal based on the Kolmogorov-Smirnov (K-S) test of normality (p > .10).

Exploratory factor analyses (EFA) in MPLUS 6.0 (Muthen and Muthen, 1998–2010) determined the latent variables displayed in Figure 1. The first EFA included the nine observed measures of “population composition” described in Table 1. VARIMAX (orthogonal) rotation was explored in order to generate uncorrelated factors, but GEOMIN (oblique) rotation was used instead because VARIMAX produced high loadings for one of the measures on two factors. Three latent variables emerged using GEOMIN rotation (CFI = .982), including “lower risk-1” (male inmate population and the proportion of inmates without histories of physical abuse), “lower risk-2” (proportions of inmates incarcerated for nonviolent crimes and with no drug use one month prior to incarceration), and “lower risk-3” (mean age, and the proportions of inmates employed at arrest, with high school diplomas, married, and who are white). Higher values on all three scales reflect lower risk populations.

The second EFA included the six observed measures of “environmental controls” (see Table 1). ML extraction with VARIMAX rotation was used to generate uncorrelated factors. Three factors initially emerged from the analysis but one of the three factors reflected only one of the observed variables (i.e. a facility’s design capacity). This variable was removed from the EFA and included in the model as an observed measure of environmental controls. (Our preference for examining facility size instead of crowding stems from more consistent findings across studies regarding the relevance of facility size for predicting misconduct levels (Wooldredge and Steiner, 2010). Findings for crowding have been mixed with only modest support for its relevance (e.g. Lahm, 2008).) The remaining five variables were factor analyzed and produced two factors (model fit χ ² = 1.8; p = .17; CFI not produced with VARIMAX rotation). These latent variables reflected a state’s “less punitive” orientation toward offenders (less severe sentencing policies and average annual costs per inmate), and prison “security” (portions of inmates housed in each of maximum security/custody and single-cells, and the ratio of officers to inmates).

The final EFA included four indicators of “remunerative controls” (see Table 1). Two latent variables emerged with GEOMIN rotation (CFI = .95), which was used instead of VARIMAX due to high loadings for one of the measures on two factors. The first latent construct includes the number of available prison programs and the proportion of inmates in paid jobs, whereas the second construct reflects the proportions of inmates with work assignments outside the facility and inmates with any job. The second latent variable might correspond with minimum security facilities for inmates serving the last year or less of their sentences (not residential treatment centers), where nearly every inmate works and each inmate might work several days per week in the community. Regarding an indicator of “coercive controls”, a separate measure of the proportion of inmates held in administrative segregation and disciplinary segregation was also included in the model as an observed variable tapping “coercive controls”.

The index of a state’s “less punitive” orientation toward offenders is a state rank based on six policies and practices. The first policy is whether a state had mandatory sentencing guidelines in 1997, reflecting the year of the inmate surveys (Steiner and Wooldredge, 2008). This three-point scale distinguished between states without mandatory guidelines (1), states with mandatory guidelines implemented after the 1980s (2), and states with mandatory guidelines implemented during the 1980s (3), because earlier schemes were more rigid. The second policy is whether states followed truth-in-sentencing in 1997, with higher ranks for states requiring completion of higher percentages of prison terms stated in court (Ditton and Wilson, 1999). This four-point scale was defined as less than 50 percent of the stated sentence served (1), exactly 50 percent (2), 85 percent (3), and 95–100 percent (4). The third policy was whether states abolished parole by 1997 (Steiner and Wooldredge, 2008), measured as a binary (yes/no) variable. The fourth policy tapped three-strikes legislation, measured as the number of defendants convicted under three-strikes laws in 1997 (Turner et al., 1999). The fifth policy captured the number of mandatory minimum prison sentences for six offender/offense groups including habitual offenders, narcotics offenses, repeat violent offenders, sex offenders, crimes involving guns, and drunk driving (ranging from 0–6) (Austin, 1996). Finally, states were ranked in ascending order on their incarceration rates per 100,000 residents during 1997 (Gilliard and Beck, 1998). Each of these scales was standardized and all of the standardized scales were summed to create a unique score for each state. These scores were then ranked in descending order so that higher values reflected more discretion and/or less punitive orientations. Although some of these laws were implemented in order to enhance the structure of legal decision making, they can also be construed as harsher laws relative to those replaced (Tonry, 1996).⁵

Models included a statistical control for average years served by inmates per facility because the outcomes were based on offenses committed since admission. It is possible that less “transient” populations were associated with higher offense levels because inmates had more time to commit these offenses.⁶

Statistical analysis

Toward the end of evaluating the direct and indirect effects displayed in Figure 1, we used multi-level structural equation modeling with robust weighted least squares estimation (available in MPLUS 6.0). Multi-level modeling was used in order to “nest” the sample of prisons within their respective states. Nesting prisons within states for the analysis is consistent with the stratified random sampling design for the original Survey of Inmates. This procedure adjusts for possible positively correlated error across prisons within each state that can result from unmeasured state-level differences that make prisons within the same state “behave” more similarly to each other (perhaps due to prison administrators in the same state sharing similar philosophies regarding use of segregation and leniency in handling inmate misconduct, or because of subjection to the same level of oversight by state departments of corrections).

Multi-level modeling also allowed us to partition within-state variance in each outcome from between-state variance while allowing the state-level effects (annual costs per inmate and sentencing practices) to be estimated with the appropriate units of analysis (n₂ = 40). This approach should have provided more reliable estimates of state-level effects (annual costs per inmate and sentencing practices) due to the smaller numbers of states relative to prisons and the uneven “spread” of state-level values across prisons based on unequal numbers of facilities across states. Grand mean-centering the prison-level effects permitted estimation of state-level effects while controlling for between-state differences in the prison factors examined (e.g. some states have more prison beds compared to other states which may, in turn, be linked to state punitiveness) (Raudenbush and Bryk, 2002).⁷

The use of structural equation modeling eliminated potential problems with multicollinearity due to the analysis of latent variables, each composed of significantly correlated observed variables. Covariances between latent variables not causally linked in Figure 1 did not reveal any problems, and the analyses did not indicate (multi)collinearity in either model.

Results

The multi-level baseline models revealed no significant differences in inmate assault rates across the 40 states (p > .10; intra-class correlation ρ = .015), but significant between-state differences in nonviolent offense rates (p < .001; ρ = .16). This implies that the nested models were most important for properly specifying prison- and state-level effects on the proportions of inmates who engaged in nonviolent offenses during incarceration.

Differences in assault levels across facilities

Table 2 displays total (zero-order), direct, and indirect effects of the latent variables on prison assault levels. Our primary interests were whether environmental controls became weaker when compositional factors were controlled, and whether the indicators of administrative controls served to buffer the effects of population composition and/or environmental controls.

Table 2.

Magnitude of within- and between-state effects on assault levels broken down by total, direct, and indirect effects

					Indirect via…
					Environmental controls						Administrative controls
	Total		Direct		< PUN.		SECURITY		LOWERCAP		REMUN1		REMUN2		TOTALSEG
Regressor	b	s.e. _b	b	s.e. _b	b	s.e. _b	b	s.e. _b	b	s.e. _b	b	s.e. _b	b	s.e. _b	b	s.e. _b
Inmate pop. composition(latent)
LOWER RISK-1	–3.04***	.76	–.04***	.01	–.00	.00	.00	.00	–.00	.00	.00	.00	.00	.00	.00	.00
LOWER RISK-2	–.56***	.09	–.06	.11	.00	.01	–.06	.05	–.09**	.03	.00	.00	–.00	.01	.02	.02
LOWER RISK-3	–.29	.30	–.46***	.09	.04	.02	–.02	.02	.03	.04	.00	.01	–.00	.03	.01	.02

Environmental controls
< PUNITIVE	–.10*	.05	–.14**	.05	–	–	–	–	–	–	.00	.00	.00	.03	.00	.00
SECURITY	.72***	.15	.09	.10	–	–	–	–	–	–	.00	.00	.00	.05	.11*	.05
LOWERCAP	–.30***	.07	–.10	.06	–	–	–	–	–	–	.01	.02	–.00	.04	–.03	.02

Administrative controls
REMUN1	–.06	.06	–.04	.04	–	–	–	–	–	–	–	–	–	–	–	–
REMUN2	–.21*	.08	–.05	.08	–	–	–	–	–	–	–	–	–	–	–	–
TOTALSEG	.26***	.04	.09*	.04	–	–	–	–	–	–	–	–	–	–	–	–

Control variable
TIMESERVED	.44***	.04	.25*	.10	–	–	–	–	–	–	–	–	–	–	–	–

Root Mean Square Error of Approximation (RMSEA) = 0.049

Proportion of within-state variance explained by prison factors = .68

Proportion of between-state variance explained by prison and state factors = .04

Notes: ML estimates derived from MPLUS 6.0; prisons weighted for the analysis with weights normalized; ***p < .001, **p < .01, *p < .05.

Comparisons of the total versus direct effects from Table 2 provide mixed support for our predictions. Evidence favoring our perspective includes the significant total effects of prison security and prison capacity, both of which dropped in magnitude and were rendered nonsignificant when controlling for population composition, and the significant direct effects of two of the three compositional variables. Moreover, the effect of lower risk-3 was by far the strongest effect in the model (even relative to time served). Controlling for population composition therefore rendered nonsignificant effects of prison security and capacity on assault levels, but did not weaken the effect of less punitive practices at the state level. Even so, the significant inverse direct effect of less punitive practices indicates that less punitive sentencing practices corresponded with lower assault levels, thus refuting the idea that assault levels are lower with more punitive practices applied to offender populations.

Regarding the findings for administrative controls, one of the two latent constructs of remunerative controls as well as the proportion of prisoners held in administrative and punitive segregation (coercive controls) maintained significant total effects on assault levels, but only remunerative controls were rendered nonsignificant when controlling for population composition. However, note the nonsignificant total effect of the first latent construct of remunerative controls (also nonsignificant in the multivariate model), in addition to the counterintuitive (positive) direct effect of segregation on assault levels. The magnitude of the segregation effect became considerably weaker and less powerful when controlling for population composition. Although the total inverse effect of remunerative controls is consistent with an administrative control perspective, the nonsignificant direct effect is consistent with our perspective that population composition is more relevant for predicting assault levels. Also, a greater use of coercive controls actually coincided with larger proportions of inmates who engaged in assaults, which counters the administrative control perspective.

The direct effect of prison security on assault levels was nonsignificant, as predicted, but even the significant total effect of this variable is inconsistent with an administrative control perspective because higher assault rates coincided with higher security levels. Considering the direct effects of all six latent constructs of environmental and administrative controls, none is consistent with the idea that more punitive practices at the state level and greater control over inmates at the facility level generate lower levels of inmate assaults once the composition of inmate populations is controlled. When considering significance in conjunction with the magnitude of these effects, the magnitude of lower risk-3 suggests that population composition is more relevant than both environmental and administrative controls for shaping assault rates, and even the significant effects of coercive controls and less punitive practices at the state level are inconsistent with administrative control theory.

The inverse direct effect of a state’s punishment orientation suggests that the prison populations of less punitive states are lower risk in general. Although somewhat of an overgeneralization, this can be seen when comparing states such as Massachusetts, Nebraska, Vermont, and Wyoming to California, the District of Columbia, Illinois, and Florida. This necessarily begs the question of whether the less punitive nature of certain governments is merely a reflection of having to deal with less dangerous criminal populations.

The indirect effects provided additional insight into the relative importance of each group of predictors. The analysis revealed only two significant indirect effects in the entire model, one of which further underscores the greater relevance of population composition and the other which refutes the importance of environmental controls for reducing assault levels. Favoring our emphasis on compositional effects, lower risk offenders (lower risk-2, including those incarcerated for nonviolent crimes and those who did not use drugs within one month prior to incarceration) were more often placed in facilities with lower capacities which, in turn, experienced lower assault levels (b = –.09; p <.01). Keeping in mind that the direct effect of prison capacity was nonsignificant, this suggests that population composition may have driven a sizeable portion of the significant and inverse total effect of prison capacity on assault levels. The significant indirect effect of lower risk-2 also means that all three latent constructs of population composition were significant predictors of assault levels (i.e. the significant direct effects of lower risk-1 and -3 as well as the significant indirect effect of lower risk-2). The significant indirect effect suggests that while these particular population characteristics may not have direct bearing on shaping assault levels, they may become more relevant for suppressing violence in smaller facilities.

The only other significant indirect effect displayed in the analysis of assault levels involved prison security, which maintained an indirect effect via coercive controls. However, the direction of the (positive) effect counters an administrative control perspective, where greater security corresponded with greater use of punitive and administrative segregation which, in turn, coincided with higher assault levels.

Worth noting are the significant compositional effects on remunerative controls that were opposite to the predicted directions (not displayed here). Facilities with proportionately more “lower risk” inmates coincided with less use of remunerative controls. These inverse effects might be explained, in part, by the lower demands for certain types of programs (special education, life skills, etc.) for lower risk populations. Regardless, remunerative controls did not mediate any of the population effects on assault levels.

Overall, population composition factors were more relevant as both direct and mediating effects on assault levels relative to the other variables in the model. The null indirect effects of population composition on assault levels via administrative controls also suggest that the relatively strong compositional effects on assault levels were primarily direct rather than indirect. Assault levels were determined primarily by the direct and indirect effects of population composition, followed by the weaker direct effects of less punitive practices at the state level and coercive controls, and finally the indirect effect of prison security via coercive controls. However, the directions of the last three significant effects (everything not involving population composition) refute the directions predicted under an administrative control perspective.

Differences in nonviolent offense levels across facilities

Findings for the total and direct effects on nonviolent offense levels (Table 3) were very similar to those from the analysis of assault levels regarding the significance and relative magnitude of these effects (although a couple of important differences are noted below). The general conclusions are even more similar: nonviolent offense levels were shaped primarily by the direct and indirect effects of population composition whereas less punitive practices at the state level and coercive controls at the prison level maintained significant but weaker effects (relative to population composition) that were in the opposite directions to predictions following administrative control theory. Generally speaking, environmental and administrative controls are not operating to actually reduce assault and nonviolent offense levels, a likely consequence of the composition of offenders being placed into facilities with more of these controls.

Table 3.

Magnitude of within- and between-state effects on nonviolent offense levels broken down by total, direct, and indirect effects

					Indirect via…
					Environmental controls						Administrative controls
	Total		Direct		< PUN.		SECURITY		LOWERCAP		REMUN1		REMUN2		TOTALSEG
Regressor	b	s.e. _b	b	s.e. _b	b	s.e. _b	b	s.e. _b	b	s.e. _b	b	s.e. _b	b	s.e. _b	b	s.e. _b
Inmate pop. composition (latent)
LOWER RISK-1	–1.87***	.50	–.02	.01	–.00	.00	.00	.00	.00	.00	.00	.00	–.00	.00	.00	.00
LOWER RISK-2	–.64***	.08	–.21*	.10	.00	.01	–.03	.07	.02	.06	–.01	.01	.01	.02	.03	.03
LOWER RISK-3	–.36	.22	–.34*	.14	–.05*	.02	–.01	.02	–.01	.02	–.00	.01	.03	.03	.01	.02
Environmental controls
< PUNITIVE	–.17*	.08	–.16**	.06	–	–	–	–	–	–	.00	.00	–.04	.04	.00	.00
SECURITY	.67***	.16	.04	.14	–	–	–	–	–	–	.00	.00	–.07	.08	.18	.10
LOWERCAP	–.25***	.06	.03	.08	–	–	–	–	–	–	.03	.02	.05	.05	–.05*	.02
Administrative controls
REMUN1	–.09	.10	.10	.07	–	–	–	–	–	–	–	–	–	–	–	–
REMUN2	–.21	.11	.09	.10	–	–	–	–	–	–	–	–	–	–	–	–
TOTALSEG	.26***	.06	.14**	.05	–	–	–	–	–	–	–	–	–	–	–	–
Control variable
TIMESERVED	.53***	.07	.32***	.10	–	–	–	–	–	–	–	–	–	–	–	–

Root Mean Square Error of Approximation (RMSEA) = 0.052

Proportion of within-state variance explained by prison factors = .57

Proportion of between-state variance explained by prison and state factors = .55

Notes: ML estimates derived from MPLUS 6.0; prisons weighted for the analysis with weights normalized; ***p < .001, **p < .01, *p < .05.

Although the total effects of all predictors on nonviolent offense and assault levels were the same in terms of statistical significance and relative magnitude, lower risk-2 maintained a significant direct effect on nonviolent offense levels whereas lower risk-1 did not (in contrast to the analysis of assaults). These were the only differences in direct effects between the two models, and they did not alter the general conclusion regarding the importance of population composition.

The two significant indirect effects in each model also differed. One of these effects involved population composition, as in the analysis of assault levels, but for lower risk-3 (which also maintained a significant direct effect on nonviolent offense levels). It appears that inmate populations with larger portions of individuals with potentially greater commitment to status quo values (older, white, employed prior to arrest, with high school diplomas, and married) are more common to states with less punitive practices which, in turn, coincide with lower nonviolent offense levels. This finding suggests that the “punitiveness” of particular state governments may be driven, in part, by the dangerousness of offender populations. States with lower crime rates might adopt less severe sentencing strategies while experiencing higher per diem costs due to lower prisoner populations, yet another manifestation of the impact of population composition on levels of inmate misconduct.

The second significant indirect effect involved a link between prison capacity and nonviolent offense levels via coercive controls. That is, facilities with lower capacities house smaller portions of their populations in segregation, and facilities with proportionately fewer inmates in segregation also experience lower levels of nonviolent offenses (i.e. an inverse effect of lower capacity on total segregation, and a positive effect of segregation on nonviolent offense levels). Although this is a different indirect effect from the model of assaults, it still counters the idea that greater control coincides with lower offense levels.

Discussion and implications

The findings from this study reflect only a partial test of our thesis, and more research is necessary before concluding that population composition is more relevant than other factors for shaping inmate offense levels across US prisons. Even so, from a theoretical perspective, the theme of our findings is consistent with a broader theme that has emerged across studies of different environmental contexts that compositional effects are more relevant than contextual effects for shaping behavioral outcomes at the aggregate level (e.g. Hauser’s (1969) and Hauser et al.’s (1974) analyses of the impact of student backgrounds versus school quality on students’ performance on vocabulary and math tests; Pickett and Pearl’s (2001) and Veenstra’s (2005) studies of neighborhood versus resident compositional effects on health risk; and Miethe and McDowall’s (1993) analysis of neighborhood versus population effects on victimization risk).

From a policy perspective, similar results in future studies might suggest a need to re-visit classification decisions based on risk assessments that ultimately shape inmate populations in particular facilities and potentially feed offense levels in those facilities. More effective risk assessment is important for both the safety and rehabilitation of prisoners, but there might be some balance in population characteristics that can be achieved without sacrificing safety and “reform”. For example, slight adjustments to the range of inmate custody scores for particular facilities and units within facilities might be monitored to examine the consequences of broadening these ranges. More specific to our findings, the rapid growth in inmate populations has resulted in noticeable differences across prisons in the proportions of inmates incarcerated for violent crimes and those with less social capital. In Ohio, for example, the proportion of inmates incarcerated for violent crimes in 2008 ranged from .04 to .72 across prisons (Steiner, 2008). Also during 2008, the proportion of inmates without high school degrees ranged from .38 to .79; the proportion not married from .69 to .91; the proportion that used drugs within one month immediately prior to incarceration from .17 to .79, and so on (Steiner, 2008). Attention might be paid to re-distributing those incarcerated for violent crimes across facilities within a state so as to prevent a critical mass of such offenders in any one facility. Efforts to reduce the numbers of incarcerated low risk offenders might be useful in this regard. Such a strategy would permit administrators the flexibility to redistribute violent offenders throughout the system and perhaps allow for more accurate assessment of risk. This does not mean shifting some of the most violent offenders into less secure units, but rather shifting smaller portions of certain types of older violent offenders into lower security units. Implementing such a procedure would demand consideration of other risk factors such as gang affiliation, criminal history, and so forth. This balance might also help to “normalize” prison environments for first-time violent offenders as opposed to feeding their anxieties by placing them in populations with higher concentrations of more serious violent offenders.

Results from our analysis suggest that a greater use of coercive controls in states with more punitive orientations toward offenders does not promote lower levels of either assaults or nonviolent offenses, contrary to an administrative control perspective. Greater use of coercive controls is placing enormous strains on corrections budgets without making these environments safer for inmates. A state’s inability to keep pace with rising prison populations is also problematic when the laws are getting tougher on offenders who may be more likely to act out in prison, such as states with gang enhancement sentencing. In California, for example, a judge can add 10 years to the sentence of someone convicted of a violent felony if that person is a street gang member. Although greater uniformity in sentencing through more structured decision making is necessary to reduce unwarranted disparities in sentences for similar offenders, sentencing based on formal rather than substantive rationality (Ulmer and Kramer, 1996) in conjunction with more punitive sentencing may only place greater strains on the infrastructure necessary to maintain safe prison environments. In short, re-visiting these policies may be necessary for enhancing inmate safety.

Inmate population composition appears to be more relevant than environmental and administrative controls for shaping facility differences in levels of violent and nonviolent offending across US prisons. Despite the statistical significance of less punitive practices at the state level and the use of segregation at the prison level, the directions of which still countered an administrative control perspective, compositional effects were significant in both models and larger in magnitude relative to these other effects. Also, the total positive effects of prison security and segregation on assault and nonviolent offense levels in conjunction with the direct inverse effects of population composition on these outcomes suggest that greater prison security actually coincides with higher levels of misconduct by nature of the confined population. Current risk assessments used by departments of corrections across the USA function to “arrange” inmates in ways that make inevitable certain levels of misconduct in certain types of facilities. Even facilities that rely more heavily on remunerative (versus coercive) controls experience misconduct levels commensurate with the types of offenders housed within them.

The null effects of remunerative controls on assault and nonviolent offense levels do not necessarily refute the potential relevance of remunerative controls at the individual level. Prisons house ample numbers of replacements for individuals who might be insulated from misconduct due to their involvement in programs or paid jobs. There are many situations conducive to violent and nonviolent crimes in prison due to common living areas, stressful conditions, and constraints on personal autonomy. Some inmates might be more prone to seizing criminal opportunities based on their backgrounds, even when most of their time is spent in relatively structured and well-monitored activities, and this may obscure any substantive aggregate level effects of remunerative controls. For example, inmates pulled from severely disadvantaged neighborhoods may be more cynical toward legal authority (e.g. Sampson and Bean, 2006), and this cultural ideology may weaken their belief in the legitimacy of correctional officers (Bottoms, 1999; see also Tyler, 2006). Inmates may believe in the substance of the rules of conduct, but mistrust of and animosity toward correctional staff may lead some inmates to violate the rules if they feel they cannot rely on legal authority for assistance in problem-solving. Rule-breaking may actually become a form of problem-solving for these individuals (e.g. Anderson, 1999; Black, 1983; Kirk and Papachristos, 2011; Toch et al., 1989). Violent offenders may be more willing to resort to violence as a means to an end in prison whereas non-violent offenders may be less inclined to consider violence as an option. As such, violent offenders may be less likely than non-violent offenders to see “legitimate” options such as prison programs and paid jobs as viable mechanisms for adaptation to prison. Levels of legal cynicism may also impact deviance if the most cynical inmates sometimes engage in rule violations as civil disobedience in order to display resistance to authority (Milovanovic and Thomas, 1989).

Greater investments in programs and jobs may be more effective for reducing misconduct levels in smaller populations where proportionately more inmates have access to these services, or if additional services are accompanied by a shift in management ideology favoring remunerative controls (Colvin, 1992). Supporting a culture of deference toward inmates rather than relying on threats of coercion may be necessary for inmates to collectively comply with facility rules and to counterbalance their reliance on each other for need satisfaction. This, in turn, may be more effective for countering inmates’ cynicism toward legal authority. Given the importance of maintaining safe and orderly environments, prison administrators might seek to develop methods to achieve more “moral” environments within their own facilities (Liebling, 2004).

Earlier, we discussed how improvements in custodial risk assessments may be the source of the current migratory loop for high risk offenders between the most economically disadvantaged urban neighborhoods and maximum security units/prisons. An important difference between these environments, however, is that high risk inmates are placed into more secure environments whereas larger numbers of individuals with deficits in social and human capital reside in neighborhoods with more criminal opportunities. This situation should raise concern over a high risk offender’s reentry into the community, especially since so many of these offenders are released into poor urban communities with limited services to facilitate successful reintegration (Petersilia, 2005). Is it realistic to expect that these individuals can refrain from crime if prisons cannot effectively alter the behaviors of high risk offenders, assuming the compositional perspective is accurate? These individuals are most likely to recidivate, suggesting a need to focus on the neighborhoods themselves and how to change community structure in order to reduce opportunities for and the temptations of recidivism.

Recent trends in US incarceration rates

Despite the punitive reputation of the United States, total admissions to state prisons in 2012 (553,800) was the lowest since 1997, reflecting the third straight year of a decline in state prison admission rates (Carson and Golinelli, 2013). Current budget woes of state governments might account for this, suggesting a possible turning point where the economically conservative politicians who once favored getting tougher with offenders are now advocating budget cuts in corrections. Attention has turned to closing existing facilities (e.g. Colorado, Kansas) and using more alternatives to prison (e.g. New Jersey, Ohio) (Steinhauer, 2009), and states such as Kentucky and Indiana are even re-visiting their sentencing statutes. The ideology of getting tougher with offenders may be losing steam out of necessity for balancing state budgets. If so, closing prisons and sending inmates elsewhere may weaken our compositional perspective because, at that point, prisons would not be filled solely with the types of offenders for which they were designed. In the long term, however, more states may transform more prisons into “mixed” security facilities in order to shut down other prisons. Inmates would then be confined within specific areas of a facility based on risk level, and the compositional perspective would again apply but on a unit by unit basis. Offense rates would then become more similar across facilities as inmate populations become more similar. This would not counter the idea that population composition is more relevant than administrative controls for reducing levels of inmate offending.

Footnotes

Funding

This project was supported by a grant from the American Statistical Association and the Bureau of Justice Statistics as part of their Special Data Collections and Statistical Methodological Studies Research Program. The data sets examined for this article were made available by the Interuniversity Consortium for Political and Social Research. The data for the 1997 Survey of Inmates in State and Federal Correctional Facilities (ICPSR 2598), the 1995 Census of State and Federal Adult Correctional Facilities (ICPSR 4021), and the 2000 Census of State and Federal Adult Correctional Facilities (ICPSR 6953) were collected by the United States Bureau of the Census. Neither the United States Bureau of the Census nor the Consortium bear any responsibility for the analyses presented here.

Notes

John Wooldredge is a professor in the School of Criminal Justice at the University of Cincinnati. His research and publications focus on institutional corrections (crowding, inmate violence, and inmate adaptation) and criminal case processing (sentencing and recidivism, extralegal disparities in case processing and outcomes).

Ben Steiner is an associate professor in the Department of Criminology and Criminal Justice at the University of Nebraska at Omaha. His research and publications focus on parole decision making and parole guidelines, juvenile court case processing and sentencing, and institutional corrections (crowding and inmate deviance).

Appendix: Observed and latent variables for the analysis

References

Anderson

(1999) Code of the Street: Decency, Violence, and the Moral Life of the Inner City, New York: WW Norton.

Andrews

Bonta

(2010) The Psychology of Criminal Conduct, New Providence, NJ: Anderson.

Association of State Corrections Administrators (2010) Violence and Safety in American Corrections: What the Research Shows, Washington, DC: Congressional Research Service.

Austin

(1996) The effect of three strikes and you’re out on corrections. In: Sichor

Sechrest

(eds) Three Strikes and You’re Out: Vengeance as Public Policy, Los Angeles, CA: SAGE, pp. 155–173.

Black

(1983) Crime as social control. American Sociological Review 48: 34–45.

Bonta

(1996) Risk-need assessment and treatment. In: Harland

(ed.) Choosing Correctional Options That Work: Defining the Demand and Evaluating the Supply, Los Angeles, CA: SAGE, pp. 18–32.

Bottoms

(1999) Interpersonal violence and social order in prison. In: Tonry

Petersilia

(eds) Crime and Justice: A Review of Research Vol. 26. Chicago, IL: University of Chicago Press, pp. 205–282.

Camp

Gaes

Langan

(2003) The influence of prisons on inmate misconduct: A multilevel investigation. Justice Quarterly 20: 501–533.

Carson

Golinelli

(2013) Prisoners in 2012: Trends in Admissions and Releases, 1991–2012, Washington, DC: Bureau of Justice Statistics.

10.

Colvin

(1992) The Penitentiary in Crisis: From Accommodation to Riot in New Mexico, Albany, NY: SUNY Press.

11.

Colvin

(2007) Applying differential coercion and social support theory to prison organizations: The case of the penitentiary of New Mexico. The Prison Journal 87: 367–387.

12.

Dietz

O’Connell

Scarpitti

(2003) Therapeutic communities and prison management: An examination of the effects of operating an in-prison therapeutic community on levels of institutional disorder. International Journal of Offender Therapy and Comparative Criminology 47: 210–223.

13.

DiIulio

(1987) Governing Prisons: A Comparative Study of Correctional Management, New York: Free Press.

14.

Ditton

Wilson

(1999) Truth in Sentencing in State Prisons, Washington, DC: Bureau of Justice Statistics.

15.

Fox

(2008) Applied Regression Analysis and Generalized Linear Models, Los Angeles, CA: SAGE.

16.

Galster

Hesser

(1981) Residential satisfaction: Compositional and contextual correlates. Environment and Behavior 13: 735–758.

17.

Gendreau

Ross

Izzo

(1985) Institutional misconduct: The effects of the UVIC program at Matsqui Penitentiary. Canadian Journal of Criminology 27: 209–217.

18.

Gilliard

Beck

(1998) Prison and Jail Inmates at Midyear 1997, Washington, DC: Bureau of Justice Statistics.

19.

Harer

Steffensmeier

(1996) Race and prison violence. Criminology 34: 323–355.

20.

Hauser

(1969) Schools and the stratification process. American Journal of Sociology 74: 587–611.

21.

Hauser R, Sewell W and Alwin D (1974) High school effects on achievement. Paper presented at the Meeting of the American Sociological Association in Montreal, Quebec (August).

22.

Huebner

(2003) Administrative determinants of inmate violence: A multilevel analysis. Journal of Criminal Justice 31: 107–117.

23.

Innes

(1997) Patterns of misconduct in the federal prison system. Criminal Justice Review 22: 157–174.

24.

Irwin

(1980) Prisons in Turmoil, Boston, MA: Little, Brown.

25.

Irwin

(2005) The Warehouse Prison: Disposal of the New Dangerous Class, Los Angeles, CA: Roxbury Press.

26.

Irwin

Cressey

(1962) Thieves, convicts and the inmate culture. Social Problems 10: 142–155.

27.

Kornhauser

(1978) Social Sources of Delinquency, Chicago, IL: University of Chicago Press.

28.

Kirk

Papachristos

(2011) Cultural mechanisms and the persistence of neighborhood violence. American Journal of Sociology 116: 1190–1233.

29.

Krivo

Peterson

(1996) Extremely disadvantaged neighborhoods and urban crime. Social Forces 75: 619–648.

30.

Lahm

(2008) Inmate-on-inmate assault: A multilevel examination of prison violence. Criminal Justice and Behavior 35: 120–137.

31.

Lahm

(2009) Educational participation and inmate misconduct. Journal of Offender Rehabilitation 48: 37–52.

32.

Land

McCall

Cohen

(1990) Structural covariates of homicide rates: Are there any invariances across time and social space? American Journal of Sociology 95: 922–963.

33.

Langan

Pelissier

(2001) The effect of drug treatment on inmate misconduct in federal prisons. Journal of Offender Rehabilitation 34: 1–18.

34.

Liebling

(2004) Prisons and Their Moral Performance: A Study of Values, Quality, and Prison Life, New York: Oxford University Press.

35.

Liska

Chamlin

(1984) Social structure and crime control among macrosocial units. American Journal of Sociology 90: 383–395.

36.

Messner

Sampson

(1991) The sex ratio, family disruption, and rates of violent crime: The paradox of demographic structure. Social Forces 69: 693–713.

37.

Miethe

McDowall

(1993) Contextual effects in models of criminal victimization. Social Forces 71: 741–759.

38.

Miller

(2010) Community in-reach through jail reentry: Findings from a quasi-experimental design. Justice Quarterly 27: 893–910.

39.

Milovanovic

Thomas

(1989) Overcoming the absurd: Prisoner litigation as primitive rebellion. Social Problems 36: 48–60.

40.

Morenoff

Sampson

Raudenbush

(2001) Neighborhood inequality, collective efficacy, and the spatial dynamics of urban violence. Criminology 39: 517–560.

41.

Mosher

Phillips

(2006) The dynamics of a prison-based therapeutic community for women offenders. The Prison Journal 86: 6–31.

42.

Muthen

(1998–2010) Mplus User’s Guide, Sixth Edition. Los Angeles, CA: Muthen & Muthen.

43.

Ostermann

(2009) An analysis of New Jersey’s Day Reporting Center and Halfway Back Programs: Embracing the rehabilitative ideal through evidence based practices. Journal of Offender Rehabilitation 48: 139–153.

44.

Petersen

(1972) The conception of criminality illustrated by a stochastic process model for deviant behavior. Journal of Research in Crime and Delinquency 9: 31–45.

45.

Petersilia

(2005) From cell to society: Who is returning home? In: Travis

Visher

(eds) Prisoner Reentry and Crime in America., New York: Cambridge University Press, pp. 15–49.

46.

Peterson

Krivo

(2010) Divergent Social Worlds: Neighborhood Crime and the Racial-Spatial Divide, New York: Russell Sage.

47.

Pickett

Pearl

(2001) Multilevel analyses of neighbourhood socioeconomic context and health outcomes: A critical review. Journal of Epidemiology and Community Health 55: 111–122.

48.

Prendergast

Farabee

Cartier

(2001) The impact of in-prison therapeutic community programs on prison management. Journal of Offender Rehabilitation 32: 63–78.

49.

Raudenbush

Bryk

(2002) Hierarchical Linear Models: Applications and Data Analysis Methods, Los Angeles, CA: SAGE.

50.

Sampson

(1986) Crime in cities: The effects of formal and informal social control. In: Morris

Tonry

(eds) Crime and Justice Vol. 8. Chicago, IL: University of Chicago Press, pp. 271–311.

51.

Sampson

Bean

(2006) Cultural mechanisms and killing fields: A revised theory of community-level racial inequality. In: Peterson

Krivo

Hagan

(eds) The Many Colors of Crime: Inequalities of Race, Ethnicity, and Crime in America, New York: New York University Press, pp. 8–36.

52.

Sampson

Wilson

(1995) Toward a theory of race, crime, and urban inequality. In: Hagan

Peterson

(eds) Crime and Inequality, Stanford, CA: Stanford University Press, pp. 37–54.

53.

Sampson

Raudenbush

Earls

(1997) Neighborhoods and crime: A multilevel study of collective efficacy. Science 277: 918–924.

54.

Sentencing Project (2012) Fact Sheet: Trends in U.S. Corrections. Available at: http://sentencingproject.org/doc/publications/inc_Trends_in_Corrections_Fact_sheet.pdf.

55.

Sparks

Bottoms

Hay

(1996) Prisons and the Problem of Order, New York: Oxford University Press.

56.

Steiner B (2008) Maintaining prison order: Understanding causes of inmate misconduct within and across Ohio Correctional Facilities. Unpublished dissertation, University of Cincinnati.

57.

Steiner

(2009) Assessing the static and dynamic influences on inmate violence levels. Crime and Delinquency 55: 134–161.

58.

Steiner

Wooldredge

(2008) Comparing state versus facility level effects on crowding in U.S. correctional facilities. Crime and Delinquency 54: 259–290.

59.

Steiner

Wooldredge

(2009) The relevance of inmate race/ethnicity versus population composition for understanding prison rule violations. Punishment and Society 11: 459–489.

60.

Steiner

Wooldredge

(2013) Implications of different outcome measures for an understanding of inmate misconduct. Crime and Delinquency 59(8): 1234–1262.

61.

Steinhauer J (2009) To cut cost, states relax prison policies. New York Times, 24 March. Available at: http://www.nytimes.com/2009/03/25/us/25prisons.html?pagewanted=all&_r=0.

62.

Toch

Adams

Grant

(1989) Coping: Maladaptation in Prisons, New Brunswick, NJ: Transaction.

63.

Tonry

(1996) Sentencing Matters, New York: Oxford University Press.

64.

Turner

Greenwood

Chen

(1999) The impact of truth-in-sentencing and three strikes legislation: Prison populations, state budgets, and crime rates. Stanford Law & Policy Review 11: 75–83.

65.

Tyler

(2006) Psychological perspectives on legitimacy and legitimation. Annual Review of Psychology 57: 375–400.

66.

Ulmer

Kramer

(1996) Court communities under sentencing guidelines: Dilemmas of formal rationality and sentencing disparity. Criminology 34: 383–407.

67.

Useem

Kimball

(1989) States of Siege: U.S. Prison Riots, 1971–1986, New York: Oxford University Press.

68.

Useem

Piehl

(2006) Prison buildup and disorder. Punishment & Society 8: 81–115.

69.

Useem

Reisig

(1999) Collective action in prisons: Protests, disturbances, and riots. Criminology 37: 735–759.

70.

Veenstra

(2005) Location, location, location: Contextual and compositional health effects of social capital in British Columbia, Canada. Social Science and Medicine 60: 2059–2071.

71.

Wacquant

(2001) Deadly symbiosis: When ghetto and prison meet and mesh. Punishment and Society 3: 95–134.

72.

Warner

(2003) The role of attenuated culture in social disorganization theory. Criminology 41: 73–97.

73.

Welsh

McGrain

Salamatin

(2007) Effects of prison drug treatment on inmate misconduct: A repeated measures analysis. Criminal Justice and Behavior 34: 600–615.

74.

Wooldredge

Steiner

(2010) Comparing methods for examining a link between prison crowding and inmate violence. Justice Quarterly 26: 795–826.