Maintaining Safety in the Prison Environment: A Multilevel Analysis of Inmate Victimisation in Assaults

Abstract

Theoretical models of victimisation emphasise the importance of context. However, few studies have assessed the influence of prison environmental variables on inmate harm in physical assaults. This study used a multilevel model approach to examine individual- and facility-level factors associated with the incidence of assaults among inmates housed at correctional centres in New South Wales, Australia. Results supported proposals that institutional routines and conditions may have an influence on risk. Inmates, who spent less time in employment, were placed in special housing arrangements such as protection, or were located in sites with higher security designations or longer routine hours out of cells were more likely to be harmed in assaults. In addition, more than 40% of variance in assaults was associated with differences across correctional centre sites. We draw on routine activities theory to explain relationships between different prison contexts, provision of guardianship, and exposure to motivated offenders in assault outcomes.

Keywords

inmate assault inmate violence victimisation prison environment routine activities institutional misconduct

A central priority of prison administrations is to safeguard inmates from physical and other harm. Although correctional systems have legal obligations to ensure the safety of inmates (e.g., Wolff & Shi, 2009), inmate violence also represents a breakdown in broader aims to maintain control within the prison environment (Griffin & Hepburn, 2013; Useem & Kimball, 1989). Victimisation may have lead on effects in undermining order in prisons, such as by aggravating cynicism towards legal authorities (Wooldredge & Steiner, 2013); encouraging inmates to adopt informal means of protecting themselves by obtaining weapons (Wolff, Blitz, Shi, Siegel, & Bachman, 2007) or through gang affiliation (Wolff, Shi, & Blitz, 2009); and generating increased workplace stress for custodial staff (Steiner & Wooldredge, 2015). There is also evidence that inmate victimisation can be detrimental to objectives of criminal justice systems as a whole by negatively affecting mental health (Listwan, Colvin, Hanley, & Flannery, 2010) and recidivism (Listwan, Sullivan, Agnew, Cullen, & Colvin, 2013) outcomes for offenders. These broader social impacts have become increasingly relevant considering the growing resort to incarceration and inmate populations in countries such as Australia and the United States (Sabol, West, & Cooper, 2009; Useem & Piehl, 2006; Weatherburn, Wan, & Corben, 2014).

Maintaining the safety of inmates is a substantial task, considering that prisons aggregate large populations of individuals with histories of violent behaviour or other antisocial tendencies in conditions of material and social deprivation (Wolff et al., 2007). Consistent with this, inmate victimisation in physical assaults is a relatively common problem, with recent research estimating a 6-month prevalence of 129 to 346 assaulted per 1,000 inmates (Wolff et al., 2009). Theories of victimisation suggest that inmate risk may be influenced by the extent to which prison routines and settings provide effective guardianship and manage interactions between suitable targets and motivated offenders (e.g., Cohen & Felson, 1979). Although a number of studies have examined individual predictors of violent misconduct, there is less research on contextual factors that may influence an individual’s risk of violent victimisation in prison (see Steiner, Ellison, Butler, & Cain, 2016 for a review). In this article, we report on a multilevel analysis of individual- and facility-level factors associated with the incidence of physical assault among inmates housed in correctional centres across New South Wales (NSW), Australia, with a focus on the influence of various institutional routines and conditions.

Literature Review

Models of victimisation have emphasised the importance of context in determining risk. Early formulations suggested that victimisation is a function of opportunity and thus more likely to occur when individuals have lifestyles or engage in activities that place them in high-risk situations (Hindelang, Gottfredson, & Garofalo, 1978). Expanding on this notion, routine activities theory (Cohen & Felson, 1979; Felson, 1986) proposes that the likelihood of victimisation is moderated by the degree of exposure that potential targets have with motivated offenders and the level of guardianship present in those situations. In this regard, the daily routines and situations that individuals engage in can have an influence on victimisation by influencing the convergence of suitable targets, motivated offenders, and absence of effective guardianship in time and space.

By definition, the prison environment has a high density of motivated offenders who have previously victimised others. Given that offenders and victims often share common characteristics (e.g., Mazerolle, Legosz, Jeffries, & Teague, 2007), prisons are also likely to house many individuals who are vulnerable to becoming targets of victimisation. Although it may be expected that the unique features of prison populations generate conditions of greater victimisation risk relative to the community in general (Wolff et al., 2007), there is substantial variance in the characteristics of both individuals and of prison conditions that may moderate the risk that any single inmate will be victimised. A number of factors may be relevant to victimisation risk in the prison environment, including (a) target suitability, (b) individual differences in routines, (c) individual and facility differences in the application of formal controls, and (d) institutional capacity for control.

Target Suitability

In correctional settings, some inmates are more likely to be viewed as suitable targets for victimisation than others. Studies have shown that a range of individual demographic and historical factors can be predictive of risk of inmate victimisation in physical assaults, including age, gender, race, mental health history, offence characteristics, and prior violent behaviour in custody (for a review, see Steiner et al., 2016).

Finkelhor and Asdigian (1996) proposed that individuals may be more likely to be perceived as targets for motivated offenders when they convey antagonism or vulnerability. Target antagonism refers to inmate behaviours or characteristics that tend to invite retaliation from others in the prison environment. For example, inmates who have previously engaged in institutional misconduct may be more likely to be subsequently victimised (Kuo, Cuvelier, & Huang, 2014; Lahm, 2009; Wooldredge & Steiner, 2012, 2013, 2014), potentially as a response to that misconduct. Similarly, factors such as a history of violent offending and inmate classification could be associated with risk because they serve as proxies for tendencies towards antagonistic behaviour (Wooldredge & Steiner, 2013).

Variation in demographic and social characteristics across inmates may also signal that an individual is an easier or more vulnerable target than others. For example, inmate race may act as a significant predictor of physical victimisation because inmates who have minority racial representation at a prison have decreased access to protection among same-race peers (e.g., Mann & Cronan, 2002; Wooldredge, 1998). Another relevant variable is that of the age of inmates. Whereas younger inmates may be vulnerable due to their limited experience of correctional processes and norms, older inmates may be vulnerable as a result of physical infirmity. A number of studies have shown that younger age is a relatively consistent predictor of increased victimisation risk (Steiner et al., 2016).

Individual Differences in Routine

Within any given correctional facility, inmates have various differences in daily routines and activities that may influence their likelihood of victimisation. For example, allocation to a work assignment, education class, or programming is likely to be accompanied by more intensive guardianship in the form of supervision by custodial officers and other staff compared with inmates who spend time on unstructured recreational activities in area yards or their own cells. However, such routines often expose inmates to a more dynamic range of contextual influences arising from requirements to move between units regularly and interact with a wider population of inmates, which may in turn increase the risk of coming into contact with motivated offenders (Wooldredge, 1998). Prison administrations may act to offset this risk by employing structured activities such as work or programs as a form of remunerative control, whereby attending inmates are selected on the basis of good behaviour and are incentivised to minimise misconduct so as to retain their position (Huebner, 2003; Steiner, 2009).

Although any engagement in structured activities has been predicted to be protective against victimisation (Wooldredge, 1998; Wooldredge & Steiner, 2013), the available evidence is mixed. For example, employment has been associated with both significant increases (Wooldredge & Steiner, 2012, 2013, 2014) and decreases (Perez, Gover, Tennyson, & Santos, 2010; Teasdale, Daigle, Hawk, & Daquin, 2015) in the likelihood of being physically assaulted. Similarly, Wooldredge (1998; Wooldredge & Steiner, 2012) found that time spent in education predicted a decreased prevalence of victimisation; however, this outcome has not been replicated in other studies (Lahm, 2009; Wooldredge, 1994; Wooldredge & Steiner, 2013, 2014). A recent meta-analysis by Steiner and colleagues (2016) concluded that across a range of individual differences in daily routines, only time spent in structured recreational activities was consistently associated with an increased likelihood of victimisation.

Formal Controls

Administrative strategies for maintaining order tend to stratify the inmate population according to considerations of security risk, vulnerability, individual needs, or other priority for higher or lower intensity supervision and resourcing. Formal controls may be relevant to victimisation risk from a routine activities perspective because they assign inmates to different environmental conditions, with attendant differences in the likelihood of exposure to motivated offenders and levels of guardianship.

At the individual level, inmates who are placed in one area of a facility may experience substantially different contextual influences compared with those in other areas of the facility. For example, disciplinary measures such as segregation are intended to mitigate the risk posed by inmates with tendencies towards misconduct by isolating them in restrictive settings (e.g., Steiner, 2009). Segregation units would, therefore, be expected to have a unique environment characterised by a higher density of motivated offenders, in addition to more intensive guardianship and other restrictions to manage these inmates, relative to the general population. Similar contextual differences may arise from allocation of vulnerable inmates to special housing arrangements (e.g., protection units, mental health or disability units) and administrative decisions relating to microclimates such as selective assignment of inmates to single or multiple occupancy cells.

Formal administrative controls also contribute to differences in environmental conditions across facilities. A prominent example of this relates to the security level of correctional centres. Prison officials employ increasingly systematic classification procedures and allocate inmates to differing security levels as a frontline strategy for maintaining order within the inmate population (e.g., Gaes & Camp, 2009). Classification is often guided by violent history and deemed risk of future misconduct (Gaes & Camp, 2009; Harer & Langan, 2001; Worrall & Morris, 2011); thus, inmates housed in higher security facilities are more likely to be exposed to motivated offenders than those in lower security facilities. The security level of sites is also associated with various operational factors that amount to differing levels of guardianship such as staff to inmate ratios, intensity of other forms of surveillance, and restrictions on the movements of and associations between inmates (Steiner, 2009). In addition to the intensity of controls, victimisation risk may also vary across sites as a function of how inmates respond to qualitative differences in the types of controls that are used across security levels (Ricciardelli & Sit, 2015).

Contrary to the aims of formal administrative controls, higher security level has been found to be one of the more consistent predictors of assault (Steiner et al., 2016). Studies by Wooldredge and Steiner (2012, 2013, 2014) have also indicated that inmates who are housed in general population are less likely to be victimised compared with those in special housing arrangements. However, it appears that general population was compared with a composite of disciplinary and administrative segregation in addition to protective custody arrangements in these studies. In contrast, Listwan, Daigle, Hartman, and Guastaferro (2014) found that having spent any amount of time in protective custody was not a significant predictor of self-reports of polyvictimisation, which was defined as including physical assaults in addition to sexual assaults, threats of violence, and theft.

Institutional Capacity for Control

According to Griffin and Hepburn (2013), administrative efforts to maintain inmate safety should be assessed in the context of that facility’s institutional capacity for control. Prisons have differing structural characteristics and resources that may influence both the types of controls employed and how successfully controls are able to be implemented. For example, inmates detained in facilities that have larger total populations are at increased risk of violent victimisation (Wooldredge & Steiner, 2012, 2013, 2014). Increasing population may contribute to social disorganisation by impeding staff efforts to maintain communications with and direct oversight of inmates (Steiner, 2009; Useem & Kimball, 1989). As the population increases, existing resources for both formal controls (e.g., places in special housing units) and remunerative controls that promote good behaviour (e.g., work or program assignments) may also become strained. The effects of population on administrative control may be more pronounced in facilities where there is crowding relative to operational capacities (Steiner, 2009); however, it is noted that crowding has been found to have a nonsignificant relationship with victimisation (e.g., Kuo et al., 2014; Lahm, 2009). The capacity of existing staff to provide effective guardianship may also be undermined in prison environments with architectural designs that promote a proliferation of blind spots (Wooldredge & Steiner, 2014) and workplace cultures that are marked by perceived laxity in rule enforcement (Wooldredge & Steiner, 2013).

The Present Study

The existing literature suggests that administrative interventions have the potential to substantially moderate violence among inmates, through the provision of guardianship and regulation of interactions between suitable targets and motivated offenders. Because prisons are typically enclosed and highly structured environments, officials have a great degree of scope to manipulate the conditions under which inmates interact so as to minimise risk. Effective use of administrative resources such as staff guardianship and formal controls may have conditioning effects on risk factors at the inmate level, by suppressing individual propensities as targets or perpetrators (Harer & Langan, 2001; Useem & Kimball, 1989; Wilcox, Land, & Hunt, 2003). On the contrary, lack of effective guardianship over inmates may allow individual factors associated with risk to express themselves (Wooldredge & Steiner, 2014). Consistent with this, Griffin and Hepburn (2013) showed that individual factors were stronger predictors of inmate misconduct in correctional centres that had low levels of environmental control (defined as above average staff vacancies, crowding, and percentage of inmates with gang affiliation) compared with those centres that had high levels of environmental control.

To date, a relatively small number of studies have examined factors associated with inmate victimisation in physical assaults (e.g., Steiner et al., 2016), and few have investigated the effects of institutional variables at the individual or facility level. A review of the literature indicated that some commonly employed administrative strategies for maintaining order have received little attention. For example, it is currently unclear whether management of vulnerable inmates through placement in protection units has a relationship with likelihood of victimisation. Similarly, while it has been recognised that the security level of a correctional centre is likely to have complex effects on behaviour as a result of various differences in settings and resources (e.g., Steiner, 2009), little research has been conducted to isolate the influence of such differences.

In addition, no previous studies have examined factors associated with inmate violence or harm in the context of Australian correctional systems to our knowledge. The vast majority of research on inmate behaviour has been conducted in the United States, and it is unclear as to the extent that existing literature is generalisable across jurisdictions. At the time of study, the NSW correctional system comprised 31 correctional centres of relatively low inmate capacity (see Table 1 for details) that often house combinations of remand and sentenced individuals. Australian jurisdictions such as NSW have been subject to recent correctional trends that are similar to other nations, including an ageing (e.g., Howard & Corben, 2018) and growing (Weatherburn et al., 2014) prison population and attendant issues with crowding. Structural components of inmate management in NSW correctional centres also show similarities to other jurisdictions, including classification to more (medium, maximum) or less secure (minimum) centres according to each inmate’s assessed security risk, behaviour, and stage of sentence; allocation to work and other programs that is contingent upon an inmate’s ongoing good behaviour; and use of disciplinary segregation procedures that isolate individuals from the inmate population and other physical contact.

Table 1.

Descriptive Statistics for Level 1 (Inmate) and Level 2 (Site) Variables and for Incidence of Assault Outcomes.

Variable	M or %	SD	Range
Level 1
Individual target characteristics
Age at census	35.83	11.71	16-90
Sex (female)	7%	0.24	0-1
Indigenous	23%	0.42	0-1
Prior custody	68%	0.46	0-1
Prior assault victimisation	19%	0.39	0-1
Prior violent misconduct	09%	0.29	0-1
History mental illness	36%	0.48	0-1
Index sex offence	11%	0.31	0-1
Index violent offence	54%	0.49	0-1
Index episode length (days)	700	1,189	1-18,543
Individual routines
Average number of visits (weeks)	0.36	0.54	0-4.72
Average hours employed (weeks)	20.78	21.59	0-125.45
Average hours of education (weeks)	0.70	1.70	0-35
Average hours of programs (week)	0.18	0.83	0-13.65
Individual formal controls
Proportion days single cell	0.24	0.36	0-1
Proportion days segregation	0.01	0.05	0-1
Proportion days protection	0.19	0.37	0-1
Level 2
Site formal controls
Medium security	23%	0.42	0-1
Maximum security	26%	0.44	0-1
Routine hours out of cells	9.29	2.95	6.07-23.6
Site capacity for control
Average population	238.47	211.01	23.46-903.65
Crowding index	0.89	0.12	0.46-0.99
Inmate turnover	6.19	5.41	0.95-30.73
Dependent variable
Number of assaults over exposure period	0.06	0.28	0-5

Note. Level 1 n = 10,484; Level 2 n = 43.

In contrast, local contextual and operational features may be expected to differ from other jurisdictions. These include the high prevalence of short-sentenced inmates, which contributes to churn in the prison population; limited use of individual disciplinary segregation that is in contrast to extensive allocation of inmates to protection or other special management units; and policies that counterbalance security priorities with individual case management models that promote widespread engagement in work, education, and other gainful occupation. In addition, while Australian inmate populations are culturally diverse and show significant overrepresentation by Aboriginal and Torres Strait Islander peoples in particular (Weatherburn, Fitzgerald, & Hua, 2003), relevant issues relating to the prevalence and management of security threat groups (STGs) may be less systemic compared with some other nations. Investigating factors that are associated with institutional outcomes across jurisdictions may assist in identifying universal risk or protective factors, or alternatively local differences in operations or policy, that can inform best practice in other settings.

The aim of this study was to examine inmate- and facility-level predictors of harm in physical assault among inmates housed in NSW correctional centres. In addition to testing individual characteristics related to target suitability, we aimed in particular to examine a range of institutional variables associated with inmate routines, individual and facility differences in the application of formal controls, and site-level factors that may influence institutional capacity for control. Our analytical approach to the study was exploratory and not informed by specific hypotheses. However, in accordance with routine activities theory, it may be expected that institutional conditions that increase guardianship by custodial staff (e.g., time in employment or other structured occupation, more secure facility, lower crowding) and/or decrease exposure to motivated offenders (e.g., time in single occupancy cells, time in protection units, routine hours out of cells) would be associated with a reduced incidence of physical victimisation.

Method

Data and Sampling

This study used official reports data to examine the incidence of assault victimisation for all inmates detained at Corrective Services NSW (CSNSW) correctional centres over an observation period spanning July 1, 2013, to June 30, 2014. The sampling method incorporated all detained individuals, including sentenced and unsentenced (remand) inmates in addition to forensic patients. This gave a total sample of 10,484 individuals.

Inmates were distributed across a total of 31 correctional centres. A complicating factor is that a number of correctional centres are not dedicated to a single security level and incorporate a combination of minimum and medium or maximum security areas. The different security areas have distinctive operations and resources and are in some cases structurally independent. To account for these differences, we treated different security areas of the same complex as distinct sites for the purpose of analysis. This resulted in a total of 43 sites.

Official data on inmates were gathered from the CSNSW Offender Integrated Management System (OIMS) database. OIMS is the central operational database maintained by CSNSW and electronically records a range of information about offenders including demographics, offence and sentence details, results of intake screening and other testing, educational and psychological program attendance, employment routines, movements, and disciplinary infractions.

Because OIMS is used to facilitate custodial operations, most of the variables of interest contained close to 100% completion rates and missing data were minimal. A total of 54 inmates (0.005%) had incomplete data for Indigenous status. More significantly, two correctional centre sites housing 843 inmates (8.04%) had missing data on employment as a result of differences in recording procedures. These missing data were not considered to be missing completely at random (MCAR) because they showed covariance with observed site-level variables. A review of operations indicated that the sites with missing data had similar employment policies and resource requirements (other than recording system) to others in the jurisdiction. To confirm this, we examined employment data for the sites that were available over a different time period, being the first 6 months of 2015. Target sites with missing data in this study had proportions of inmates engaged in minimal weekly employment and average weekly duration of employment that were within tolerances of the average for comparable sites across the jurisdiction. We concluded that the missing data may be considered to meet statistical assumptions for being missing at random (MAR), in that missingness for the variable was not associated with the value of that variable after adjusting for other observed variables.

Missing data were estimated by multiple imputations using the NORM program (Schafer, 1997). The expectation maximisation (EM) algorithm converged in 10 iterations; therefore, we imputed 10 data sets for the missing values. Results were pooled using the hierarchical linear modeling 7 (HLM 7) multiple imputation function (Raudenbush, Bryk, & Congdon, 2011).

Measures

Outcome variable

The outcome variable of interest was the incidence of physical assault victimisations that each inmate was officially reported as being involved in during the exposure period. Assault victimisations were defined as all events in which the inmate was identified as sustaining observable physical harm from an inmate-on-inmate assault or fight. Assaults were recorded by custodial staff using official critical incident reporting modules that follow standardised templates and uniform reporting requirements across all correctional centres in the jurisdiction. Acts of assault included physical assaults only and sexual assaults were excluded.

The exposure period was permitted to vary across inmates to account for variation in custodial episode length and movements across locations. To calculate exposure period, inmates were first assigned a current site of residence based on their location at the data census date of June 30, 2014. Exposure period was then defined as the number of days since the inmate had entered the current site of residence, capped to a maximum of 365 days over the observation time frame. This approach was applied to allow for inclusion of all inmates regardless of their stage or duration of imprisonment, while addressing model criteria that each observation at the individual level is assigned to an observation at the facility level.

Inmate-level predictors

Individual- and site-level predictors used in this study are listed in Table 1. Predictor variables were selected on the basis of a review of the empirical literature on inmate victimisation, in addition to consideration of individual and site-level factors that may influence inmates’ exposure to motivated offenders and level of guardianship in accordance with routine activities theory.

Individual variables were first categorised as relating to target suitability and included demographic and historical variables such as age, gender, Indigenous status, prior imprisonment, prior involvement in institutional violence, mental health treatment history, index offence, and index custodial episode. The remainder of predictor variables at the individual level were categorised as relating to inmates’ routine activities (time spent in education, employment and programming, visits) or their exposure to localised formal controls (time spent in single cells, protection, and segregation).

Inmate age was measured as a continuous variable and calculated at the time of the census date. Indigenous status indicated whether the inmate self-identified as Aboriginal or Torres Strait Islander (0 = no; 1 = yes). Prior imprisonment was coded to indicate whether the inmate had been detained in custody with CSNSW prior to the index episode (0 = no; 1 = yes). Prior assault victimisation was derived from records of being harmed in an assault during their index custodial episode and prior to the exposure period (0 = no; 1 = yes), whereas prior violent misconduct was coded from records of receiving official charges for physically violent behaviour in the index episode prior to the exposure period (0 = no; 1 = yes). Index custodial episode was calculated as number of days detained in any correctional centre as part of their current imprisonment. Index offence variables for sex offence and violent offence were determined from Australian and New Zealand Standard Offence Classification (ANZSOC; Australian Bureau of Statistics, 2011) codes that define various offence categorisations. Mental health treatment history was a dichotomous variable and was obtained from an intake screening questionnaire that asked inmates whether they had ever received treatment for clinical disorder prior to the current imprisonment (0 = no; 1 = yes).

Variables for time spent in different structured occupations (employment, education, and rehabilitative programs) were derived from official operations data and calculated as average number of hours per week over the exposure period. Number of visits was similarly calculated as average frequency per week over the exposure period. Time spent in protection refers to the proportion of days over the exposure period that inmates were placed in units termed Special Management Area Placements (SMAP), which are used to house groups of inmates requesting protection or who are otherwise deemed vulnerable separate from the general population. Time spent in individual segregation as a disciplinary measure and time spent in single occupancy cells (as opposed to multiple occupancy or dorm-style cells) were also calculated as proportion of days in the exposure period.

Site-level predictors

A number of institutional characteristics were entered into analyses at the site level to examine associations between formal controls (security level, routine hours out of cells) or factors that influence institutional capacity for control (population, crowding, inmate turnover) and outcomes for inmates at the different facilities.

The security level of each site was modelled using two dummy variables with minimum security as the reference group and medium security (1 = medium security; 0 = other) or maximum security (1 = maximum security; 0 = other) as the comparison groups. Routine hours out of cells was used to indicate facilitywide policies on the number of hours inmates are allowed out of their cells and was derived from annual reports on average release, containment, and lockdown routines by site area. Population was calculated as the average daily inmate count for each site over the observation period. A measure of crowding was obtained by dividing average population by the average total number of beds (operational capacity) available at that site over the observation period. Inmate turnover was conceived as a measure of instability of the inmate population at that site and was calculated as the total number of inmates entering the site as new placements over the financial year divided by the operational capacity of that site.

Analytic Plan

Predictors of inmate assault at the individual- and site-level were assessed using the multilevel modelling analytical approach. Multilevel or hierarchical linear models are a robust method of analysing hierarchically structured data in which lower order units (in this case, inmates) are nonrandomly clustered across higher order units (in this case, correctional centres). By adjusting for shared variance between variables at the individual level that are associated with clustering at higher order levels, these models are robust to assumptions of dependence of observations that may be violated in traditional regression approaches. In this regard, multilevel modelling allows for simultaneous analysis of both individual- and site-level predictors on an outcome that may alternatively require separate models or derive biased results from nonhierarchical regression approaches.

Multilevel model data were analysed using HLM 7. Because the outcome variable consisted of the count or number of assaults reported for each inmate, data were modelled using a Poisson distribution adjusted for overdispersion. To account for differences in the period of measurement across inmates, the exposure period variable was controlled for as an offset in the model.

Variables at Levels 1 and 2 were simultaneously added to multilevel models using the entry method so as to minimise issues with overfitting. Variables were individually selected for parsimony and to limit multicollinearity where possible. Consistent with this, multicollinearity diagnostics indicated that correlations did not exceed moderate sizes for all predictors at the offender level (rs < .41), at the site level (rs < .44), and for interactions between offender and site-level variables (rs < .40).

Each of the Level 1 variables were grand mean centred prior to entry to adjust for compositional differences in these variables across the sites. Grand mean centring serves a similar function to adding aggregate versions of the same variable at Level 2 while preserving the limited degrees of freedom available at this level (Raudenbush & Bryk, 2002; see also Wooldredge & Steiner, 2014). We specified intercepts as outcomes models so that all predictors were entered as fixed effects and only the Level 1 intercepts were permitted to vary as random effects across sites. The models are represented as follows:

Level 1:

\log (\frac{E (\frac{Y_{i j}}{β_{j}})}{exposure}) = β_{0 j} + Σ β X_{i j}

Level 2:

β_{0 j} = γ_{00} + Σ γ W + μ_{0 j}

The βX in the Level 1 equation represents the effects of the sum of the fixed individual-level predictor variable coefficients. The intercepts in the Level 1 equation (β_0j) were modelled from the Level 2 regression equation, which includes the grand mean across Level 2 sites (γ₀₀) in addition to the effects of the sum of the site level predictor variable coefficients (γW) and the random error term or residual variance at the site level.

Results

Sample Characteristics

Table 1 provides descriptive statistics for the inmates and sites included in the study, which were calculated from original values prior to multiple imputation. Over the exposure period, a total of 632 assaults were recorded and 5.1% of inmates in the sample were harmed in one or more assaults (range = 0-5 incidents). The duration of the calculated exposure period was 143.2 days on average. Overall, inmates had been detained in custody for almost 2 years on average at the time of the census date (M = 700 days; range = 1-18,543 days). Almost one in five (19.4%) had experienced assault victimisation during their index episode prior to the exposure period, and around one in 10 (9.3%) had been charged with prior violent misconduct during their index episode.

At the time of the data census date, the average age of inmates was 35.8 years. A total of 6.5% of inmates were female (n = 685) and were largely housed (83.9%) in three dedicated female facilities. Almost one quarter (23.7%) of the sample identified as Indigenous. The majority of inmates (67.6%) had been imprisoned prior to the index episode. More than half of inmates had an index violent offence (54%), whereas a smaller number had been charged with a sexual offence (10.7%). At the time of intake screening, 36.3% of inmates reported that they had a history of treatment for mental health matters.

In terms of individual routines, inmates tended to spend more time on average in employment (M = 20.78 hr per week) compared with education (M = .70 hr) or offender treatment programming (M = .18 hr). Similarly, fewer inmates were recorded as spending no time engaged in employment (40.2%) compared with those who had not engaged in any education (59.3%) or offender programs (85.9%). On average, inmates received a visit around once every 3 weeks (M = 0.36 visits per week); two in five inmates (40%) had not received any visitors throughout the exposure period. Inmates spent an average of 0.5% of the exposure period in disciplinary segregation and 18.6% of the exposure period in protection units. Most inmates did not spend any of their exposure period in protection (76.6%) or in disciplinary segregation (96.1%). On average, inmates spent around one quarter (24.9%) of their exposure period housed in single occupancy cell placements.

At the facility level, 22 of the 43 sites were designated as minimum security, 10 were designated as medium security, and 11 as maximum security. The average population density of sites ranged between 23.46 and 903.65 inmates (M = 238.47; SD = 211.01). This corresponded with a level of crowding relative to operational capacity that averaged at 89.1%; it is noted that no site had a population that exceeded operational capacity (range = 46.1%-98.6%). Over the measurement period, sites had a mean inmate turnover of 6.19, indicating that each available bed at that site was vacated and filled with another resident more than 6 times on average (range = .95-30.73). On average, inmates were able to leave their cells for more than 9 hr per day (M = 9.29; SD = 2.95; range = 6.07-23.6 hr).

Multilevel Model Analyses

The first step of our multilevel model was to run an unconditional (null) model to establish whether variance in assaults occurred at the site level. The null model returned a significant variance component at Level 2 (see Table 2), indicating that outcomes varied significantly across sites after controlling for within-group variance. Intraclass correlation (ICC) statistics showed that 40.4% of the total variance in incidents was accounted for by differences between sites, whereas the remaining 59.6% of variance was accounted for by differences between individuals. Because there was significant variation in assault incidents across sites, the results justified further use of a multilevel modelling approach to the data.

Table 2.

Variance Components and ICC Statistics for the Unconditional (Null) Model.

Random effect	Variance component	ICC	df	χ²	p value
Intercept	0.84	.403	42	456.57	<.001
Level 1 residual	1.24	.596

Note. ICC = intraclass correlation.

Next, a random intercepts model was conducted to test predictors of outcome at the individual level. Coefficients and incidence risk ratios (IRRs) for each of the Level 1 predictors are provided in Table 3. A comparison of residual variance components with the null model indicated that the predictors entered into this model accounted for a 17.7% reduction in unexplained variance across individuals. A number of individual demographic and history variables were shown to be significant predictors, including age, prior assault victimisations, prior violent misconduct, and index violent offence. Age had a significant inverse relationship with outcome, whereby each additional year of inmate age was associated with an expected 3% decrease in the number of assault victimisations. Being assaulted or the perpetrator of violent misconduct in custody prior to the exposure period was associated with increases in the incidence of victimisation by 46% and 36%, respectively. Similarly, those inmates who were imprisoned in relation to a violent offence showed a 36% increase in the expected incidence of assaults, relative to those who did not have an index violent offence.

Table 3.

Multilevel Model Results for Level 1 (Inmate) and Level 2 (Site) Variable Associations With Incidence of Physical Assaults.

Variable	b	SE	IRR	95% CI
Level 1
Individual target characteristics
Age at census	−0.03**	0.01	0.97	[0.96, –0.99]
Female	0.64	0.42	1.90	[0.83, 4.37]
Indigenous	−0.01	0.11	0.99	[0.79, –1.23]
Prior custody	0.16	0.12	1.17	[0.93, 1.48]
Prior assault victimisation	0.37**	0.08	1.46	[1.24, 1.72]
Prior violent misconduct	0.31**	0.10	1.36	[1.12, 1.66]
History mental illness	0.21	0.12	1.23	[0.97, 1.56]
Index sex offence	−0.14	0.12	0.87	[0.72, 1.05]
Index violent offence	0.29**	0.11	1.36	[1.08, 1.65]
Index episode length (days)	−0.0002	0.0001	0.99	[0.99, 1.0]
Individual routines
Average number of visits (week)	−0.05	0.14	0.94	[0.72, 1.24]
Average hours employed (week)	−0.01**	0.003	0.99	[0.98, 0.99]
Average hours of education (week)	−0.01	0.02	0.99	[0.95, 1.04]
Average hours of programs (week)	0.01	0.08	1.01	[0.87, 1.18]
Individual formal controls
Proportion days single cell	−0.07	0.32	0.96	[0.50, 1.74]
Proportion days segregation	2.08**	0.51	8.05	[2.96, 21.87]
Proportion days protection	0.33*	0.16	1.40	[1.03, 1.90]
Level 2
Site formal controls
Medium security	1.94**	0.34	6.95	[3.49, 13.84]
Maximum security	1.75**	0.34	5.75	[2.89, 11.40]
Routine hours out of cells	0.16**	0.04	1.17	[1.07, 1.29]
Site capacity for control
Average population	0.00002	0.0004	1.00002	[0.99, 1.001]
Crowding	−0.60	1.87	0.55	[0.01, 24.04]
Inmate turnover	−0.004	0.03	0.99	[0.94, 1.06]

Note. Level 1 n = 10,484; Level 2 n = 43. IRR = incidence risk ratio; CI = confidence interval.

p < .05. **p < .01.

In terms of individual-level institutional variables, both inmate routine activities (employment) and formal controls (time spent in disciplinary segregation, time spent in protection units) were significant predictors of outcome. As inmates spent an increasing number of hours per week in employment on average, their incidence of assault victimisations decreased. In contrast, increasing proportion of the exposure period spent in disciplinary segregation or in protection units was associated with an increased incidence of victimisation.

Finally, means as outcomes model was estimated to test predictors of assault incidents at the site level. Results are given in Table 3. Level 2 predictors entered into the model were found to account for a 52.4% reduction in unexplained variance between sites, relative to variance components derived from the null model. Formal institutional controls relating to security level and routine hours out of cells were significant predictors of assault victimisation. The expected incidence of assault victimisations was almost 7 times higher (IRR = 6.95; 95% confidence interval (CI) = [3.49, 13.84]) for inmates in medium security sites and 6 times higher (IRR = 5.75; 95% CI = [2.89, 11.40]) for those in maximum security sites compared with inmates in minimum security sites. Increasing time spent out of cells was also associated with a significant increase in the incidence of victimisation. In contrast, factors that could potentially affect institutional capacity for control such as population, crowding, and inmate turnover were not significant predictors of outcome.

Discussion

The common objective of correctional centres is to maintain security and the safety of inmates across all prison environments. In an ideal system, no inmates would be at risk of harm in assaults regardless of their individual characteristics or degree of exposure to motivated offenders (Wooldredge & Steiner, 2014). Our study contributes to these objectives by examining how both individual- and site-level variance in prison routines and conditions are associated with victimisation in physical assaults. In addition to individual target suitability characteristics that were predictive of outcomes (younger age, index violent offence, prior history of victimisation, prior history of assault misconduct), the results of our study identified a number of contextual and institutional variables that may be relevant to the incidence of victimisation.

To begin, results indicated that individual differences in daily routines and activities were associated with assault victimisation. In particular, increasing time spent in employment predicted a significantly lower expected incidence of physical assault. This outcome replicates previous research (Perez et al., 2010; Teasdale et al., 2015) although is in opposition to other findings (Wooldredge & Steiner, 2012, 2013, 2014). In accordance with routine activities theory, disparate findings across studies may be attributed to differences in the levels of guardianship and other management of inmates that accompanied work conditions. For example, work assignments at the prisons studied here may have been more intensively supervised by custodial and other staff, or planned to occur in settings that minimised risk, compared with those examined in previous research. Consistent with this, Wooldredge and Steiner (2014) reported observing workplace conditions where inmates were supervised by few custodial staff and appeared to be out of visual range of staff at times. Another possibility is that in the current context, administrative officials were better able to mitigate risk using formal (e.g., inmate classification) or remunerative (e.g., history of good behaviour) controls to select inmates for employment, thus reducing the likelihood of conflict within work groups (Huebner, 2003; Ricciardelli & Sit, 2015).

In contrast, time spent in other structured activities such as education, treatment programming, or visits was not associated with outcomes. One account for this is that inmates did not spend enough time in these activities on average to influence risk, relative to employment. In the event that structured routines are protective by having proximal effects on levels of guardianship (Wooldredge, 1998), it is reasonable to suggest that such effects would be cumulative as a function of the amount of time spent in those conditions. One implication of the null effects is that engaging in education or treatment programs or receiving visits did not appear to have detrimental impacts on victimisation risk for those inmates by aggravating perceptions of target suitability in motivated offenders.

The results of this study also indicated that the formal controls employed by prisons can have significant interactions with risk of physical assault. Contrary to expectations, however, some controls appeared to have effects that were opposite to those intended. In particular, placement in disciplinary segregation and protection units were both associated with significant increases in the incidence of assaults. Although these results are consistent with previous findings (Wooldredge & Steiner, 2013, 2014), it is unclear why frontline administrative strategies for maintaining order and protecting vulnerable inmates could have such unintended effects. Steiner (2009) found that protective custody was associated with higher sitewide rates of violent misconduct and reasoned that such special treatment of some inmates may raise perceptions of inequality in the rest of the population. Resentment at preferential treatment may also operate at the individual level by provoking hostility towards the recipients of protection. Similarly, spending time in protection may confer changes in status (e.g., that the inmate is a sex offender or informant) that affects perceptions of target suitability when they return to general population. A limitation of this account is that prior placement in protection was not found to be predictive of subsequent victimisation in a previous study (Listwan et al., 2014), which suggests that time in protection may not confer increased target vulnerability. It also appears unlikely that placement in disciplinary segregation would similarly generate perceptions of target suitability among other inmates. Another possibility is that inmates who enter segregation or protection are at greater risk of victimisation when in those environments because placement policies tend to aggregate individuals with conduct problems or potential motivated offenders to a greater extent than in the general population.

An alternative account of the results is that associations between special housing arrangements and victimisation are attributable to simultaneity effects, where the outcome has a prior causal effect on the predictor (e.g., Worrall & Morris, 2011). For example, inmates in protection or segregation may have been placed in these arrangements because they had already been victimised or responded to an assault by fighting back, respectively. There is circumstantial support for this account, in that time in segregation had a strong association with assaults despite the relative underutilisation and brevity of this form of control. A significant limitation of this and other studies of correlates of inmate behaviour is that they are often unable to establish the temporal direction between dynamic contextual factors and outcomes. These considerations will be explored in greater detail in the “Limitations” section.

At the site level, formal controls related to intensity of security was also associated with an increased individual incidence of assaults. Wooldredge and Steiner (2014) suggested that facility security designation may be a more faithful indicator of the composition of inmates at that site than of the level of guardianship provided at that site. Given that security-level allocations are partly determined by violent offending or institutional misconduct (e.g., Gaes & Camp, 2009; Worrall & Morris, 2011), the implication is that higher security sites may increase exposure to motivated offenders without necessarily counterbalancing that risk with commensurate protective controls. Considering the substantial differences in environments, resources, and operations across security levels (Steiner, 2009), it is difficult to isolate factors that may contribute to this outcome. From the model, it does appear that site-level policies about the number of hours inmates are routinely allowed out of cells are an important operational factor that could influence risk of victimisation. However, a common formal control that corresponds to increasing security level is less time out of cells (e.g., Ricciardelli & Sit, 2015), which suggests that higher security prisons are associated with greater victimisation risk after adjusting for the use of such controls.

It is intuitive that site-level policies designed to isolate inmates in their cells for longer periods would act to inhibit risk of victimisation. Across jurisdictions there appears to be increasing preference for this strategy for maintaining order as the inmate population grows (Paget, 2015). However, the utility of such policies may be best considered in conjunction with indications that time spent in employment (presumably out of their own cells) is also protective. These factors may have cross-level interaction effects whereby time spent out of cells is less likely to aggravate risk at sites where there is a high level of inmate participation in structured activities (McCorkle, Miethe, & Drass, 1995). This may be particularly likely because inmate routines such as employment are a function of both individual willingness to engage in the activity and the resources available at that site. Because structured activities are often disproportionately available on the basis of security level (Ricciardelli & Sit, 2015), lower security prisons would be expected to have concentrations of low-risk individuals engaged in structured time out of their cells, whereas higher security prisons would have inmates engaged in unstructured, less intensively supervised time out of cells in an environment that is also likely to house greater densities of motivated offenders. In the case of higher security prisons, containment may be necessary to reduce the incidence of victimisation because fewer alternative methods for structuring interactions between inmates are available. However, the results of this study suggest that properly managed activities such as employment can have similarly positive effects on outcome, while also reducing the boredom (Rocheleau, 2013) and other conditions of deprivation that may undermine safety relative to more coercive controls (Ricciardelli & Sit, 2015).

Finally, features of the prison environment relevant to institutional capacity for control, such as inmate population size, crowding, and inmate turnover, did not have significant relationships with outcome. Although crowding has similarly been found to have no association with victimisation in other studies (Kuo et al., 2014; Lahm, 2009), population size is a relatively consistent predictor of risk (Steiner et al., 2016). A relevant consideration is that across the facilities sampled in this study, populations averaged at some 240 inmates and none of the sites exceeded 100% of operational capacity. In contrast, recent studies of prisons in the United States have reported average populations in excess of 1,000 inmates (e.g., Wooldredge & Steiner, 2013) and crowding that approached 140% of capacity in some cases (e.g., Lahm, 2008). Inmate population may have an increasing impact on inmate behaviour after reaching a certain threshold of density relative to the prison environment. Alternatively, inmate population may be more strongly associated with victimisation when it acts as a proxy for administrative resource deficiencies or other difficulties adapting to the increasing population over time. It has previously been proposed that population factors influence inmate behaviour by disrupting the application of formal controls (Griffin & Hepburn, 2013; Steiner, 2009). Paget (2015) similarly suggested that the growing inmate population in NSW may compromise safety by straining prison capacities to support access to treatment programs, employment, and other remunerative or rehabilitative activities.

Limitations

Some limitations of the study are noted. Significantly, the analytical approach employed in this and most other studies of inmate behaviour assess covariation between variables of interest and outcomes, and it is often not possible to determine temporal or causal relationships. Although temporal relationships are easily identified for static or historical predictors (e.g., inmate gender), this becomes more challenging in the case of dynamic factors such as institutional routines and conditions. For example, it is uncertain whether employment reduces victimisation risk or inmates who are not victimised are more likely to maintain employment. A related issue is that of simultaneity, where change in the status of a predictor is directly influenced by change in the status of the outcome. An example of this is placement in protective custody as a result of being victimised in general population. There is the potential for temporal relationships to be more clearly specified using point-in-time measures of variables as a predictor of subsequent outcomes; however, this approach is vulnerable to spontaneous changes in factors such as employment status, classification, or area of placement over the observation period. Application of longitudinal or experimental designs in future research would be beneficial to clarify the causal relationships between dynamic institutional factors and victimisation outcomes.

Second, we acknowledge that measurement of inmate victimisation is subject to various limitations. Our dependent variable incorporated a definition of victimisation that focused on the outcome of physical harm, with precipitating events including both unilateral attacks and fights with other inmates. Although this definition has been applied to other research on victimisation (e.g., Teasdale et al., 2015) and is relevant to understanding assault-related harms, it is unclear how victims of the different forms of inmate violence overlap in terms of individual characteristics and trajectories of risk. For example, being victimised in an assault predicts an increased risk of further victimisation (Teasdale et al., 2015), whereas inmates who respond to threat by fighting back may be subsequently less likely to be victimised (Kuo et al., 2014). In the event that physical harm is a product of qualitatively distinct inmate behaviours, it may be more difficult to identify consistent predictors of risk (Steiner & Wooldredge, 2009).

In addition, official data on victimisation is prone to underreporting (Wolff et al., 2007) and may also be biased by the resources and practices that are used to detect inmate behaviours across prison environments. Higher security facilities in particular may report more frequent victimisations because they have more intensive monitoring resources to detect incidents. However, it is noted that a similar relationship between security level and risk has been found in studies of inmate self-reports (e.g., Wooldredge & Steiner, 2012, 2013, 2014).

Finally, the complexity of analyses was limited to some degree by the relatively small sample of sites available for study. As a result, it was necessary to specify regression models with a limited subset of facility-level predictors and without cross-level interaction terms. Given that this study incorporated all available sites in the state of NSW, future study may aim to improve the power of multilevel analyses by aggregating data across multiple jurisdictions. As discussed, research employing increased power at the site level would be beneficial to examine interactions between site-level controls (e.g., routine hours out of cells) and individual routine activities; cross-level interaction designs would also assist in understanding how institutional factors such as crowding moderate the expression of individual risk factors (e.g., Lahm, 2008; Wooldredge & Steiner, 2014). A related limitation was that while this study used a wide range of data on inmates and prison operations, it was not possible to obtain some variables of theoretical interest that may further an understanding of factors associated with victimisation, including staff experience and ratios or inmate perceptions of features of the unique social climate of prisons.

Conclusion

The results of this study contribute to the international literature on factors associated with physical assaults in prisons and give support for the generalisability of routine activities theory by finding that a range of micro- and macroinstitutional variables were significantly associated with victimisation outcomes. From both theoretical and policy perspectives, there is the implication that physical assaults may be moderated by structured activities and other conditions that allow inmates to maintain regular prosocial routines, while providing effective guardianship and management of interactions with other inmates (Cohen & Felson, 1979; Felson, 1986; Wooldredge, 1998). It is, therefore, likely that formal activities and controls would have differing effects according to various structural and resourcing factors including the consistency of routines, availability, and level of staff oversight. Comparison of operational cases in which activities such as employment have had varying associations with behavioural outcomes across countries and jurisdictions (e.g., Wooldredge & Steiner, 2012, 2013, 2014) may be instrumental in further defining and implementing the qualitative conditions under which routine activities can act as a protective factor for victimisation.

Conversely, the results are consistent with cross-jurisdictional research by indicating that traditional formal controls for managing risk of physical harm in prisons often have counterintuitive relationships with victimisation outcomes. In particular, placement in protective custody or in more secure prison environments had positive associations with the incidence of assaults. Considering that inmates in this study who were previously harmed in physical assaults were also more likely to be assaulted over the measurement period (see also Teasdale et al., 2015), there appear to be outstanding challenges relating to the utility of existing formal responses for managing inmates who are vulnerable to harm or have been victimised within the prison system.

Whereas security level was not found to have the intended effects of improving inmate safety, there was some indication that containment policies commonly applied in such settings may be effective in managing risk. Perhaps unsurprisingly, inmates who spent less time out of their cells in accordance with site-level standards had a lower incidence of harm in physical assaults. However, in consideration of the available theoretical and empirical literature, it may be important to balance the potential effectiveness of coercive controls such as containment in cells with the benefits of remunerative controls in overall management of inmate behaviour (Huebner, 2003; Ricciardelli & Sit, 2015).

Inmate victimisation in prison has been a relatively understudied area of inquiry to date, and we acknowledge that the evidence base informing the complex causal interactions between individual characteristics, routine activities, and institutional conditions in mediating outcomes continues to develop. Although the current study identifies a number of institutional factors that are relevant to the incidence of physical assaults, there is a need for further research to elaborate on these findings and establish mechanisms of effect so as to better inform correctional policy and practice towards maintaining safety in the prison environment.

Footnotes

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: The authors conducted the research in this article while employed by Corrective Services NSW.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Mark V. A. Howard

References

Australian Bureau of Statistics. (2011). Australian and New Zealand Standard Offence Classification (ANZSOC) (3rd ed.). Canberra, Australian Capital Territory: Author.

Cohen

L. E.

Felson

(1979). Social change and crime rate trends: A routine activity approach. American Sociological Review, 44, 588-608. doi:10.2307/2094589

Felson

(1986). Linking criminal choices, routine activities, informal control, and criminal outcomes. In Cornish

Clarke

R. V.

(Eds.), The reasoning criminal (pp. 119-128). New York, NY: Springer.

Finkelhor

Asdigian

(1996). Risk factors for youth victimization: Beyond a lifestyles/routine activities theory approach. Violence and Victims, 11, 3-19.

Gaes

G. G.

Camp

S. D.

(2009). Unintended consequences: Experimental evidence for the criminogenic effect of prison security level placement on post-release recidivism. Journal of Experimental Criminology, 5, 139-162. doi:10.1007/s11292-009-9070-z

Griffin

M. L.

Hepburn

J. R.

(2013). Inmate misconduct and the institutional capacity for control. Criminal Justice and Behavior, 40, 270-288. doi:10.1177/0093854812457920

Harer

M. D.

Langan

N. P.

(2001). Gender differences in predictors of prison violence: Assessing the predictive validity of a risk classification system. Crime & Delinquency, 47, 513-536. doi:10.1177/0011128701047004002

Hindelang

M. J.

Gottfredson

M. R.

Garofalo

(1978). Victims of personal crime: An empirical foundation for a theory of personal victimization. Cambridge, MA: Ballinger.

Howard

M. V. A.

Corben

(2018). Forty is the new thirty (for recidivism): Trends in offender age, reimprisonment, and time to desistance among the New South Wales custodial population. Sydney, New South Wales, Australia: Corrective Services NSW.

10.

Huebner

B. M.

(2003). Administrative determinants of inmate violence: A multilevel analysis. Journal of Criminal Justice, 31, 107-117. doi:10.1016/s0047-2352(02)00218-0

11.

Kuo

S. Y.

Cuvelier

S. J.

Huang

Y. S.

(2014). Identifying risk factors for victimization among male prisoners in Taiwan. International Journal of Offender Therapy and Comparative Criminology, 58, 231-257. doi:10.1177/0306624X12465272

12.

Lahm

K. F.

(2008). Inmate-on-inmate assault: A multilevel examination of prison violence. Criminal Justice and Behavior, 35, 120-137. doi:10.1177/0093854807308730

13.

Lahm

K. F.

(2009). Physical and property victimization behind bars: A multilevel examination. International Journal of Offender Therapy and Comparative Criminology, 53, 348-365. doi:10.1177/0306624X08316504

14.

Listwan

S. J.

Colvin

Hanley

Flannery

(2010). Victimization, social support, and psychological well-being: A study of recently released prisoners. Criminal Justice and Behavior, 37, 1140-1159. doi:10.1177/0093854810376338

15.

Listwan

S. J.

Daigle

L. E.

Hartman

J. L.

Guastaferro

W. P.

(2014). Poly-victimization risk in prison: The influence of individual and institutional factors. Journal of Interpersonal Violence, 29, 2458-2481. doi:10.1177/0886260513518435

16.

Listwan

S. J.

Sullivan

C. J.

Agnew

Cullen

F. T.

Colvin

(2013). The pains of imprisonment revisited: The impact of strain on inmate recidivism. Justice Quarterly, 30, 144-168. doi:10.1080/07418825.2011.597772

17.

Mann

C. D.

Cronan

J. P.

(2002). Forecasting sexual abuse in prison: The prison subculture of masculinity as a backdrop for “deliberate indifference.” The Journal of Criminal Law and Criminology, 92, 127-185. doi:10.2307/1144209

18.

Mazerolle

Legosz

Jeffries

Teague

(2007). Breaking the cycle: A study of victimisation and violence in the lives of non-custodial offenders. Brisbane, Queensland, Australia: Crime and Misconduct Commission.

19.

McCorkle

R. C.

Miethe

T. D.

Drass

K. A.

(1995). The roots of prison violence: A test of the deprivation, management, and “not-so-total” institution models. Crime & Delinquency, 41, 317-331. doi:10.1177/0011128795041003003

20.

Paget

J. R.

(2015). Full house: The growth of the inmate population in NSW. Sydney, New South Wales, Australia: Inspector of Custodial Services.

21.

Perez

D. M.

Gover

A. R.

Tennyson

K. M.

Santos

S. D.

(2010). Individual and institutional characteristics related to inmate victimization. International Journal of Offender Therapy and Comparative Criminology, 54, 378-394. doi:10.1177/0306624X09335244

22.

Raudenbush

Bryk

(2002). Hierarchical linear models: Applications and data analysis methods. Thousand Oaks, CA: Sage.

23.

Raudenbush

Bryk

Congdon

(2011). HLM 7: Hierarchical linear and nonlinear modeling. Lincolnwood, IL: Scientific Software International.

24.

Ricciardelli

Sit

(2015). Producing social (dis)order in prison: The effects of administrative controls on prisoner-on-prisoner violence. The Prison Journal, 96, 210-231. doi:10.1177/0032885515618362

25.

Rocheleau

A. M.

(2013). An empirical exploration of the “pains of imprisonment” and the level of prison misconduct and violence. Criminal Justice Review, 38, 354-374. doi:10.1177/0734016813494764

26.

Sabol

W. J.

West

H. C.

Cooper

(2009). Prisoners in 2008. Washington, DC: Bureau of Justice Statistics, U.S. Department of Justice.

27.

Schafer

J. L.

(1997). Analysis of incomplete multivariate data. London, England: Chapman & Hall.

28.

Steiner

(2009). Assessing static and dynamic influences on inmate violence levels. Crime & Delinquency, 55, 134-161. doi:10.1177/0011128707307218

29.

Steiner

Ellison

J. M.

Butler

H. D.

Cain

C. M.

(2016). The impact of inmate and prison characteristics on prisoner victimization. Trauma, Violence, & Abuse, 18, 17-36. doi:10.1177/1524838015588503

30.

Steiner

Wooldredge

J. D.

(2009). Implications of different outcome measures for an understanding of inmate misconduct. Crime & Delinquency, 59, 1234-1262. doi:10.1177/0011128709335151

31.

Steiner

Wooldredge

J. D.

(2015). Individual and environmental sources of work stress among prison officers. Criminal Justice and Behavior, 42, 800-818. doi:10.1177/0093854814564463

32.

Teasdale

Daigle

L. E.

Hawk

S. R.

Daquin

J. C.

(2015). Violent victimization in the prison context: An examination of the gendered contexts of prison. International Journal of Offender Therapy and Comparative Criminology, 60, 995-1015. doi:10.1177/0306624X15572351

33.

Useem

Kimball

(1989). States of siege: U.S. prison riots, 1971-1986. New York, NY: Oxford University Press.

34.

Useem

Piehl

A. M.

(2006). Prison buildup and disorder. Punishment & Society, 8, 87-115. doi:10.1177/1462474506059141

35.

Weatherburn

Fitzgerald

Hua

(2003). Reducing Aboriginal over-representation in prison. Australian Journal of Public Administration, 62, 65-73. doi:10.1046/j.1467-8500.2003.00338.x

36.

Weatherburn

Wan

Corben

(2014). Why is the NSW prison population growing? (Crime and Justice Statistics Bureau Brief No. 95). Sydney, Australia: NSW Bureau of Crime Statistics and Research.

37.

Wilcox

Land

Hunt

(2003). Criminal circumstance: A dynamic multicontextual criminal opportunity theory. New York, NY: Aldine de Gruyter.

38.

Wolff

Blitz

Shi

Siegel

Bachman

(2007). Physical violence inside prisons: Rates of victimization. Criminal Justice and Behavior, 34, 588-599. doi:10.1177/0093854806296830

39.

Wolff

Shi

(2009). Feelings of safety inside prison among male inmates with different victimization experiences. Violence and Victims, 24, 800-816. doi:10.1891/0886-6708.24.6.800

40.

Wolff

Shi

Blitz

(2009). Understanding physical victimization inside prisons: Factors that predict risk. Justice Quarterly, 26, 445-475. doi:10.1080/07418820802427858

41.

Wooldredge

J. D.

(1994). Inmate crime and victimization in a southwestern correctional facility. Journal of Criminal Justice, 22, 367-381. doi:10.1016/0047-2352(94)90083-3

42.

Wooldredge

J. D.

(1998). Inmate lifestyles and opportunities for victimization. Journal of Research in Crime and Delinquency, 35, 480-502. doi:10.1177/0022427898035004006

43.

Wooldredge

J. D.

Steiner

(2012). Race group differences in prison victimization experiences. Journal of Criminal Justice, 40, 358-369. doi:10.1016/j.jcrimjus.2012.06.011

44.

Wooldredge

J. D.

Steiner

(2013). Violent victimization among state prison inmates. Violence and Victims, 28, 531-551. doi:10.1891/0886-6708.11-00141

45.

Wooldredge

J. D.

Steiner

(2014). A bi-level framework for understanding prisoner victimization. Journal of Quantitative Criminology, 30, 141-162. doi:10.1007/s10940-013-9197-y

46.

Worrall

J. L.

Morris

R. G.

(2011). Inmate custody levels and prison rule violations. The Prison Journal, 91, 131-157. doi:10.1177/0032885511404380