A Time-Varying Model of Risk for Predicting Recidivism Among Women Offenders Over 3 Years Following Their Release From Jail

Abstract

This study examines time-varying gender-responsive and gender-neutral predictors of recidivism over 3 years using baseline and quarterly follow-up interviews with 477 women released from a county jail. Of the 55 time-varying predictors tested in the longitudinal analysis, 39 were significant predictors of recidivism (new arrest or incarceration) even after controlling for baseline fixed predictors. Stepwise multivariate analysis simplified the model to 12 significant variables, including three time-varying variables associated with reduced risk of recidivism (custody of one’s children, self-help activity, environmental support), eight time-varying variables associated with increased risk of recidivism (illegal activity, type of crime, problems with probation/parole, days in jail/prison, number of sexual partners, past-year trauma, problem orientation, external pressure), and the composite measure of risk from baseline. These findings support the development of post-release re-entry services tailored for female offenders that address both gender-responsive and gender-neutral criminogenic risk factors.

Keywords

women offenders substance use disorders recidivism time-varying covariates

This study examines the time-varying effect of risk factors for recidivism over 3 years among a sample of women offenders following their release from a jail-based substance use program to the community. We build on a prior study (Scott, Grella, Dennis, & Funk, 2014), in which we systematically examined baseline predictors of recidivism over 36 months, with a focus on determining the predictive value of variables derived from gender-responsive and gender-neutral criminogenic recidivism models. Several variables measured at baseline, such as younger age, greater substance use frequency, loss of child custody, arrest for prostitution, and indicators of criminogenic thinking, were predictive of recidivism within a multivariate framework, but primarily during the first 3 to 12 months following release with diminished significance and accuracy over subsequent years.

These findings raise questions about the predictive value of baseline participant characteristics and behaviors for understanding longer term risks for recidivism following incarceration and in-custody treatment. Specifically, the changing behaviors and circumstances of participants during the years following their release from jail may be more informative about the time-varying predictors of recidivism and highlight the need for measurement at multiple time periods and ongoing or longer term monitoring.

Recidivism among Female Offenders

Few systematic studies have examined predictors of recidivism among women offenders with substance use problems, and even fewer have measured the predictors multiple times and looked at the relationship between these factors and recidivism longitudinally. Most studies have relied on baseline variables measured prior to release within the framework of the Risk–Needs–Responsivity (RNR) model of rehabilitation and recidivism, which has been used to guide correctional assessment and placement (Taxman & Marlowe, 2006; Van Voorhis, 2005, 2012). In the RNR model, variables pertaining to criminogenic risk, which are presumed to be most proximate to criminal activity, are assessed to predict risk of future re-offending; these include criminal behavior history and indicators of antisocial personality and beliefs. The need dimension focuses on substance use problems that are likely to precipitate a return to crime to obtain substances, crime committed under the influences of substances, or both. Some RNR analyses have also looked at less proximate predictor variables related to home, work, school, and leisure activity (Andrews et al., 2012). A meta-analysis of five large studies found that a risk–needs assessment was generally “gender-neutral” across eight risk/need domains (i.e., Criminal History, Substance Use, Companions, Procriminal Attitude, Antisocial, Education/Employment, Family Marital, Leisure/Recreation), although the substance use factor was more strongly related to the recidivism of female offenders than males (Andrews et al., 2012).

Although few studies to date have examined post-release predictors of recidivism, some studies have noted that there is often limited access to treatment and other services following release from correctional settings and that this likely increases the risk of recidivism (Blitz, Wolff, Pan, & Pogorzelski, 2005; Schram, Koons-Witt, Williams, & McShane, 2006; Scott, Coleman-Cowger, & Funk, 2014). This is particularly critical for women offenders, who have a higher prevalence of co-occurring mental health disorders (Green, Miranda, Daroowalla, & Siddique, 2005; Lynch et al., 2014; Sacks, 2004; Scott, Dennis, & Lurigio, 2013), often stemming from histories of trauma and abuse (Grella, Lovinger, & Warda, 2013; Salisbury & Van Voorhis, 2009; Scott, Coleman-Cowger, & Funk, 2014). In the absence of adequate treatment, such women are at high risk of substance use relapse following their release, and ultimately recidivism (Sarteschi & Vaughn, 2010).

Some factors that may mitigate the likelihood of recidivism pertain to access to resources and assets, such as income or education, as well as supportive social networks; however, access to these resources may be highly variable and change over time following release. These may be characterized within the framework of “social capital,” which typically refers to resources including one’s network of personal contacts and associations that may propel social mobility (Granfield & Cloud, 2001, p. 1566). When applied to the development of criminal behavior careers, social capital has been examined as part of a life course perspective, or as a background characteristic, measured at the time of incarceration. For example, Reisig, Holtfreter, and Morash (2002) utilized social capital to examine the resources and social networks available to women offenders at the time of their incarceration. They found a clear distinction between women with more education, income, and access to resources through their social networks, and women who were younger and had less education and prior legal income. Furthermore, women offenders released from prison who were impoverished had nearly 5 times the risk of recidivism, and approximately 13 times the risk of a supervision violation, over a 6-month follow-up period, compared with other women (Holtfreter, Reisig, & Morash, 2004). Conversely, receipt of public assistance (e.g., housing, income support) reduced the risk of recidivism among this sample.

Within a life course context, Salisbury and Van Voorhis (2009) identified three distinct pathways to incarceration among women corresponding to the central roles of childhood victimization, relationship dysfunction, and limited social and human capital in the areas of education, family support, and self-efficacy. Although unemployment is a robust predictor of recidivism for women offenders (Matheson, Doherty, & Grant, 2011), women offenders typically have more limited employment skills and work histories (Langan & Pelissier, 2001; Pelissier & Jones, 2005), which increases their susceptibility to criminal behavior re-involvement following release. The role of social capital in recidivism was confirmed in an 8-year follow-up study, which found that less education, as well as drug dependence and more extensive criminal involvement, was associated with shorter time to recidivism among women released from prison (Huebner, DeJong, & Cobbina, 2010). However, these characteristics were measured at baseline, and do not indicate how changes in social capital factors following release from prison may influence recidivism.

Building on this developmental approach, Brennan, Breitenbach, Dieterich, Salisbury, and Van Voorhis (2012) further elaborated a set of multiple pathways to criminal involvement among women that identified a “superordinate category” related to social marginalization and characterized by lack of educational–vocational assets, unstable housing, and social networks supporting crime and drug use (Brennan et al., 2012, p. 1499). A related concept, the notion of “recovery capital,” refers to “resources that can be brought to bear on the initiation and maintenance of substance misuse cessation” (Granfield & Cloud, 2008, p. 1983). These include assets or resources that individuals with substance use problems can use to cope with stressors and sustain recovery, such as having access to treatment services and supportive family, friends, and social networks, including 12-step groups (Granfield & Cloud, 2001; Laudet & White, 2008). For example, in a 10-year follow-up study of a treatment-based sample of mothers, their perceived neighborhood safety was associated with access to resources, and was also highly predictive of recovery outcomes (Evans, Li, Buoncristiani, & Hser, 2014). A recent study with women offenders found that pre-incarceration relationship characteristics served as positive (parents) or negative (peers) influences on women’s drug use and HIV-risk behaviors (Staton-Tindall, Frisman, et al., 2011). In a similar manner, factors associated with recovery capital may lessen the likelihood of recidivism among women whose criminal behavior is distinctly associated with their drug use, yet prior research has not examined the time-varying effects of recovery capital on recidivism outcomes among women offenders.

We apply these conceptual frameworks of social and recovery capital and life course to an examination of time-varying predictors of recidivism among women offenders following their release to the community from jail, as well as factors that derive from both the RNR and gender-responsive theoretical frameworks, as described below.

Findings from the Prior Article

This study builds on the findings from a prior study (Scott, Grella, et al., 2014) that examined baseline predictors of risk among study participants. This prior study tested the relative strength of baseline variables derived from (a) the gender-neutral RNR model (Andrews, Bonta, & Wormith, 2006), such as socio-economic status (employment, housing, education, income sources), substance use history and severity, history and nature of criminal behavior involvement (e.g., type of crime, age at initiation of criminal behavior, or contact with criminal justice system), indicators of criminal thinking and other personality disorders, and environmental or community-related factors, such as neighborhood crime or poverty, and (b) the gender-responsive model (Bloom, Owen, & Covington, 2003; Spjeldnes & Goodkind, 2009), which includes factors that are presumed to be more predictive of recidivism for women offenders, such as parental status and family relationships, mental health status, trauma history, and relational factors (e.g., substance use or criminal behavior involvement among family/social network).

Area under the curve (AUC) in a receiver operator characteristic analysis was used to compare several models in terms of their ability to predict recidivism. In this kind of analysis, an AUC = .50 would be considered to be chance, and higher scores would be better prediction, with AUC = 1.0 being perfect. The most recent meta-analyses of widely used measures of recidivism show that their AUC varied between .53 and .83 (Andrews et al., 2012; Kroner & Loza, 2001; Kroner & Mills, 2001; Schwalbe, 2007). Although more than a dozen factors were related to recidivism in the univariate analysis in our prior study (Scott, Grella, et al., 2014), the multivariate analysis in the same shows that recidivism can be reliably predicted (AUC = .90) with only four factors: age, no custody of children, substance use frequency, and number of substance problems. Further analysis shows that both univariately and multivariately, most of the variables primarily predicted early recidivism (1 to 12 months post release) and contributed much less to the prediction of later recidivism (13 to 36 months post release). This led us to the current article, examining whether information from each observation could be used as additional time-varying predictors to further improve the prediction of recidivism in subsequent observations.

Significance of Studying Women in Jail

From 2000 to 2010, about half of incarcerated women were held in local jails and the number of women in local jails has increased by 30% (Minton, 2011). Because most are incarcerated for only a few weeks or months, they also represent the majority of women offenders re-entering the community. Consistent with the reasons for their arrests, 70% of women entering jail in 2002 reported using alcohol or other drugs weekly in the month before their arrest (Adams et al., 2011; Bureau of Justice Statistics, 2005). Despite the high concentrations of substance-abusing women who are incarcerated in jails, most prior research on relapse and recidivism among women offenders has focused those who are released from prison (Staton-Tindall, Duvall, McNees, Walker, & Leukefeld, 2011).

Current Article

In the present longitudinal study, we examined the ability of time-varying predictors measured quarterly during the 3-year post-release follow-up period better predict recidivism after controlling for fixed predictors from the baseline (the focus of the first article). We used a wide range of time-varying predictors drawn from both gender-neutral and gender-responsive models. We also specifically hypothesized that variables pertaining to reductions in substance use and improvements in social and recovery capital may be protective against recidivism over time. Although coming from jail is the most common source of women re-entering the community, it is a relatively understudied population and represents a critical point in time for changing the course of the chronic and cyclical nature of their condition.

Method

Data Source and Design of Parent Study

Data are from the Recovery Management Checkups for Women Offenders (RMC-WO) Experiment (Scott & Dennis, 2012). The purpose of this subject by treatment experiment is to examine the long-term effects of specialized probation supervision and RMCs on substance use drug treatment participation and subsequent reductions in substance use, HIV-risk behaviors, and recidivism. Women were recruited from the Division of Women Justice Services (DWJS) of Cook County Jail in a two-step process: (a) at the time of entering DWJS/jail and (b) at the time of release to the community. The 480 women were randomly assigned to either an experimental group participating in quarterly assessments and RMC for 3 years (n = 238) or a control group participating in only quarterly assessments for 3 years (n = 242). Of these 480 women, more than 90% completed interviews at each 3-month interval from 6 to 36 months post release. In this article, we are focused on using the baseline (aka fixed predictors) and repeated measures (aka time-varying predictors) data to develop a model for predicting the time to recidivism (regardless of random assignment), so that it can be used as a covariate in the main findings.

Participation was voluntary after providing informed consent and conducted in accord with the standards of the Committee on Human Experimentation of the institution in which the experiments were done or in accord with the Helsinki Declaration of 1975. It was also conducted under the protection of a Certificate of Confidentiality from the NIDA and under the supervision of the Chestnut Human Subject Institutional Review Board (IRB) and an independent Data Safety Monitoring Board (DSMB).

Participants

Participants were recruited between 2008 and 2010 from the DWJS in Illinois’ Cook County Jail, which operates jail-based (residential) and furlough-based (outpatient) treatment programs for women offenders with drug problems and nonviolent charges. Cook County Jail is the largest single site jail, and DWJS is one of the largest jail-based treatment programs for women in the United States.

The target population for the experimental trial consisted of adult female offenders who were re-entering the community from the jail substance use treatment program. Recruitment was done in two phases: (a) during the initial detention and (b) at release. In Phase 1 (initial detention), women were deemed ineligible if they had not used substances in the 90 days before detention, had no substance use disorder symptoms in the year before detention, were under age 18, lived or planned to move outside Chicago within the next 12 months, were fluent in neither English nor Spanish, were cognitively unable to provide informed consent, or were released before their 14th day in DWJS. Over a 20-month recruitment period, a total of 866 women were eligible and 810 (93%) initially agreed to participate and completed the initial interview. In the second phase (time of release to the community after an average of 2.4 months in jail), we also excluded women who (a) were transferred from the jail to a prison in the Illinois Department of Corrections (n = 186) or (b) were still in jail at the time recruitment ended (n = 132). Of the remaining 492 women released to the community, 480 (98%) agreed to participate in the post-release randomized experiment and 3-year follow-up. Of the 12 planned quarterly follow-ups, on average, women completed 11.3 (SD = 1.8). For the purpose of this longitudinal analysis (which requires multiple observations), we further limited the sample to the 477 (99%) women with 4 (1/3rd) or more of the 12 follow-up interviews completed during the 3-year follow-up. In practice, 80% of the women had all 12 waves, 18% had nine to 11 waves, and 2% had four to eight waves.

The study sample was 84% African American, 7% White, 5% Hispanic, and 4% of other race/ethnicity. The average age was 37.1 (SD = 10.4), and most (71%) had never been married. About two thirds of the sample had at least one child under age 21, although 21% had lost custody of at least one child. Half the sample had completed less than 11 years of education, and only about 16% reported any employment in the 90 days prior to study intake. More than 90% self-reported criteria for lifetime dependence (three or more of seven symptoms on the Global Appraisal of Individual Needs [GAIN]), including for one or more of the following: opiates (46%), cocaine (42%), alcohol (14%), and cannabis (10%). Many (42%) had a past-year mental disorder, most commonly a mood disorder. Half had five or more arrests in their lifetime.

Measures

Dependent Measure

Recidivism was defined as the time from release to re-arrest or re-incarceration based on a combination of data from three sources: (a) Cook County Jail’s Incarceration Management and Cost (CCJ-IMAC) system, (b) the State of Illinois’s Law Enforcement Agencies Data System (IL-LEADS), and (c) self-report during follow-up on the quarterly GAIN. To minimize the impact of method bias and missing data, information from all three sources were combined using a Longitudinal, Expert, All Data (LEAD) standard (Kranzler, Tennen, Babor, Kadden, & Rounsaville, 1997; Pilkonis, Heape, Ruddy, & Serrao, 1991; Scott & Dennis, 2012; Spitzer, 1983). These indicators were largely in agreement (kappa = .64), but each captured some unique variance and none was sufficient. Relative to recidivism in the next 12 months based on any source of information using the LEAD standard, the sensitivity (% of cases identified) of the three individual sources was 84% for CCJ-IMAC, 88% for IL-LEADS, and 82% for GAIN.

Our decision to use either re-arrest or re-incarceration for the primary outcome was based on two things: context and analytic best practice. The context is that the program we recruited from included both a residential jail-based program (where re-incarceration would be the typical outcome in meta-analyses) and an outpatient furlough-based program (where re-arrest would be the typical outcome in meta-analyses). Moreover, women frequently moved between the programs; thus, it made sense to look at both outcomes. In terms of analyses, both criminological/substantive (e.g., Andrews et al., 2012; Scott, Grella, et al., 2014; Steadman, Redlich, Callahan, Robbins, & Vesselinov, 2011; Taxman & Marlowe, 2006) and methodological (e.g., Harris, Lockwood, Mengers, & Stoodley, 2011; Kranzler et al., 1997; Pilkonis et al., 1991; Scott & Dennis, 2012; Spitzer, 1983) experts have recommended that when both sources of information are available, it is preferable to use both. In practice, re-arrest is driving the results presented in this article because the days to next arrest and days to next incarceration were highly correlated (Spearman’s rho = .71) with the arrest preceding incarceration 56% of the time, both on the same day 39% and following incarceration only 5% of the time. In terms of any recidivism over 3 years, the kappa is high (.625) with 83 of 85 (98%) of the cases where they did not agree being due to the women who were re-arrested and not re-incarcerated.

Fixed Predictors Measured at Baseline

As part of earlier work with this sample (Scott, Grella, et al., 2014), we developed a model for predicting the time from release to recidivism based on four variables measured at baseline admission to the jail: age (odds ratio [OR] = 0.83), no custody of children (OR = 1.56), baseline substance frequency scale score (OR = 1.12), and baseline substance problem scale score (OR = 1.14). Taken together as a composite measure, they provide an excellent (AUC = .90) baseline predictor of the time to recidivism over 3 years.

Time-Varying Predictors Measured Multiple Times

Time-varying predictors comprise a broad range of variables that have been identified in prior research as relevant to substance use, criminal behavior involvement, and treatment outcomes among women, consistent within the gender-responsive, social/recovery capital, and life course frameworks reviewed previously. Table 1 lists the time-varying predictors that were re-assessed either annually or quarterly. It includes the variable’s psychometric properties (see below), frequency of assessment, and a short definition. These measures come from a battery of instruments administered at intake and quarterly thereafter, including a study-specific enhanced version of the GAIN (Dennis, Titus, White, Unsicker, & Hodgkins, 2003), Readiness for Change in Substance Use (Scott, Dennis, & Foss, 2005; Scott, Foss, & Dennis, 2005), and Relapse Coping Inventory (Dennis, Foss, & Scott, 2007; Moos, 1993). These include severity of alcohol and drug use, mental health symptoms, exposure to traumatic events, parental status, family and social relationships, criminal behaviors and criminal justice interactions, coping strategies, self-help and treatment participation, attitudes regarding treatment and recovery, and coping responses. If the measure in Table 1 is a scale or count of symptoms for a common underlying disorder/problem, the psychometrics included Cronbach’s alpha as a measure of the degree to which they are inter-correlated at the same point in time. For all time-varying predictors, the psychometrics included the interclass correlation coefficient (ICC) as a measure of the extent to which the values are inter-correlated over time within individual.

Table 1:

Time-Varying Predictors

Scale Name (α or ICC if Applicable): Description
Demographics
Lived with own children (ICC = .52): Past-year yes/no item for having lived with your own children.
Lived with other children (ICC = .16): Past-year yes/no item for having lived with children other than their own.
Custody (ICC = .93): Past-year categorical variable with group for having no children, not having custody of any of your children, and having custody of at least some of your own children.
Days worked (ICC = .39): In past 90 days, how many days worked for money.
FIS (α = .64, ICC = .30): Proportional average (past 90-day items divided by range) of recency and days of money problems and gambling. Higher values indicate increased levels of problems with money and/or frequency of gambling for money, drugs, sex, or other things.
Recency of family member sent to jail (ICC = .20): Based on single item, asked annually, on when last time experienced trauma from a family member being sent to jail.
Criminological risk factors
IAS (α = .95, ICC = .22): Five past-90 day items scored as average of proportional items (divided by their range) for the recency and days (during the past 90 days) of illegal activity and supporting oneself financially with illegal activity.
General Crime Scale (α = .78, ICC = .18): 19 past-90 day items across different types of illegal activities endorsed by the respondent.
Any violent crime (ICC = .11): Any report of violent crime in the past 90 days: Aggravated assault, armed robbery, rape, murder, or arson.
Any property crime (ICC = .15): Any report of property crime in the past 90 days: Vandalism, fencing, forgery, shoplifting, burglary, or auto theft.
Any AOD crime (ICC = .12): Any report of an alcohol or other drug (AOD) crime in the past 90 days: Driving under the influence, sold, distributed, or made illegal drugs.
Any prostitution (ICC = .21): Any report of trading sex for drugs or money in the past 90 days.
Any problems with probation or parole (ICC = .09): Report of any days in the past 90 days they were in trouble with probation or parole officer.
Any other crime (ICC = .06): Any report in the past 90 days of being gang member, gambling illegally, or other illegal activity.
Days incarcerated the prior wave (ICC = .28): Total days in jail or prison in the preceding 90 days divided into three groups: None, and then a median split on the rest: 1 to 53 and 54 to 90 days.
Substance use
SFS (α = .80, ICC = .52): The SFS is an eight past 90-day item measure that averages the percent of days reported of any AOD use, days of heavy AOD use, days of problems from AOD use, days of alcohol, marijuana, crack/cocaine, and heroin/opioid use and ranges from 0 (no use) to 1.
SPS (α = .95, ICC = .44) is a 16-item count of past-month symptoms of substance abuse, dependence, or substance induced disorders and is based on DSM-IV (APA, 1994, 2000).
Days of treatment (ICC = .17): Sum of days of substance use treatment in the past 90 days (residential, outpatient, intensive outpatient, methadone) maxed at 90 days.
SHAS (α = .98, ICC = .54): A count of past 90-day self-help activities endorsed out of 19: Talked at a meeting, have a sponsor, actively worked the 12 steps, etc.
HIV-risk behaviors
NPS (α = .89, ICC = .67): Count of the nine past-year activities endorsed, related to using a needle to inject drugs that increase the risk of transmission of blood-borne diseases.
NFS (α = .73, ICC = .49): Proportional past 90-day scale based on recency and days of needle use, days of sharing needles, and number with whom needles were shared. Higher values indicate greater, more recent needle use.
SXRS (α = .82, ICC = .29): Count of 40 past 90-day items, of the number of sexual activities that might put the client at risk of sexually transmitted diseases or other difficulties.
Number of sexual partners (ICC = .13): Number of sexual partners in the past 90 days, including their main partner.
TSXS (α = .88, ICC = .25): Count of 13 past-month sex trading items endorsed (trading sex for food, clothing, drugs, money, to support children, pimping, etc.).
Mental health
EPS (α = .78, ICC = .30): Average of proportional past 90-day items (divided by their range) for the recency and days (during the past 90) bothered by, and days kept from responsibilities because of emotional problems; recency of and days disturbed by memories; and recency and days having problems paying attention or with self-control.
Any past-year trauma exposure: (Kilpatrick, Acierno, Resnick, Saunders, & Best, 1997; Kimerling et al., 1999; McHugo et al., 2005; Scott, Coleman-Cowger, & Funk, 2014; ICC = .30) Endorsed any past-year exposure from 25 types of life-threatening situations (e.g., major disaster, very serious accident or fire, being physically assaulted, or raped), seeing another person killed or dead, or badly hurt, or hearing about something horrible that has happened to someone they are close to.
IMDS (α = .95, ICC = .28): 43 past-year items that measure internalizing disorders including somatic, anxiety, depression, traumatic stress, and suicidal/homicidal thoughts.
BCS (α = .95, ICC = .20): 33 past-year items related to externalizing disorders including attention deficit, hyperactivity/impulsivity, and conduct disorder based on the DSM-IV (APA, 1994; 2000).
ADHDS (α = .96, ICC = .22): 18 past-year items associated with attention deficit and hyperactivity disorder.
CDS (α = .85, ICC = .15): 15 past-year items related to conduct disorder symptoms based on DSM-IV.
Readiness for change in substance use (Scott, Dennis, & Foss, 2005; Scott, Foss, & Dennis, 2005)
SES (α = .84, ICC = .46): A modified GAIN scale with five items on whether the participant currently believes she or he could avoid thinking about or using substances in different settings.
PRS (α = .94, ICC = .53): A Texas Christian University (TCU) scale with nine items on whether the participant currently recognizes that she or he has a problem.
POI (ICC = .36): A modified GAIN scale with five items on whether the participant currently sees her or his problems as predictable and solvable (an inverse of learned helplessness).
DHS (α = .88, ICC = .58): A TCU scale with seven items on whether the participant currently wants help with her or his substance-related problems.
TRI (ICC = .42): A summative index from the GAIN with five items on different current issues that would make it more difficult to attend treatment (e.g., other responsibilities, too demanding, doesn’t think it will be help, hard to resist use in current environment, friends likely to encourage use).
IMS (α = .89, ICC = .57): A combination of the (modified) GAIN Treatment Motivation Index and the TCU Treatment Readiness Index with five current items that suggest internal motivation to be in treatment (e.g., believes treatment can help, thinks treatment is needed or last chance, thinks needs treatment for a month or one or more times).
ETPS (α = .84, ICC = .44): A combination of the (modified) GAIN Treatment Motivation Index and the TCU External Pressure Scale with six current items that measure different types and sources of external pressure to be in treatment.
TXPI (ICC = .12): Count of eight current sources of pressure to be in treatment or to go to treatment. Higher scores represent more diverse sources of external pressure on the client to go to/remain in treatment.
Relapse Coping Inventory (Dennis, Foss, & Scott, 2007; Moos, 1993)
RCPCU (α = .84, ICC = .36). Measure of past 90-day support received from family, friends, self-help, and other groups, to be clean and sober.
RCECU (α = .84, ICC = .27). Measure, in the past 90 days, of how much certain environmental situations have helped you to be clean and sober, such as your health, money situation, place and people you live with, etc.
LA (α = .87, ICC = .33). A past 90-day measure of a cognitive, approach coping style that “attempts to understand and prepare mentally for a stressor and its consequences.”
PR (α = .89, ICC = .32). A past 90-day measure of a cognitive, approach coping style that “attempts to construe and restructure a problem in a positive way while still accepting the reality of the situation.”
SG (α = .81, ICC = .32). A past 90-day measure of a behavioral, approach coping style that “attempts to seek information, guidance, or support.”
PS (α = .89, ICC = .32). A past 90-day measure of a behavioral, approach coping style that “attempts to take action to deal directly with the problem.”
CA (α = .85, ICC = .36). A past 90-day measure of a cognitive, avoidance coping style that “attempts to avoid thinking realistically about a problem.”
AR (α = .81, ICC = .39). A past 90-day measure of a cognitive, avoidance coping style that “attempts to react to the problem by accepting it.”
SR (α = .83, ICC = .36). A past 90-day measure of a behavioral, avoidance coping style that “attempts to get involved in substitute activities and create new sources of satisfaction.”
ED (α = .83, ICC = .34). A past 90-day measure of a behavioral, avoidance coping style that “attempts to reduce tension by expressing negative feelings.”

Source. Measures adapted from the GAIN (Dennis, Titus, White, Unsicker, & Hodgkins, 2003) and/or other sources noted.

Note. ICC = intra-class correlation coefficient; FIS = Financial Instability Scale; IAS = Illegal Activity Scale; SFS = Substance Frequency Scale; SPS = Substance Problem Scale; DSM-IV = Diagnostic and Statistical Manual of Mental Disorders (4th ed.; DSM-IV); APA = American Psychiatric Association; SHAS = Self-Help Activity Scale; NPS = Needle Problem Scale; NFS = Needle Frequency Scale; SXRS = Sexual Risk Scale; TSXS = Trading Sex Scale; EPS = Emotional Problem Scale; IMDS = Internal Mental Distress Scale; BCS = Behavior Complexity Scale; ADHDS = Attention Deficit/Hyperactivity Disorder Scale; CDS = Conduct Disorder Scale; SES = Self-efficacy Scale; GAIN = Global Appraisal of Individual Needs; PRS = Problem Recognition Scale; POI = Problem Orientation Index; DHS = Desire for Help Scale; TRI = Treatment Resistance Index; IMS = Internal Motivation Scale; ETPS = External Pressure Scale; TXPI = Treatment Pressure Index; RCPCU = Relapse/Coping Personal Support; LA = Logistic Analysis; PR = Positive Reappraisal; SG = Seeking Guidance and Support; PS = Problem Solving; CA = Cognitive Avoidance; AR = Acceptance or Resignation; SR = Seeking Alternative Rewards; ED = Emotional Discharge.

Data Analysis

Design

This longitudinal study examines the ability of 55 time-varying predictors to improve our ability to predict the time to recidivism over 3 years after controlling for baseline risk. At each quarter, the most recent version of the variable (from the past quarter or past year) will be used to predict the days to recidivism in the next quarter. The value of these predictors can change every quarter, the longitudinal analysis establishes temporal precedence, and all results are above and beyond factors measured at baseline that were the focus of the prior article (and most recidivism analyses). Bivariate analysis looked at the effect of each variable individually (controlling only for baseline risk). Stepwise multivariate analysis was then used to simplify the model. Figure 1 shows the relationship between the dependent variable of recidivism and how it is predicted from fixed measures at baseline and time-varying measures that were re-assessed either annually or quarterly. On the top half of the figure, the dashed line represents the cumulative percent re-arrested or re-incarcerated over the time since release. The blue histogram shows the number of women who have not been arrested and who are remaining in the analysis at a specific point in time. In this analysis, baseline predictors are measured once (at the time of entering the jail) and maintain the same value at all time points. The value of time-varying predictors changes each time they are measured, which can be either annually or quarterly depending on the variables. For example, in the 18-month analysis, future arrest will be predicted for 192 women using a composite measure of variables measured pre-release, annual variables using values measured from the 12-month observation, and quarterly variables using values measured at the 18-month interview.

Figure 1:

Design for Time-Varying Model of Risk for Predicting Recidivism

Analyses

Survival analysis was done with Cox regression, modeling the time from release to recidivism as a function of the baseline composite predictor as fixed predictors (FP) and the repeated measures as time-varying predictors (TVP). Because our focus is on the marginal improvement associated with adding repeated measurement of predictors, all analyses control for baseline predictors. In the first set of analyses, each of the potential TVPs were tested one at a time while controlling only for baseline risk. All significant TVPs (p < .05) from this first step were retained for the multivariate stepwise analyses unless there was multi-collinearity (discussed further below).

In the stepwise model, the baseline aggregate risk variable along with the TVPs were entered into the model using the stepwise procedure, entering the significant (p < .05) variable with the lowest p value at each step, and then testing all variables in the model and removing any that were no longer significant (p < .05). All statistical analyses were done using IBM SPSS Version 22.

Multi-Collinearity

When two or more predictors were significant in the bivariate analysis, but were also dependent (e.g., number of crimes and categorization of number of crimes) or highly correlated (e.g., multiple measures of substance use), we ran intermediate stepwise Cox regression analyses on the subset of variables with multi-collinearity and kept only those that stayed in the model (i.e., the best version).

Missing Data Replacement

When using TVPs in a Cox regression, the model requires complete data in all of the time periods. Maximum likelihood is not available as an option when using Cox regression with time-varying predictors (aka covariates). Instead, we carried forward the prior value within individual or used weighted hot deck across individuals as recommended by methodological experts (Little & Rubin, 2014; Schafer & Graham, 2002). For follow-up waves that were missing data, if there was a valid value for that variable in the previous follow-up wave, the missing data were replaced with the previous wave for the same individual. For the first follow-up quarterly and annual wave where there is no time for recidivism post baseline, a weighted hot-deck procedure was used across individuals. This involved sorting by the baseline value of the predictor variable, then using the median (for categorical) or mean (for continuous) of the four surrounding cases (the two above and the two below) to replace missing data. In practice, this had little impact, as the proportion of missing data in the 55 variables used in these analyses ranged from 0% to 4.9% with 52 (95%) variables missing under 1%. This replacement method allowed us to include 447 or 93% of the total study participants in the final multivariate analyses.

Results

Bivariate Analysis

Table 2 shows TVP measured over the follow-up period, either quarterly or annually, the mean and standard deviations for continuous predictors, and the overall percent for categorical predictors across all observations. The results of the bivariate (controlling only for baseline predictors) are displayed as OR and their 95% confidence intervals. Continuous predictors were converted to z scores, so that ORs are based on a change of one standard deviation. Per above, all the bivariate analyses controlled for the baseline risk of recidivism. Of the 55 variables tested in the bivariate analyses, 39 variables were significant even after controlling for baseline risk.

Table 2:

Cox Regression to Predict Time to Recidivism 3 Years Post Release (n = 447)

Time-varying predictors	When measured	% or M (SD)	Univariate controlling for baseline risk		Multivariate: Forward stepwise with days in jail
Time-varying predictors	When measured	% or M (SD)	OR	95% CI	OR	95% CI
Control variables
Baseline risk	Baseline	0.70 (0.12)	11.72	[1.78, 28.72]	4.08	[1.60, 10.43]
Demographics
Lived with own children in the past year	Y	40%	0.78	[0.62, 0.98]	0.79	[0.62, 0.99]
Lived with other children in the past year	Y	10%	0.91	[0.62, 1.33]
Custody
No children	Y	41%	1.00	NA
No custody of own children	Y	20%	1.24	[0.93, 1.65]
Custody of any of own children	Y	38%	0.96	[0.75, 1.23]
Any days worked (past 90)^a	Q	20%	0.71	[0.53, 0.96]
Financial Instability Scale	Q	0.15 (0.19)	1.13	[1.02, 1.24]
Recency of family member in jail^b
Never	Y	58%	1.00	NA
1+ years ago		29%	0.76	[0.59, 0.98]
Past year		13%	1.12	[0.83, 1.50]
Criminological risk factors
IAS	Q	.07 (0.12)	1.37	[1.28, 1.46]	1.12	[1.02, 1.23]
GCS	Q	0.2 (0.8)	1.27	[1.18, 1.36]
Any violent crime	Q	3%	2.61	[1.76, 3.85]
Any property crime	Q	5%	3.41	[2.45, 4.75]	1.85	[1.24, 2.77]
Any AOD crime	Q	4%	3.36	[2.35, 4.81]
Any prostitution	Q	2%	3.15	[1.99, 4.98]
Any problems with probation or parole	Q	1%	7.16	[3.78, 13.54]	5.91	[3.01, 11.62]
Any other crime	Q	3%	2.73	[1.89, 3.94]	1.69	[1.11, 2.58]
Days in a jail/prison (prior Q)
0 days	Q	82%	1.00		1.00
1-53 days		9%	4.04	[2.50, 6.54]	4.16	[2.50, 6.93]
54-90 days		9%	3.43	[1.36, 8.64]	3.90	[1.51, 10.06]
Substance use
Substance Frequency Scale	Q	0.15 (0.19)	1.21	[1.09, 1.33]
Substance Problem Scale	Y	4.7 (5.5)	1.33	[1.19, 1.48]
Days of treatment	Q	7.23 (21.1)	1.06	[0.95, 1.148]
Self-Help Activity Scale	Q	3.0 (5.2)	0.86	[0.77, 0.96]	0.83	[0.73, 0.94]
HIV-risk behaviors
Needle Problem Scale	Y	0.1 (0.5)	1.18	[1.08, 1.29]
Needle Frequency Scale	Q	0.01 (0.05)	1.11	[1.02, 1.22]
Sexual Risk Scale	Q	2.7 (3.5)	1.15	[1.04, 1.26]
Number of sexual partners	Q	1.1 (4.2)	1.22	[1.14, 1.31]	1.14	[1.03, 1.26]
Trading Sex Scale (past month)	Q	0.1 (0.8)	1.13	[1.03, 1.24]
Mental health
Emotional Problem Scale	Q	0.10 (0.14)	1.15	[1.05, 1.26]
Any past-year trauma exposure	Y	47%	1.49	[1.19, 1.86]	1.32	[1.04, 1.66]
IMDS	Y	2.9 (6.5)	1.01	[0.91, 1.13]
BCS	Y	1.8 (5.0)	1.10	[0.99, 1.22]
Attention Deficit/ Hyperactivity Disorder Scale	Y	1.3 (3.8)	1.08	[0.97, 1.19]
Conduct Disorder Scale	Y	0.5 (1.6)	1.13	[1.03, 1.24]
Readiness for change in substance use
Self-Efficacy Scale (current)	Q	19.2 (3.8)	0.86	[0.77, 0.95]
Problem Recognition Scale (current)	Q	23.4 (8.6)	1.41	[1.27, 1.56]
Problem Orientation Index (current)	Q	17.9 (2.9)	1.17	[1.05, 1.31]	1.19	[1.06, 1.34]
Desire for Help Scale (current)	Q	21.8 (6.1)	1.32	[1.18, 1.47]
Treatment Resistance Index (current)	Q	11.5 (3.1)	1.19	[1.08, 1.32]
Internal Motivation Scale (current)	Q	19.8 (6.1)	1.37	[1.23, 1.53]
External Pressure Scale (current)	Q	14.1 (4.3)	1.46	[1.32, 1.62]	1.32	(1.18, 1.47)
Treatment Pressure Index (current)	Q	0.1 (0.4)	1.18	[1.08, 1.28]
Relapse Coping Inventory
Personal support	Q	8.5 (4.8)	1.06	[0.96, 1.18]
Environmental support	Q	6.2 (3.7)	0.86	[0.78, 0.96]	0.84	[0.75, 0.94]
Logistic analysis	Q	10.2 (4.7)	0.90	[0.81, 0.99]
Positive reappraisal	Q	11.6 (4.8)	0.90	[0.81, 0.99]
Seeking guidance and support	Q	10.1 (4.5)	0.91	[0.82, 1.01]
Problem solving	Q	11.5 (4.8)	0.87	[0.79, 0.96]
Cognitive avoidance	Q	10.0 (4.8)	1.03	[0.93, 1.15]
Acceptance or resignation	Q	8.8 (4.7)	1.01	[0.91, 1.12]
Seeking alternative rewards	Q	8.6 (4.8)	0.85	[0.77, 0.95]
Emotional discharge	Q	7.7 (4.2)	1.03	[0.92, 1.15]

Note. Y indicates the predictor was measured annually, and a Q indicates it was measured every quarter. Bold indicates p < .05. Significant increase in fit comparing the multivariate model to baseline risk alone, χ²(13) = 175.74, p < .001. OR = odds ratio; CI = confidence interval; IAS = Illegal Activity Scale; GCS = General Crime Scale; IMDS = Internal Mental Distress Scale; BCS = Behavior Complexity Scale.

Used employment activity scale based on intermediate stepwise regression. ^bUsed any past-year trauma exposure based on intermediate stepwise regression.

Stepwise Cox Regression Model

A stepwise multivariate Cox regression analysis simplified the model to 12 significant variables. Further to the right in Table 2, the results of the multivariate (controlling for everything in the column) Cox regression analyses are shown in terms of adjusted odds ratios (AOR) and their 95% confidence intervals. Variables that were associated with significant reductions in the odds of recidivism included (a) having lived with one’s children in the past year (AOR = 0.79), (b) Self-Help Activity Scale (AOR = 0.83), and (c) Environmental Support Scale (AOR = 0.84).

The following variables (measured quarterly unless noted) predicted a significant increase in the odds of recidivism over time: (a) Illegal Activity Scale (AOR = 1.12), (b) any property crime (AOR = 1.85), (c) other types of crime (AOR = 1.69), (d) other problems with probation or parole (AOR = 5.91), (e) days in jail or prison in the preceding quarter, with 1 to 53 days (AOR = 4.16) and 54 to 90 days (AOR = 3.90) relative to none, (f) number of sexual partners (AOR = 1.14), (g) any past-year trauma exposure (AOR = 1.32), (h) Problem Orientation Scale (AOR = 1.19), (i) External Pressure Scale (AOR = 1.32), and (j) the composite baseline risk variable (AOR = 4.08). In most cases, the strength of the predictors remained approximately the same or decreased in the multivariate model, as compared with the bivariate, particularly with regard to the baseline risk variable, which was reduced by more than half (from OR = 11.7 to AOR = 4.08).

Post Hoc Correlation Analysis

Although measures are already based on four- to 40-item scales designed to measure specific constructs, these scales have some collinearity. Using an alpha method factor analysis, we can collapse the 55 scales into 11 factors with eigenvalue more than 1. With varimax rotation and minimal loss of statistical information, we can further simplify it to four orthogonal factors: (a) substance use, substance problems, readiness for change, and time in treatment; (b) illegal activities, conduct disorder symptoms, trauma exposure, and time in jail; (c) number of sexual partners, sexual risk, trading sex, and prostitutions; and (d) as seeking alternative rewards, logistic analysis, Self-Help Activity Scale, recovery coping environment, and last, alcohol- or other drug-related crime. We chose not to use the factors for two reasons. First, they have to collapse across items that are measured at different frequencies (e.g., intake only, annual, quarterly). Second, each factor has some negative loadings (the last items of each factor), so summary factor score alone would hide key differences in their impact on recidivism. Instead, we addressed these issues in Table 3 by examining the correlations among the 12 predictors that entered into the final multivariate model with variables that were significantly related to recidivism at the bivariate level, but not in the multivariate model. Such correlations help reveal shared variance that was subsumed by one variable while maintaining the recognizable constructs.

Table 3:

Zero-Order Correlations Between Time-Varying Predictors From the Final Model (Columns) by Other Significant Predictors

Direction	Significant in bivariate	Risk factors									Protective factors
Direction	Significant in bivariate	Illegal activity scale	Any property crime	Any other crime	Number of sexual partners	Any past-year trauma exposure	Grouped days incarcerated	Problem orientation index	External pressure scale	Any problems w/ probation/parole	Lived with own children in past year	Self-help activity scale	Environmental support
Risk	General Crime Scale	.61	.73	.39	.15	.17	.05	.05	.16	.11	−.01	−.05	−.01
Risk	Any drug crime	.51	.34	.26	.04	.11	.05	.04	.12	.11	−.04	−.03	−.02
Risk	Any violent crime	.32	.37	.27	.08	.14	.03	.02	.08	.06	.01	−.04	−.01
Risk	Any prostitution	.55	.21	.12	.42	.11	.06	.08	.16	.06	−.03	−.02	−.01
Risk	Trading sex in past month	.41	.14	.10	.36	.10	.00	.05	.15	.04	−.04	−.05	−.01
Risk	Sexual Risk Scale	.22	.16	.10	.20	.17	−.12	.03	.07	.05	−.03	−.07	.00
Risk	Needle Frequency Scale	.09	.07	−.01	.10	.06	.02	.05	.06	.01	−.05	.02	−.02
Risk	Conduct Disorder Scale	.30	.35	.11	.02	.22	.11	.08	.19	.08	.00	.01	.04
Risk	Substance Problem Scale	.28	.18	.11	.11	.22	.17	.21	.44	.15	−.12	.07	.03
Risk	Substance Frequency Scale	.26	.17	.10	.15	.02	−.18	.03	.23	.04	−.08	−.31	−.10
Risk	Emotional Problem Scale	.26	.21	.16	.08	.35	.06	.10	.20	.07	−.01	.04	.04
Risk	Problem Recognition Scale	.24	.17	.09	.12	.08	.17	.31	.58	.13	−.08	−.04	−.01
Risk	Financial Instability Scale.	.20	.17	.12	.10	.28	−.03	.05	.16	.10	−.01	.01	.02
Risk	Needle Problem Scale	.01	.01	−.01	.02	.05	.10	.05	.02	−.03	−.08	−.01	.01
Risk	Internal Motivation Scale	.21	.14	.09	.09	.15	.17	.32	.71	.13	−.09	.05	.06
Risk	Desire for Help Scale	.19	.13	.08	.09	.12	.19	.45	.60	.13	−.12	.19	.09
Risk	Treatment Pressure Index	.09	.07	.07	.00	.13	.07	.08	.37	.16	−.02	.13	.04
Risk	Treatment Resistance Index	.13	.09	.05	.08	.06	.02	.04	.31	.03	−.03	−.20	−.10
Protective	Seeking alternative rewards	−.04	−.03	−.03	−.04	.06	−.06	.18	.05	−.01	.00	.31	.30
Protective	Problem solving	−.03	−.02	−.01	−.04	.11	−.06	.20	.02	.01	−.01	.28	.29
Protective	Logistic analysis	.04	.04	.02	.00	.14	−.02	.22	.09	.02	−.02	.23	.27
Protective	Positive reappraisal	.03	.03	.02	−.01	.15	−.02	.22	.08	.01	.01	.22	.28
Protective	Self-Efficacy Scale	−.22	−.15	−.07	−.13	−.04	−.03	.08	−.32	−.06	.10	.21	.13
Protective	Days worked (past 90)	−.07	−.05	−.01	−.03	−.10	−.15	−.07	−.20	−.06	.01	.05	.12

Note. Bold: p < .05.

Any Property Crime is a significant predictor of recidivism at the multivariate level; it is positively correlated with Any Drug Crime, Any Violent Crime, and the General Crime Scale (r = .34, .37, and .73, respectively), indicating a high degree of clustering among criminal behaviors. Similarly, Number of Sex Partners was a significant predictor of recidivism in the multivariate model, but was also correlated with Any Prostitution (r = .42), the Trading Sex Scale (r = .36), and the Sexual Risk Scale (r = .20). This indicates that sexual risk behaviors are strongly, but not exclusively, associated with prostitution and sex trading. However, the key point is that the general measure of sexual risk was found to be a better predictor of recidivism than other more traditionally used measures. Furthermore, Number of Sexual Partners had a small, but significant, correlation with Needle Frequency Scale (r = .10), indicating a shared pattern of risk behaviors. Past-year trauma, which increased the odds of recidivism by more than 30%, was significantly correlated with the past-year Substance Problem Scale (r = .22), the Emotional Problems Scale (r = .35), and the Financial Instability Scale (r = .28), indicating the clustering of problems associated with traumatic exposure.

The External Pressure Scale was another significant predictor of recidivism in the multivariate model, but was significantly correlated with several measures indicating internal motivation or problem recognition, including the Internal Motivation Scale (r = .71), the Desire for Help Scale (r = .60), the Problem Recognition Scale (r = .58), the Substance Problem Scale (r = .44), the Treatment Pressure Index (r = .37), the Treatment Resistance Index (r = .31), and the Problem Orientation Index (r = .22); it was negatively associated with the Self-Efficacy Scale (r = −0.32) and the number of days worked in past 90 (r = −.20). Thus, there are complex relationships among factors that may facilitate treatment or problem recognition, rather than simple inverse relationships among internal and external components.

Among the variables having a protective effect (i.e., inversely related to the time to recidivism), the Self-Help Activity Scale was significantly associated (inversely) with the Substance Frequency Scale (r = −.31) and the Treatment Resistance Index (r = .20); it was positively associated with several other protective factors, including Seeking Alternative Rewards (r = .31), Problem Solving (r = .28), Logistic Analysis (r = .23), Positive Reappraisal (r = .22), and the Self-Efficacy Scale (r = .21). Similar correlations were found with Relapse/Coping Environmental Support and other indicators of positive coping. Therefore, self-help involvement is a critical factor that is protective against recidivism, but is also associated with a range of other cognitive, behavioral, and environmental factors that contribute to this protective effect.

Because the follow-up data were collected in the context of an experiment, we also tested the model with and without the randomization. This did not change the direction or magnitude of the effects in the final model.

Discussion

This study extends prior research predicting recidivism from baseline data (Scott, Grella, et al., 2014) conducted with a sample of women offenders who participated in a drug treatment program in Cook County Jail by examining the relative contributions of multiple post-release time-varying predictors of recidivism over 3 years. Although more than half of the sample returned to custody within the first year following release, another 20% had recidivated by the end of the 3-year observation period. Beyond the baseline variables that addressed behaviors and circumstances prior to initial detention, recidivism was a function of changes in a range of variables subsequent to release to the community. These included time-varying factors related to their parental status, sex-related risk behaviors, recovery capital (e.g., involvement in recovery activities, cognitive processes related to treatment orientation, and environmental support), criminal behavior, and substance use.

Drawing on the area of gender-responsive factors, both parental status and risky sexual behaviors were associated with recidivism in inverse ways. Women who lived with their children in the prior year had reduced odds of recidivism within the next year over the duration of follow-up. This finding is consistent with a study involving participants in a boot camp intervention that found that having more children reduced the risk of recidivism among women (not specific to substance users; Benda, 2005). Other studies have shown that the presence of children is a protective factor among women offenders with substance use disorders in that it has been associated with higher levels of treatment motivation or participation (Grella & Rodriguez, 2011; Saum, Hiller, Leigey, Inciardi, & Surratt, 2007) and lower risk of substance use relapse following release (Saxena, Grella, & Messina, 2015). Yet, one study found that women who had received treatment for opioid use were less likely to be using heroin at a 3-year follow-up if they had children in their care at treatment intake, but were more likely to be using other substances, including alcohol and other opioids (Comiskey, 2013). It is important to note that the relationship between parental status and recidivism may be bi-directional, in that women who are less likely to relapse and/or recidivate may be more likely to retain custody of their children; thus, their parental status is not necessarily the “cause” of their lower risk of recidivism, but a function of it. Nevertheless, interventions for women offenders that address coping strategies for parental stress, parenting skills, and parental support may be beneficial (Dowden & Andrews, 1999), particularly for women who are currently involved with child welfare services.

Having a greater number of sexual partners was predictive of recidivism. Furthermore, this variable was associated with both sex work involvement and injection risk. This finding is important, as HIV-risk behaviors among parolees may be ignored within re-entry treatment planning (Belenko, Langley, Crimmins, & Chaple, 2004). In our prior studies with this sample, involvement in prostitution was a strong predictor of returning to custody within the first 90 days following release. In the current study, although prostitution following release was significant at the bivariate level, after controlling for baseline risk, it was not retained in the final model. Thus, a greater number of sex partners may be a more significant predictor, as it may encompass sex partners in both sex exchanges and personal relationships. Interventions to reduce risky sexual behaviors and associated recidivism therefore need to encompass the broader range of sexual partners, rather than solely focusing on sex work interactions. Prior research has shown that women offenders have differential perceptions of the power dynamics within these kinds of relationships, and their relationship with risky sexual behaviors (Knudsen et al., 2008). A recent multi-site experimental trial found that a relationship-focused intervention was more effective in reducing unprotected sexual encounters in the initial period following release from prison among women offenders (Knudsen, Staton-Tindall, Oser, Havens, & Leukefeld, 2014).

Although prior studies have established that a history of childhood and adult trauma exposure is associated with women’s involvement in criminal behavior (Messina & Grella, 2006; Moloney, van den Bergh, & Moller, 2009), the current study extends this literature by demonstrating that trauma exposure subsequent to release is a significant predictor of recidivism. Nearly half of the study sample reported any past-year trauma exposure over the follow-up period, which increased the risk of recidivism by approximately one third. The mechanisms through which trauma exposure leads to criminal behavior recidivism needs further explication, although the post hoc correlational analysis showed that trauma exposure was associated with financial, substance use, and emotional problems during the follow-up period. As noted previously, Salisbury and Van Voorhis (2009) identified a criminal behavior pathway to incarceration among women that is characterized by trauma exposure and associated substance use and mental health problems. Whether this same constellation is explanatory of post-release criminal behavior involvement still needs additional exploration.

The study findings further contribute to the literature by demonstrating that recovery capital was strongly associated with a reduced risk of recidivism, as seen in the protective effects of greater self-help involvement and recovery-oriented environmental support. Furthermore, self-help involvement was associated with several indicators of functional cognitive and coping strategies and higher self-efficacy. Continuing care interventions aimed at this population can incorporate a focus on providing “recovery-support services” (Laudet & Humphreys, 2013, p. 126) for women offenders. Post-release interventions that include linkage to 12-step groups and 12-step facilitation may help to promote engagement in self-help activities within the community. Innovative interventions, such as cellphone-based applications, may be a promising means for delivering recovery-support services to this population (Scott, Johnson, & Dennis, 2013).

Finally, the multivariate findings showed the predictive effects of several time-varying predictors that derive from the RNR model, which included variables that typify the “gender-neutral” model. These include a higher degree of criminal involvement, as seen in the overall Illegal Activity Scale, Property Crime Involvement, and other types of criminal involvement. Having problems with one’s probation or parole status significantly reduced the time to recidivism (by a factor of six), but this may indicate the effects of closer supervision, which then triggers a return to custody. Although a powerful predictor, we note that only about 1% of the sample indicated an affirmative response to this item, averaged over the follow-up period. Similarly, having spent any time in prison or jail in the previous quarter was strongly associated with recidivism. Higher scores on the measure of External Pressure also reduced time to recidivism; because this measure indicates pressure from others to enter treatment, it may serve as an indicator of problem severity.

Limitations

This study utilized a cohort of women sampled from a jail in Cook County, Illinois. As such, the study cohort is not representative of populations outside this setting, and this may limit generalizability of study findings. However, the sample displayed characteristics of substance use, criminal behavior history, parental status, trauma exposure, and mental health problems that have been well-established in prior studies of women offenders, in both jail and prison settings (Green et al., 2005; Moloney et al., 2009; Pelissier & Jones, 2005; Zlotnick, 1997). Another potential concern is the large number of variables used in the analysis. In a longitudinal analysis, the typical rule of thumb is to have at least 10 observations per variable. Based on the number of observations that are free to vary (shown in the “n surviving” row of Figure 1), there are 2,819 observations where we can predict potential recidivism in the next wave. (Note this excludes Wave 36 because there is no subsequent period.) Thus, there is an average of 51 observations per variable (2,819 observations divided by 55 variables). Even adjusting the observations for the correlation of observations on the same person over time, there are still more than 10 observations per variable.

Conclusion

Taken collectively, the study findings demonstrate that the recidivism of women offenders with drug problems is related not only to their status at baseline (the focus of most research to date), but to what happens to them post release after community re-entry. These findings highlight the need for ongoing intervention tailored to their specific needs in the areas of parenting, trauma exposure, and recovery support to reduce their risk of returning to custody. Providing these services is likely to help the women to maintain their recovery and thereby reduce recidivism and its associated costs to their communities and society.

Footnotes

Acknowledgements

The authors thank the women who participated in the interviews, the Cook County Jail Division 17 staff who accommodated the researchers and helped extract all the recidivism data, and Art Lurigio and Janet Titus for feedback on the article.

This article was supported by National Institute on Drug Abuse (NIDA) Grants DA021174 and DA016383.

The opinions are those of the authors and do not reflect official positions of the government.

Christy K. Scott, PhD, is the director of the Lighthouse Institute and the Illinois Survey Laboratory at Chestnut Health Systems in Chicago, Illinois. Her research focuses on understanding and predicting how people move through cycles of substance use, crime, treatment, incarceration, and periods of recovery, as well as how to experimentally test strategies for improving recovery management over time. She has published widely on recovery management, how to achieve better than 90% follow-up rates in longitudinal studies, intensive data collection with smartphones, and developed and tested different interventions for managing addiction over time.

Christine E. Grella, PhD, is professor-in-residence in the Department of Psychiatry and Biobehavioral Sciences at University of California, Los Angeles (UCLA) and co-director of the UCLA Integrated Substance Abuse Programs. Her research focuses on the intersection of multiple service delivery systems, including substance abuse treatment, mental health, child welfare, health services, HIV services, and criminal justice, and the relationship of service delivery with treatment outcomes. She has published her work widely in the areas of addiction, mental health, health services, and evaluation research.

Michael L. Dennis, PhD, is a senior research psychologist at the Lighthouse Institute and director of the GAIN Coordinating Center at Chestnut Health Systems in Normal, Illinois. His research also focuses on understanding and predicting how people move through cycles of substance use, and crime, treatment, incarceration, and periods of recovery, as well as how to experimentally test strategies for improving recovery management over time. He has published widely on recovery management, integrating clinical and research assessment, measurement, intensive data collection with smartphones, and evaluation research.

Rodney R. Funk is a research associate at the Lighthouse Institute of Chestnut Health Systems in Normal, Illinois. He holds SPSS’s highest certification for data mining methods and modeling. His research interests are in the analysis of complex hierarchical and intensive data analytic models, as well as using clustering techniques to interpret heterogeneous samples and trends over time.

References

Adams

S. M.

Peden

A. R.

Hall

L. A.

Rayens

M. K.

Staten

R. R.

Leukefeld

C. G.

(2011). Predictors of retention of women offenders in a community-based residential substance abuse treatment program. Journal of Addiction Nursing, 22, 103-116. doi:10.3109/10884602.2011.585719

American Psychiatric Association. (1994). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: Author.

American Psychiatric Association. (2000). Diagnostic and statistical manual of mental disorders (4th ed., text rev.). Washington, DC: Author.

Andrews

D. A.

Bonta

Wormith

S. J.

(2006). The recent past and near future of risk and/or need assessment. Crime & Delinquency, 52, 7-27. doi:10.1177/0011128705281756

Andrews

D. A.

Guzzo

Raynor

Rowe

R. C.

Rettinger

L. J.

Brews

Wormith

J. S.

(2012). Are the major risk/need factors predictive of both female and male reoffending? A test with the eight domains of the level of service/case management inventory. International Journal of Offender Therapy and Comparative Criminology, 56, 113-133. doi:10.1177/0306624X10395716

Belenko

Langley

Crimmins

Chaple

(2004). HIV risk behaviors, knowledge, and prevention education among offenders under community supervision: A hidden risk group. AIDS Education and Prevention, 16, 367-385. doi:10.1521/aeap.16.4.367.40394

Benda

B. B.

(2005). Gender differences in life-course theory of recidivism: A survival analysis. International Journal of Offender Therapy and Comparative Criminology, 49, 325-343. doi:10.1177/0306624X04271194

Blitz

C. L.

Wolff

Pan

K. Y.

Pogorzelski

(2005). Gender-specific behavioral health and community release patterns among New Jersey prison inmates: Implications for treatment and community reentry. American Journal of Public Health, 95, 1741-1746. doi:10.2105/AJPH.2004.059733

Bloom

Owen

Covington

(2003). Gender-responsive strategies: Research, practice, and guiding principles for women offenders (NIC018017). Washington, DC: U.S. Department of Justice, National Institute of Corrections.

10.

Brennan

Breitenbach

Dieterich

Salisbury

E. J.

Van Voorhis

(2012). Women’s pathways to serious and habitual crime: A person-centered analysis incorporating gender responsive factors. Criminal Justice and Behavior, 39, 1481-1508. doi:10.1177/0093854812456777

11.

Bureau of Justice Statistics. (2005). Prison statistics: Summary findings. Washington, DC: U.S. Department of Justice.

12.

Comiskey

C. M.

(2013). A 3 year national longitudinal study comparing drug treatment outcomes for opioid users with and without children in their custodial care at intake. Journal of Substance Abuse Treatment, 44, 90-96. doi:10.1016/j.jsat.2012.04.002

13.

Dennis

M. L.

Foss

M. A.

Scott

C. K.

(2007). An eight-year perspective on the relationship between the duration of abstinence and other aspects of recovery. Evaluation Review, 31, 585-612. doi:10.1177/0193841X07307771

14.

Dennis

M. L.

Titus

J. C.

White

Unsicker

Hodgkins

(2003). Global Appraisal of Individual Needs (GAIN): Administration guide for the GAIN and related measures (Version 5). Bloomington, IL: Chestnut Health Systems. Retrieved from www.gaincc.org/gaini

15.

Dowden

Andrews

D. A.

(1999). What works for female offenders: A meta-analytic review. Crime & Delinquency, 45, 438-452. doi:10.1177/0011128799045004002

16.

Evans

Buoncristiani

Hser

Y. I.

(2014). Perceived neighborhood safety, recovery capital, and successful outcomes among mothers 10 years after substance abuse treatment. Substance Use & Misuse, 49, 1491-503. doi:10.3109/10826084.2014.913631

17.

Granfield

Cloud

(2001). Social context and “natural recovery”: The role of social capital in the resolution of drug-associated problems. Substance Use & Misuse, 36, 1543-1570. doi:10.1081/JA-100106963

18.

Granfield

Cloud

(2008). Conceptualizing recovery capital: Expansion of a theoretical construct. Substance Use & Misuse, 43, 1971-1986. doi.org/10.1080/10826080802289762

19.

Green

B. L.

Miranda

Daroowalla

Siddique

(2005). Trauma exposure, mental health functioning, and program needs of women in jail. Crime & Delinquency, 51, 133-151. doi:10.1177/0011128704267477

20.

Grella

C. E.

Lovinger

Warda

U. S.

(2013). Relationships among trauma exposure, familial characteristics, and PTSD: A case-control study of women in prison and in the general population. Women & Criminal Justice, 23, 63-79. doi:10.1080/08974454.2013.743376

21.

Grella

C. E.

Rodriguez

(2011). Motivation for treatment among women offenders in prison-based treatment and longitudinal outcomes among those who participate in community aftercare. Journal of Psychoactive Drugs, 43(Suppl. 1), 58-67. doi:10.1080/02791072.2011.602275

22.

Harris

P. W.

Lockwood

Mengers

Stoodley

B. H.

(2011). Measuring recidivism in juvenile corrections. OJJDP Journal of Juvenile Justice, 1(1), 1-16.

23.

Holtfreter

Reisig

M. D.

Morash

(2004). Poverty, state capital, and recidivism among women offenders. Criminology & Public Policy, 3, 185-208. doi:10.1111/j.1745-9133.2004.tb00035.x

24.

Huebner

B. M.

DeJong

Cobbina

(2010). Women coming home: Long-term patterns of recidivism. Justice Quarterly, 27, 225-254. doi:10.1080/07418820902870486

25.

Kilpatrick

D. G.

Acierno

Resnick

H. S.

Saunders

B. E.

Best

C. L.

(1997). A 2-year longitudinal analysis of the relationships between violent assault and substance use in women. Journal of Consulting and Clinical Psychology, 65, 834-847. doi:10.1037/0022-006X.65.5.834

26.

Kimerling

Calhoun

K. S.

Forehand

Armistead

Morse

. . . Clark

(1999). Traumatic stress in HIV-infected women. AIDS Education and Prevention, 11, 321-330.

27.

Knudsen

H. K.

Leukefeld

Havens

J. R.

Duvall

J. L.

Oser

C. B.

Staton-Tindall

. . . Inciardi

J. A.

(2008). Partner relationships and HIV risk behaviors among women offenders. Journal of Psychoactive Drugs, 40, 471-481. doi:10.1080/02791072.2008.10400653

28.

Knudsen

H. K.

Staton-Tindall

Oser

C. B.

Havens

J. R.

Leukefeld

C. G.

(2014). Reducing risky relationships: A multisite randomized trial of a prison-based intervention for reducing HIV sexual risk behaviors among women with a history of drug use. AIDS Care, 26, 1071-1079. doi:10.1080/09540121.2013.878779

29.

Kranzler

H. R.

Tennen

Babor

T. F.

Kadden

R. M.

Rounsaville

B. J.

(1997). Validity of the longitudinal, expert, all data procedure for psychiatric diagnosis in patients with psychoactive substance use disorders. Drug and Alcohol Dependence, 45, 93-104. doi:10.1016/S0376-8716(97)01349-5

30.

Kroner

D. G.

Loza

(2001). Evidence for the efficacy of self-report in predicting nonviolent and violent criminal recidivism. Journal of Interpersonal Violence, 16, 168-177. doi:10.1177/088626001016002005

31.

Kroner

D. G.

Mills

J. F.

(2001). The accuracy of five risk appraisal instruments in predicting institutional misconduct and new convictions. Criminal Justice and Behavior, 28, 471-489. doi:10.1177/009385480102800405

32.

Langan

N. P.

Pelissier

(2001). Gender differences among prisoners in drug treatment. Journal of Substance Abuse, 13, 291-301. doi:10.1016/S0899-3289(01)00083-9

33.

Laudet

A. B.

Humphreys

(2013). Promoting recovery in an evolving policy context: What do we know and what do we need to know about recovery support services? Journal of Substance Abuse Treatment, 45, 126-133. doi:10.1016/j.jsat.2013.01.009

34.

Laudet

A. B.

White

W. L.

(2008). Recovery capital as prospective predictor of sustained recovery, life satisfaction, and stress among former poly-substance users. Substance Use & Misuse, 43, 27-54. doi:10.1080/10826080701681473

35.

Little

R. J.

Rubin

D. B.

(2014). Statistical analysis with missing data. Hoboken, NJ: John Wiley.

36.

Lynch

S. M.

DeHart

D. D.

Belknap

J. E.

Green

B. L.

Dass-Brailsford

Johnson

K. A.

Whalley

(2014). A multisite study of the prevalence of serious mental illness, PTSD, and substance use disorders of women in jail. Psychiatric Services, 65, 670-674. doi:10.1176/appi.ps.201300172

37.

Matheson

F. I.

Doherty

Grant

B. A.

(2011). Community-based aftercare and return to custody in a national sample of substance-abusing women offenders. American Journal of Public Health, 101, 1126-1132. doi:10.2105/AJPH.2010.300094

38.

McHugo

G. J.

Caspi

Kammerer

Mazelis

Jackson

E. W.

Russell

. . . Kimerling

(2005). The assessment of trauma history in women with co-occurring substance abuse and mental disorders and a history of interpersonal violence. Journal of Behavioral Health Services and Research, 32, 113-127. doi:10.1007/BF02287261

39.

Messina

Grella

C. E.

(2006). Childhood trauma and women’s health in a California prison population. American Journal of Public Health, 96, 1842-1848. doi:10.2105/AJPH.2005.082016

40.

Minton

T. D.

(2011). Jail inmates at midyear 2010—Statistical tables. Washington, DC: Bureau of Justice Statistics. Retrieved from http://www.bjs.gov/index.cfm?ty=pbdetail&;iid=2375

41.

Moloney

K. P.

van den Bergh

B. J.

Moller

L. F.

(2009). Women in prison: The central issues of gender characteristics and trauma history. Public Health, 123, 426-430. doi:10.1016/j.puhe.2009.04.002

42.

Moos

R. H.

(1993). Coping responses inventory: Adult form. Odessa, FL: Psychological Assessment Resources.

43.

Pelissier

Jones

(2005). A review of gender differences among substance abusers. Crime & Delinquency, 51, 343-372. doi:10.1177/0011128704270218

44.

Pilkonis

P. A.

Heape

C. L.

Ruddy

Serrao

(1991). Validity in the diagnosis of personality disorders: The use of the LEAD standard. Psychological Assessment: A Journal of Consulting and Clinical Psychology, 3, 46-54. doi:10.1037/1040-3590.3.1.46

45.

Reisig

M. D.

Holtfreter

Morash

(2002). Social capital among women offenders: Examining distribution of social networks and resources. Journal of Contemporary Criminal Justice, 18, 167-187. doi:10.1177/1043986202018002004

46.

Sacks

(2004). Women with co-occurring substance use and mental disorders (COD) in the criminal justice system: A research review. Behavioral Sciences & the Law, 22, 449-466. doi:10.1002/bsl.597

47.

Salisbury

E. J.

Van Voorhis

(2009). Gendered pathways: A quantitative investigation of women probationers’ paths to incarceration. Criminal Justice and Behavior, 36, 541-566. doi:10.1177/0093854809334076

48.

Sarteschi

C. M.

Vaughn

M. G.

(2010). Double jeopardy: A review of women offenders’ mental health and substance abuse characteristics. Victims & Offenders: An International Journal of Evidence-Based Research, Policy, and Practice, 5, 161-182. doi:10.1080/15564881003640728

49.

Saum

Hiller

Leigey

Inciardi

Surratt

(2007). Predictors of substance abuse treatment entry for crime-involved, cocaine-dependent women. Drug and Alcohol Dependence, 91, 253-259. doi:10.1016/j.drugalcdep.2007.06.005

50.

Saxena

Grella

C. E.

Messina

N. P.

(2015). Continuing care and trauma in women offenders’ substance use, psychiatric status and self-efficacy outcomes. Women and Criminal Justice, 26, 1-23.

51.

Schafer

J. L.

Graham

J. W.

(2002). Missing data: Our view of the state of the art. Psychological Methods, 7, 147-177. doi:10.1037/1082-989X.7.2.147

52.

Schram

P. J.

Koons-Witt

B. A.

Williams

F. P.

McShane

M. D.

(2006). Supervision strategies and approaches for female parolees: Examining the link between unmet needs and parolee outcome. Crime & Delinquency, 52, 450-471. doi:10.1177/0011128705281845

53.

Schwalbe

C. S.

(2007). Risk assessment for juvenile justice: A meta-analysis. Law and Human Behavior, 31, 449-462.

54.

Scott

C. K.

Coleman-Cowger

V. H.

Funk

R. R.

(2014). Trauma and posttraumatic stress disorder among substance-using women entering Cook County Jail. Women & Criminal Justice, 24, 44-61. doi:10.1080/08974454.2013.842518

55.

Scott

C. K.

Dennis

M. L.

(2012). The first 90 days following release from jail: Findings from the Recovery Management Checkups for Women Offenders (rMCWO) experiment. Drug and Alcohol Dependence, 125, 110-118. doi:10.1016/j.drugalcdep.2012.03.025

56.

Scott

C. K.

Dennis

M. L.

Foss

M. A.

(2005). Utilizing recovery management checkups to shorten the cycle of relapse, treatment reentry, and recovery. Drug and Alcohol Dependence, 78, 325-338. doi:10.1016/j.drugalcdep.2004.12.005

57.

Scott

C. K.

Dennis

M. L.

Lurigio

A. J.

(2013). Women’s participation in a jail-based treatment program in a large urban setting: A process evaluation. Offender Programs Report, 17(4), 49-59.

58.

Scott

C. K.

Foss

M. A.

Dennis

M. L.

(2005). Pathways in the relapse–treatment–recovery cycle over 3 years. Journal of Substance Abuse Treatment, 28(2), S63-S72. doi:10.1016/j.jsat.2004.09.006

59.

Scott

C. K.

Grella

Dennis

M. L.

Funk

R. R.

(2014). Predictors of recidivism over 3 years among substance-using women released from jail. Criminal Justice and Behavior, 41, 1257-1289. doi:10.1177/0093854814546894

60.

Scott

C. K.

Johnson

Dennis

M. L.

(2013). Using mobile phone technology to provide recovery support for women offenders. Telemedicine and e-Health, 19, 767-771. doi:10.1089/tmj.2012.0309

61.

Spitzer

R. L.

(1983). Psychiatric diagnosis: Are clinicians still necessary? Comprehensive Psychiatry, 24, 399-411. doi:10.1016/0010-440X(83)90032-9

62.

Spjeldnes

Goodkind

(2009). Gender differences and offender reentry: A review of the literature. Journal of Offender Rehabilitation, 48, 314-335. doi:10.1080/10509670902850812

63.

Staton-Tindall

Duvall

McNees

Walker

Leukefeld

(2011). Outcomes following prison and jail-based treatment among women residing in metro and non-metro communities following release. Journal of Drug Issues, 41, 197-216. doi:10.1177/002204261104100203

64.

Staton-Tindall

Frisman

Lin

Leukefeld

Oser

Havens

. . . Clarke

(2011). Relationship influence and health risk behavior among re-entering women offenders. Women’s Health Issues, 21-3, 230-238. doi:10.1016/j.whi.2010.10.006

65.

Steadman

H. J.

Redlich

Callahan

Robbins

P. C.

Vesselinov

(2011). Effect of mental health courts on arrests and jail days: A multisite study. Archives of General Psychiatry, 68, 167-172. doi:10.1001/archgenpsychiatry.2010.134

66.

Taxman

F. S.

Marlowe

(2006). Risk, needs, responsivity: In action or inaction? Crime & Delinquency, 52, 3-6. doi:10.1177/0011128705281757

67.

Van Voorhis

(2005). Classification of women offenders: Gender-responsive approaches to risk/needs assessment. Community Corrections Report, 12(2), 19-27.

68.

Van Voorhis

(2012). On behalf of women offenders: Women’s place in the science of evidence-based practice. Criminology & Public Policy, 11, 111-145. doi:10.1111/j.1745-9133.2012.00793.x

69.

Zlotnick

(1997). Posttraumatic stress disorder (PTSD), PTSD comorbidity, and childhood abuse among incarcerated women. Journal of Nervous and Mental Disease, 185, 761-763.