LEAD Program Evaluation: The Impact of LEAD on Housing, Employment and Income/Benefits Seema L. Clifasefi Heather S. Lonczak Susan E. Collins Harm Reduction Research and Treatment Center University of Washington – Harborview Medical Center March 31, 2016 This report was prepared by the University of Washington LEAD Evaluation Team with important contributions from the LEAD Evaluation Advisory Committee and others acknowledged on the back page. Executive Summary • • • Background: Seattle’s Law Enforcement Assisted Diversion (LEAD) program offers people suspected of low-level drug and prostitution offenses legal assistance and harm-reductionoriented case management as an alternative to prosecution and incarceration. Purpose: This report describes findings for LEAD participants in terms of their housing, employment, and income/benefits both prior and subsequent to their referral to LEAD. Primary analyses also tested whether the number of contacts LEAD participants had with their case managers contributed to these findings. Additional, secondary analyses examined the associations between positive participant outcomes (i.e., attainment of housing, employment and income/benefits) and involvement in the criminal justice system (i.e., recidivism). Findings: o Primary Analyses: Participants were significantly more likely to obtain housing, employment and legitimate income in any given month subsequent to their LEAD referral (i.e., during the 18-month follow-up) compared to the month prior to their referral (i.e., baseline). § Housing: • LEAD participants were over twice as likely to be sheltered (e.g., permanent housing, temporary housing, emergency shelter, motel/hotel) versus unsheltered (e.g., sleeping on the streets, in abandoned buildings) during the follow-up. Further, each contact participants had with LEAD case managers was associated with a 2% increase in the likelihood of obtaining shelter during the follow-up. • Participants were 89% more likely to obtain permanent housing during the follow-up, and each contact they had with their LEAD case manager translated to a 5% higher likelihood of being housed during follow-up. § Employment: LEAD participants were 46% more likely to be on the employment continuum (i.e., in vocational training, employed in the legitimate market, retired) at follow-up versus baseline. § Income/benefits: LEAD participants were 33% more likely to have income/benefits at follow-up versus baseline. o Secondary Analyses: Additional analyses exploring the association between recidivism and obtaining housing, employment and income/benefits showed that housing and employment obtained during LEAD involvement was associated with less recidivism as measured by arrests during the 6-month follow-up. • • Interpretation of findings: o Prior reports showed Seattle’s LEAD program reduces recidivism as well as utilization of and costs for the criminal justice and legal systems. The present report adds to these initial findings by showing that LEAD participants have a higher likelihood of attaining housing, improving employment outcomes, and obtaining legitimate income/benefits after their referral to the LEAD program. o Obtaining housing and employment is associated with less recidivism. o Interpretation of the present findings is limited by the fact that we did not have housing, employment and income data for the control group. Thus, the promising associations between LEAD participation and positive participant outcomes cannot be assumed to be causal. There could be other factors involved in producing the observed LEAD effects. o That said, the fact that increased contact with LEAD case managers predicted better housing outcomes increases our confidence that those effects may be attributable to LEAD. Next Steps: This report is one in a series being prepared by the University of Washington LEAD Evaluation Team over a two-year period. A final report will be released in the summer of 2016 and will document LEAD participants’ perceptions of the LEAD program and its effect on their lives. Introduction Background The US imprisons more of its population than any other country in the world, and incarceration rates, particularly among drug offenders, have increased exponentially since 2008.1-3 Given that up to 3.5 million Americans experience homelessness in any given year,4 homelessness also represents a large and growing problem in America.5,6 Although the relationship between homelessness and criminal recidivism is not well understood, the two appear to be inexorably linked,5-7 with one representing a risk factor for the other.8-10 For example, the prevalence of homelessness among incarcerated offenders is 7.5 to 11.3 times that found in the general population.11 Further, incarceration is disproportionately high among homeless individuals.5 For example, a recent study indicated that nearly one-quarter of homeless and marginally housed individuals had a history of incarceration.8 Collectively, studies have shown that people with unstable housing are more frequently arrested, are incarcerated longer, and are re-arrested at higher rates than people with stable housing.12,13 Both homeless and incarcerated populations share certain characteristics that are noteworthy, including housing instability, unemployment, poverty, lack of job skills/training, and substance-use problems.6,11,14-18 Substance-use problems and ensuing drug offenses represent a particular challenge to policy makers because traditional policing efforts have not been found to improve public safety or decrease recidivism for drug offenders.3,19-21 Drug offenders instead often cycle through the criminal justice system with such frequency that this phenomenon is referred to as a “revolving door.”22 A recent study has indicated that the standard approach of prosecution and incarceration may contribute to the revolving door phenomenon by decreasing opportunities to obtain housing, employment and legitimate income/benefits, thereby confining offenders to continued work in illegal markets.23 These individuals thus lack the resources needed to stop the revolving door of homelessness and incarceration.8 Considering the numerous undesirable consequences of repeat drug offending at both the individual and societal level, there have been calls for innovative programs to engage repeat offenders and help them stop this detrimental cycle.22 Over the past decade, studies have shown that case management, which entails connecting individuals with community services to help them meet their basic needs, is an essential component of programs that precipitate significant reductions in recidivism. For example, one study of a re-entry program offering case management (i.e., housing assistance, job training, financial assistance) to male offenders showed significant reductions in arrest rates.24 Additionally, an evaluation of a pretrial diversion program featuring intensive, client-centered case management reported reduced arrest and incarceration rates.25 Connecting individuals to services is a key component of case management; however, the nature of the relationship between clients and case managers also appears to be important to successful program outcomes. In a large-scale population-based study, Chinman and colleagues showed that a stronger therapeutic alliance between homeless clients and case managers was associated with fewer days of homelessness and better quality of life.26 Taken together, research to date highlights the importance of building a strong case management relationship and connecting individuals with services to maximally impact housing, employment and income/benefit outcomes for repeat offenders. Seattle’s LEAD Program An understanding of this literature, the needs of this population, and potential viable solutions led to the development of Seattle’s LEAD program. LEAD is a collaborative, prebooking diversion program that offers individuals suspected of low-level drug and prostitution offenses legal assistance and harm-reduction-oriented case management instead of prosecution and incarceration. The primary aim of LEAD is to reduce criminal recidivism. Secondary aims include reductions in criminal justice and legal system utilization and associated costs as well as improvements in outcomes directly impacting participants’ lives, including housing, employment and legitimate income/benefits. Because LEAD is the first known prebooking diversion program of its kind in the United States, an evaluation is critically needed to inform key stakeholders, policy makers, and other interested parties of its impact. For evaluation purposes, the implementation phase of this project occurred from October 2011 through January 2014. The Seattle Police Department’s officer shifts were randomly divided into ‘red- and greenlight’ shifts. Offenders who were encountered during greenlight shifts were screened for project eligibility by officers on duty and, provided they met inclusion criteria and completed the intake process, they were offered LEAD at the point of arrest instead of undergoing standard criminal booking and prosecution. Additional participants were referred by officers as ‘social contacts.’ Social contacts were individuals who were eligible for LEAD but were recruited by officers outside of a criminal incident. Participants encountered during redlight shifts were randomized to the “system-as-usual” control group. As the original catchment area’s potential participant population dwindled, additional control participants were recruited through an adjacent catchment area where they were encountered by the same officers who made referrals in the original LEAD catchment area. Only LEAD-eligible individuals were included in the control group. Participants were then referred to a LEAD case manager to complete an intake assessment. After completing the intake process, participants received legal assistance as well as case management through Evergreen Treatment Services’ (ETS) REACH homeless outreach program. The REACH Program LEAD case managers are trained and supervised by ETS’s REACH program. As part of ETS, which is a Western Washington-based nonprofit organization that delivers addiction treatment services, the REACH program provides outreach and harm-reduction-oriented case management to individuals experiencing homelessness and substance use disorders. The LEAD priority population that REACH serves includes a high percentage of individuals who have been repeatedly involved in the criminal justice system and are considered vulnerable and ‘hard-toreach.’ REACH is guided by the mission of “joining with individuals through outreach, relationship building, advocacy, and bridging gaps to reduce harm and support healing” (K. Craig, personal communication, February 8, 2016). REACH espouses a trauma-informed, harm-reduction approach, which entails meeting participants ‘where they are at’ in their communities and in their own motivation to change. The program’s case management model is highly individualized and uses a nonjudgmental, collaborative approach in which the client’s own needs and priorities are the primary focus of attention. In this model, the goals are to engage and retain individuals in services by listening attentively to clients’ needs and connecting them with appropriate community resources, such as housing placement, medical care, legal advocacy, job training, mental health counseling, and chemical dependency treatment. To do this work, REACH employs a diverse and interdisciplinary team of professionals and paraprofessionals with backgrounds in nursing, social work, chemical dependency counseling and related disciplines. REACH case managers emphasize building and maintaining a trusting and supportive relationship with clients. Case management is provided on the streets, in clients’ living situations, and onsite at REACH’s home office. In the context of LEAD, case managers also have access to funds for the fulfillment of participants’ basic needs (e.g., motel stays during cold weather, food, clothing, treatment). Overall, REACH’s client-centered, theoretically grounded approach promotes self-efficacy and motivation to change by facilitating access to services and developing a flexible and compassionate outreach relationship. Overall Program Evaluation Aims The overall program evaluation was designed to assess the LEAD program in meeting the following objectives. • Specific aim 1 is to test the relative effectiveness of the LEAD program compared to the ‘system-as-usual’ control condition in reducing criminal recidivism (i.e., arrests and charges). • Specific aim 2 is to test the effectiveness of the LEAD program compared to the ‘systemas-usual’ control condition in reducing publicly funded legal and criminal justice service utilization and associated costs (i.e., prosecution, public defense, jail, prison) subsequent to evaluation entry. • Specific aim 3 is to test within-subjects differences on housing, employment and income variables subsequent to LEAD program entry. • Specific aim 4 is to explore LEAD participants’ perceptions of the program in their own words. Findings from specific aims 1 and 2 were released in reports in March 2015 and June 2015, respectively. The current report (specific aim 3) reviews housing, employment and income/benefits outcomes for LEAD participants subsequent to LEAD involvement. A final report documenting qualitative findings for specific aim 4 will be released in Summer 2016. Purpose and Methods Design This report documents changes for LEAD participants on housing, employment and income/benefit outcomes after their entry into the LEAD program. Because data were only available for LEAD participants, the design is a single-arm, within-subjects analysis of outcomes for the one month prior to the LEAD program referral (baseline) and for any given month of the 18 months subsequent to the LEAD program referral (follow-up). Participants Participants were adults who were suspected of low-level drug or prostitution offenses and were offered and diverted to LEAD instead of booking and prosecution as usual. Based on whether law enforcement contact was made during a red- or greenlight shift and whether it occurred in the LEAD catchment area, participants were either assigned to the LEAD (n = 203) or control (i.e., booking as usual; n = 115) conditions. At the time of referral, 146 of the LEAD participants were under arrest, and 57 were suspected of qualifying criminal activity but were referred outside of an alleged criminal incident as social contacts. Because the data for the current report were collected by REACH case managers in the course of their work with LEAD participants, only LEAD participants--not control participants--were included in the present analyses. Further, housing, employment and income data were available for a smaller subset of the original, intent-to-treat sample; thus, 176 participants made up the sample featured in the present report. All LEAD participants were suspected of recent violations of the uniform controlled substances act (VUCSA) and/or prostitution offenses and were deemed eligible for the program by SPD officers. SPD considered individuals ineligible if they met any of the following criteria: • The amount of drugs involved exceeded 3 grams, except where an individual was arrested for delivery of or possession with intent to deliver marijuana or possession, delivery or possession with intent to deliver prescription controlled substances (pills). • The individual did not appear amenable to diversion. • The suspected drug activity involved delivery or possession with intent to deliver (PWI), and there was reason to believe the suspect was dealing for profit above a subsistence income. • The individual appeared to exploit minors or others in a drug dealing enterprise. • The individual was suspected of promoting prostitution. • The individual had a disqualifying criminal history as follows: o Without time limitation: Any conviction for murder 1 or 2, arson 1 or 2, robbery 1, assault 1, kidnapping, Violation of the Uniform Firearms Act (VUFA) 1, any sex offense, or attempt of any of these crimes. o Within the past 10 years: Any conviction for a domestic violence offense, robbery 2, assault 2 or 3, burglary 1 or 2, or VUFA 2. o The individual was already involved in King County Drug Diversion Court or Mental Health Court. This exclusion criterion served to ensure the LEAD program was not combined with other models of intervention and case management. Measures The evaluation team obtained all necessary IRB exemptions and data sharing agreements from the appropriate entities for the purposes of conducting these analyses. Demographic data were obtained via SPD and REACH case management records. Case management contacts were defined as any phone or in-person communications between a REACH case manager and a LEAD participant lasting at least 5 minutes. Contact data were logged by case managers and stored in the REACH database (AGENCY Software, Seattle, WA). Baseline housing, employment and income/benefit statuses were based on participants’ retrospective self-report to REACH case managers at their intake into the LEAD program. Ongoing housing, employment and income/benefit data were obtained by REACH case managers throughout their work with clients and were documented in the REACH database. Housing outcomes entered into the REACH database at any given time point were coded using the federal definition of homelessness (i.e., lacking a fixed, regular and adequate nighttime residence; having a primary nighttime dwelling that is not a regular sleeping accommodation; living in a supervised shelter or transitional housing; exiting an institution that served as temporary residence when the individual had previously resided in a shelter or place not meant for human habitation; or facing imminent loss of housing when no subsequent residence is identified and insufficient resources/support networks exist.27 For the housed versus unhoused outcome, this variable was recoded, where 1 = permanent housing and 0 = homelessness for any given month during the baseline and 18-month follow-up period. For the sheltered versus unsheltered outcome, this variable was recoded, where 1 = sheltered and either housed or homeless (e.g., permanent housing, temporary housing, emergency shelter, motel/hotel) and 0 = unsheltered homeless (e.g., sleeping on the streets, in abandoned buildings). Employment outcomes for any given time point included being in vocational training/internship; in legitimate, paid employment; retired from legitimate employment; unemployed; and unable to work. Participants were classified as ‘unable to work’ based on formal legal and medical determination. Employment data were recoded for the employed versus unemployed outcome, such that 1 = part or full time legitimate employment and 0 = all others. Employment data were recoded for the employment continuum versus nonemployment continuum outcome, where 1 = being in vocational training/internship; in legitimate, paid employment; or retired from legitimate employment, and 0 = unemployed or unable to work. Income/benefits outcomes for any given time point included AFDC/TANF; Aged, blind, or disabled (ABD) funding; supplemental security income (SSI); social security disability insurance (SSDI); income from legitimate full or part time employment; pensions; unemployment compensation; veterans benefits; or no legitimate income. Income/benefits data were recoded, such that 1 = any legitimate income/benefits and 0 = no legitimate income/benefits. Data Analysis Plan Using SPSS 19 and Stata 13, descriptive analyses were conducted to describe the sample, ascertain the nature of the data distributions, and detect potential outliers. Primary analyses. Population-averaged generalized estimating equations28 (GEEs) were used in primary analyses. GEEs model marginal effects and may be used to accommodate alternative distributions (e.g., binomial) and correlated data (e.g., data collected on the same participant over time). In this evaluation, GEEs were used to test the relative prediction of housing, income and employment outcomes by: a) time (0=baseline, 1=follow-up), which accounted for overall, pre- to post referral longitudinal effects; b) case management contacts; and c) the two-way time x case management contacts interaction. The interaction effect reflects changes on the outcomes over time as a function of the intensity of a participant’s exposure to case management. Because outcomes were dichotomous, we specified Bernoulli distributions with the logit link. We assumed an exchangeable correlation structure to accommodate repeated measures on one individual, which served as the sole clustering variable.29 To enhance model interpretability, resulting effect sizes were exponentiated and reported as odds ratios (ORs), where ORs < 1 indicate an inverse association, ORs = 1 indicate no association, and ORs > 1 indicate a positive association. Alphas were set to p = .05, indicating statistically significant results. Confidence intervals were set to 95%. Two sets of models were used. The first set of models included main effects of covariates (i.e., age, gender, race/ethnicity and death during the study), time and case management contacts. The second set of models included the time x case management contact interactions, to test whether there were differential pre- to post referral effects as a function of participants’ exposure to LEAD case management. The bestfitting model was determined by the lowest quasilikelihood under the independence model information criterion (QICu) score, with lower scores indicating the more parsimonious model.29 Secondary analyses. Additional analyses were conducted linking the housing, income and employment data from the current report to recidivism data from the initial March 2015 report. Given the well-documented associations between housing status and involvement in the criminal justice system,5,8,9 we tested months participants spent in housing, were engaged on the employment continuum, and had legitimate income/benefit sources during the 6-month follow-up as correlates of recidivism (arrest, charges) during the 6-month follow-up. Associations were tested using logistic regression models that controlled for baseline recidivism, demographic characteristics and death. Alphas were set to p = .05, indicating statistically significant results. Confidence intervals were set to 95%. Results Overall Sample Description Participants in this phase of the project (N = 176) had an average age of 42.62 (SD = 11.01) years and were predominantly male (39.20% female; n = 69). The racial and ethnic diversity of the overall sample is shown in Figure 1. Figure 1. Ethnic/racial backgrounds of parOcipants European American 26% African American 57% Hispanic or Laono/a 4% Other 2% Naove American/ Alaska Naove/ Pacific Islander 6% Asian American 1% More than one race 4% Comparing Arrest Diversion and Social Contact Participants Of the baseline demographic and outcome variables (i.e., housing status, employment, income/benefits), the arrest diversion and social contact groups significantly differed on participant age (p = .02) and housing status (i.e., housed versus unhoused; p = .01; other ps > .18). Specifically, arrest diversion participants were younger (M = 41.36, SD = 10.94) than social contact participants (M = 45.55, SD = 10.71) and more likely to be housed than social contact participants (24% versus 6%, respectively). Thus, age was included, along with other demographic variables, as a covariate in all analyses, and group was included as a covariate in the housed status analyses. Pre- and Post-referral Descriptive Statistics of Outcomes by Group Descriptive statistics for raw, unadjusted housing, income and employment outcomes were calculated prior and subsequent to referral to LEAD (see Table 1). Table 1. Unadjusted descriptive statistics for primary outcomes Outcome Measures Pre-LEAD referral Housing Sheltered versus unsheltered Housed versus unhoused Employment Employed versus not employed On employment continuum versus not on employment continuum Income Having legitimate income/benefits versus not 48.30% 17.61% 7.43% 8.57% Post-LEAD referral 65.83% 28.49% 9.03% 11.83% 51.76% 57.45% Note: This table features unadjusted values. Postreferral values are comprised of the percentage of individuals fitting that category averaged over each month of the 18-month follow-up period. Primary Analyses Housing status. The model for sheltered status was significant, Wald Χ2(6, N = 176) = 39.51, p < .001, QICu = 4221. After controlling for sociodemographic variables, LEAD participants were over twice as likely to have been sheltered in any given month during the follow-up versus baseline (OR = 2.08, robust SE = .25, p < .001). The interaction model was also significant, Wald Χ2(7, N = 176) = 43.93, p < .001, QICu = 4215, and showed that each contact with a case manager was associated with a 2% higher likelihood of being sheltered in any given month during the follow up compared to baseline (OR = 1.02, robust SE = .01, p < .001). See Figure 2 below for the relationship of contact with case managers and sheltered status from baseline to follow-up. See Appendix A for full output. Note: For the purposes of graphing, number of contacts was split using the interquartile range, where a low number of contacts th th th th is less than the 25 percentile, medium is between the 25 and 75 percentiles, and high is greater than the 75 percentile. The model for housing status was likewise significant, Wald Χ2(7, N = 176) = 36.70, p < .001, QICu = 3823. After controlling for demographic variables and social contacts/arrest diversion study entry, LEAD participants were 89% more likely to have been housed at some point during the follow-up versus baseline (OR = 1.89, robust SE = .24, p < .001). The interaction model was also significant, Wald Χ2(8, N = 176) = 33.96, p < .001, QICu = 3802, and showed that each contact with a case manager was associated with a 5% higher likelihood of being housed during the follow-up compared to baseline (OR = 1.05, robust SE = .01, p < .001). See Figure 3 below for the relationship of contact with case managers and housing status from baseline to followup. See Appendix A for full output. Note: For the purposes of graphing, number of contacts was split using the interquartile range, where a low number of contacts th th th th is less than the 25 percentile, medium is between the 25 and 75 percentiles, and high is greater than the 75 percentile. Employment status. The model predicting being employed versus not employed was significant, Wald Χ2(6, N = 176) = 18.56, p = .005, QICu = 1871. There were, however, no significant changes in employment over time or as a function of case management contacts (ps > .14). Similarly, the interaction effect was not significant (p = .89) in the full model, Wald Χ2(7, N = 176) = 19.49, p = .007, QICu = 1873. The model predicting being on the employment continuum (i.e., legitimately employed, in vocational training, or retired from legitimate employment) was significant, Wald Χ2(6, N = 176) = 14.03, p = .029, QICu = 2261. After controlling for demographic variables, LEAD participants were 46% more likely to have been on the employment continuum at some point during the follow-up versus at baseline (OR = 1.46, robust SE = .22, p = .013). The interaction model was, however, not significant (p > .05, QICu = 2263). Case management contact was not a significant predictor in either model (ps > .51). Income/benefits. The main effects model for having legitimate income/benefits was significant, Wald Χ2(6, N = 176) = 21.60, p = .001, QICu = 4327. After controlling for demographic variables, LEAD participants were 33% more likely to have received legitimate income/benefits during the follow-up versus at baseline (OR = 1.33, robust SE = .12, p = .002). The interaction model was significant, Wald Χ2(7, N = 176) = 22.09, p = .003, QICu = 4332; however, neither case management contacts nor the time x case management interaction were significant predictors in either model (ps > .28). Associations between Housing, Employment and Income/Benefits and Recidivism Outcomes The omnibus model for likelihood of arrest was significant, χ2(8, N = 176) = 24.24, p = .002. After controlling for baseline recidivism and demographic variables, the number of months housed (OR = .83, SE = .07, p = .01) and the number of months spent on the employment continuum (OR = .68, SE = .12, p = .02) were significant predictors of recidivism. For each additional month housed, participants were 17% less likely to have been arrested during the 6month follow-up. For each additional month spent on the employment continuum (i.e., in job training, legitimately employed or retired), participants were 33% less likely to have been arrested. This pattern held even after warrant arrests were removed, χ2(8, N = 176) = 22.92, p = .004. For each additional month housed, participants were 17% less likely to have been arrested during the 6-month follow-up (OR = .83, SE = .08, p = .045). For each additional month spent on the continuum to employment (i.e., in job training, legitimately employed or retired), participants were 41% less likely to have been arrested (OR = .59, SE = .15, p = .041). The omnibus models for overall charges and felony charges were not significant (ps > .05). Discussion This report documents housing, employment, and income/benefits outcomes subsequent to individuals’ referral to LEAD as well as the potential additive effects of the amount of contact with LEAD case managers. Secondary analyses tested associations between participants’ outcomes (i.e., housing, employment and income/benefits) and recidivism (i.e., arrests, charges) following LEAD referrals. Overall, findings indicated that participants’ housing, employment, and income/benefits outcomes improved subsequent to LEAD program involvement. Moreover, obtaining housing and employment was related to reduced recidivism, as measured by arrests. Housing Outcomes Housing analyses tested LEAD participants’ likelihood of being sheltered (versus unsheltered) and housed (versus unhoused) at baseline versus any given month in the year and a half after their LEAD referral. Findings indicated that participants were twice as likely to have been sheltered, and were 89% more likely to have obtained permanent housing after their LEAD referral. Given that approximately 82% of the current sample was homeless at baseline, achieving better housing outcomes was a key goal of LEAD case managers. This goal appears to have been fulfilled, with a 62% increase in participants housed over the course of the study. This outcome is particularly impressive when considering Seattle’s limited housing stock and the state of emergency regarding homelessness declared by both the City of Seattle and King County in November 2015.30 The present findings are also in line with those of other studies, which have shown that both the amount of contact and the perceived therapeutic alliance established during homeless outreach and case management are essential predictors of positive housing outcomes.26,31,32 Additionally, each contact—via phone or in person—that participants had with case managers was associated with an additional 2% higher likelihood of being sheltered and a 5% higher likelihood of being housed subsequent to LEAD referral. The fact that housing outcomes improved as a function of the number of contacts with case managers increases our confidence that REACH case management is a key factor in predicting LEAD participants’ positive housing outcomes. That said, the current evaluation design does not comprise a control group and thus precludes causal attributions for this association. This means that other factors (e.g., participants’ own motivation for change) may have accounted for the number of case management contacts and are thus responsible for the observed effects. Further study is needed to understand the nature of the relationship between case management contacts and housing outcomes and to provide definitive recommendations regarding an ideal number of case management contacts. It is notable that LEAD was associated with improved shelter and housing outcomes despite various systemic challenges. First, Seattle and King County have experienced recent, dramatic increases in homelessness, which have placed considerable stress on local resources for this population. Second, overall housing stock is insufficient, and there are limited housing options for substance-involved individuals with criminal histories. Third, LEAD entailed neither housing ‘set asides’ nor preferential access to housing for LEAD participants,1 which makes it difficult to compare LEAD with programs that do have priority access to housing stock. Instead, the present findings show what can be accomplished within the existing system with the benefit of additional legal assistance, case management, and monies to support case management efforts (e.g., for treatment, emergency shelter, rent assistance). Employment Outcomes The raw percentages of employed LEAD participants were low: 7.4% and 9% of LEAD participants had full- or part-time employment prior and subsequent to their LEAD referrals, respectively. Although this percentage increased slightly over the course of the study, this increase was not statistically significant. When we expanded this analysis to include individuals who fell along the legitimate employment continuum (i.e., participating in vocational training/internships, being employed, being retired from legitimate employment), however, LEAD participants were 46% more likely to be on the employment continuum subsequent to their LEAD referral. The overall numbers of people joining the employment continuum increased from 9% to 12% or by 33%. This finding echoes those of other studies in the literature, which have indicated that criminal justice diversion and case management programs can help individuals make positive steps towards legitimate employment.33,34 Joining the employment continuum is a more realistic outcome to consider for the LEAD priority population than achievement of full or part time employment alone. First, 82% of LEAD 1 This “nondisplacement” principle of LEAD stemmed from the understanding that the program was meant to achieve community-wide gains in public order and safety if taken to scale. If LEAD participants had gained access to services with a wait list, thus driving other similarly situated people further down or off the wait list, the net impact on the community might have been neutral or negative, even though results for individuals in LEAD were positive. participants were homeless. Considering the vulnerability of this priority population, many individuals were ineligible to work due to chronic physical or mental health disabilities. Further, as repeated drug and prostitution offenders, LEAD participants had been working in illegal markets and regularly cycling in and out of the street-to-jail-to-street revolving door for some time. Taken together, these population characteristics likely complicated and slowed participants’ complete reintegration into mainstream, legitimate employment. Thus, participants’ significant movement along the employment continuum is both realistic and highly encouraging. We did not observe effects of case management contacts on employment outcomes. In retrospect, this lack of a significant finding is not surprising given the various factors influencing whether a person is ready, willing and/or able to engage along the employment continuum.14,3538 Many of these factors, including the severity of existing disabilities, income status, client motivation, job readiness and availability of suitable positions, are outside of a case manager’s control. This evaluation also featured a follow-up period of 18 months. Such a relatively short period of time may be adequate for simpler and more achievable goals, such as obtaining shelter and housing. It may not, however, be adequate for case managers to help participants fully achieve such multistep, complex tasks as attaining and maintaining full time employment. For example, employment attainment trajectories may involve case managers helping LEAD participants obtain housing (most employers require a permanent address), secure a stateissued identity card, complete vocational training, acquire appropriate interview attire, attend interviews, complete hiring paperwork, find transportation to work, and perform adequately on the job. Income/benefits Outcomes LEAD participants were 33% more likely to be connected to income/benefits subsequent to their LEAD involvement. Sources of income/benefits included income stemming from legitimate employment (e.g., wages, unemployment benefits, military pensions) as well as income from state and federal sources (e.g., ABD, SSI, SSDI, TANF). We did not, however, observe a significant association between number of contacts with case managers and income/benefits outcomes. Similar to employment, however, there are many mediating factors that determine an individual’s ability to secure income and benefits that are not within a case manager’s immediate control. Thus, the further outcomes move away from those that case managers can directly influence, the less of a direct effect we may expect to see. Associations Between Recidivism and Participants’ Housing, Employment and Income/Benefits Status Additional, secondary analyses showed that housing and employment obtained during participants’ LEAD involvement was associated with significantly less recidivism as measured by arrests. In other words, housing and employment appear to serve as independently predictive and protective factors against arrests. This finding corresponds to existing literature showing that employment and housing is associated with reduced risk of recidivism.39,40 Limitations The limitations of this evaluation should be noted. First, administrative datasets often feature missing data and clerical errors. That being said, considerable effort was made by the evaluation team to follow up with administrative sources and ensure we obtained complete and accurate data. Second, specific features of the geographical location of this work likely influenced key characteristics of Seattle’s LEAD program and the resulting evaluation. For example, 82% of LEAD participants in this evaluation were homeless, which certainly influenced the nature of the participants’ needs and the resulting approaches used for case management and legal assistance. Thus, the present findings may not easily generalize to other communities where the LEAD priority population and existing systems (e.g., case management services, housing stock, criminal justice system) may differ. Because we lack a control group for these particular analyses, the present evaluation design is not sufficient to demonstrate causality of the observed effects. In other words, we cannot be sure the changes we have observed are due to the LEAD program versus other confounding factors or statistical phenomena, such as regression to the mean. Thus, causal conclusions cannot be drawn based on these findings. Fortunately, we can conclude that all effects are moving in a positive direction, and thus, LEAD does not appear to have iatrogenic or negative effects for participants. Further, our confidence that the observed effects are attributable at least in part to LEAD is increased by the fact that the number of contacts with case managers predicted positive housing outcomes above and beyond what we would expect due to statistical regression to the mean. Conclusions and Future Directions Findings indicated improvements for LEAD participants across housing, employment, and income/benefit outcomes. Case management appears to play a significant role in ameliorating housing outcomes. Secondary analyses showed that better housing and employment outcomes are associated with reduced recidivism among LEAD participants. Further study of LEAD programs is necessary to understand whether these effects are generalizable to other communities and to draw causal conclusions as to the programmatic components that are driving the observed LEAD effects. Further, future studies should include assessment and analysis of other relevant participant outcomes to elucidate LEAD’s impact on various aspects of participants’ lives. Because these individuals and their communities are particularly impacted by substance use, we are consulting with REACH on data collection protocols to assess new LEAD participants’ substance use at baseline and allow for future evaluation of the effects of LEAD on substance use and substance-related harm.2 Taken together, the findings from our three reports to date suggest positive findings for LEAD and collectively indicate that LEAD is slowing the street-to-jail-to-street revolving door for low-level drug and prostitution recidivating offenders. This report is one in a series being 2 When LEAD was designed, its stated goal was to reduce recidivism of individual participants. Reducing substance use and substance-related harm were not explicitly stated goals. As a result, the original evaluation consultants engaged by the LEAD Policy Coordinating Group prior to the program’s launch suggested methods needed to assess LEAD’s effect on recidivism. The original evaluation consultants were not retained to conduct the current evaluation and were subsequently replaced by the current authorship team. Moving forward, we are consulting with the LEAD Policy Coordinating Group and REACH to establish substance-use assessment protocols for new LEAD participants that will allow future analysis of the effects of LEAD on substance use and substance-related harm. prepared by the University of Washington LEAD Evaluation Team. The next and final report, which we plan to release in Summer 2016, will explore LEAD participants’ perceptions of the program in their own words. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. References Schmitt J, Warner K, Gupta S. The High Budgetary Cost of Incarceration. Washington, DC: Center for Economic and Policy Research; 2010. Motivans M. Federal justice statistics, 2011-statistical tables. 2015; http://www.bjs.gov/index.cfm?ty=pbdetail&iid=1745. Accessed May 29, 2015. Walmsley R. World Population List 10 ed. London: International Centre for Prison Studies; 2013. National Law Center on Homelessness and Poverty. Homelessness in the United States and the Human Right to Housing. 2015; http://www.mplp.org/Resources/mplpresource.2006-06-13.0349156065/file0. Accessed March 7, 2016. Metraux S, Roman C, Cho R. Incarceration and Homelessness. Washington, DC: US Department of Housing & Urban Development; 2008. Shelton K, Taylor P, Bonner A, van den Bree M. Risk factors for homelessness: evidence from a population-based study. Psychiatry Services. 2009;60(4):465-472. Metraux S, Culhane D. Homeless shelter use and reincarceration following prison release. Criminology and Public Policy. 2004;3(2):139-160. Kushel M, Hahn J, Evans J, Bangsberg D, Moss A. Revolving doors: Imprisonment among the homeless and marginally housed population. American Journal of Public Health. 2005;95:1747-1752. Metraux S, Culhane D. Recent incarceration history among a sheltered homeless population Crime and Delinquency. 2006;52:504-517. National Health Care for the Homeless Council. Incarceration & homelessness: A revolving door of risk. In Focus: A Quarterly Research Review of the National HCH Council. 2013;2(2):1-5. Greenberg G, Rosenheck R. Jail incarceration, homelessness, and mental health: A national study. Psychiatric Services. 2008;59:170-177. McNiel D, Binder R, Robinson J. Incarceration associated with homelessness, mental disorder, and cooccurring substance abuse. Psychiatric Services. 2005;56(7):840-860. Weiser S, Neilands T, Comfort M, et al. Gender-specific correlates of incarceration among marginally housed individuals in San Francisco. American Journal of Public Health. 2009;99(8):1459-1463. Burt M, Laudan Y, Aron T, Valente J, Lee E, Iwen B. Homelessness: Programs and the people they serve: Findings of the National Survey of Homeless Assistance Providers and Clients. Washington DC: The Urban Institute; 1999. Conklin T, Lincoln T, Tuthill R. Self-reported health and prior health behaviors of newly admitted correctional inmates. American Journal of Public Health. 2000;90:1939-1941. Freudenburg N. Jails, prisons, and the health of urban populations: A review of the impact of the correctional system on community health. Special Feature: Continuity of Care from Corrections to Community. Journal of Urban Health. 2001;78(2):214-235. Langan P, Levin D. Recidivism of Prisoners Released in 1994. Washington, DC: U.S. Department of Justice, Bureau of Justice Statistics; 2002. Western B, Beckett K. How unregulated is the US labor market? The penal system as a labor market institution. American Journal of Sociology. 1999;104(4):1030-1060. Drug Policy Alliance. A brief history of the drug war 2014; http://www.drugpolicy.org/new-solutionsdrug-policy/brief-history-drug-war. Accessed March 2, 2016. National Research Council. The Growth of Incarceration in the United States: Exploring Causes and Consequences. Washington, DC: The National Academies Press; 2014. Wormith JSO, M. Offender treatment and attrition, and its relationship with risk, responsivity and recidivism. Criminal Justice and Behavior. 2002;29:447-471. Warner T, Kramer J. Closing the revolving door? Substance abuse treatment as an alternative to traditional sentencing for drug-dependent offenders. Criminal Justice and Behavior. 2009;36:89-109. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. Fletcher DR. Offenders in the post-industrial labour market: Lubricating the revolving door? People, Place and Policy. 2013;1:80-89. Braga A, Piehl A, Hureau D. Controlling Violent Offenders Released to the Community: An Evaluation of the Boston Reentry Initiative. Delinquency. 2009;46(4):411-436. Loveland D, Boyle M. Intensive case management as a jail diversion program for people with a serious mental illness: A review of the literature. International Journal of Offender Therapy and Comparative Criminology. 2007;51(2):130-150. Chinman M, Rosenheck R, Lam J. The case management relationship and outcomes of homeless persons with serious mental illness. Psychiatric Services. 2000;51(9):1142 1147. The McKinney-Vento Homeless Assistance Act. 42 USC Section 11302, (2009). Zeger SL, Liang KY. Longitudinal data analysis for discrete and continuous outcomes. Biometrics. 1986;42:121-130. Hardin JW, Hilbe JM. Generalized estimating equations. Boca Raton, FL: Chapman & Hall/CRC; 2003. Beekman D, Broom J. Mayor, county exec declare ‘state of emergency’ over homelessness. The Seattle Times. November 2, 2015. Bybee D, Mowbray CT, Cohen E. Short versus longer term effectiveness of an outreach program for the homeless mentally ill. American Journal of Community Psychology. 1994;22:181-210. Bybee D, Mowbray CT, Cohen E. Evaluation of a homeless mentally ill outreach program: Differential short-term effects Evaluation and Program Planning. 1995;18:13-24. Leukefeld C, Webster JM, Staton-Tindall M, Duvall J. Employment and work among Drug Court clients: 12-Month outcomes. Substance Use and Misuse. 2007;42(7):1109-1126. Zlatic J, M, Wilkerson D, C, McAllister S, M. Pretrial diversion: The over-looked pretrial services evidence-based practice Federal Probation 2010;74 (1):28-33. Acosta O, Toro P. Let’s Ask the Homeless People Themselves: A Needs Assessment Based on a Probability Sample of Adults. American Journal of Community Psychology. 2000;28(3):343-366. Dachner N, Tarasuk V. Homeless “squeegee kids”: Food insecurity and daily survival. Social Science & Medicine. 2002;54(7):1039–1049. Ferguson K, M, Bender K, Thompson S, J, Maccio E, M, Pollio D. Employment status and income generation among homeless young adults: Results from a five-city, mixed-methods study. Youth and Society. 2012;44(3):385–407. National Alliance to End Homelessness. 2016; http://www.endhomelessness.org/pages/economicsecurity. Accessed March 3, 2016. Morenoff J, D, Harding D, J. Final technical report: Neighborhoods, recidivism, and employment among returning prisoners. Ann Arbor, MI: National Institute of Justice, Office of Justice Programs, U.S. Department of Justice;2011. Lowenkamp C, T , Latessa E, J. Increasing the effectiveness of correctional programming through the risk principle: identifying offenders for residential placement. Criminology and Public Policy. 2005;4(2):263–290. ACKNOWLEDGMENTS The authorship team for this report included Drs. Seema L. Clifasefi, Heather S. Lonczak, and Susan E. Collins. We would like to thank the current and former members of the LEAD Evaluation Advisory Committee, including Mark Baird, Mary Barbosa, Mark Cooke, Kelley Craig, Clifton Curry, Lisa Daugaard, Ian Goodhew, Chloe Gale, Ron Jackson, Jutta Joesch, Anita Khandelwal, Kris Nyrop, Amy Tsai, Christa Valles, Natalie WaltonAnderson, and Mike West, for their valuable contributions to this report. We would also like to acknowledge Kenneth Tanzer of AGENCY Software for extracting and preparing the administrative data from the REACH database. We would like to thank Evergreen Treatment Services-REACH management, program staff, and especially, LEAD case managers for helping us clean and manage the LEAD program data. Finally, we thank all of our Harm Reduction Research and Treatment (HaRRT) Center staff for their additional administrative and data management contributions. This evaluation is supported by a grant from the Laura and John Arnold Foundation to Susan E. Collins and Seema L. Clifasefi. LEAD services and project management have been funded by the Ford Foundation, the Open Society Foundations, the RiverStyx Foundation, the Vital Projects Fund, the Massena Foundation and the City of Seattle. Appendix A: Analysis Output Key for abbreviations used in this output qic dsheltered Age … contacts_number , i(id) t(time) family(bin) link(logit) corr(exch) eform robust Analysis type: Outcome: generalized d=dichotomous estimating sheltered = equations with outcome type QICu test of model parsimony/fit Covariates/primary predictors: t=time contacts_number= number of case management contacts txcontacts_number = interaction effect Time point Distribution Case Link identifier as a unique type (binomial) function (participant identifier (logit) ID number) within ID Correlation Requests Robust structure for exponentiated standard panel data coefficients errors to (exchangeable) (ORs) account for correlated data structure Primary analyses . qic dsheltered Age reEthGrp Gender Died t contacts_number if include==1, i(Client_ID) t(time) fam(bin) link(logit) corr(exch) robust eform Iteration 1: tolerance = 6.567e-14 GEE population-averaged model Group variable: Link: Family: Correlation: Client_ID logit binomial independent Scale parameter: 1 Pearson chi2(3287): Dispersion (Pearson): 3286.58 .9998716 Number of obs Number of groups Obs per group: min avg max Wald chi2(6) Prob > chi2 = = = = = = = 3287 176 3 18.7 19 50.05 0.0000 Deviance Dispersion = = 4204.65 1.279176 --------------------------------------------------------------------------------dsheltered Odds Ratio Std. Err. z P> z [95% Conf. Interval] ----------------+---------------------------------------------------------------Age 1.005058 .0038429 1.32 0.187 .9975541 1.012618 reEthGrp 1.04215 .0593434 0.73 0.468 .9320945 1.1652 Gender .7056628 .0608479 -4.04 0.000 .5959363 .8355927 Died 1.370062 .229892 1.88 0.061 .9860777 1.903571 t 2.089565 .3266443 4.71 0.000 1.538136 2.838685 contacts_number 1.004872 .0020531 2.38 0.017 1.000856 1.008904 _cons .7566102 .1807105 -1.17 0.243 .4737719 1.208301 --------------------------------------------------------------------------------Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration 1: 2: 3: 4: 5: 6: 7: 8: 9: tolerance tolerance tolerance tolerance tolerance tolerance tolerance tolerance tolerance = = = = = = = = = .07320543 .01382931 .00331801 .0005853 .0001375 .00002691 6.067e-06 1.259e-06 2.742e-07 GEE population-averaged model Group variable: Client_ID Link: logit Family: binomial Correlation: exchangeable Scale parameter: 1 Number of obs Number of groups Obs per group: min avg max Wald chi2(6) Prob > chi2 = = = = = = = 3287 176 3 18.7 19 39.51 0.0000 (Std. Err. adjusted for clustering on Client_ID) -------------------------------------------------------------------------------- Robust dsheltered Odds Ratio Std. Err. z P> z [95% Conf. Interval] ----------------+---------------------------------------------------------------Age 1.006172 .0141849 0.44 0.663 .9787505 1.034361 reEthGrp 1.053863 .2269067 0.24 0.807 .6910542 1.607148 Gender .6612762 .2099359 -1.30 0.193 .354937 1.232011 Died 1.254462 .5943718 0.48 0.632 .4956248 3.175134 t 2.079075 .2478541 6.14 0.000 1.645869 2.626304 contacts_number 1.002769 .0068408 0.41 0.685 .9894506 1.016267 _cons .7708743 .5631298 -0.36 0.722 .184151 3.226956 --------------------------------------------------------------------------------QIC and QIC_u ___________________________________________ Corr = exch Family = bin Link = logit p = 7 Trace = 73.265 QIC = 4353.047 QIC_u = 4220.517 ___________________________________________ . qic dsheltered Age reEthGrp Gender Died i(Client_ID) t(time) fam(bin) link > (logit) corr(exch) robust eform t contacts_number txcontacts_number if include==1, Iteration 1: tolerance = 6.741e-14 GEE population-averaged model Group variable: Link: Family: Correlation: Client_ID logit binomial independent Scale parameter: Pearson chi2(3287): Dispersion (Pearson): 1 3287.77 1.000233 Number of obs Number of groups Obs per group: min avg max Wald chi2(7) Prob > chi2 = = = = = = = 3287 176 3 18.7 19 55.92 0.0000 Deviance Dispersion = = 4197.64 1.277043 ----------------------------------------------------------------------------------dsheltered Odds Ratio Std. Err. z P> z [95% Conf. Interval] ------------------+---------------------------------------------------------------Age 1.005079 .003844 1.32 0.185 .9975727 1.012641 reEthGrp 1.041748 .0593576 0.72 0.473 .9316706 1.164832 Gender .7049413 .0608486 -4.05 0.000 .5952229 .8348843 Died 1.368318 .2296272 1.87 0.062 .9847838 1.901224 t 1.261022 .3134189 0.93 0.351 .7747505 2.052501 contacts_number .9835599 .0084478 -1.93 0.054 .9671411 1.000257 txcontacts_number 1.023068 .0090423 2.58 0.010 1.005498 1.040945 _cons 1.217427 .3658469 0.65 0.513 .6755384 2.193997 ----------------------------------------------------------------------------------Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration 1: tolerance = .07847544 2: tolerance = .01199298 3: tolerance = .00202913 4: tolerance = .00082846 5: tolerance = .00013771 6: tolerance = .00006402 7: tolerance = .00001555 8: tolerance = 5.458e-06 9: tolerance = 1.537e-06 10: tolerance = 4.915e-07 GEE population-averaged model Group variable: Client_ID Link: logit Family: binomial Correlation: exchangeable Number of obs Number of groups Obs per group: min avg max Wald chi2(7) = = = = = = 3287 176 3 18.7 19 43.93 Scale parameter: 1 Prob > chi2 = 0.0000 (Std. Err. adjusted for clustering on Client_ID) ---------------------------------------------------------------------------------- Robust dsheltered Odds Ratio Std. Err. z P> z [95% Conf. Interval] ------------------+---------------------------------------------------------------Age 1.006227 .0142321 0.44 0.661 .9787155 1.034511 reEthGrp 1.045478 .2259723 0.21 0.837 .6844382 1.596967 Gender .6651334 .2130229 -1.27 0.203 .3550533 1.246017 Died 1.249776 .5959637 0.47 0.640 .4908298 3.182241 t 1.248178 .2145253 1.29 0.197 .89121 1.748128 contacts_number .9848146 .0090115 -1.67 0.094 .9673098 1.002636 txcontacts_number 1.023268 .0073278 3.21 0.001 1.009006 1.037731 _cons 1.174034 .8598203 0.22 0.827 .2794415 4.932535 ----------------------------------------------------------------------------------QIC and QIC_u ___________________________________________ Corr = exch Family = bin Link = logit p = 8 Trace = 76.115 QIC = 4351.152 QIC_u = 4214.923 ___________________________________________ . qic dhoused Age reEthGrp Gender Died TxSbGrp t contacts_number if include==1, i(Client_ID) t(tim > e) fam(bin) link(logit) corr(exch) robust eform Iteration 1: tolerance = 2.217e-11 GEE population-averaged model Group variable: Link: Family: Correlation: Client_ID logit binomial independent Scale parameter: 1 Pearson chi2(3287): Dispersion (Pearson): 3344.48 1.017486 Number of obs Number of groups Obs per group: min avg max Wald chi2(7) Prob > chi2 = = = = = = = 3287 176 3 18.7 19 113.41 0.0000 Deviance Dispersion = = 3772.99 1.147851 --------------------------------------------------------------------------------dhoused Odds Ratio Std. Err. z P> z [95% Conf. Interval] ----------------+---------------------------------------------------------------Age 1.027583 .0043394 6.44 0.000 1.019113 1.036124 reEthGrp .7721077 .0481137 -4.15 0.000 .6833379 .8724093 Gender .898736 .0827759 -1.16 0.246 .7502989 1.07654 Died 1.163846 .1895691 0.93 0.352 .845766 1.60155 TxSbGrp 1.868499 .1772735 6.59 0.000 1.551441 2.250352 t 1.898937 .3880648 3.14 0.002 1.272213 2.834401 contacts_number .9937413 .0022499 -2.77 0.006 .9893413 .9981608 _cons .0816158 .0248214 -8.24 0.000 .0449679 .148131 --------------------------------------------------------------------------------Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration 1: 2: 3: 4: 5: 6: 7: 8: tolerance tolerance tolerance tolerance tolerance tolerance tolerance tolerance = = = = = = = = .20523242 .01682038 .00445555 .00069221 .00006518 .00002024 1.347e-06 5.265e-07 GEE population-averaged model Group variable: Client_ID Link: logit Family: binomial Correlation: exchangeable Scale parameter: 1 Number of obs Number of groups Obs per group: min avg max Wald chi2(7) Prob > chi2 = = = = = = = 3287 176 3 18.7 19 36.70 0.0000 (Std. Err. adjusted for clustering on Client_ID) -------------------------------------------------------------------------------- Robust dhoused Odds Ratio Std. Err. z P> z [95% Conf. Interval] ----------------+---------------------------------------------------------------Age 1.027529 .0165709 1.68 0.092 .9955585 1.060526 reEthGrp .7620567 .1814025 -1.14 0.254 .4779302 1.215095 Gender .8710026 .2876841 -0.42 0.676 .4559062 1.664039 Died 1.318378 .7318231 0.50 0.619 .4441644 3.913239 TxSbGrp 1.343767 .4634255 0.86 0.392 .683549 2.641668 t 1.888952 .2442628 4.92 0.000 1.466057 2.433834 contacts_number 1.002144 .0084416 0.25 0.799 .9857342 1.018826 _cons .090044 .0742976 -2.92 0.004 .0178693 .4537353 --------------------------------------------------------------------------------QIC and QIC_u ___________________________________________ Corr = exch Family = bin Link = logit p = 8 Trace = 96.215 QIC = 3999.661 QIC_u = 3823.232 ___________________________________________ . . qic dhoused Age reEthGrp Gender Died TxSbGrp t contacts_number txcontacts_number if include==1, > i(Client_ID) t(time) fam(bin) link(logit) corr(exch) robust eform Iteration 1: tolerance = 6.881e-13 GEE population-averaged model Group variable: Link: Family: Correlation: Client_ID logit binomial independent Scale parameter: 1 Pearson chi2(3287): Dispersion (Pearson): 3325.64 1.011756 Number of obs Number of groups Obs per group: min avg max Wald chi2(8) Prob > chi2 = = = = = = = 3287 176 3 18.7 19 114.78 0.0000 Deviance Dispersion = = 3762.56 1.14468 ----------------------------------------------------------------------------------dhoused Odds Ratio Std. Err. z P> z [95% Conf. Interval] ------------------+---------------------------------------------------------------Age 1.027668 .0043411 6.46 0.000 1.019194 1.036212 reEthGrp .7715287 .0481256 -4.16 0.000 .6827422 .8718614 Gender .8978951 .0827747 -1.17 0.243 .7494721 1.075711 Died 1.165068 .1898619 0.94 0.348 .8465206 1.603484 TxSbGrp 1.869697 .1776399 6.59 0.000 1.552025 2.252392 t .8284745 .2679885 -0.58 0.561 .4394799 1.561778 contacts_number .9475423 .0165083 -3.09 0.002 .9157328 .9804567 txcontacts_number 1.050068 .0184354 2.78 0.005 1.01455 1.08683 _cons .1817139 .0703702 -4.40 0.000 .0850654 .3881711 ----------------------------------------------------------------------------------Iteration Iteration Iteration Iteration 1: 2: 3: 4: tolerance tolerance tolerance tolerance = = = = .20047132 .03206022 .00331681 .00014459 Iteration 5: tolerance = .00004303 Iteration 6: tolerance = 2.572e-06 Iteration 7: tolerance = 5.401e-07 GEE population-averaged model Group variable: Client_ID Link: logit Family: binomial Correlation: exchangeable Scale parameter: 1 Number of obs Number of groups Obs per group: min avg max Wald chi2(8) Prob > chi2 = = = = = = = 3287 176 3 18.7 19 33.96 0.0000 (Std. Err. adjusted for clustering on Client_ID) ---------------------------------------------------------------------------------- Robust dhoused Odds Ratio Std. Err. z P> z [95% Conf. Interval] ------------------+---------------------------------------------------------------Age 1.016433 .0167207 0.99 0.322 .9841836 1.049739 reEthGrp .8459414 .2118144 -0.67 0.504 .5178557 1.381885 Gender .9617023 .324652 -0.12 0.908 .4962398 1.863759 Died 1.259143 .6877876 0.42 0.673 .4316375 3.673084 TxSbGrp 1.400699 .496634 0.95 0.342 .6991039 2.806391 t .866046 .1222359 -1.02 0.308 .6567502 1.142041 contacts_number .9556347 .0128106 -3.39 0.001 .9308535 .9810757 txcontacts_number 1.045332 .0097721 4.74 0.000 1.026354 1.064662 _cons .2620723 .2196451 -1.60 0.110 .0507016 1.35463 ----------------------------------------------------------------------------------QIC and QIC_u ___________________________________________ Corr = exch Family = bin Link = logit p = 9 Trace = 99.626 QIC = 3982.883 QIC_u = 3801.631 ___________________________________________ . qic employed Age reEthGrp Gender Died t contacts_number if include==1, > i(Client_ID) t(time) fam(bin) link(logit) corr(exch) robust eform Iteration 1: tolerance = 1.556e-11 GEE population-averaged model > 3323 Group variable: > 176 Link: > 6 Family: > 18.9 Correlation: > 19 > 6.02 Scale parameter: > 0000 Pearson chi2(3323): Dispersion (Pearson): Client_ID Number of obs = Number of groups = logit Obs per group: min = binomial avg = independent max = 1 3190.18 .9600299 Wald chi2(6) = 11 Prob > chi2 = 0. Deviance Dispersion = = 1850.75 .5569503 --------------------------------------------------------------------------------employed Odds Ratio Std. Err. z P> z [95% Conf. Interval] ----------------+---------------------------------------------------------------Age .965192 .0063071 -5.42 0.000 .9529091 .9776333 reEthGrp 1.485465 .1417824 4.15 0.000 1.232022 1.791046 Gender 5.542965 .9947291 9.54 0.000 3.899305 7.87947 Died .0673952 .0678419 -2.68 0.007 .009371 .4846995 t 1.259666 .3788828 0.77 0.443 .6986037 2.271329 contacts_number 1.008603 .0033476 2.58 0.010 1.002063 1.015185 _cons .0407806 .0176256 -7.40 0.000 .0174808 .0951365 --------------------------------------------------------------------------------Iteration Iteration Iteration Iteration Iteration Iteration 1: 2: 3: 4: 5: 6: tolerance tolerance tolerance tolerance tolerance tolerance = = = = = = .12638868 .02549085 .00163697 .00003672 5.528e-06 7.185e-07 GEE population-averaged model Group variable: Client_ID Link: logit Family: binomial Correlation: exchangeable Scale parameter: 1 Number of obs Number of groups Obs per group: min avg max Wald chi2(6) Prob > chi2 = = = = = = = 3323 176 6 18.9 19 18.56 0.0050 (Std. Err. adjusted for clustering on Client_ID) -------------------------------------------------------------------------------- Robust employed Odds Ratio Std. Err. z P> z [95% Conf. Interval] ----------------+---------------------------------------------------------------Age .9602728 .0271553 -1.43 0.152 .9084974 1.014999 reEthGrp 1.323081 .4827561 0.77 0.443 .6471516 2.704995 Gender 8.206686 7.201811 2.40 0.016 1.469577 45.8293 Died .104982 .104393 -2.27 0.023 .0149516 .7371247 t 1.258215 .1970998 1.47 0.143 .9255799 1.710393 contacts_number 1.010044 .0158691 0.64 0.525 .9794148 1.04163 _cons .0434965 .0536631 -2.54 0.011 .0038752 .4882149 --------------------------------------------------------------------------------QIC and QIC_u ___________________________________________ Corr = exch Family = bin Link = logit p = 7 Trace = 99.867 QIC = 2057.204 QIC_u = 1871.471 ___________________________________________ . qic employed Age reEthGrp Gender Died t contacts_number txcontacts_number if include==1, i(Clien > t_ID) t(time) fam(bin) link(logit) corr(exch) robust eform Iteration 1: tolerance = 1.556e-11 GEE population-averaged model Group variable: Link: Family: Correlation: Scale parameter: Pearson chi2(3323): Dispersion (Pearson): Client_ID logit binomial independent 1 3189.83 .9599257 Number of obs Number of groups Obs per group: min avg max Wald chi2(7) Prob > chi2 = = = = = = = 3323 176 6 18.9 19 116.03 0.0000 Deviance Dispersion = = 1850.74 .5569476 ----------------------------------------------------------------------------------employed Odds Ratio Std. Err. z P> z [95% Conf. Interval] ------------------+---------------------------------------------------------------Age .9651927 .0063071 -5.42 0.000 .9529099 .9776339 reEthGrp 1.485459 .1417815 4.15 0.000 1.232017 1.791037 Gender 5.542969 .994729 9.54 0.000 3.89931 7.879474 Died .0673963 .067843 -2.68 0.007 .0093711 .4847072 t 1.215398 .5853597 0.41 0.685 .4728935 3.123734 contacts_number 1.007152 .0157317 0.46 0.648 .9767858 1.038462 txcontacts_number 1.001507 .0159833 0.09 0.925 .9706648 1.033328 _cons .042198 .0237063 -5.63 0.000 .0140313 .1269072 ----------------------------------------------------------------------------------Iteration Iteration Iteration Iteration Iteration Iteration Iteration 1: 2: 3: 4: 5: 6: 7: tolerance tolerance tolerance tolerance tolerance tolerance tolerance = = = = = = = .15829 .02657027 .00164058 .00048231 .00003658 5.406e-06 4.594e-07 GEE population-averaged model Group variable: Client_ID Link: logit Family: binomial Correlation: exchangeable Scale parameter: 1 Number of obs Number of groups Obs per group: min avg max Wald chi2(7) Prob > chi2 = = = = = = = 3323 176 6 18.9 19 19.49 0.0068 (Std. Err. adjusted for clustering on Client_ID) ---------------------------------------------------------------------------------- Robust employed Odds Ratio Std. Err. z P> z [95% Conf. Interval] ------------------+---------------------------------------------------------------Age .9602364 .0270505 -1.44 0.150 .9086555 1.014745 reEthGrp 1.328144 .4824041 0.78 0.435 .6517402 2.706548 Gender 8.120708 6.96926 2.44 0.015 1.510378 43.66186 Died .1111601 .1108473 -2.20 0.028 .015745 .7847931 t 1.228539 .3385433 0.75 0.455 .7158579 2.10839 contacts_number 1.008 .0200402 0.40 0.689 .9694776 1.048054 txcontacts_number 1.000988 .0068942 0.14 0.886 .9875668 1.014592 _cons .0458041 .0572597 -2.47 0.014 .003952 .5308729 ----------------------------------------------------------------------------------QIC and QIC_u ___________________________________________ Corr = exch Family = bin Link = logit p = 8 Trace = 100.375 QIC = 2057.869 QIC_u = 1873.118 ___________________________________________ . qic employmentpath Age reEthGrp Gender Died t contacts_number if include==1, i(Client_ID) t(time > ) fam(bin) link(logit) corr(exch) robust eform Iteration 1: tolerance = 2.138e-13 GEE population-averaged model Group variable: Link: Family: Correlation: Scale parameter: Pearson chi2(3323): Dispersion (Pearson): Client_ID logit binomial independent 1 3223.12 .9699419 Number of obs Number of groups Obs per group: min avg max Wald chi2(6) Prob > chi2 = = = = = = = 3323 176 6 18.9 19 118.54 0.0000 Deviance Dispersion = = 2223.82 .6692193 --------------------------------------------------------------------------------employmentpath Odds Ratio Std. Err. z P> z [95% Conf. Interval] ----------------+---------------------------------------------------------------Age .9669292 .0056383 -5.77 0.000 .9559412 .9780435 reEthGrp 1.363536 .116634 3.63 0.000 1.153073 1.612415 Gender 4.194338 .6292662 9.56 0.000 3.125789 5.628168 Died .2035582 .1041321 -3.11 0.002 .0746876 .55479 t 1.429118 .4006374 1.27 0.203 .8249791 2.475671 contacts_number 1.007125 .0030009 2.38 0.017 1.001261 1.013024 _cons .0680841 .0265655 -6.89 0.000 .0316898 .1462759 --------------------------------------------------------------------------------Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration 1: 2: 3: 4: 5: 6: 7: 8: tolerance tolerance tolerance tolerance tolerance tolerance tolerance tolerance = = = = = = = = 1.0925738 .44537486 .25302544 .01161685 .0003632 .00008009 2.976e-06 1.928e-07 GEE population-averaged model Group variable: Client_ID Link: logit Family: binomial Correlation: exchangeable Scale parameter: 1 Number of obs Number of groups Obs per group: min avg max Wald chi2(6) Prob > chi2 = = = = = = = 3323 176 6 18.9 19 14.03 0.0293 (Std. Err. adjusted for clustering on Client_ID) -------------------------------------------------------------------------------- Robust employmentpath Odds Ratio Std. Err. z P> z [95% Conf. Interval] ----------------+---------------------------------------------------------------Age .9622265 .021895 -1.69 0.091 .9202559 1.006111 reEthGrp 1.221131 .3846112 0.63 0.526 .6586658 2.263911 Gender 6.394073 3.71943 3.19 0.001 2.04472 19.99499 Died .9192846 .9348925 -0.08 0.934 .1252554 6.746888 t 1.457666 .2211492 2.48 0.013 1.082726 1.962446 contacts_number 1.008021 .0123355 0.65 0.514 .9841315 1.03249 _cons .0680509 .0651343 -2.81 0.005 .0104259 .4441751 --------------------------------------------------------------------------------QIC and QIC_u ___________________________________________ Corr = exch Family = bin Link = logit p = 7 Trace = 81.401 QIC = 2409.628 QIC_u = 2260.827 ___________________________________________ . qic employmentpath Age reEthGrp Gender Died t contacts_number txcontacts_number if include==1, i > (Client_ID) t(time) fam(bin) link(logit) corr(exch) robust eform Iteration 1: tolerance = 8.993e-13 GEE population-averaged model Group variable: Link: Family: Correlation: Client_ID logit binomial independent Number of obs Number of groups Obs per group: min avg max = = = = = 3323 176 6 18.9 19 Scale parameter: 1 Pearson chi2(3323): Dispersion (Pearson): 3223.36 .9700144 Wald chi2(7) Prob > chi2 = = 118.59 0.0000 Deviance Dispersion = = 2223.79 .6692108 ----------------------------------------------------------------------------------employmentpath Odds Ratio Std. Err. z P> z [95% Conf. Interval] ------------------+---------------------------------------------------------------Age .9669306 .0056382 -5.77 0.000 .9559428 .9780447 reEthGrp 1.363534 .1166333 3.63 0.000 1.153072 1.612411 Gender 4.194328 .6292625 9.56 0.000 3.125786 5.62815 Died .2035662 .1041359 -3.11 0.002 .0746907 .5548105 t 1.347827 .5995905 0.67 0.502 .5636001 3.223275 contacts_number 1.004696 .0148277 0.32 0.751 .9760501 1.034182 txcontacts_number 1.00252 .015087 0.17 0.867 .9733817 1.03253 _cons .0720149 .0368288 -5.14 0.000 .026431 .1962146 ----------------------------------------------------------------------------------Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration 1: 2: 3: 4: 5: 6: 7: 8: tolerance tolerance tolerance tolerance tolerance tolerance tolerance tolerance = = = = = = = = 1.0636059 .44276926 .24713603 .00950001 .00037574 .00005536 3.449e-06 1.244e-07 GEE population-averaged model Group variable: Client_ID Link: logit Family: binomial Correlation: exchangeable Scale parameter: 1 Number of obs Number of groups Obs per group: min avg max Wald chi2(7) Prob > chi2 = = = = = = = 3323 176 6 18.9 19 13.98 0.0515 (Std. Err. adjusted for clustering on Client_ID) ---------------------------------------------------------------------------------- Robust employmentpath Odds Ratio Std. Err. z P> z [95% Conf. Interval] ------------------+---------------------------------------------------------------Age .9622353 .0218021 -1.70 0.089 .9204388 1.00593 reEthGrp 1.22379 .3869852 0.64 0.523 .6584777 2.27443 Gender 6.480529 3.771941 3.21 0.001 2.070977 20.27896 Died .903193 .9103588 -0.10 0.920 .1252638 6.512315 t 1.407254 .3140728 1.53 0.126 .9086576 2.179439 contacts_number 1.00483 .0180363 0.27 0.788 .970094 1.04081 txcontacts_number 1.001525 .0068884 0.22 0.825 .9881145 1.015117 _cons .0720976 .0704904 -2.69 0.007 .0106095 .4899445 ----------------------------------------------------------------------------------QIC and QIC_u ___________________________________________ Corr = exch Family = bin Link = logit p = 8 Trace = 85.246 QIC = 2417.708 QIC_u = 2263.216 ___________________________________________ . qic d_income Age reEthGrp Gender Died t contacts_number if include==1, i(Client_ID) t(time) fam( > bin) link(logit) corr(exch) robust eform Iteration 1: tolerance = 6.741e-13 GEE population-averaged model Group variable: Link: Family: Correlation: Client_ID logit binomial independent Scale parameter: 1 Pearson chi2(3338): Dispersion (Pearson): 3392.79 1.016414 Number of obs Number of groups Obs per group: min avg max Wald chi2(6) Prob > chi2 = = = = = = = 3338 176 18 19.0 19 231.45 0.0000 Deviance Dispersion = = 4306.47 1.290134 --------------------------------------------------------------------------------d_income Odds Ratio Std. Err. z P> z [95% Conf. Interval] ----------------+---------------------------------------------------------------Age 1.05441 .0041219 13.55 0.000 1.046362 1.06252 reEthGrp .7201798 .0406838 -5.81 0.000 .6446968 .8045005 Gender .6258397 .0531937 -5.51 0.000 .5298032 .7392845 Died .8923293 .1387254 -0.73 0.464 .6579511 1.210199 t 1.283302 .210286 1.52 0.128 .9307829 1.769333 contacts_number 1.004211 .002019 2.09 0.037 1.000262 1.008177 _cons .2567375 .0625232 -5.58 0.000 .159293 .4137919 --------------------------------------------------------------------------------Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration 1: 2: 3: 4: 5: 6: 7: 8: 9: tolerance tolerance tolerance tolerance tolerance tolerance tolerance tolerance tolerance = = = = = = = = = .19709543 .01501042 .00493148 .00041922 .00018174 .00002399 7.231e-06 1.188e-06 3.012e-07 GEE population-averaged model Group variable: Client_ID Link: logit Family: binomial Correlation: exchangeable Scale parameter: 1 Number of obs Number of groups Obs per group: min avg max Wald chi2(6) Prob > chi2 = = = = = = = 3338 176 18 19.0 19 21.60 0.0014 (Std. Err. adjusted for clustering on Client_ID) -------------------------------------------------------------------------------- Robust d_income Odds Ratio Std. Err. z P> z [95% Conf. Interval] ----------------+---------------------------------------------------------------Age 1.056536 .0177347 3.28 0.001 1.022342 1.091873 reEthGrp .67787 .1592577 -1.65 0.098 .4277269 1.074302 Gender .6632276 .2230023 -1.22 0.222 .3431281 1.281944 Died .716222 .4365778 -0.55 0.584 .2168663 2.365392 t 1.330762 .1197888 3.17 0.002 1.115525 1.587528 contacts_number 1.001602 .0079514 0.20 0.840 .986138 1.017308 _cons .2560845 .1887893 -1.85 0.065 .0603761 1.08618 --------------------------------------------------------------------------------QIC and QIC_u ___________________________________________ Corr = exch Family = bin Link = logit p = 7 Trace = 99.047 QIC = 4511.362 QIC_u = 4327.267 ___________________________________________ . qic d_income Age reEthGrp Gender Died i(Clien t contacts_number txcontacts_number if include==1, > t_ID) t(time) fam(bin) link(logit) corr(exch) robust eform Iteration 1: tolerance = 3.949e-13 GEE population-averaged model Group variable: Link: Family: Correlation: Client_ID logit binomial independent Scale parameter: Pearson chi2(3338): Dispersion (Pearson): 1 3392.93 1.016456 Number of obs Number of groups Obs per group: min avg max Wald chi2(7) Prob > chi2 = = = = = = = 3338 176 18 19.0 19 231.86 0.0000 Deviance Dispersion = = 4305.98 1.289989 ----------------------------------------------------------------------------------d_income Odds Ratio Std. Err. z P> z [95% Conf. Interval] ------------------+---------------------------------------------------------------Age 1.05442 .0041222 13.55 0.000 1.046371 1.06253 reEthGrp .7201523 .0406834 -5.81 0.000 .6446702 .8044724 Gender .62573 .0531882 -5.52 0.000 .529704 .7391638 Died .8927664 .138806 -0.73 0.466 .6582552 1.210825 t 1.117646 .2876345 0.43 0.666 .674903 1.850834 contacts_number .998393 .0085549 -0.19 0.851 .9817657 1.015302 txcontacts_number 1.006159 .0088603 0.70 0.486 .9889422 1.023675 _cons .292594 .0898969 -4.00 0.000 .1602291 .5343053 ----------------------------------------------------------------------------------Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration Iteration 1: tolerance = .21640148 2: tolerance = .01861824 3: tolerance = .0088202 4: tolerance = .00110106 5: tolerance = .00047854 6: tolerance = .00007863 7: tolerance = .00002694 8: tolerance = 5.220e-06 9: tolerance = 1.564e-06 10: tolerance = 3.344e-07 GEE population-averaged model Group variable: Client_ID Link: logit Family: binomial Correlation: exchangeable Scale parameter: 1 Number of obs Number of groups Obs per group: min avg max Wald chi2(7) Prob > chi2 = = = = = = = 3338 176 18 19.0 19 22.09 0.0025 (Std. Err. adjusted for clustering on Client_ID) ---------------------------------------------------------------------------------- Robust d_income Odds Ratio Std. Err. z P> z [95% Conf. Interval] ------------------+---------------------------------------------------------------Age 1.057014 .0180636 3.24 0.001 1.022196 1.093017 reEthGrp .6673578 .1589182 -1.70 0.089 .4184674 1.06428 Gender .6675479 .2269162 -1.19 0.234 .3428766 1.299652 Died .7066382 .4298465 -0.57 0.568 .214493 2.32799 t 1.161906 .1476405 1.18 0.238 .9057555 1.490497 contacts_number .9949732 .0089586 -0.56 0.576 .9775687 1.012688 txcontacts_number 1.005979 .0055659 1.08 0.281 .9951288 1.016947 _cons .2983638 .2210347 -1.63 0.103 .0698484 1.274487 ----------------------------------------------------------------------------------QIC and QIC_u ___________________________________________ Corr = exch Family = bin Link = logit p = 8 Trace = 100.896 QIC = 4517.593 QIC_u = 4331.801 ___________________________________________ Secondary analyses logistic dpostarresttot6 dprearresttot6 fu6_employmentpathmo, or reEthGrp Died Gender Age fu6_incomemo Logistic regression Number of obs LR chi2(8) Prob > chi2 Pseudo R2 Log likelihood = -103.79436 = = = = fu6_housedmo 176 24.24 0.0021 0.1045 -------------------------------------------------------------------------------------dpostarresttot6 Odds Ratio Std. Err. z P> z [95% Conf. Interval] ---------------------+---------------------------------------------------------------dprearresttot6 2.037231 .7329022 1.98 0.048 1.006508 4.123477 reEthGrp .5758409 .153921 -2.06 0.039 .3410188 .9723592 Died .6701924 .4912991 -0.55 0.585 .1592971 2.819624 Gender 1.360886 .5280627 0.79 0.427 .6361087 2.91147 Age .9913533 .0178873 -0.48 0.630 .9569075 1.027039 fu6_incomemo 1.007277 .0607965 0.12 0.904 .8948964 1.133771 fu6_housedmo .8282983 .0708764 -2.20 0.028 .7004071 .9795419 fu6_employmentpathmo .6759141 .1158121 -2.29 0.022 .4831093 .9456657 _cons 2.212062 1.852901 0.95 0.343 .4283521 11.42335 -------------------------------------------------------------------------------------. logistic dwpostarrest6 dwprearrest6 > employmentpathmo, or reEthGrp Died Gender Age fu6_incomemo Logistic regression Log likelihood = -96.219073 Number of obs LR chi2(8) Prob > chi2 Pseudo R2 = = = = fu6_housedmo fu6_ 176 22.92 0.0035 0.1064 -------------------------------------------------------------------------------------dwpostarrest6 Odds Ratio Std. Err. z P> z [95% Conf. Interval] ---------------------+---------------------------------------------------------------dwprearrest6 1.839091 .7331828 1.53 0.126 .8418956 4.017429 reEthGrp .5746563 .1608483 -1.98 0.048 .332012 .9946322 Died .4790478 .3998218 -0.88 0.378 .0933149 2.459274 Gender 1.443567 .5843754 0.91 0.364 .6529189 3.191647 Age 1.000858 .0189336 0.05 0.964 .9644282 1.038663 fu6_incomemo 1.004087 .0633391 0.06 0.948 .8873121 1.13623 fu6_housedmo .8316587 .0763416 -2.01 0.045 .6947196 .9955904 fu6_employmentpathmo .5903784 .1525002 -2.04 0.041 .355842 .9794984 _cons 1.167799 1.017767 0.18 0.859 .2116041 6.444838 -------------------------------------------------------------------------------------. logistic dpostcrime6_tot dprecrime6_tot > fu6_employmentpathmo, or Logistic regression Log likelihood = -98.391185 reEthGrp Died Gender Age fu6_incomemo Number of obs LR chi2(8) Prob > chi2 Pseudo R2 = = = = fu6_housedmo 176 7.48 0.4856 0.0366 -------------------------------------------------------------------------------------dpostcrime6_tot Odds Ratio Std. Err. z P> z [95% Conf. Interval] ---------------------+---------------------------------------------------------------dprecrime6_tot 2.153892 .8389657 1.97 0.049 1.003855 4.621434 reEthGrp .6971475 .1916792 -1.31 0.189 .4067136 1.19498 Died 1.059312 .7672641 0.08 0.937 .2561502 4.380793 Gender 1.207777 .4918772 0.46 0.643 .5436605 2.683154 Age 1.00401 .0186995 0.21 0.830 .9680205 1.041337 fu6_incomemo 1.002502 .0630669 0.04 0.968 .8862102 1.134055 fu6_housedmo .9350532 .0779717 -0.81 0.421 .7940663 1.101072 fu6_employmentpathmo .9297929 .1054397 -0.64 0.521 .7444901 1.161217 _cons .4902335 .4382952 -0.80 0.425 .084993 2.827633 -------------------------------------------------------------------------------------- . logistic fu6_housed dpostfelonycharge6_tot dprefelonycharge6_tot fu6_employmentpath, or Logistic regression Log likelihood = -50.426755 reEthGrp Died Gender Age fu6_income Number of obs LR chi2(8) Prob > chi2 Pseudo R2 = = = = 176 15.32 0.0532 0.1319 ---------------------------------------------------------------------------------------dpostfelonycharge6_tot Odds Ratio Std. Err. z P> z [95% Conf. Interval] -----------------------+---------------------------------------------------------------dprefelonycharge6_tot 2.349346 1.805953 1.11 0.267 .5207443 10.59911 reEthGrp .4186286 .1825915 -2.00 0.046 .1780589 .9842241 Died 2.416419 2.208981 0.97 0.334 .4027564 14.4978 Gender 2.320975 1.449299 1.35 0.178 .6825749 7.892065 Age .9956924 .0267366 -0.16 0.872 .9446448 1.049499 fu6_income 1.026323 .605601 0.04 0.965 .3228621 3.262506 fu6_housed .0932402 .1033629 -2.14 0.032 .0106167 .8188722 fu6_employmentpath .231206 .33338 -1.02 0.310 .0136974 3.902668 _cons .5359406 .6765428 -0.49 0.621 .0451446 6.362503 ----------------------------------------------------------------------------------------