Changing Places: The Role of Household and Community Context in Long-Term Patterns of Recidivism

Abstract

The period of reentering society after prison is a critical time in the life-course and presents numerous challenges. With whom and where individuals live can determine aspects of social support, as well as bonds and informal social control that may predict whether a person “succeeds” by abstaining from crime and avoiding technical violations of parole. We use detailed administrative data from a release cohort in Missouri to construct monthly variation in household and community context at two levels (neighborhood disadvantage and county-level rural-urban designation). We follow the local-life circumstances approach, decomposing changing context into within and between person variation (Horney et al., 1995). We find important selection effects but also intervention effects – with households including family as particularly beneficial. This work further highlights the needs for multiple measures of recidivism and a broadening of the geographic locales researchers and policymakers consider.

Keywords:

• Reentry
• households
• recidivism
• community
• social support
• hybrid model

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Notes

1 Further, individuals without any address changes would not contribute to the within-person variation in the models.

2 All measures of recidivism have limitations (NASEM, 2022). While new convictions reflect discretion of criminal legal system actors at multiple stages, they may counteract some of the discretionary practices at the stages of arrest and charging.

3 Among months living with an intimate partner, it was almost evenly split between spouses (44.9%) and non-marital relationships (55.1%). See online Appendix D for model results with measurement of intimate partners split by marital status. Results are consistent with main models.

4 Due to multicollinearity of the Census tract items, the primary measure of immediate community context used in the analyses is concentrated disadvantage. See Appendix D for model results using the alternate Census measures.

5 The original nine categories were: urban at populations of 1) at least 1 million residents, 2) 250,000 to under 1 million residents, and 3) under 250,000 residents; rural and urban adjacent at populations of 4) greater than 20,000 residents, 5) 2,500 to under 20,000 residents, and 6) under 2,500 residents; and rural but not adjacent to urban areas at populations of 7) greater than 20,000 residents, 8) 2,500 to under 20,000 residents, and 9) under 2,500 residents.

6 The reference category is primarily white (less than 1% were identified as another race). Ethnicity was a separate field with only 1% identified as of Hispanic origin and was, therefore, omitted from analyses.

7 We use xtlogit in Stata 16 with robust standard errors (vce(robust)). We also include the intpoints(30) option to ensure accurate estimates (Rabe-Hesketh & Skrondal, 2012).

8 The pattern of failure events across the reentry period follows a cubic time trend. See figures in Appendix B.

9 Appendix C also includes nested models including the null model for the intraclass correlation coefficient (ICC), a model with time trends added, and a model with the decomposed time varying reentry context predictors added, as well as the full model with all the between-person controls reported that are in the Table 2 models.

10 An alternate model that used counts of misconduct (range 0-30) rather than an indicator of having institutional misconduct showed similar results. Those with more misconduct events were significantly more likely to have technical violations. The same within-individual time varying factors remained significant predictors of technical violations: months living with intimate partners and family were protective, as were months living in rural counties that were not urban adjacent.

11 An alternate model that used counts of misconduct (range 0-30) rather than an indicator of having institutional misconduct showed similar results. Those with more misconduct events were significantly more likely to have new convictions. The same within-individual time varying factor remained a significant predictor of convictions: months living with family were protective.

12 Hausman test results for technical return model χ²(df = 9) = 241.39, p<.0001 and new conviction model χ²(df = 9) = 69.54, p<.0001.

13 We thank a reviewer for raising the importance of gender differences in reentry households and suggesting we conduct gender-specific analyses.

14 Although prison populations rose for about another decade nationally and in Missouri. See the BJS National Prisoner Statistics (NPS) Program: https://bjs.ojp.gov/data-collection/national-prisoner-statistics-nps-program