Developing a Dynamic Social-Reinforcement Framework for Repeat Offending

Thesis Abstract

This study investigates the dynamics of social reinforcement processes in repeat offending, addressing the gap between individual criminogenic factors and situational peer-group influences that sustain repetitive criminal behavior. The problem addressed centers on how social reinforcement mechanisms—such as peer validation, normative alignment, and perceived rewards—interact with individual propensity, routine activities, and environmental cues to maintain cycles of reoffending. The aim is to develop a dynamic social-reinforcement framework that integrates micro-level psychological processes with meso-level social networks to predict and explain repeat offending trajectories. Specific objectives are to (1) conceptualize a dynamic model that operationalizes social reinforcement as time-variant inputs affecting decision-making; (2) identify key social-network characteristics (peer density, tie strength, sanction sensitivity) associated with escalation or desistance; (3) quantify the relative contributions of personal criminogenic traits, routine activity patterns, and social reinforcement signals to recurrence risk; (4) test the model on a longitudinal dataset to examine temporal ordering and feedback loops; and (5) derive policy-relevant interventions targeting social reinforcement pathways to reduce recidivism. A mixed-methods design is employed, combining a longitudinal cohort survey with nested qualitative interviews. The population comprises released offenders from five medium- to high-security correctional facilities in a metropolitan region, with a target sample of 1,200 individuals followed over 24 months, including quarterly survey waves and an annual in-depth interview subset of 180 participants. Quantitative data are collected via structured instruments assessing criminogenic needs (e.g., LSI-R domains), social network characteristics (egocentric network surveys measuring size, density, contact frequency, and perceived recency of reinforcement), protective factors, routine activity patterns, and self-reported recidivism events corroborated by official criminal records. Qualitative data are gathered through semi-structured interviews focusing on lived experiences of peer influence, perceived rewards or sanctions from social groups, and contextual factors shaping decision-making under social pressure. Validity and reliability procedures include pilot testing of instruments, Cronbach’s alpha analyses for internal consistency, test-retest reliability checks, and triangulation across data sources. Analytical approaches combine time-series and multilevel modeling to capture within-person changes and between-person heterogeneity in social reinforcement effects. Specifically, hierarchical generalized linear models (HGLM) will assess how time-varying social reinforcement indicators predict recidivism events, controlling for baseline criminogenic risk, employment status, housing stability, and mental health. Dynamic structural equation modeling (DSEM) will be employed to estimate feedback loops between reinforcement signals and offending behavior across multiple time points, allowing for lagged effects and reciprocal causation. The qualitative data will undergo thematic analysis to identify salient reinforcement narratives, cross-validated with quantitative findings to elucidate mechanisms driving observed patterns. A conceptual model will be iteratively refined through model comparison metrics (AIC, BIC) and sensitivity analyses. Key expected findings include evidence that social reinforcement intensity and perceived legitimacy of peer norms significantly amplify reoffending risk, particularly when reinforcement is intermittent but highly salient, and when individual resilience resources are weak. The study anticipates identifying distinct offender subgroups characterized by differential susceptibility to social reinforcement, moderated by network structure (e.g., dense, sanctioning networks vs. sparse, permissive ones) and individual traits (impulsivity, social learning propensity). Findings are expected to demonstrate temporal precedence of reinforcement signals preceding relapse events, supporting a dynamic framework over static risk models. The study contributes to knowledge by delivering a theoretically grounded, empirically tested dynamic social-reinforcement framework that integrates social network theory, social learning theory, and routine activity theory to explain repeat offending. It advances measurement of social reinforcement as a time-variant construct and provides actionable insights for interventions targeting peer-group processes, post-release supervision, and community-based support networks. Practical recommendations include structured peer-support programs that reframe reinforcement cues, enhanced monitoring of high-risk social ties, and policy instruments to promote prosocial reinforcement environments in reintegration contexts. The main conclusion is that reformulating recidivism risk through dynamically evolving social reinforcement signals yields superior predictive accuracy and informs targeted, context-sensitive interventions to disrupt reinforcement-driven relapse cycles.