Modeling Peer Networks and Behavior Change to Understand the Role of Peers in the Development of Problem Behavior

Empirical evidence of peer influence is mixed due to the multifaceted nature of influence and the inability of traditional analytic methods to fully capture the dynamic, multilevel, and bidirectional interactions between peer relationships and behavior. This dissertation aimed to (1) clarify how peers may shape the development of problem behavior and promote the diffusion of intervention effects and (2) address several measurement challenges that arise in longitudinal studies of peer relationships and behavior.Study 1 used actor-oriented social network models to test hypotheses about selection and influence while controlling for other processes that impact friendship and aggression. Data were from 480 youth followed biannually from 6th to 7th grade. After controlling for selection in other domains, there was no evidence for selection with respect to aggression, but there was evidence of influence. Girls were more likely to select aggressive friends and low status youth were more susceptible to influence. Rejected youth were not more likely to select or be influenced by aggressive friends.Study 2 was motivated by the possibility that effects from family-based interventions may diffuse through peer networks allowing non-participants to benefit from the intervention. Indices that capture network-level features of diffusion potential were identified within a sample of 33 6th grade networks participating in an intervention trial. These indices were uncorrelated with non-network measures of diffusion potential (e.g. participation rate), demonstrating discriminant validity. Diffusion potential indices varied considerably across networks, suggesting that some networks were more likely to support diffusion than others.Study 3 demonstrated that the Graded Partial Credit Model for Repeated Measures (GPCM-RM) could recover reasonable estimates of latent trait and change scores across multiple conditions. When the GPCM-RM was applied to data from the National Youth Survey, inter-individual differences in delinquency were highly stable across adolescence with population-level increases over time. Observed average scores indicated that delinquency was moderately stable, increasing then decreasing over time. Reasons for these discrepancies are explored.Taken together, these studies demonstrate how different measurement models can inform our understanding of peer relationships and problem behavior. Continued research is needed to clarify how different analytic frameworks shape conclusions about development.