Accounting for Projection Bias in Models of Delinquent Peer Influence: The Utility and Limits of Latent Variable Approaches

Abstract

Projection effects have been shown to bias respondent perceptions of peer delinquency, but network data required to measure peer delinquency directly are unavailable in most existing datasets. Some researchers have therefore attempted to adjust perceived peer behavior measures for bias via latent variable modeling techniques. The present study tested whether such adjustments render perceived peer coefficients equal to direct peer coefficients, using original data collected from 538 young adults (269 dyads). After first replicating projection effects in our own data and examining the degree to which measures of personal, perceived peer, and direct peer violence represent empirically distinct constructs, we compared coefficients derived from two alternative models of personal violence. The first model included an error-adjusted latent measure of perceived peer violence as a predictor, whereas the second substituted a latent measure of directly-assessed, peer-reported violence. Results suggest that personal, perceived peer, and direct peer measures each reflect fundamentally separate constructs, but call into question whether latent variable techniques used by prior researchers to correct for respondent bias are capable of rendering perceived peer coefficients equal to direct peer coefficients. Research cannot bypass the collection of direct peer delinquency measures via latent variable modeling adjustments to perceived peer measures, nor should models of deviance view perceived peer and direct peer measures as alternative measures of the same underlying construct. Rather, theories of peer influence should elaborate and test models that simultaneously include both peer measures and, further, should attempt to identify those factors that account for currently unexplained variance in perceptions of peer behavior.