Objectives

Research indicates respondents overestimate the similarity between their own deviance and that of their peers. Extending Rebellon and Modecki’s (J Quant Criminol 30:163–186, 2014) study, we examine if item-level error correlations in structural models reduce bias for non-peer-based, theoretically derived covariates such as self-control. Our specific interest lies in investigating the theoretical implications and practical value of using the correlated error technique in ‘everyday’ structural equation modeling.

Methods

Using dyadic data and multiple constructs of deviance, we present three sets of structural equation analyses. The first assesses the relationship between peer behavior and deviance via perceptual measures. The second uses identical constructs, but estimates item-level error correlations between perceptual and deviance items. The third replaces perceptions of peer deviance with items measuring peers’ self-reported behavior.

Results

Self-control and demographic variables have equivalent effects in perceptually-based correlated error models and models controlling peer self-reported deviance. However, latent variable adjustments to perceptions of peer behavior fail to bring perceived peer deviance coefficients into line with corresponding coefficients from models using peer self-reports, indicating that perceptions and peer self-reports are distinct constructs.

Conclusion

Researchers cannot use item-level error-correlations to model peer effects without collecting data from peers. They may, however, use these correlations to control for peer effects even when peer self-reports are not available. Because we find strong effects of self-control while maintaining social learning theory’s emphasis on perceptions, we argue that the technique is a form of theoretical reconciliation and recommend criminologists adopt the use of correlated errors in all social influence-based structural models.