Is the Aggregated Delinquency Rate of a School Class a Suitable Indicator for Determining Criminogenic Peer Effects? Selected Problems and Solutions: Ausgewählte Probleme und Lösungsmöglichkeiten

In criminological research, criminogenic peer effects are sometimes estimated using multilevel models for school class based self-report data, whereby the aggregated crime prevalence rate of a class is employed as contextual-level predictor of an individual's crime involvement. This approach assumes that, first, young people recruit their friends from the pool of their classmates, and second, that the aggregated delinquency rate of a school class is a proper indicator for estimating the delinquent peer effect. This article examines the second assumption and shows that the approach described above leads to biased results. It is faced with two problems: The adolescent whose behavior is explained is included in the calculation of the predictor variable (ego-bias problem) and the mutual influence between the adolescent and his classmates is neglected (simultaneity problem). Both problems may lead to inflated effect statistics. The present work introduces potential remedies and illustrates their application at the example of adolescent shoplifting activity. Both instrumental-variables regression and non-recursive structural equation models reveal significant effects of classmates' delinquency rate on young people's shoplifting involvement.