Using Longitudinal Self-Report Data to Study the Age–Crime Relationship

Abstract

Objectives: Given the growing reliance on longitudinal self-report data for making causal inferences about crime, it is essential to investigate whether the within-individual change in criminal involvement exists and is not a measurement artifact driven by attrition or survey fatigue—a very real possibility first identified by Lauritsen (Soc Forces 77(1):127–154, 1998) using the National Youth Survey (NYS). The current study examines whether the same threats to the validity of within-individual change in criminal involvement exist in the National Longitudinal Survey of Youth 1997 cohort (NLSY97). Methods: We first estimate cohort-specific growth curve models of general crime, arrest, and substance use, and then test the difference between the age–crime curves of adjacent cohorts. We take a general approach to test cohort differences in the growth curve models, which advances the existing method separately modeling for each pair of adjacent cohorts. To explore the sources of cohort differences, we also estimate separate growth curve models by individual crime item and by demographic group. Results: We document non-standard cohort differences between the age–crime curves of adjacent cohort pairs that are consistent with the findings of Lauritsen (1998) on measures of self-reported offending. However, the size of the cohort effects in the NLSY97 is substantially smaller than those in the NYS. We also found that the cohort effects were only evident in some of the survey items. Moreover, we did not identify any similar cohort issues in the longitudinal measure of arrest. Conclusions: The findings of cohort effects localized in a certain crime items and demographic groups may mitigate concerns over the limited validity of longitudinal self-report data. We discuss how the survey techniques used in the NLSY97 might explain our findings and suggest an area of future study to explicate remaining cohort differences.