Methodological Issues in Longitudinal Research Designs: An Assessment of the National Youth Survey's Delinquency Measures and the Age-Crime Curve

David S. Kirk, University of Chicago

No issue is more critical to the study of crime and delinquency than the issue of measurement. At the heart of this have been ongoing debates over the relative advantages and disadvantages of self-report measures of delinquency. While it is widely accepted that crime rates increase during adolescence to a peak close to 17 years of age, findings from self-report measures of multiple cohorts from the National Youth Survey show declining trends in delinquency regardless of age. Various researchers have debated the reasons for these discrepant findings, including period effects, testing effects, and changing content validity of delinquency measures to name a few. The present paper engages in these debates and augments previous findings in two ways: 1) by using multi-level modeling to account for the presence of missing time points in longitudinal designs, treating these missing data as latent variables in multivariate outcome models, and 2) by relaxing the classical assumptions of independent errors and equal variances of these errors. One of the key benefits of multi-level modeling is that it utilizes all available longitudinal data even if a respondent misses a time point, thereby letting researchers assess the effects of sample attrition across waves of longitudinal data collections.

(Return to Program Resources)

Updated 05/20/2006