Synthetic Data as a Pathway to Reproducible Criminology

Abstract

Abstract Open science practices remain rare in criminology, and scholars commonly point to sensitive data as reason to restrict public access. Synthetic data can expand access, provided their analytic utility and disclosure risks are assessed. We reproduce the data and results of a previously published survey experiment using the synthpop package in R. The procedure retains experimental assignment patterns and jointly models the original nested outcome responses. We compare five mixed-effects models using coefficient differences, and standard-error ratios, alongside multivariate and disclosure diagnostics. Of 20 treatment coefficients, 90% differ from the original estimates by less than two original standard errors, and the mean absolute standardized coefficient difference is 1.03. The results support the use of synthetic data to improve data sharing in criminology, while emphasizing the need for purpose-specific validation and disclosure review.