Abstract

Statistical models are often used to answer scientific questions. When discussing the validity of empirical results, the focus is usually on the interpretation of the models, including testing the model assumptions. However, the reproducibility of the results using the same data and the same statistical technique is rarely discussed but taken for granted. However, reproducibility is a requirement of valid research findings. This study examines the reproducibility of statistical models of four important topics of social research: fertility, marriage, divorce, and regional mobility. To assess whether reproducibility has improved over time, articles from two time periods, 1975–1985 and 2010–2020, were selected from high-impact journals. The majority of models included in the articles could not be reproduced exactly. No significant differences in reproducibility between older and newer articles could be found. The fact that most statistical analyses could not reproduce the published estimates or reconstruct the datasets used raises concerns about the credibility and generalizability of research findings in the social sciences. This finding calls for a reassessment of research practices and the demand for improving the reproducibility of statistical modelling.