An empirical assessment of data sharing and computational reproducibility in sociology

Abstract

Abstract Public sharing of data and analysis code enables independent verification of published findings. While audit studies have examined data and code sharing in psychology, economics and political science, comparable evidence for sociology remains limited. We examine 730 empirical articles published between January 2019 and June 2024 in three leading general-interest journals: American Sociological Review (ASR), American Journal of Sociology (AJS) and Social Forces (SF). For each article, we record whether a publicly accessible replication package is available and, when materials are shared, attempt to computationally reproduce the main-text tables and figures. Overall, 9.9% of articles provide replication packages (12.2% among quantitative articles), with substantial variation across journals (5.2–22.9%) and research types (0% for qualitative studies, 30% for experiments). These rates fall well below those reported for political science and economics. Of the 72 packages examined in detail, more than half cannot be verified owing to missing or incomplete materials. At the same time, no article is wholly non-reproducible and 22% are fully or largely reproducible. Sociology is not unusually prone to errors when materials are shared and runnable—it simply shares them far less often. We discuss implications for journal policies and transparency standards that reflect the field's methodological diversity.