What Happened to Computational Social Science? Methods as Boundary Objects in Computational Social Science

Abstract

Computational social science (CSS) was introduced to integrate the social and computational sciences. Yet, two decades after its introduction, it remains unclear whether CSS has become an integrated intellectual field or whether CSS researchers only draw on partially overlapping methodological toolkits. We examine this question using 45,290 empirical CSS papers published between 2005 and 2023 with retrievable open-access full text, drawn from a 140,365-paper CSS corpus identified with positive–unlabeled classification. Using an LLM-based pipeline, we extract approximately 545,000 instances of method use. We find two patterns: First, the methodological toolkits of CSS papers converge over time. The mean pairwise Jaccard overlap of the top 50 method sets across broad disciplinary domains aggregated from OpenAlex subfields rose from 0.11 in 2005–2009 to 0.20 in 2020–2023, while the number of represented broad domains increased from five to eight. But this use of shared methods does not imply engagement with the same set of sources. Among 47 widely used methods for which we curated foundational sources, the average gap between the highest and lowest citation rates in different disciplines is 19.83 percentage points. We interpret this pattern through Star and Griesemer’s boundary-object framework. Many CSS methods circulate widely enough to support cross-disciplinary coordination, yet remain sufficiently plastic to be cited and used differently across disciplinary communities. CSS appears integrated through the use of shared methods more than through the lineages of these methods.