Discovering Emergent Constructs: A Computational Epistemology to Advance the Science of Crime
Year: 2026
Type: preprint
Abstract: Quantitative research in criminology generally follows a model in which researchers design measures that capture theoretically identified constructs. Accordingly, researchers largely measure and theorize about previously identified constructs, limiting the range of scientific inquiry. We may then consider whether knowledge could be advanced through a complementary approach that allows criminologists to consider the theoretical relevance of new patterns. This study introduces a framework informed by computer science advances in representation learning which illustrates a widely applicable four-stage procedure allowing researchers to identify 'candidate constructs' in complex data which are not defined a priori yet are statistically associated with outcomes of interest. Abductive interpretation of candidate constructs can yield four outcomes: theoretical corroboration, theoretical refinement, novel construct discovery, and higher-order discovery. In a case study demonstrating this framework, AI-generated visual descriptions of 187,048 georeferenced street-view images from Chicago were semantically embedded and analyzed through a series of procedures to identify nine candidate constructs predicting micro-spatial differences in robbery beyond traditional controls and standard computer vision measures. Abductive interpretation of these indicators produced a higher-order discovery that crime-associated visual patterns, as measured through multiple computer vision techniques, manifest as configurations of systematically co-occurring environmental features that cross traditional theoretical boundaries.
Source: CrimRxiv
Author: Riley Tucker
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7225863625
