Learning from Noise: Applying Sample Complexity for Social Science Research

Résumé fourni par la source

While statistical learning bridges the gap between theoretical concepts and complex empirical realities, the question of what constitutes ``good enough" data for social scientists remains understudied. In this article, we introduce the Probably Approximately Correct model and present sample complexity bounds, which take advantage of researcher-specified estimates of labeling error to guarantee the sample size required for a minimum level of accuracy. We develop a simulation-based approach to demonstrate its feasibility and provide the scR R package, offering a computationally efficient way to implement the proposed methods. We aim to improve standard practice by providing a general-purpose tool to validate the quality of measures when fuzzy measurement boundaries make generating data with ground-truth labels infeasible.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.