Incorporating Model Uncertainty into Social Science Research: A Computational Estimator for Model Robustness

• In Event: 558 - Regular Session. Quantitative Methodology

Abstract

Model uncertainty is pervasive in social science. A key question is how robust empirical results are to sensible changes in model specification. We present a comprehensive estimator that can test robustness across sets of possible controls, variable definitions, standard errors, and functional forms. This allows researchers to become meta-analysts of their own empirical strategies, focusing on model ingredients and distributions of results rather than on specific estimates. We emphasize (1) the difference between statistical significance and model influence, and (2) the concept of the influence regression, showing how model ingredients affect the coefficient of interest. Using a series of social science data sets, we demonstrate the core functionality and practical value of the robustness estimator.

Authors

• Cristobal Young, Stanford University

• Kathy Kroeger, Stanford University