Reflections on Learning from Observational Data

Abstract

The social sciences study various aspects of human behaviour – social, economic and political – based on observational data. Observational data are inaccurate and subject to simultaneity, seasonality, structural breaks, random variation and too many interlocking variables masking the underlying causal patterns. During the past two decades or so, the use experimental data (RCTs) has become widely popular across the social sciences, creating a tension between the supporters and critics of the new and the old methodologies. In this paper, we first review these methodologies, both observational and experimental, focusing on how economists and other social scientists try to learn about the underlying causal relationships from the correlations contained in the data. We then reflect on whether the new or the old methodologies should be the way forward from a purely statistical and a broader policy and development perspectives.