Machine Learning for Experiments in the Social Sciences

Machine learning is increasingly used in causal inference, with applications in survey experiments and robust causal inference. This Element provides theoretical and practical introductions to machine learning for social scientists, demonstrating two specific methods for characterizing treatment effect heterogeneity and testing its robustness to out-of-sample prediction.

Format: Paperback / softback
Length: 75 pages
Publication date: 13 April 2023
Publisher: Cambridge University Press

Machine learning and causal inference are often treated as distinct methodological traditions within the social sciences. However, the applications of machine learning in causal inference have been on the rise. This Element aims to provide both theoretical and practical insights into machine learning for social scientists interested in applying these methods to experimental data.

We begin by exploring how machine learning can be utilized for conducting robust causal inference. We then establish a theoretical framework that researchers can use to understand and apply emerging methods in this rapidly evolving field. To illustrate the practical applications, we demonstrate two specific methods: the prediction rule ensemble and the causal random forest. These methods are designed to characterize treatment effect heterogeneity in survey experiments and test the robustness of such heterogeneity to out-of-sample prediction.

In conclusion, we discuss the limitations and tradeoffs of these methods while providing readers with a guide to additional related methods available on the Comprehensive R Archive Network (CRAN). By leveraging the power of machine learning, social scientists can gain new insights into the complex relationships between variables and improve our ability to draw valid causal conclusions.

Weight: 133g
ISBN-13: 9781009168229