Browsing by Author "Morucci, Marco"

The development of interpretable causal estimation methods is a fundamental problem for high-stakes decision settings in which results must be explainable. Matching methods are highly explainable, but often lack the accuracy of black-box nonparametric models for causal effects. In this work, we propose to investigate theoretically the statistical bias and variance of Almost Matching Exactly (AME) methods for causal effect estimation. These methods aim to overcome the inaccuracy of matching by learning on a separate training dataset an optimal metric to match units on. While these methods are both powerful and interpretable, we currently lack an understanding of their statistical properties. In this work we present a theoretical characterization of the finite-sample and asymptotic properties of AME. We show that AME with discrete data has bounded bias in finite samples, and is asymptotically normal and consistent at a root-n rate. Additionally, we show that AME methods for matching on networked data also have bounded bias and variance in finite-samples, and achieve asymptotic consistency in sparse enough graphs. Our results can be used to motivate the construction of approximate confidence intervals around AME causal estimates, providing a way to quantify their uncertainty.

Causal inference is a fundamental tool of empirical political science. The existing methodologies used to perform causal analyses are, however, sometimes hard to adapt to subfields of the discipline in which data is scarce, populations are hard to reach, and experimentation is impossible. In this work, we aim to extend the reach of causal methodology to such subfields by proposing methods aimed at addressing several existing shortcomings of causal inference tools. First, we introduce Credible Assumption Mixtures, a methodology for sensitivity analysis of observational results that enables researchers to assess the sensitivity of their results to many different assumptions, both separately and all at once, producing a complete and rich picture of sensitivity in applied cases. Second, we introduce a methodology for measurement of quantities of interest to political scientists in image data: our approach is based on contemporary deep-learning tools and can quickly and cheaply annotate large sets of images, thus enabling researchers in all subfields to take advantage of image data regardless of their resources. Finally, we propose Matched Machine Learning: a methodology that boosts the interpretability of non-parametric causal estimates by combining matching with powerful machine learning black-box models. In this way, causal estimates are very accurate but also easily interpretable and explainable. In turn, this interpretability should enable researchers in those fields in which causal inference is hard to better develop, test and assess models for their causal quantities of interest. Finally, we apply all our method to answering a causal question of interest in empirical political science: we study the effect of electoral success on public good allocation, whether violence can be measured from images, and whether presence of police makes violence more likely in civil protest settings.