Causality, Decision Making and Data Science

Catalog description

Policymakers often need to make decisions when the implications of those decisions are not known with certainty. In many cases they rely in part on statistical evidence to guide these decisions. This requires statistical methods for estimating causal effects, that is the impact of these interventions. In this course we study how to analyze causal questions using statistical methods. We look at five specific causal questions in detail. For each case we study various statistical and econometric methods that may shed light on these questions. We discuss what the critical assumptions are that underlie these methods and how to assess whether the methods are appropriate for the settings at hand. We then analyze data sets, partly in class, and partly in assignments, to see how much we learn in practice. We start by looking for the first two weeks at randomized experiments. Traditionally experiments have been viewed as the most credible settings for estimating causal effects. We discuss how to analyze randomized experiments and what their limitations are. We do this in the context of randomized experiments for labor market programs that target unemployed individuals. The remainder of the course we study causal questions in {\bf observational settings} where assignment to treatment was not randomized. This is one of the foundational questions in econometrics, and we look at some of the classic causal questions studied in that literature using modern methods. These include looking at (i) the effect of a basic universal income on labor supply, and, relatedly Milton Friedman's permanent income hypothesis, (ii), the effect of schooling on earnings, (iii), the estimation of demand and supply functions in competitive markets, and (iv), the effect of military service on labor market outcomes. In all cases we study (a) simple models of economic behavior to guide the empirical models…