CAUSAL GRAPHS IN POLITICAL METHODOLOGY

Abstract

Political scientists increasingly use causal graphs, specifically directed acyclic graphs (DAGs), to communicate identification assumptions for causal inference, but are reluctant to treat them as formal models. Their relationship to so-called “potential outcomes” has been largely unclear in both the applied as well as the methodological literature. This dissertation suggests that political scientists, as well as other empirical researchers, use causal graphs to communicate crucial assumptions, and in a second step to derive counterfactual and other independence assumptions from them. In Chapter 2, I show that our understanding of existing analyses can be improved by using formal concepts from the causal graph literature. Specifically, I discuss how to systematically and transparently derive observable as well as a counterfactual assumptions from a given graph, and I apply these tools to four examples of published research. Here, I show how DAGs allow us to formally justify specification tests in causal mediation analysis, relax assumptions for complex observational studies as well as panel analysis, and illuminate the substantive content of assumptions in compliance modeling. When using instrumental variables, researchers often assume that causal effects are only identified conditional on covariates. In Chapter 3—co-authored with Adam Glynn and Miguel Rueda—we show that the role of these covariates in applied research is often unclear, and that there exists confusion regarding their ability to mitigate violations of the exclusion restriction. We explain how existing adjustment strategies may lead to bias. We then discuss assumptions that are sufficient to identify various treatment effects, some of which are new, when the exclusion restriction only holds conditionally. In general, these assumptions are highly restrictive, albeit they sometimes are testable. We also show that other existing tests are generally misleading. Then, we introduce an alternative sensitivity analysis that uses information on variables influenced by the instrument to gauge the effect of potential violations of the exclusion restriction. We illustrate it by reanalyzing Spenkuch and Tillmann (2017)’s analysis of Catholicism and voting in the Weimar Republic. Finally, we summarize our results in easy-tounderstand guidelines. In Chapter 4 Peter Selb and I demonstrate how DAGs can be used to encode and communicate theoretical assumptions about nonprobability samples and survey nonresponse, determine whether typical population parameters of interest to survey researchers can be identified from a sample, and support the choice of adjustment strategies. Following an introduction to basic concepts in graph and probability theory, we discuss sources of bias and assumptions for eliminating it in selection scenarios familiar from the missing data literature. We then introduce and analyze graphical representations of multiple selection stages in the data collection process, which highlights the strong assumptions implicit in using only design weights. Furthermore, we show that the common practice of evaluating adjustment variables based on correlations with sample selection or survey outcomes is ill-justified. Finally, we identify areas for future survey methodology research that can benefit from advances in causal graph theory. The dissertation concludes with a discussion of these insights in relationship to parametric assumptions, robustness tests, political science theory, as well as the so-called “credibility revolution”.