While the field of political science may seem staid to outsiders, it has evolved significantly in terms of research methods over the last 40 years. The behaviorally based studies that dominated in the 1970s gave rise to the subfield of American Political Development (APD) in the 1980s as a way to more fully realize and incorporate the study of history and institutions. APD scholars made narrative-based causal arguments to understand the history of American politics.
Over the past decade, a trend toward more data-oriented studies of causal relationships has emerged. This “causal inference movement” puts a premium on research designs that identify empirical evidence consistent with causally based arguments. Accepted designs include approaches like difference-in-difference, regression discontinuities, and instrumental variables. However, causal inference (CI) has not made much headway within the subfield of APD.
How can the study of American Political Development benefit from the research designs and innovations of political methodologists with respect to causal inference? How can causal inference take advantage of the deep historical insights of APD?
Earlier this month, Jeffery A. Jenkins (USC), Nolan McCarty (Princeton), and Charles Stewart III (MIT) organized a conference where proponents of CI and APD could come together to discuss how scholarship and approaches in one area could complement scholarship and approaches in the other. The Society for Political Methodology, the Dornsife College of Letters, Arts and Sciences at USC, and the Unruh Institute at USC funded the Causal Inference and American Political Development Conference. The Bedrosian Center at USC handled the logistical planning. A number of top scholars from major research universities in the country came together for two days in a lively and intellectually engaging format to look at the future of research in Political Science.
David Bateman (Cornell) shared a paper on the application of the potential outcomes framework of causal inference to historical research. The idea behind the potential outcomes framework is that each unit would have differing “potential outcomes” based on counterfactual conditional statements, or assigned conditions. Per this view, “each unit can be theorized as having multiple potential outcomes depending on the treatment it receives,” since we can only treat something once, we can see only one outcome. Bateman argues that historical information should be used to construct a “timeline of plausible counterfactual[s]” in order to neatly identify causation with historical data akin to the way scholars can with more contemporary data.
Sanford Gordon (NYU) also discussed the potential outcomes framework, noting that it is silent with respect to external validity, the specification of appropriate counterfactuals, and mechanisms. To better explain the causal relationship (rather than just the effect) between phenomena, he argued that “historical transformations permit us to illuminate novel causal mechanisms or relationships in need of empirical examination, provide the setting in which to adjudicate empirically among competing causal mechanisms, and facilitate accounting for different forms of selection bias that may emerge under those mechanisms.”
Corrine McConnaughy (George Washington) argued that variation over time and across states offers APD scholars important opportunities to leverage causal inference tools to draw conclusions. With a more literal interpretation of “states as laboratories,” she proposed that multi-level theory building will increase the observable implications of political theories, and allow researchers to identify moments in history or across space that allow for rigorous causal testing.
Sean Gailmard (UC Berkeley) argued that APD scholars and game theorists need to communicate better with each other. He stated, “The potential for formal theory’s contribution to APD exists because both fields focus on institutions.” Gailmard explained that game theorists view institutions either as a game form or as equilibria, “stable, repeated patterns of behavior.” As such, scholars of APD would do well to understand and utilize game theory and formal modeling in their theories of political development.
Sara Chatfield (Denver) presented applications of causal inference to APD with some of her work on survey quotas and the Tennessee Valley Authority. To identify causality, she relied on historical knowledge gleaned from an APD approach. That knowledge then informed the design and implementation of her analysis. Chatfield concluded by arguing that dynamic processes, like the ones at the heart of APD, are rarely accounted for by just one estimated parameter. Rather, knowledge about APD will be built by “triangulation with quantitative (descriptive and causal) and qualitative evidence.”
Anna Harvey (NYU) and Gregory DeAngelo (Claremont) presented applications of regression discontinuity design (RDD) to APD. RDD is a pretest-posttest comparison group strategy instrument within the causal inference toolkit. Harvey and DeAngelo’s presentation of the tool for use in APD focused on the causal effects of the Civil Rights Act of 1875 and of financial incentives on law enforcement.
Greg Wawro (Columbia) shared work on applying causal inference methods to APD and historical research more generally. He focused on researchers’ abilities to take advantage of and utilize randomization present in observational data to augment causal inferences and research designs that use historical data.
Abby Wood (USC Law) and Christian Grose (USC) focused on causal inference and political reform, by exploiting a random FEC audit prior to the 1978 election to investigate how voters reacted when candidates and elected officials violated the law. They find that the audit increased transparency and revealed legislator noncompliance in campaign funding. Moreover, their paper highlights the potential for APD researchers to take advantage of government audits as leverage for randomization.
Sarah Binder (GWU; Brookings) focused on institutional interdependence, which is a concern for researchers studying Congress and history. Binder argued that institutional interdependence is difficult to disentangle, and used the relationship between Congress and the Federal Reserve regarding monetary policy as an example. In spite of this entanglement, Binder suggests researchers should continue attempting to answer these types of questions through design-based inferences.
Joshua Clinton (Vanderbilt) echoed the importance of making a distinction between identification and causal identification. He argued that despite recent majority-party transitions providing observed differences in roll call votes, the average effect across time is not significant. In all, being mindful of what we as researchers can measure and not measure is important.
Shigeo Hirano (Columbia) shared an application, using a novel dataset to show that the introduction of civil service coverage in large cities does not necessarily translate into benefits for residents. However, he provides some evidence that cities without civil service coverage have also experienced close mayoral elections see increased costs.
Finally, Avidit Acharya (Stanford) presented his co-authored book, Deep Roots: How Slavery Still Shapes Southern Politics. Acharya showed that Southern states that had higher proportions of slaves led to increased racial sentiment and support for Republicans in modern politics. The book advances the concept of Behavioral Path Dependence, which suggests political behaviors existing in a community become entrenched and therefore challenging to reverse.
Conference participants left with a heightened awareness of what causal inference can do for APD research in the future, as well as how APD can provide for better causal inferences. Moreover, because causal inference is rife with issues of complicated causality, the conference was an opportunity for researchers to think more deeply about applying various causal inference methods and measurement techniques to American Political Development.