The Partisan Politics of COVID-19

This image is from a new working paper (draft to come soon, joint with Shana Gadarian and Sara Goodman) on the partisan politics of the COVID-19 (coronavirus) pandemic in the United States.

Many of us have seen the survey “top-lines” comparing Americans’ views of COVID-19 and focusing on partisan differences. What these simple comparisons don’t tell us, though, are whether or not these differences persist when we account for demographic or geographic factors. And we might wonder how representative the surveyed populations are, or whether these differences are the result of data-mining across a wide range of different survey responses.

Our analysis, by contrast, is a pre-registered, IRB-approved analysis of the partisan politics of COVID-19. The figure above adjusts flexibly for a wide range of demographic and geographic differences, and adopts a very conservative Bonferroni correction to adjust for multiple comparisons.

In our other analysis, we adopt a regularized regression approach to sifting through our wide range of predictors of health attitudes and behaviors, and find that respondents’ partisan affiliations are the most commonly selected predictor of health behaviors, attitudes, and policy preferences. Partisanship is more consistently predictive than education, news consumption, income, or anything else.

That the politics of COVID-19 are partisan is perhaps not surprising given the condition of American politics, but that mass public health behavior is more consistently predicted by partisanship than by anything else we measured has profound and distressing implications for public health in the coming months.

Historical Persistence and Nazi Legacies in Contemporary Germany

One of the most exciting developments in comparative politics and political economy has been renewed attention to history—specifically, to the historical foundations of contemporary politics, and how historical legacies continue to affect us today. Although few social scientists would ever have denied that “history matters,” a wealth of recent works have provided compelling quantitative evidence that features of politics from long ago predict features of politics today. I have done some of this myself, looking at how colonial migration patterns are related to to contemporary local governance patterns in Java. Other work has looked at the political legacies of slavery in the United States, the legacies of colonial land tenure policies in India, and a host of other outcomes.

What makes such arguments so interesting is their attention to local variation in historical experiences. But linking local variation over long time scales raises difficult methodological issues. Morgan Kelly, in an important recent paper, has shown that many strong correlations between spatially-defined historical variables and spatially-defined contemporary outcomes are the product of spatial autocorrelation.

Inspired in part by such concerns, in a new paper, Sara Wallace Goodman, Conrad Ziller, and I take a fresh look at an important recent article examining the historical legacies of the Holocaust on contemporary German attitudes. The authors find a strong and statistically significant correlation between how far Germans live from a former Nazi concentration camp and their out-group tolerance: people who live closer to former concentration camps are less tolerant than those who live further away. They also show that this distance also predicts support for the far-right Alternative für Deutschland party.

Yet because Nazi concentration camps were not randomly assigned across space, and because neither are contemporary political attitudes, it could be that these analyzes are inadvertently picking up correlations that have nothing to do with the legacies of the institutions of mass slaughter. And indeed, we show that those results are not robust to statistical approaches that recognize the differences across contemporary German states (Länder). There is lots to dig into in this paper: German administrative divisions; Prussian reforms under Weimar; post-treatment variables, colliders, and bias-amplifiers; Hausman tests and reweighting estimators for heterogenous effects; and multi-level modeling of non-nested hierarchical structures. Our approach, we hope, will provide a template of best practices for future scholars who are interested in establishing the causal relationships between spatial historical variables and contemporary outcomes.

Rather uniquely (we believe), this paper is both a replication and a pre-analysis plan. One of the data sources that we wish to replicate is not publicly available, so we are unable to replicate it at present. Still, we are able to specify the analyses that we plan to conduct when we do have access to those data, and will register these at OSF (these plans are currently in the Supplemental Appendix).

You can read the paper here. And you can replicate our analysis here. Watch this space for more updates. And we wish to offer our thanks to the original authors for releasing their own replication materials, which allows us to conduct our analysis with full transparency and assurance that any errors or misunderstandings on our part can be quickly addressed.