by Amanda Pearson
Their recent book won the American Political Science Association’s 2019 William H. Riker Book Award “for the best book on political economy published during the past three calendar years.” How did they do it?
In Deep Roots: How Slavery Still Shapes Southern Politics, IQSS faculty affiliates Matthew Blackwell and Maya Sen (with co-author Avidit Acharya from Stanford University) examine a sample of more than 40,000 white Southerners, along with historical census records, to evaluate how political attitudes vary in counties that had more slaves back in 1860. They argue that the political outlook of Southerners today can be explained by the ideas, norms, and behaviors of their slave-owning predecessors, or what they call “path dependence” of political attitudes.
Matthew Blackwell. (Photo credit: Stephanie Mitchell)
How can it be the case that this deep historical past—more than 150 years ago—is a key determinant of political attitudes and behaviors today? To make their claim causal, the three authors say they “collected probably two dozen different data sets” and used a variety of statistical techniques and other designs.
“Data was a huge part of our approach,” Blackwell and Sen told me. Thanks to large data sets such as the Cooperative Congressional Election Study (CCES), “we were able to investigate how public opinion on important partisan and policy topics varies from one county to the next.”
Maya Sen. (Photo credit: Stephanie Mitchell)
Deep Roots is an important contribution to a large body of scholarship that shows how racial attitudes are transmitted across generations through history, culture, and institutions. Blackwell, Associate Professor of Government at Harvard University, and Sen, Professor of Public Policy at Harvard Kennedy School, discuss their award-winning book with IQSS.
IQSS: How did you decide to make political attitudes and behaviors the subjects of your observations?
Blackwell and Sen: We came at it from two directions. First, we were really interested in seeing whether a huge and impactful part of American history—slavery—would have any kind of effect on contemporary politics. Of course, slavery has been fundamental to American history—it has shaped everything from basic American demographics to how the United States was established in a federal system and even to which states were admitted to the union and when. We were extremely interested in seeing if this important historical institution—which has clearly profoundly affected American democracy from the ground up—has any kind of lasting effect on people's political hearts and minds.
Second, political scientists have long known that certain parts of the American South are among the most conservative parts of the United States. The Southern "Black Belt" has a long and important political history. For example, during the 1930s and 1940s, congressional representatives from the Black Belt were at the forefront of fighting against FDR's New Deal. This included fighting against protections for workers, general redistributive policies, and other types of labor and agricultural programs that could have potentially helped African Americans. Some of the most bitter confrontations during the Civil Rights Movement took place in the Black Belt. We wanted to know if the Black Belt's historical prevalence of slavery played a key role in driving these events.
Slavery’s effects on Democratic vote shares, 1844–2008
Effect of proportion slave on vote for Democratic presidential candidates in the South over time. Each point is the effect of a 25-percentage-point increase in proportion slave from separate IV models of county-level Democratic share of the presidential vote on proportion slave, using cotton suitability as an instrument. (Figure courtesy of the authors.)
IQSS: Your book relies on big data to tell the quantitative side of your argument. How did it come about that some of your IQSS colleagues shared data with you for your analyses?
Blackwell and Sen: Data was a huge part of our approach. Many historians, sociologists, and economists working before us have noted the strong impact slavery has had on policy and politics. But they have largely done so looking at qualitative accounts or historical data on economic indicators. Part of this gap has been data-related: even as little as 15 years ago, we had no public opinion data of the scope needed to properly assess geographic differences in political attitudes, which is necessary to assess the historical impact of geographically prevalent institutions such as slavery. Thanks to large data sets such as the Cooperative Congressional Election Study (CCES), the American National Election Studies (ANES), and other data sets, however, we were able to investigate how public opinion on important partisan and policy topics varies from one county to the next.
IQSS colleagues such as Steve Ansolabehere and Jim Alt were invaluable in sharing data and resources with us. And, of course, the infrastructure at IQSS enabled us to create and maintain public repositories for the data that we generated. These public repositories—made possible through the Dataverse Project at IQSS—are now helping other scholars investigate the long-term impact of slavery on a wider variety of contemporary outcomes.
IQSS: As you write in your appendix, “one obstacle to using historical data is that county borders have shifted over time.” How did you overcome the methodological problem of using historical data on the prevalence of slavery from the 1860 U.S. Census with contemporary county borders?
Blackwell and Sen: Our work relies not just on contemporary "big data" but also on data less commonly used by social scientists, including historical sources such as county-level data from the 1860 U.S. Census and voting across various counties on slavery-related initiatives. However, American counties have shifted over time, which presents a challenge to quantitative social scientists. To map historical counties onto contemporary counties, we relied on a kind of areal interpolation that simply makes some assumptions about how populations are distributed within a county and divides historical populations into current-day counties. We also brought in some techniques new to the social sciences that were helpful in assessing the causal effects of historical forces on contemporary variables. Here, work by IQSS affiliates such as Tyler Vanderweele and Jamie Robins was very helpful.
Map of slavery in 1860 using modern county boundaries. Source: 1860 Census. (Figure courtesy of the authors.)
IQSS: You find that the prevalence of slavery in 1860 correlates with present-day Republican affiliation and several measurements for political attitudes and behavior (e.g., opposition to affirmative action). With this kind of correlation, it's always possible there are some other causes driving the findings. What confounding variables did you control for that could explain the link between slavery and modern political views? And, why might this matter for contemporary politics?
Blackwell and Sen: We argue that slavery has had a causal impact on today's politics, which is particularly tricky given the time frames involved—some 150 years! To make the claim causal, we used a variety of statistical techniques, including controlling for observable data, instrumental variables analyses, and other designs. For example, for one of our analyses, we took into account a variety of economic and political indicators—including a county's demographics, political features, geographic features, farm holdings, levels of religiosity, etc. Basically, we accounted for whatever we could get our hands on that was relevant to the prevalence of slavery. In addition, we used an instrumental variable that leveraged a county's suitability for growing cotton. We learned through our qualitative research that some counties in the South are very suitable for growing cotton, but others are not; however, a county's soil conditions, rainfall, and dew point predict whether the county went on to become reliant on slavery in the 1860s. This strategy enabled us to detect the causal effects of slavery on contemporary attitudes more crisply.
In all, we collected probably two dozen different data sets and conducted close to a thousand statistical analyses. Only some of these ended up in the final book.
IQSS: Although the scope of your book is limited to the U.S. South, how might your findings apply to other global contexts?
Blackwell and Sen: Our work actually builds on a large body of scholarship showing that historical institutions that no longer exist can have quantifiable and important downstream consequences. Other researchers currently or formerly at IQSS have shown similar effects in Eastern Europe, the former Soviet Union, Africa, and Latin America. Given that we see the lasting impact of historical institutions in other contexts, it is no surprise that we also see it in the United States. In sum: we think these results speak to a broader pattern about the importance of historical institutions having very long downstream effects on people's attitudes and politics.
IQSS: As with many scholarly research projects that unfold over time as authors incorporate new research and peer reviews into their revisions, what is one example of how you gradually developed your argument about the historical persistence of political attitudes?
Blackwell and Sen: We have had wonderful engagement with scholars and members of the public. Between all three co-authors (including Avi Acharya, at Stanford) we have presented this research probably two to three dozen times. Each time we present, we get new ideas for extensions and statistical checks. Probably the most helpful pushback came early on from members of the IQSS community. Dustin Tingley, Gary King, Ken Shepsle, Jim Alt, and Jeff Frieden were really helpful in pushing on us in terms of honing down the empirical strategies and discussions of causal mechanisms.
IQSS: The journal article that preceded Deep Roots generated some criticism. How do you handle such critiques in conducting research on sensitive topics?
Blackwell and Sen: Criticism is par for the course for academics, especially if you're working on important topics. The criticisms of our work were very discordant. Some people said our findings were obvious, while others said our findings were obviously wrong. Ultimately, unless the feedback spoke to our research methodology or helped us in terms of pointing us to resources or new ideas for extensions or checks, we mostly just listened and learned.
IQSS: What lessons have you learned about how to make your scholarship resonate with a public audience?
Sen: I think the most important thing is to choose important topics. Scholars are often engaged in their own little academic bubbles; but it’s very important to draw scholarly inspiration from the real world and from the struggles and experiences of real people. Sometimes graduate students ask me how I get ideas, and I tell them it’s from reading the news, talking to people, and finding out what matters to them.
Blackwell: I learned how hard it is to take complicated statistical material and make it accessible to both other academics and the public at large. Those of us who do quantitative work tend to focus on the latest and greatest methods (which are really important!), but it's also important for us to understand how to communicate the results of these methods and their strengths and weaknesses.
IQSS: What are each of you working on now, and what topics are on your wish lists for future exploration?
Sen: I’m finishing a paper looking at the long-term impact of Japanese American internment during World War II (with former IQSS affiliate Mayya Komisarchick). We show in that paper that people who were interned are less likely to be politically engaged today than people of Japanese ancestry who weren’t interned. It also shows, like our slavery project, how important historical forces are in understanding contemporary politics.
Blackwell: I've been working on many projects related to statistical methodology in general and causal inference in particular. These papers try to estimate causal effects in complicated causal situations. Many of these are natural methodological extensions of what we did in our book.
