By David Alexander Bateman and Eric Schickler
There is little doubt that slavery continues to shape American life, through its legacies on the country’s institutions, economic and social practices, ideologies, and distribution of resources.
Avidit Acharya, Matthew Blackwell, and Maya Sen’s Deep Roots has brought a great deal of clarity to one area in which slavery’s legacies might still be with us. They find that slavery and its abolition continue to have a direct and quantifiable effect on the geographical distribution of white racial attitudes within the contemporary South.
Deep Roots is a leading example of the causal inference revolution’s impact on political science. Political scientists have long approached history as a source of data or as an opportunity for comparative analyses. But many also turned to history to uncover the origins of contemporary politics, treating it as a process through which often contingent choices or developments were stabilized and reproduced over time. This work has often been qualitative, given its emphasis on contingent agency and on sequencing.
The causal inference revolution has brought renewed interest in statistically identifying the origins of contemporary political patterns. The research strategy often identifies historical interventions or features of the landscape that were plausibly related to the probability of “treatment” but which were (conditionally) uncorrelated with outcomes of interest before treatment: The drawing of Vichy France’s borders, the siting of Nazi concentration camps, or (in the case of Deep Roots) the soil and climatic conditions most conducive to cotton agriculture. So long as critical assumptions are met, such exogenous discontinuities can be leveraged to return causal estimates of large-scale phenomena across decades, centuries, or even millennia.
Deep Roots makes an important substantive and theoretical contribution. But in recently published work, we find that a critical assumption needed to identify its specific causal model cannot be sustained.
Deep Roots’ Causal Story
To understand why, it is important to clarify Deep Roots’ causal story. An older but still active literature on racism’s origins debated whether it pre-existed slavery and contributed to its growth, by leading to selection into the slave trade and slaveholding; or whether it was produced alongside slavery’s emergence as a legal status nearly exclusive to persons of African descent. Alongside this was a related question about whether racism was the result of a deliberate strategy by the slaveholding class, or of a more bottom-up process through which racial identities were drawn and given meaning. A separate literature on the contemporary distribution of racial attitudes argued that were an effect of local context, in particular the Black proportion of the population. So far as the historical presence of enslavement in a locality was related to contemporary white racial attitudes, it was generally assumed to reflect slavery’s legacy on an area’s current racial composition. Deep Roots both made the relationship of slavery to contemporary attitudes systematically clear, and provided an alternative explanation.
Its intervention was framed primarily around the literature on racial threat. But Deep Roots spoke also to the origins debate. The authors argue that the geographic divergence in southern whites’ racial attitudes emerged as a result of the crises surrounding slavery from the immediate onset of the Civil War to the tumult of emancipation and Reconstruction. Before the critical juncture of the Civil War and abolition, they claim, there was no geographic correlation in the South between levels of enslavement and white racist attitudes.
This uncorrelatedness claim was critical to the specific model of slavery’s causal effect. The geographic divergence in white racist attitudes was the result of how the shock of emancipation impacted regions with large numbers of enslaved persons versus those without. The causal treatment was the localized efforts of the former slaveholding elite to reconstitute their economic and political power. Absent the assumption of uncorrelatedness before the Civil War, their model cannot establish the causal effect of the shock of abolition from alternatives, such as a pre-1860 process of selection into and out of slaveholding or the overall effect of exposure to slavery before 1860.
This is a theoretically clear account with important implications: Racism’s geographic distribution in the South was the result of exposure to local conditions rather than selection into those, and it reflected deliberate elite agency at a specific moment, rather than a more diffuse, societal-wide process of association between “race” and the civic status of being enslaved that might have played out over centuries.
The most compelling evidence to sustain this claim is the authors’ analysis of the Southern Rights elections in Georgia and Mississippi, in which the more radical pro-slavery side called for rejection of the Compromise of 1850. What potentially makes these such valuable cases is that the old Democratic and Whig parties seemingly were absent from the contest. The old parties had conspired to take slavery off the agenda, and so long as they succeeded any subregional differences over it would be suppressed from the legislative and electoral record. If these parties were irrelevant, the 1851 elections could be interpreted as “a popular referendum on slavery” (2018, p. 235).
In replicating Deep Roots’ finding that voting in these elections was uncorrelated with slaveholding, we raise serious doubts that 1. the elections were not structured by partisanship, and 2. that the local prevalence of enslavement was uncorrelated with the results. In fact, both historians and election data make clear that the seemingly new parties were largely replications of the old ones, with Democrats rebranding as Southern Rights and Whigs as Unionists. As we see in the top row of Figure 1, previous vote share for the old parties explains most of the subsequent vote share for the new ones. The local deviations from partisan loyalties (bottom row of Figure 1), moreover, were largely in the direction we hypothesize: the districts most likely to shift towards the Unionist position were those with low levels of enslavement. An even more striking demonstration was in South Carolina (Figure 2), which had never developed a two-party system. While the positions here were more extreme, there was a clear preference of high enslavement districts for the more radical, uncompromising position.
In short, the data that Acharya, Blackwell, and Sen present to justify their claim of uncorrelatedness is either inadequate to the task or suggests there was a pre-existing relation between racial policy questions and the local prevalence of slavery before the war.
We also present what we think is more compelling evidence: legislative votes on the rights of free Black southerners. We focus on North Carolina and Tennessee (where free Black men could vote until the 1830s), and Virginia (where both abolition and restriction of free Black rights were debated in the same decade). These votes occurred before the Civil War, and before the emergence of the second party system suppressed southern dissent over slavery. The results are consistent: variation in legislative voting closely tracked the local prevalence of enslavement. Regions with higher enslavement rates were much more likely to support more extreme restrictions on free Black rights than districts with fewer enslaved persons. Figure 3 shows the probability of a legislator voting yea across a series of votes in Virginia, as a function of the local prevalence of slavery. The first two votes (the left and center panels of the top row) concern restrictions on the rights of free Blacks to learn to read or write. The third vote (top right panel) was on a resolution declaring it expedient for the state to consider a plan of abolition. The left and center panels of the bottom row show votes on a bill to remove free persons of color from the state, the first amending the bill to require the consent of the persons and the second passing the bill. Finally, the bottom right panel shows the vote on a “police bill” that would have further restricted free Black rights. In all cases but final passage of the removal bill, the more persons held in slavery, the higher the probability that a legislator would take the anti-Black position. Even the exception sustains our hypothesis: the previous vote that made removal conditional on consent transformed the bill into a voluntary colonization measure, which some slaveowners opposed as a prelude to emancipation. On the critical vote to require consent, the low-enslavement regions took the “pro” Black rights side.
So what? Does the IV approach need this assumption?
One of Deep Roots’ many strengths is its clear account of how slavery might have shaped racial attitudes, one that emphasizes former slaveholders’ investment in ideologies and practices of white supremacy to regain power. If the geography of racism was already correlated with slavery before this moment, then the causal effect of this model can’t be distinguished from other possible ways in which slavery might have shaped racial attitudes.
This holds broader lessons for the “historical legacies” literature. A research design’s identifying assumptions are evaluated with reference to a particular causal model. The critical identifying assumption of an instrumental variable approach is the exclusion restriction, namely that the instrument (in this case local suitability for cotton agriculture) not impact the outcome through any pathway other than the treatment. So long as one buys the exclusion restriction in the case of cotton suitability – and we haven’t questioned it – then one should buy the estimated effect of “slavery” on later outcomes as a causal one.
But the hypothesized treatment in Deep Roots is not simply “slavery,” i.e.., all possible avenues by which the complex institution might have become related to racial attitudes at any moment during or after its existence, from selection into slaveholding or high enslaving counties, exposure to it, elite-led defenses of it, bottom-up hostility to its victims, etc. These are not simply mechanisms, of secondary importance to the causal claim; they concern the nature of the treatment itself.
As is true of much work in the historical legacies literature, slavery had no discrete moment of treatment, but a long period in which it was established, spread, abolished, commemorated and contested, with related forms of labor and political oppression emerging in its wake. It was such a pervasive and long-lived institution that possible causal pathways proliferate. It is this which makes the claim that slavery continues to matter so profoundly true and analytically under-specified.
Acharya, Blackwell, and Sen quite admirably made a more well-specified claim about how the treatment of slavery operated, one whose theory and empirical modeling requires that there be no correlation between slavery and racial attitudes before the Civil War. But this argument is not sustained.
What are we left with?
Deep Roots is an important intervention in the study of historical legacies. It should shift expectations about the significance of locally enduring cultures in shaping white racial attitudes, especially relative to “racial threat.” It also provides compelling evidence that slavery, in its broadest terms, has left an imprint in the geographic distribution of white attitudes.
The project has also made important theoretical contributions. The Deep Roots argument can be separated into several components, including a claim that racist attitudes would be most intense in areas where the need for an ideology justifying coercive labor and political relations was strongest, and a claim that it was the moment of abolition when this need became sufficiently pressing to generate a geographic divergence. We provide additional evidence for the first, but against the second. We don’t discount the importance of Reconstruction. But its specific effects are not empirically distinguished from the longer run consequences of slavery that preceded it.
Ultimately, neither our agreement about part of the causal story, nor our shared expectation that Reconstruction had an important effect, hinges on the project’s claims causal identification. We agree with Acharya, Blackwell, and Sen on these points because the first part of the causal story is theoretically persuasive and supported by a wide range of scholarship across different disciplines, albeit little of it causally identified in its now common meaning.
It is theory and triangulation from multiple sources of evidence, none so dispositive that they should be given primacy, that remains the most compelling reason to believe that slavery, from its inception to abolition, produced an enduring geographic organization to white racism.