2. Turnout and Republican Vote Share by Region: A Century of Change

In the aftermath of the Civil War, the key question for national leaders in the United States was how to integrate the 11 states that seceded back into the Union. President Andrew Johnson initially took charge of Reconstruction with a plan to return the ex-Confederate states in a manner that would have all but replicated the antebellum social and political order, with only the elimination of slavery being the difference (McKitrick Reference McKitrick1960). So-called “Radical” Republicans in Congress had a different idea. By 1867, they had wrestled control of Reconstruction from President Johnson and sought to elevate the ex-slaves by granting them citizenship, the franchise, and civil rights protections (Jenkins and Peck Reference Jenkins and Peck2021; Wang Reference Wang1997).Footnote ⁴ The Reconstruction Act of 1867 forced the revision of state constitutions to enfranchise Black Americans. And the Fifteenth Amendment (1870) prohibited race, color, or previous condition of servitude from being used to deny voting rights anywhere in the country going forward.

By 1870, all ex-Confederate states were back in the Union, and citizens from the entire nation cast votes in the 1872 presidential election. It is from that point on—through 1968—that we track both turnout and Republican vote share in presidential elections. In figure 1, we show national turnout in presidential elections, with Southern states compared to all other states. For the entire period, the South had lower turnout than the rest of the United States. However, this difference ballooned in the 1890s and early 1900s and did not return to prior levels until the 1960s. In figure 2, we show a similar graph, but with the Republican Party’s vote share in each election as the outcome of interest. The patterns are less stark in figure 2 than in figure 1, but the Republicans achieved less success in the South than in the rest of the United States in all but one presidential election during the time period. Save for a couple of elections driven by unique candidate selection, Republican success in the South—relative to the rest of the country—similarly declined in the first half of the twentieth century.

Figure 1 Turnout in Presidential Elections, South and Non-South, 1872–1968

Source: Burnham (Reference Burnham2010).

Figure 2 Republican Vote Share in Presidential Elections, South and Non-South, 1872–1968

Source: Burnham (Reference Burnham2010).

In 1872, the incumbent Republican presidential nominee, Ulysses S. Grant, won eight of 11 Southern states, and carried popular-vote majorities in both the South and non-South. This gave Republican leaders hope that a Republican South, built on freedmen’s votes, was taking hold despite violence by Democrat-backed terror groups like the Ku Klux Klan and Red Shirts (Heersink and Jenkins Reference Heersink and Jenkins2020). In 1873, however, a financial panic swept the nation and ushered in a lengthy recession that was especially hard on the overleveraged Southern states. Republican governments throughout the South were blamed for mismanagement and fraud, and Democrats used these recriminations as a pretext for ramping up their insurgency (Donald, Baker, and Holt Reference Donald, Baker and Holt2001). By 1876, only three Southern states remained marginally Republican, and by 1877 the entire ex-Confederacy was “redeemed” by white Democrats.

Between 1880 and 1916, the Republicans won no Electoral College votes in the South. During this time, a profound change occurred within the Southern electorate. Once in power, white Democrats in the South continued to use violence and intimidation to dampen Black voting power. But between 1877 and 1890, Southern states adopted no significant legal measures to disenfranchise African Americans. Through 1888—when Republican Benjamin Harrison was elected president—Southern turnout was still greater than 60% (Northern turnout exceeded 80%). With Harrison in the White House, the Republican Party also enjoyed majorities in both congressional chambers, and there was a serious push (the first since Reconstruction) to create new voting rights protections for African Americans in the South. This effort—the Federal Elections Bill—fell just short thanks to a filibuster in the Senate led by Southern Democrats (Calhoun Reference Calhoun2006; Welch Reference Welch1965). As a result, beginning with Mississippi in 1890, white Democrats in the South began a process of disenfranchising African Americans (and some poor whites) through a variety of legal techniques, like poll taxes and literacy tests, which we discuss fully in section 3. The changes in turnout that followed, as presented in figure 1, were stark.

Southern turnout in presidential elections dipped below 60% starting in 1892. By 1900, when a number of Southern states had disenfranchising provisions in place, turnout fell below 45% —while remaining above 80% outside the South. Republican vote share, by comparison, fell to about 35% in the South, while remaining comfortably above 50% outside the South. A sizable drop in Southern turnout began in 1904, when all states but Georgia had their full set of disenfranchising provisions in place. Between 1904 and 1948, Southern turnout averaged under 26%, with only two elections with greater than 30%. Non-Southern turnout dropped during the same period, but averaged 67%. The Republican Party won about 28% of Southern votes in presidential elections between 1904 and 1948—with an aberration of 47.4% in 1928, when New York Governor Al Smith, the Democratic nominee, faced a significant anti-Catholic backlash throughout the nation.

Beginning in 1952, turnout and Republican vote share in the South both began increasing—with Republican vote share rising substantially with General Dwight D. Eisenhower’s selection as the Republican presidential nominee. Eisenhower’s popularity would result in a more than 20 percentage-point increase in Republican vote share in the South, as he won four Southern states.Footnote ⁵ Republican vote share hovered over 45% for the next three presidential elections—with Eisenhower, Richard Nixon, and Barry Goldwater winning five, three, and five Southern states, respectively—before sinking to 35% in 1968, when independent candidate George Wallace cut significantly into Richard Nixon’s potential voting base (although Nixon still won five Southern states). In 1960, turnout in the South was 41%; in 1968, it crossed the 51% mark.

3. Methods of Shaping Electorates

There are two primary tactics in shaping a voting electorate—changing the eligibility to vote and changing the propensity of those who are eligible to cast a vote.Footnote ⁶ The former directly prohibits targeted groups from voting. A prominent example is the historical restriction prohibiting women from voting. Within a model of turnout, this effectively adds an infinite cost to voting. Because turnout is a fraction where the numerator is the set of people who vote and the denominator is the set of people who are legally eligible to vote, these direct, formal restrictions affect turnout by removing from the denominator and the numerator simultaneously (so long as at least one member of the excluded group would have voted). The effect on turnout can be positive, negative, or zero, depending on the rate at which the excluded group would have voted relative to the rate at which the rest of the electorate voted.Footnote ⁷ The effect on the partisan outcome of the election can also vary in either direction, depending on how the excluded group compared to the nonexcluded group.Footnote ⁸ Direct policies may be more sharply defined with the intended targets—such as women, children, and noncitizens—spelled out in the legislation.

The second method does not formally prohibit someone from voting, but instead leaves them eligible to vote and raises (in a finite way) their costs of voting, thus making it less likely they will do so. The poll tax and literacy tests are two examples: they did not make people ineligible to vote. Instead, they added a cost (in literal terms of money in the case of the poll tax and in terms of effort and human capital in literacy tests) that made voting difficult for many poor and uneducated eligible voters. Notably, these policies do not affect the denominator, which remains the same, and instead reduce only the numerator. In the naïve sense, indirect effects can only lower turnout, not increase it.Footnote ⁹ These indirect policies may shape partisan outcomes in either direction, depending on the relative support for a given party in the excluded group compared to the nonexcluded group.Footnote ¹⁰ Indirect policies may often be less well targeted, capturing intended groups as well as those collaterally disenfranchised. For example, many policies that targeted the formerly enslaved used their high rates of poverty and low rates of education as targeting methods. But these polices also captured many poor and uneducated whites at the same time.

There are, of course, other methods that raise costs to voting without formal policy, such as the terroristic practices that we discussed in the preceding section.Footnote ¹¹ Intimidation and violence can be used to prevent people from even attempting to jump through the hoops created to enable voting. Thus, these informal practices continue and do influence outcomes even in the presence of formal institutions. In this paper, we are exclusively focused on the results of formal policies, but we recognize that informal violence and intimidation and other social factors continued to affect voting outcomes in the United States during the period of Jim Crow laws in the South. Despite these facts, states did choose to pass and apply such disenfranchising laws, and so our inquiry is about their effects. For the purposes of our analysis, variation in enforcement and exercised discretion is the most relevant informal effort of disenfranchisement. All extant evidence indicates significant differences in enforcement between the South and the rest of the country in ways that sent strong signals to African Americans that they would not be welcome to vote. We evaluate that possibility thoroughly in this paper.

Specific Policies We Consider

We consider four primary, racially motivated, formal policies that indirectly altered turnout and Republican success during the period between 1870 and 1970. Most historians agree that the poll tax and the literacy test were the two primary techniques of indirect disenfranchisement. We also analyze the impact of various forms of felon and ex-felon disenfranchisement. Finally, we consider the argument that the Australian ballot was strategically used to disenfranchise African Americans. In addition, we account for a significant direct eligibility change, the granting of suffrage to women, and an informal practice that may also have affected African American political participation, lynching.

The Poll Tax

The poll tax was a fee one needed to pay to vote, with the cost ranging between $1 and $2 annually.Footnote ¹² It had to be paid months in advance—sometimes up to a year—and a citizen needed to present a receipt at election time to vote. Some states allowed these taxes to accumulate, with prospective voters required to pay off several years’ worth of unpaid poll taxes to vote. Adjusted for inflation, a poll tax around the turn of the century was equivalent to about $30 or $60 in 2022 US dollars, depending on the amount of the poll tax. In figure 3, we present the set of states that used a poll tax at any point between 1870 and 1970, which is coterminous with the region we define as the “South.” As the figure illustrates, the poll tax was used exclusively in the 11 ex-Confederate states. As an indirect disenfranchisement tool, the poll tax should have reduced turnout by making those eligible to vote far less likely to do so due to the cost. At the same time, the poll tax should have reduced Republican success in the South by excluding African Americans and poorer whites who were more likely to support the Republican Party through much of the period we analyze.

Figure 3 States that Employed Poll Taxes at Some Point, 1870–1970

Source: Rusk (Reference Rusk2001, table 2-15).

The era of poll taxes began to unwind in the early to mid-twentieth century. Three Southern states—North Carolina (1920), Louisiana (1934), and Florida (1938)—were early (re)movers with Georgia (1945), South Carolina (1951), and Tennessee (1953) following suit by the early 1950s. The remaining five Southern states held out, even as Congress tried repeatedly (and unsuccessfully) in the 1940s to push for a federal law banning the poll tax (Jenkins and Peck Reference Jenkins and Peck2013).Footnote ¹³ Eventually, the civil rights movement of the 1950s and 1960s led Congress to adopt legislation (in 1962) to constitutionally prohibit the use of poll taxes in federal elections. The Twenty-Fourth Amendment was ratified by the requisite number of states in 1964, and its provisions were extended to all elections by the Supreme Court in Harper v. Virginia Board of Elections in 1966.

Literacy Tests

Literacy tests were evaluations that citizens had to pass before they could vote. That these tests could be passed or failed, in theory, by anyone, but would disproportionately exclude some groups, makes them a classic indirect method of disenfranchisement. Though commonly called “literacy” tests, they are more accurately thought of as language assessments that could take the form of reading or writing tests. States varied in their implementation: some required passing in one or the other, while others required both reading and writing skill. Moreover, election registrars were the sole judges of whether a would-be voter “passed” their test, which could be rigged by local whites seeking to disenfranchise Blacks who attempted to vote.

Seven of the 11 Confederate states (and one border state, Oklahoma) would adopt literacy tests.Footnote ¹⁴ Additionally, at various points, literacy tests were enacted outside the South, mainly in the Northeast and West, as a way to dampen voting by immigrants or to maintain the Progressive idea that voters needed to be literate to be informed and responsible citizens (Rusk Reference Rusk2001, 18). These non-Southern initiatives were typically initiated by Republicans. In figure 4, we present the set of states that employed literacy tests at some point between 1870 and 1970, labeled in green. Those states in yellow never used a literacy test during this time frame. Unlike the poll tax, a significant number of northeastern and western states employed literacy tests at some point. Like poll taxes, literacy tests should have reduced turnout, though perhaps by only small amounts if they were not rigorously enforced. We would also expect that they would have the effect of limiting Republican success in the South, but improving Republican success outside the South, where the voting coalitions were quite different.

Figure 4 States that Employed Literacy Tests at Some Point, 1870–1970

Source: Rusk (Reference Rusk2001, tables 2-17 and 2-19).

Literacy tests saw no within-region erosion as the poll tax had seen. Congress began to sweep aside literacy tests in the Civil Rights Act of 1964, stipulating that anyone who had completed at least six years of formal education must be presumed literate. In the Voting Rights Act of 1965, Congress suspended the use of literacy tests in all jurisdictions in which less than half of voting-age residents were registered as of November 1, 1964, or had voted in the 1964 presidential election. In 1970, Congress amended the act and expanded the ban on literacy tests to the entire country, which the Supreme Court upheld in Oregon v. Mitchell later that year (Rusk Reference Rusk2001).

Ex-Felon Disenfranchisement

We also consider the role of felon disenfranchisement—prohibiting convicted felons from voting—which started at the state level in the eighteenth century and thus is almost as old as the country itself (Brooks Reference Brooks2005). Over time, it became a near-universal policy that continues to the present day. We focus on a more impactful form of the policy that has greater variation: ex-felon disenfranchisement. The key difference between felon and ex-felon disenfranchisement is whether a person convicted of a felony regains the right to vote after serving their sentence. In some states, the right to vote is recovered while in others it is permanently forfeited. This has a more substantial effect as the pool of people who have been convicted of felonies grows very large over time, much larger than the pool who are still serving their sentences at any given point in time.

As ex-felon disenfranchisement is a formal restriction that removes voters from eligibility, it thus reduces both the numerator and denominator of the turnout fraction.Footnote ¹⁵ Because felons have historically voted at lower rates than the rest of the population, these laws generally increase turnout. In the South, we expect these laws had negative effects on Republican success.

In figure 5, we display the set of states that used ex-felon disenfranchisement at some point between 1870 and 1970.Footnote ¹⁶ In total, 40 states restricted the voting rights of ex-felons during this period. Although ex-felon laws affected both whites and Blacks, a change in the prison population in the post-Civil War era—predominantly in the South—suggests that Blacks bore the brunt. Behrens, Uggen, and Manza (Reference Behrens, Uggen and Manza2003), for example, note that “[i]n many Southern states, the percentage of nonwhite prison inmates nearly doubled between 1850 and 1870.” Alabama was an extreme case, as Manza and Uggen (Reference Manza and Uggen2006, 57) remark that “nonwhites made up just 2% of the prison population in 1850, but 74 percent by 1870.” This expansion is unsurprising given the presence of slavery in 1850, but shows that the effect of felon disenfranchisement provisions would very quickly shift from primarily working against white voters to heavily limiting African American voting after the Civil War. Variations on felon and ex-felon disenfranchisement remain widespread in the United States today, and thus our consideration of this method of disenfranchisement may be informative as to present-day politics.

Figure 5 States that Employed Ex-Felon Disenfranchisement at Some Point, 1870–1970

Sources: Behrens, Uggen, and Manza (Reference Behrens, Uggen and Manza2003); Keyssar (Reference Keyssar2000, 376–86).

The Australian Ballot

Between 1888 and 1920, party ballots—in which Republican and Democrat operatives created and distributed ballots in elections—were replaced by the Australian ballot, which was administered by state governments (Engstrom and Roberts Reference Engstrom and Roberts2020). The Australian ballot listed all candidates for office, not just those of a particular party, and was standardized to allow for secrecy in the voting process. The shift from party ballots to the Australian ballot occurred in every state in the Union, but the change came at different points in time. In figure 6, we illustrate four groupings. Most states (or territories) went to the Australian ballot by 1891, but there was also staggered adoption during the 1890s and throughout the twentieth century.

Figure 6 State Adoption Dates of the Australian Ballot

Source: Engstrom and Roberts (Reference Engstrom and Roberts2020).

The move to the Australian ballot is typically framed as part of the more general Progressive movement that swept the nation and sought to dampen the power of political parties. However, some Democratic politicians in the South also saw the adoption of the Australian ballot as a way to exclude Black voters, as it necessitated literacy—the ability to read—to be able to vote, something that was not required in the party ballot days when a variety of shortcuts (like color-coded ballots and personal relationships with party operatives) allowed illiterate citizens to exercise their franchise (Kousser Reference Kousser1974; Perman Reference Perman2001). As illiteracy was highest in the South, and Black illiteracy exceedingly high in that region, Democrats could emphasize the progressive nature of the Australian ballot while using it as a tool of disenfranchisement.Footnote ¹⁷ Because the Australian ballot did not render anyone ineligible to vote, but changed the cost of voting (potentially in a positive or negative way depending on the voter),Footnote ¹⁸ there is no uniform expectation for its effect, though the historical literature suggests that its adoption reduced turnout. The Australian ballot, like felon disenfranchisement provisions, remains in use today, and the extra costs it places on voters go largely unmentioned.Footnote ¹⁹

4. Data and Empirical Analysis

We estimate the relationship between electoral institutions—poll taxes, literacy tests, Australian ballot laws, and ex-felon disenfranchisement laws—and key electoral outcomes from 1870 to 1970 across the entire United States. We analyze presidential contests at the state level as well as gubernatorial elections. By including gubernatorial elections, we expand our dataset and include state elections, which may plausibly have different patterns than federal elections. The majority of gubernatorial elections occur in years without presidential elections, filling in the time gaps between presidential contests.

We begin our national analysis in 1870 with the return of competitive politics after the Civil War and end our analysis a century later in 1970, one election cycle after the Voting Rights and Civil Rights acts, and coinciding with the final end of literacy-test laws. In total, this provides a 101-year period of analysis, with at least one election in each year and an average of 28.8 elections per year. This is an admittedly long period in which certain factors certainly changed. We might alternatively conceive of two periods. First, the 51 years following the Civil War, ending in 1920, which saw the adoption of disenfranchisement laws, the expansion of the right to vote to women, and the adoption of the Australian ballot in almost all states.Footnote ²² A second period of 50 years, culminating in 1970 with the final abolition of literacy tests, is the period of voting liberalization that saw a series of voting restrictions crumble as well as a universal or near-universal use of the Australian ballot and broad voting eligibility for women. However, our results in these two periods are sufficiently similar—both to each other and to the overall results for the 101-year period—that we focus on the entire range from 1870 to 1970 and report the results of the sub-analyses by smaller time periods in the appendix (tables A1 and A2).

We use two dependent variables. First, we are interested in turnout as an indirect measure of informal disenfranchisement. For each presidential election and gubernatorial election, we define turnout as the number of votes cast in the statewide election divided by the voting-eligible population in the state that year, presented in a 0–100 percentage format (Burnham Reference Burnham2010). Thus, if one thousand people were eligible to vote and 563 votes were cast, this would take the value 56.3 in our dataset. The best available estimates for the voting-eligible population are imperfect. They are conditioned on rules about age, sex, race, and nationality status, based on census measures, but fail to exclude more granular groups, such as those removed by ex-felons. Turnout ranged from about 1%Footnote ²³ to 101%,Footnote ²⁴ and averaged 56.5%. The unique case of a theoretically impossible 101% turnout rate was the result of imperfections in the estimates of the voting-eligible population.

Turnout, even if measured perfectly, is limited as it includes both too little and too much to fully capture disenfranchisement. First, it does not account for formal disenfranchisement. For example, in the period before the Nineteenth Amendment and earlier state-specific female suffrage laws, women were not part of the voting-eligible population and thus did not influence turnout percentages. Thus, turnout does not tell us about the tens of millions of women who were disenfranchised for more than a century. In addition, turnout does give us information about informal disenfranchisement—methods to keep people from voting who had the de jure right to vote. Such people are in the numerator but not the denominator and thus systematically push the turnout variable toward zero. However, other things also influence turnout, including the type of election, the appeal of the contest to voters, variable enforcement of election laws, and larger social trends. A perfect measure of disenfranchisement is not available, and turnout offers a strong connection. But we move forward cognizant of the limitations of turnout as a measure.

While election turnout is interesting as an indication of civic liberty and political engagement, we are also interested in the extent to which targeted reduction of turnout through disenfranchisement works to secure power for particular interests. In our case, we are most interested in the ability of Southern Democrats in the post-Reconstruction period to exclude Republican-voting Black voters and thus suppress any Republican power in the state. Thus, our second dependent variable is Republican vote share, which ranged from 0% to about 92%, with an average of 45.6%.Footnote ²⁵

Our key independent variables are a series of dummy variables indicating whether a particular electoral institution was “turned on” at the time of a given election. These variables include poll tax, literacy test, Australian ballot, and ex-felon disenfranchisement. In each case, the variable takes the value “1” when the institution was on the books and “0” when it was not. We count poll taxes as existing if there was any fee applied to voting, and do not differentiate by amount, whether they accumulated, or other small variations. We count a state as having a literacy test if it had either a reading or a writing test to vote, and do not differentiate based on small interstate variations.Footnote ²⁶ We count Australian ballot as “1” if the state utilized any form of the Australian ballot, which presented voters with competing candidates rather than party-prepared straight-ticket ballots.Footnote ²⁷ Finally, we count ex-felon disenfranchisement as “1” if the state restricted the right to vote after the completion of the incarceration component of the sentence on top of the near-universal restrictions on felons serving their sentence.

Additionally, we include a set of control variables. The first is women’s suffrage, which takes the value “1” if the state allowed women to vote in that type of election—presidential or gubernatorial—and when universal enfranchisement occurred with the Nineteenth Amendment. We also include lynchings, which is a count of the number of lynchings of African Americans that occurred in that state in that election year.Footnote ²⁸ Finally, we also control for the electoral context, noting when given elections were gubernatorial elections (in a presidential year and in a non-presidential year). The base category is a presidential election.

We are unable to specify how—and how intensely—each policy was enforced in any given year, and this was surely heterogenous. The level of enforcement of the literacy test, for example, varied based on the preferences of state and local administrators. It is extremely difficult or impossible for us to measure the varied intensities of enforcement. That said, formal state policies empowered local administrators and should have introduced observably systematic differences where they influenced actual turnout.Footnote ²⁹ The historical record provides evidence that the ex-Confederate states, places with larger Black populations, and those with more extreme records of institutionalized white supremacy aggressively used electoral institutions to disenfranchise and shape the electorate. Thus, as a first approach to measuring variation, we include South, which we define as the 11 ex-Confederate states, in an interaction with each of the electoral institution variables. In a subsequent section, we analyze similar models with a different variable—Black percentage of the population—in place of South.

Our estimation strategy is a two-way fixed-effects ordinary least squares (OLS) model, with year and state fixed effects, allowing our model to estimate, through institutions turning on and off at specific times, the association of different electoral institutions with changes in turnout and Republican vote share. We note that electoral institutions are not added and removed at random, and thus we should be cautious in interpreting the results.Footnote ³⁰ Our findings are suggestive, but not definitive. We argue that they are descriptively identified, but not fully causally identified.

We present our results in table 1, first for turnout (models 1 and 2) and then for Republican vote share (models 3 and 4). Models 1 and 3 do not contain the South interaction, while models 2 and 4 do.

Table 1 The Effect of Electoral Institutions on Turnout and Republican Vote Share

Notes: Numbers in cells are OLS coefficients with clustered robust standard errors in parentheses. ^ = p < 0.1; * = p < 0.05; ** = p < 0.01.

A threshold question is whether there is sufficient variation between the South and the rest of the country to merit exclusively evaluating models 2 and 4. If the models yield similar results, then the simpler models in tables 1 and 3 would be preferable. Instead, we find significant differences between model 1 and model 2, indicating that there was substantial variation between the South and non-South in the relationship between institutions and turnout. To a lesser—but still meaningful—extent, we observe the same between model 3 and model 4. The models with South interactions explain considerably more of the variation and tell a very different, and more nuanced, story. Thus, we analyze the results in models 2 and 4. For ease of understanding, we calculate the region-specific relationship by combining the interaction and base terms appropriately for ex-Confederate states and present them side by side with non-South states, as well as the difference between the two. We present these in table 2 for the turnout dependent variable and in table 3 for the Republican vote share dependent variable. We reiterate that tables 2 and 3 do not contain new models, but rather are derived from models 2 and 4 in table 1.

Table 2 Regional Estimates Based on Model 2 (Turnout) in Table 1

Table 3 Regional Estimates Based on Model 4 (Republican Vote Share) in Table 1

First, we find that the poll tax, used exclusively in the South, was most strongly associated with reducing turnout. All else being equal, when Southern states imposed a poll tax, their turnout was about 18 percentage points lower. And this had the intended effect: about a 9 percentage-point reduction in Republican vote share. This is confirmatory evidence that the poll tax substantially contributed to disenfranchising poor (and largely African American) voters and guaranteeing the (white) supremacy of the Democratic Party in these states.

Second, we find that literacy tests outside the South were not significantly associated with any change in turnout. Our best estimate is that they reduced turnout by less than 1 percentage point, with a 95% confidence interval between −5 and +3 percentage points. This likely reflects that, though these laws remained “on the books” in many non-Southern states into the mid-twentieth century, with near-universal literacy and weak application, there was no discernable impact on turnout. Outside the South, where the Republicans were far more likely to be in power, we find evidence that the Republican Party benefited from the imposition of literacy tests by about 3 percentage points of the vote share.

Literacy tests in Southern states were another matter altogether. Southern literacy tests, likely enforced with much more vigor and discriminatory intent against African American voters, are associated with about an 8 percentage-point reduction in turnout and about a 6 percentage-point reduction in Republican vote share. In Southern states, literacy tests proved an effective secondary tool to limit voter turnout by those who might vote against the dominant Southern Democrats.Footnote ³¹

We find no discernable relationship between adopting the Australian ballot and turnout or with Republican vote share. Adopting the Australian ballot, all else being equal, did not drastically change the composition of the electorate or the electoral success of either major party. Interestingly, we do still find a statistically significant difference between the South and the non-South. While the Australian ballot is associated with an increase in turnout outside the South, it is associated with a decrease in the South of about equal size. So, while neither alone is significantly distinguishable from zero, the difference between them is distinguishable—giving some evidence of a different impact of the switch to the Australian ballot.Footnote ³² While we cannot speak to long-term changes in political culture that the Australian ballot may have contributed to, we find little evidence that adoption of the Australian ballot directly impacted turnout or voting.

Finally, ex-felon disenfranchisement policies had no meaningful impact on turnout or partisan results outside the South. We estimate both coefficients to be less than 2 percentage points and with confidence intervals that overlap zero. Again, however, the South is a different story. There, ex-felon disenfranchisement policies are strongly associated with reduced turnout (by about 10 percentage points) and reduced Republican vote share (by about 8.5 percentage points). This points to ex-felon disenfranchisement as an impactful policy on the level of, or exceeding, literacy tests within the South. Excluding felons and ex-felons from voting, coupled with an aggressive criminal justice system unfairly employed against African Americans, yielded substantial drops in turnout and in the Republican vote in the South.

In sum, we find robust evidence that Southern efforts at voter disenfranchisement through formal policies, unsurprisingly, helped to maintain white supremacy. Though other variations and institutions were attempted, the poll tax, ex-felon disenfranchisement laws, and literacy tests formed an arsenal that was extremely effective at depriving African Americans (and poor whites) of their right to vote and guaranteeing Democratic control.

5. Digging Deeper: Race as a Driving Factor for Southern Electoral Outcomes

Our main model relies on a blunt South-versus-the-rest-of-the-country coding, where we estimate an association of each electoral institution for both Southern and non-Southern states. This is decidedly limited and fails to get at what might drive the significant differences between the two regions. There is nothing magical about the designation “South” that changes the impact of electoral institutions. Instead, there must be a more complete explanation that differentiates the ex-Confederate states from the rest of the country. With no novelty, we focus on the importance of race in Southern electoral politics.

The story of disenfranchisement in America is largely about race and specifically the removal of African Americans from the electorate—something formally prohibited by the Fifteenth Amendment, but informally and indirectly possible through alternative electoral institutions. One possibility is that “South” is effectively a proxy for places with large African American populations and an entrenched culture of white supremacy that was powerful enough not only to support slavery but to participate (through secession) in a bloody civil war to preserve it. Though racism was present across the country, with segregation and blatant discrimination common through the mid-twentieth century, it is plausible that the ex-Confederate states had a political culture among white elites that would tolerate more extreme applications of white supremacy to maintain power. It is also true that the large African American populations in the South were potentially winning coalitions, likely for Republican candidates for much of the time period we analyze. Thus, white Democrats in the South had more to gain from aggressive exclusion.

State demographics are important, not just for a cultural or political explanation, but also in simple numbers. Even if there was a pointed effort to remove the Black population from politics across all states equally, then the disenfranchising institutions should have been more impactful in places with a larger African American population share, namely the post-Confederate South. Thus, we reestimate our models with Black percentage of the population as an alternative interactive variable.Footnote ³³ Because the African American share of the population varies within states over time—unlike South—this variable is included both on its own and as part of interaction terms. Across our entire dataset, the average of Black percentage of the population was 9.66%, ranging from 0.06% to 60.7%. The average was 33.6% and 2.7% among Southern and non-Southern states, respectively. This confirms the stark difference between the two regions.

We present the results for models with turnout as the dependent variable in table 4. We see that racial demographics offer a substantial explanation for the unique impact of Southern electoral institutions. In model 1, we find a strong relationship between electoral institutions, racial demographics, and turnout. Because these are interactive models, it is important to evaluate the coefficients at appropriate levels of the component variables. For example, neither poll tax nor the interaction, poll tax × Black percentage of the population, have a significant coefficient. However, the marginal effect of poll tax becomes significant at the p < 0.05 level around the point when the Black share of the population crosses 5%. In the average Southern state, this amounts to about a 19.57 percentage-point reduction in turnout, in line with estimates in prior models in this paper.

Table 4 The Effect of Electoral Institutions and Racial Demographics on Turnout

Notes: Numbers in cells are OLS coefficients with clustered robust standard errors in parentheses. ^ = p < 0.1; * = p < 0.05; ** = p < 0.01.

To account for this interactive relationship for all the main electoral institutions we analyze, we estimated the effect of each variable at four relevant levels of Black percentage of the population, based on model 1 in table 4. We use the minimum observed percentage (0.06%), the average in Non-Southern states (2.7%), the average in Southern states (33.15%), and the maximum observed percentage (60.7%). The results are presented in table 5.

Table 5 Marginal Effects Estimated from Model 1 in Table 4

In addition to the large impact of the poll tax, we also find that in the average Southern state (by Black share of the population), the Australian ballot and ex-felon disenfranchisement are each associated with a significantly reduced turnout (by about 5.5 and 8 percentage points, respectively). The literacy test’s relationship with turnout does not appear to have been mediated by the Black share of the population. The simplest explanation for this is that the poll tax and ex-felon disenfranchisement substantially reduced the number of African Americans who could have been disenfranchised by the literacy test.Footnote ³⁴ Our results suggest that literacy tests may have been used in Southern states to remove those who would not have been as thoroughly removed by other institutions—likely a predominantly white set of voters, such as populists.

In table 6, we continue our analysis, by replicating table 4 with Republican vote share as the dependent variable. We then replicate table 5 via table 7, showing the interactive effects for representative levels of state racial diversity. All other features of the models remain the same.

Table 6 The Effect of Electoral Institutions and Racial Demographics on Republican Vote Share

Notes: Numbers in cells are OLS coefficients with clustered robust standard errors in parentheses. ^ = p < 0.1; * = p < 0.05; ** = p < 0.01.

Table 7 Marginal Effects Estimated from Model 1 in Table 6

As the historical record informs us, the goals of disenfranchising institutions were multifold. In addition to the white supremacist opposition to Blacks participating in politics, there was the instrumental goal of guaranteeing Democratic control in the Southern states. Thus, we should observe that not only did these electoral institutions exclude prospective voters, but they also reduced the Republican Party’s vote shares in these elections. And that is what we find. The poll tax and ex-felon disenfranchisement are strongly associated with reduced support for Republicans in the South. In the average Southern state (by Black share of the population), the poll tax and ex-felon disenfranchisement were each associated with a significantly reduced Republican vote share (by about 8.5 and 7.5 percentage points, respectively). And at the highest observed Black population shares, those reductions tripled and doubled, respectively. Literacy tests in these Southern contexts, though, added almost nothing on top of that.

6. County-Level Analysis

State-level analyses offer descriptions—and some explanations—of the differences between the units at which most electoral institutions were set: states. But these state analyses struggle to aid explanation of why the differences emerged or how they functioned in practice. For example, while we observe a strong correlation between the share of the population that was Black in a given state and turnout reductions associated with restrictive electoral institutions, these institutions were not adopted at random and full causal identification is not possible. Southern states had uniquely high shares of their population made up of African Americans, in ways that make it difficult to compare to other states. But if we move our analysis to the county level, other possibilities emerge. Some non-Southern counties had relatively high racial diversity and some Southern counties were almost entirely white. If it is true that the restrictive electoral institutions primarily worked in the South against African Americans, then we should observe the greatest turnout effects in places that had very large Black populations, and far less in the almost entirely white Appalachian counties of the South.

We investigate this by constructing a dataset of county-level election turnout in presidential elections from 1872 to 1968 (Clubb, Flanigan, and Zingale Reference Clubb, Flanigan and Zingale2006). For each county, in each state, in each election, we use the estimate of turnout based on the total number of votes cast in the county and the estimated county adult population of those broadly eligible to vote (such as the adult male citizen population before the expansion of suffrage to women). We note that this is not exactly equivalent to the voting-eligible population, as the share of noncitizens and those otherwise legally ineligible to vote are not precisely estimable. This gives us an estimate of voter turnout (county) for each county election observation.

Additionally, we rely on US Census data from each decennial census between 1870 and 1970 to estimate the percentage of the population that was Black in each county as of the year of a given election. For the presidential elections held in census years—1880, 1900, 1920, 1940, and 1960—we use the estimate of that year’s census without changes. For elections falling between censuses, we linearly impute the percentage from a combination of the decennial census before and after the election. Thus, each county in 1872 received a percentage equivalent to the 1870 census value for that county plus 20% of the change from 1870 to 1880. In 1876, the county receives the 1870 percentage plus 60% of the change from 1870 to 1880. This is replicated through the entire dataset to produce an estimate for each county election from 1872 to 1968. We call this variable Black percentage of the population (county).

Otherwise, we replicate our variables from prior models, with the same set of institutions and controls, with fixed effects now set at the election and county levels. We initially attempt the three main models of prior tables: a noninteractive model, a model interacted with a South dummy variable showing consistent differences in the South, and finally an interactive model with Black percentage of the population (county), which shows to what extent it is county-level demographics that explain variation. We present the results in table 8.

Table 8 The Effect of Electoral Institutions on Turnout and Republican Vote Share

Notes: Numbers in cells are OLS coefficients with clustered robust standard errors in parentheses. ^ = p < 0.1; * = p < 0.05; ** = p < 0.01.

These county-level analyses of presidential elections broadly conform to our state-level results in prior models with two exceptions: while state models had indicated that literacy tests and the adoption of the Australian ballot were not strongly associated with a race-based effect, our county-level models indicate otherwise. We find that the effect of literacy tests was almost entirely associated with the percentage of a county that was Black, with no apparent relationship otherwise. This indicates that, at least in presidential elections, while state demographics may not have strongly correlated with the role of literacy tests, within states, it was in counties with large Black populations that they had their strongest association with turnout. Similarly, we find that the Australian ballot’s relationship with turnout was also strongest in counties with the largest Black shares of the population.

Otherwise, we observe broadly similar patterns, with poll taxes and ex-felon disenfranchisement showing strong negative relationships with turnout, especially in counties with large Black populations. Much as in the state-level analyses, we come away with the strong takeaway that the poll tax was the most potent formal institution of disenfranchisement, with other institutions having a significant relationship, but one that only reached comparable levels to the poll tax in counties that were heavily African American.