Turnout

In light of the stakes being decided in the secession conventions, historians have noted that turnout in the elections for delegates was surprisingly low, particularly in the Lower South (e.g., Johnson Reference Johnson1999; Barney Reference Barney2004; Freehling Reference Freehling2007). We quantify this decline by taking the 1860 presidential election turnout as a benchmark, which provides a lower bound for the number of eligible voters in each county. This election occurred less than two months prior to all the elections for delegates and provides a plausible counterfactual for the potential participation under a statewide referendum.Footnote ¹⁶

Figure 1 shows the mean county-level support for secessionist (“Sec”) and cooperationist (“Coop”) candidates, respectively, as a share of 1860 presidential turnout. The third bar in each panel shows our measure of abstention (“Abst”), defined as the shortfall between the total number of votes cast for delegates and the total number of votes for president (as a share of the latter). Observations are weighted by the total number of presidential votes to account for the size of county electorates.Footnote ¹⁷ In our Online Appendix, we present these figures separately by state.

Figure 1

PARTICIPATION AND COMPETITIVENESS: ELECTION OF CONVENTION DELEGATES

Notes: Each bar represents the average district-level votes in the elections for convention delegates in AL, GA, and LA as a share of the 1860 presidential turnout (weighted by the latter). Sec and Coop are the vote shares received by secessionist and cooperationist candidates, respectively. Abst is the average percentage point difference in the total number of votes cast between the 1860 presidential and convention elections. The middle panel only takes the sample of districts in which the highest vote was obtained by a secessionist candidate(s). The right panel are districts in which cooperationist candidates obtained the highest vote.Sources: See Online Appendix Table 1.

As shown, secessionist candidates obtained, on average, only 40 percent of the votes cast in the preceding presidential election. Moreover, on average, there is an abstention of 23 percent in the election of delegates compared to the turnout observed in this presidential election. Figure 1 also shows these results when disaggregated by counties won by secessionist and cooperationist candidates, respectively. In the secessionist counties (middle panel), which combined elected slightly more than half of the delegates, there is an approximately 30-percentage-point difference in the number of votes cast between the 1860 presidential and respective convention election. This is approximately twice the average abstention observed in the counties won by cooperationists (right panel), which together elected 46 percent of the delegates.Footnote ¹⁸

As a consequence of this steep decline in participation, particularly in counties won by secessionists, the majorities in these conventions were elected with only 27 percent of the votes cast in the preceding presidential election.Footnote ¹⁹

An additional factor associated with the low turnout in the elections of delegates is the high number of districts with candidates running unopposed. This potentially had a negative impact on participation as the strategic choice by candidates of whether or not to run should affect the number of votes cast, as the opposition voters have no candidates to vote for. In total, 12 percent of the districts of these three conventions were uncontested. Of these, nearly three-fourths were won by secessionists. In the key states of AL and GA, candidates for either side ran unopposed in more than 13 percent of the districts. The share of uncontested races was more than twice as high for districts won by secessionist candidates (16 percent were uncontested) as for those districts won by cooperationists (6 percent).

Local Competitiveness

While many causes have been offered for the decline in turnout, such as poor weather (Johnson Reference Johnson1999), voter intimidation (Barney Reference Barney2004, pp. 212–19; McCurry Reference McCurry2010, pp. 49–54; Williams Reference Williams2008), resignation that secession was inevitable (Barney Reference Barney2004; Sinha Reference Sinha2003), and collective action problems inherent to the heterogeneous cooperationist coalition (Freehling Reference Freehling2007; Johnson Reference Johnson1999, p. 24), none has been able to account for the observed variation in participation across counties. While approximately 10 percent of counties experienced a decline in turnout of more than 50 percent, around 10 percent also witnessed an increase in turnout.

The electoral incentives at the local level partially explain this variation in participation. As is well known, the use of plurality voting to select representatives out of geographic districts is associated with low turnout (Jackman Reference Jackman1987), particularly in non-competitive districts (e.g., Eggers Reference Eggers2015).Footnote ²⁰ This effect is relevant to the participation in the election of delegates given the high spatial concentration of slaves, especially in AL and GA. Namely, in districts having a high share of slaves, the likelihood of electing a secessionist candidate was high. Similarly, the likelihood of electing a cooperationist representative was high in the lower slave-share districts. This implies that many of these local races were unlikely to be competitive. The electoral returns bear this out. The average margin of victory was 44 percent, and in only 14 percent of counties was the election decided by 10 percentage points or less.Footnote ²¹ As a result, the total number of effective votes (i.e., only those votes needed to win the election) across districts electing secessionist candidates was equivalent to only 11 percent of the turnout in the 1860 presidential election. To put this in perspective, the number of effective votes in districts won by secessionists (27,081) was less than half of the number of slaveholders living in those districts (58,058).

In Figure 2, we provide systematic evidence of the positive association between local competitiveness and turnout in these elections. Again, we use turnout in the 1860 presidential election as a benchmark. The left panel presents a partial regression plot with the residual values of the percentage change in participation on the y-axis against the residuals of a measure of district competitiveness on the x-axis.Footnote ²² The effect of several factors is partialed out. First, due to the salience of slavery to secession, we control for the share of the district’s population who were slaves. Second and similarly, we include total district population, as the likelihood of any voter being pivotal is decreasing in district size. Last, to control for unobserved state-level factors, we include state fixed effects.Footnote ²³

Figure 2

CHANGE IN PARTICIPATION AND LOCAL ELECTORAL COMPETITIVENESS

Notes: Partial regression plots with the residuals of the change in voter participation between i) the elections for convention delegates in AL, GA, and LA (Panel A) or ii) the secession referenda in NC, TN, and VA (Panel B) and the 1860 presidential election, against the residuals of district competitiveness measured as 1 – abs (district-level margin of victory). In each model, we partial out the effect of slave population (over total district population) and (ln) total district population. Both models include a full set of state fixed effects and allow for arbitrary heteroscedasticity at the district level. Plots produced in Stata with the added-variable plot command (avplot).Sources: See Online Appendix Table 1 and main text.

As shown in Panel A, there is a positive and highly significant correlation between district competitiveness and the change in turnout between these elections. The slope coefficient of 0.51 (S.E. = 0.06) implies that a one-standard-deviation increase in the competitiveness of a district is associated with an increase in participation of approximately 16 percentage points. Compared to the mean decline in turnout in these states (28 percent), the magnitude of this increase is substantial.Footnote ²⁴

Further evidence that the electoral system used to select delegates conditioned voter behavior during the crisis comes from the Upper South states, which held state referendums on questions pertaining to secession. In NC, for instance, where voters decided not to hold a secession convention, turnout declined, on average, 5 percent compared to the 1860 presidential election. In a similar referendum in TN, turnout declined, on average, 12 percent.Footnote ²⁵ Overall, the average decline in turnout in the referendums for which data are available (NC, TN, and VA) was below 10 percent. More importantly, since every vote in a referendum adds to the aggregate state tally, there should not be an association between local competitiveness and turnout. This is precisely what we find in Panel B of Figure 2. As can be seen, the relationship between the closeness of the district-level result and the change in turnout after controls are taken into account is not statistically different from zero.

In our Online Appendix, we explore the robustness of these conditional correlations. For example, we expand the set of controls and account for additional factors such as the pre-crisis level of electoral competitiveness proxied by the number of parties in the 1860 presidential election and the incidence of slaveholders in the district (see Online Appendix Figure 2). We also show that these results are not driven by any outliers; in fact, once we exclude outliers, the slopes are more precisely estimated (Online Appendix Figure 3).

Figure 3

CONVENTION REPRESENTATION AND SLAVEHOLDERS INCIDENCE

Notes: Partial regression leverage plots with the residuals of the (ln) convention RRI against the residuals of the slaveholding population over the district’s electorate. The electorate is defined as the AWM population minus the AWM foreign-born population. Each dot represents a single district. The model in Panel A partials out the effect of total district population in 1860 and state fixed effects. The model in Panel B controls in addition for the (ln) mean value of farms per acre and the (ln) population density (total population/district area). Estimated standard errors allow for arbitrary heteroscedasticity at the district level. Plots produced in Stata with the added-variable plot command (avplot).Sources: See Online Appendix Table 3.

Did Abstainers Oppose Secession?

Our evidence suggests that if the Lower South states would have allowed referendums—either asking whether a convention should be held or to ratify a convention’s choice—overall turnout would have increased and perhaps even a majority would have voted against secession. This counterfactual is consistent with a historical literature downplaying the popularity of the movement and arguing that only a minority in each state supported it (e.g., Escott Reference Escott1978; Freehling Reference Freehling2007; Potter Reference Potter1942).Footnote ²⁶

Historical accounts provide suggestive evidence that abstainers in the high slave-share counties opposed secession (McCrary, Miller, and Baum Reference McCrary, Miller and Baum1978) and did not vote because of intimidation by secessionists (e.g., Barney Reference Barney2004; Sinha Reference Sinha2003; Williams Reference Williams2008).Footnote ²⁷ For instance, a cooperationist in MS said, “I knew many who were in favor of the Union, who were intimidated by threats, and by the odium attending it, from voting at all” (as cited in Barney Reference Barney2004, p. 269). Similarly, a cooperationist in TX wrote that secession was achieved by the coercion of a “ferocious minority … I only dared protest as strongly and I thought prudent … accordingly I did not vote” (as cited in Freehling Reference Freehling2007, p. 459). Hence, some historians have claimed that the elected pro-secession majorities in the Lower South states reflected substantial voter fraud and intimidation (e.g., McCurry Reference McCurry2010; Merritt Reference Merritt2017).

The argument that abstention was systematically influenced by coercion fits broadly with the more conventional view that the slaveholding elite used “the traditional powers of their planter oligarchy” to overcome the will of the majority during the crisis (Donnelly Reference Donnelly1965, p. 81).Footnote ²⁸ While these tactics may have helped slaveholders to obtain pluralities in the high slave-share districts, it would have been much more costly, if not impossible, to employ broadly in the lower slave-share districts.Footnote ²⁹ Moreover, since secessionists only needed to win pluralities in districts comprising a majority of delegates, the geographic scale of where they may have needed to employ these tactics was significantly reduced. That is, the conventions localized political competition and significantly lowered the share of voters needed in order to obtain a majority of delegates.

Political Inequality in the Conventions

An important source of distortion in the representation of slave interests originated in the apportionment rules of the conventions. We demonstrate this using a measure based on the number of delegates per district. This information was coded from the individual pieces of legislation passed in late 1860 and early 1861 stipulating how each convention’s delegates would be elected. In each state, the delegates were elected out of the existing electoral districts for one or both chambers of the state legislature. For instance, representation in the FL, GA, LA, and SC conventions equaled the combined number of representatives and senators apportioned to the state legislature. In AL and MS, the delegates were elected using only the basis of apportionment for each state’s lower house.

Following Ansolabehere, Gerber, and Snyder (Reference Ansolabehere, Gerber and Snyder2002), we use a metric that is relative to the “fair” level of representation in each particular state (the “relative representation index” (RRI)). Formally, this measure is

(1) $$RR{I_i} = {{{D_{j(i)}}/{N_{j(i)}}} \over {{D_j}/{N_j}}}$$

where the subscript j(i) indicates that district i is located in state j. D is the number of delegates and N denotes the voting population. This index creates a common metric across conventions by normalizing the representation of each locality by the voting power specific to each state. Individuals in districts with an index of less (more) than one were under-represented (over-represented) in their convention. A value close to one corresponds to a level of representation consistent with the “one person, one vote” principle.

To estimate the electorate, we take the AWM population as a proxy for the number of registered voters and subtract the adult white foreign-born males. This latter group was largely ineligible for failing to meet citizenship or residency requirements (Keyssar Reference Keyssar2001).Footnote ³⁰ These variables are constructed from the 1860 Census.Footnote ³¹ We focus the analysis on the founding members of the Confederacy (AL, FL, GA, LA, MS, and SC), whose choice to secede was made solely by elected representatives in state conventions. In our Online Appendix, we show the close association between legislative and convention over-representation in these states (Online Appendix Figure 4).

In Figure 3, we present two scatterplots exploring the relationship between the incidence of slaveholders in the electorate and the relative representation of counties in the conventions. Namely, these figures show two partial regression plots with the residuals of the convention RRI (in natural logs) on the y-axis against the residuals of the share of slaveholders in the electorate on the x-axis. The latter captures the breadth of slave ownership among each county’s electorate. We take the natural log of the RRI to reduce the weight of outliers and the right skew of this variable. Both models include state fixed effects; therefore, the slope coefficients have a within-state interpretation. Each marker corresponds to a single district in the corresponding state.

In Panel A, we first show a parsimonious model that controls only for (ln) total county population and state fixed effects. The estimated slope coefficient on the local incidence of slaveholders is positive and highly significant. The point estimate of 1.65 (S.E. = 0.14) in Panel A implies that holding all other things equal, a district where half of its voters are slaveholders will have a predicted RRI score about 25 percent greater than that of the average district in the sample (where slaveholders comprised roughly one-third of the electorate). This difference represents approximately 48 percent of the standard deviation observed in the empirical distribution of the log RRI for these conventions. The difference in relative representation with a district one standard deviation below the mean, where slaveholders represented 19 percent of the electorate, is more than 65 percent.

In Panel B, we control for population density, which is usually associated with legislative malapportionment and is positively correlated with slavery.Footnote ³² To calculate this measure, we take the county size and 1860 boundaries from the Atlas of Historical County Boundaries (Newberry Library, Chicago). In addition, to account for the possibility that over-representation is driven by land wealth, which is also positively correlated with slave labor incidence, we include the district’s farmland value per acre as a control.Footnote ³³ As is evident from Panel B, the positive association between over-representation and the share of slaveholders in the electorate is largely unchanged when these covariates are partialed out.

In our Online Appendix, we explore the robustness of these results. First, we expand the set of controls to include other economic characteristics of constituencies (e.g., land inequality). Second, we use alternative measures for the local-level importance of slavery: the share of the electorate who were planters, defined as slaveholders owning 20 or more slaves (Wright Reference Wright1978). Last, we perform a sensitivity analysis excluding outliers and taking each convention separately. Overall, the positive and highly significant association between slaveholders and over-representation in the conventions is robust to the inclusion of additional controls and across specifications (Online Appendix Figures 5–8).

The Political Geography of Slavery

Political scientists have demonstrated that in electoral systems with plurality voting, the degree to which representation reflects the preferences of the electorate depends on the spatial distribution of voters (Cox and Katz Reference Cox and Katz2002). This was particularly relevant in the election of convention delegates, as secession divided the electorate across slave-ownership lines (see, e.g., Lipset 1960; Key 1984; Johnson Reference Johnson1999). In particular, the concentration of slaves meant that slaveholders were spatially distributed such that despite being a minority overall, they could obtain pluralities in enough districts to elect a majority of delegates.

In Table 1, we illustrate this electoral bias. In Column (1), we report the average district non-slaveholder–to–slaveholder ratio in each of the original Confederate states.Footnote ³⁴ Although slaves constituted nearly 50 percent of the total population in these states, there were, on average, more than three non-slaveholders for every one slaveholder. We then sort counties from lowest to highest slave share and create two groups, each containing approximately half of the state’s non-slaveholders. In the low slave-share districts (Column (2)), which elected 36 percent of the convention delegates, there were, on average, approximately six non-slaveholders for every slaveholder. Column (3) shows that in the remaining counties, there were, on average, only 1.58 non-slaveholders for every slaveholder. As reported in Column (4), around 74 percent of the slaveholding population lived in these highest slave-share counties. This uneven spatial distribution of slave ownership was particularly pronounced in the three Lower South states with the largest white populations (AL and GA). The electoral advantage of this is shown in Column (5), as these most enslaved counties selected, on average, almost two-thirds (64 percent) of the delegates.

Table 1

SPATIAL DISTRIBUTION OF SLAVE OWNERSHIP AND REPRESENTATION

Notes: The ratio of non-slaveholders to slaveholders (Columns (1)–(3) and (6)), the share of slaveholders (Column (4)), and the share of voters (Column (7)) are calculated from the 1860 Census. The set of “Minimum Winning” districts (Columns (6) and (7)) is the smallest set of counties electing a simple majority of delegates in each convention, sorting counties from highest to lowest slave share (slave population/total county population). Number of delegates obtained from the official journal for each convention.

Sources: 1860 Census and convention’s journal.

Finally, in Columns (6) and (7), we combine these demographics with the apportionment of each convention. Specifically, we construct a set of “minimal winning districts,” defined as a subset of the most enslaved districts electing a simple majority of delegates. As shown in Column (6), the non-slaveholders–to–slaveholders ratio of these districts was significantly lower than the overall average, and while they contained approximately 47 percent of the electorate in these states (Column (7)), they had 63 percent of its slaveholders and more than 80 percent of its planters. Hence, it is not surprising that these localities overwhelmingly elected secessionist candidates. Indeed, as we show in the next section, nearly 90 percent of the delegates from these counties voted for secession in their convention. Since secessionists only needed local majorities in these minimal winning districts, the choice of secession could have been made by slaveholders alone. To wit, in the three states for which we have electoral returns and census information, the slaveholders in the districts won by secessionists (58,058) represented a substantial share of the total votes received by pro-secession candidates (67,181).

Was the Bias in Representation Pivotal?

While the evidence indicates that secessionist leaders believed that using conventions and avoiding referendums was crucial to the success of their movement, can we conclude from the existing electoral returns that the bias to representation was pivotal to the election of pro-secession majorities in the conventions? We now explore this question using a simple basis for fair representation to construct a counterfactual delegation for the three conventions for which we have district-level electoral returns.

To conduct this analysis, we simplify matters by using the rules of apportionment employed in each state. This means, for instance, that we use the total number of delegates the legislatures specified for each convention (e.g., 100 delegates in AL, 301 in GA, and 130 in LA). We then reapportion each convention closest to the “one (adult white) man, one vote” principle. Specifically, we calculate the state’s electorate per delegate and assign delegates to each district based on the size of their electorate. In each state, we use the existing electoral districts. For AL and GA, counties were not combined to make electoral districts and, thus, each delegate was wholly elected out of a single county. Hence, we reapportion delegates entirely from one county to another.

The choice of the LA legislature to select delegates out of both the existing lower and upper house districts complicates the analysis. For one, most senate districts were composed of multiple counties. Furthermore, the delegates elected out of senatorial districts comprised only 32 of the convention’s total delegation of 130. As a result, our counterfactual exercise in this state primarily entails equalizing representation between the delegates elected out of house and senatorial districts. Finally, to calculate the counterfactual for each convention, we assume that voters would have voted the same under this new apportionment (i.e., voting behavior is invariant to the number of local representatives).

Our findings are presented in Table 2. For each convention, we divide counties into quartiles sorted from the highest slaveholder proportion to the lowest. Thus, Q1 (Q4) consists of counties whose electorate had the highest (lowest) proportion of slaveholders within each state. Columns (1)–(3) report the share of the state’s slaves, slaveholders, and electorate each quartile contains. Column (4) reports the number of delegates these counties were actually apportioned and Column (5) the number of those who were elected as secessionists. For example, in AL, the top quartile counties contained more than half the state’s slaves, 40 percent of its slaveholders, and around 23 percent of its electorate, and they were apportioned 23 percent of the delegates. As shown in Column (5), all of these delegates were elected as secessionists. Column (6) reports the total number of delegates this group would receive under fair apportionment. In Column (7), we use the electoral returns to calculate the number of counterfactual delegates who would have been elected as secessionists. The top two quartiles lose three delegates to reapportionment and, as a result, the total number of elected secessionist delegates declines from 53 to 50 (out of 100).Footnote ³⁵ Table 2 shows a similar pattern in our counterfactual GA convention: The highest slaveholder-proportion counties lose seats. As a result, secessionist delegates actually become a minority in this crucial state.

Table 2

COUNTERFACTUAL APPORTIONMENT BY SLAVEHOLDER-SHARE QUARTILES

Notes: Quartiles in each state are obtained sorting counties by highest-to-lowest slaveholder share in the electorate. Columns (1)–(3) report the share of each state’s slaves (1), slaveholders (2), and electorate (3) in each group. Columns (4) and (5), respectively, show the number of actual delegates apportioned to the counties in each quartile (4) and the number of these in which Secessionist candidates were elected (5). Columns (6) and (7), respectively, report the counterfactual number of delegates apportioned to each county under fair apportionment (6) and then the number of secessionist candidates elected under this apportionment (7) using the same observed patterns in the election of delegates.

Source: See Online Appendix Table 3.

In LA, because the senatorial apportionment was more biased against low slave-share districts, this analysis resulted in a roughly 5 percent increase in the number of cooperationist delegates. Yet, the slim secessionist majority in the low slave-share and under-represented Orleans Parish (i.e., New Orleans), a surprising outcome largely attributed to low turnout among anti-secessionists (e.g., Gary Reference Gary2004; McCrary, Miller, and Baum Reference McCrary, Miller and Baum1978), is possibly the reason why our counterfactual analysis does not result in a larger cooperationist delegation in this state.Footnote ³⁶

While the lack of county-level returns for delegates means that we can only perform this analysis in three states, these calculations demonstrate the importance of the biased system of representation in manufacturing the observed pro-secession majorities in at least two conventions. Furthermore, the analysis for LA shows how the separate influences of the convention on turnout and representation each contributed to the observed secessionist majorities in this state.