Patronage Politics and the Development of the Welfare State: Confederate Pensions in the American South

Abstract

Beginning in the 1880s, southern states introduced pensions for Confederate veterans and widows. They expanded these programs through the 1920s, while states outside the region were introducing cash transfer programs for workers, poor mothers, and the elderly. Using pension application records and county-level electoral data, we argue that political considerations guided the distribution of these pensions. We show that Confederate pension programs were funded during years in which Democratic gubernatorial candidates were threatened at the ballot box. Moreover, we show that pensions were disbursed to counties in which these candidates had lost ground to candidates from alternative parties.

During the late nineteenth and early twentieth centuries, U.S. states introduced cash transfer programs for the needy. Such programs included Mothers' Pensions for poor mothers, Workers' Compensation for employees who became disabled on the job, and Old Age Pensions for the elderly. By the early 1920s, most states had passed aid legislation and begun administering cash transfers. The notable exception was the South, which overwhelmingly failed to administer aid to mothers, workers, or the elderly, even in rare cases when aid programs were approved by state legislatures. ¹ Instead, southern states widely enacted and funded Confederate pension programs, which were modest income support programs for Confederate veterans and widows. These programs were typically enacted during the late 1880s and early 1890s, but southern state legislatures continued to expand them through the 1920s. Using newly compiled data on pension legislation, expenditures, and applications, we explore the political factors that motivated southern states to introduce, fund, and disburse Confederate pensions. As such, we offer new information about the development of the welfare state in the South.

A major contribution of this article is to provide the first cross-state accounting of the passage and distribution of Confederate pensions. More broadly, this study offers new insight into the economic history of income redistribution in the region. The leading explanation for southern states' reluctance to adopt the types of welfare programs typical to other parts of the country is that public policy in the South was dictated by the white, rural elite. Lee Alston and Joseph Ferrie (1993, 1999) write that paternalism—in which workers trade loyalty to their employer for nonmarket goods like old-age assistance without an explicit contract—reduced turnover costs and increased work effort in cotton agriculture. They argue that rural elites in the South actively blocked the adoption of welfare programs because these programs would have disrupted the paternalistic employment contracts they had with their primarily black workers. Moreover, southern politicians did not need to cater to low income voters by enacting income redistribution programs because the Democratic party had effectively suppressed all political opposition, through measures taken to disenfranchise poor and black voters (Kousser 1974; Margo 1990; Naidu 2012).

In contrast, state legislatures outside the South were highly responsive to political pressure to enact and fund redistributive programs. In Theda Skocpol's (1992) seminal account of early American welfare programs, she proposes a “polity-centered” view, arguing that politicians introduced social programs when well-organized and vocal groups of citizens demanded them. The Union Army pension developed into a large social insurance program because Union Army veterans were a clearly identifiable and politically mobilized group. In a similar vein, Skocpol argues that the most successful Progressive Era programs were those aimed at helping poor women rather than non-veteran men. Mothers' Pensions—which were state welfare programs for women—passed broadly while other forms of social insurance for working men did not. ² She argues that this occurred because of organized pressure from women's clubs, which, unlike men's organizations, were united across class boundaries. Skocpol also describes the widespread use of income redistribution programs for political patronage, especially during the late nineteenth century: key bureaucratic posts in the Union Army pension system, such as Commissioner of the Pension Bureau, went to Republican party loyalists. Similarly, Shari Eli (2015) shows that Union Army pensions were used explicitly to shore up support for Republican congressmen, showing that they were redirected to contested congressional districts. ³

We argue that the existence of Confederate pensions demonstrates that the post-Reconstruction South had more in common with the rest of the United States, in terms of social provision, than the existing literature suggests. Much like the federal Union Army pension, which provided aid to veterans living mainly in the North and Midwest, Confederate pensions appear to have been used to buy votes for governing state parties; this means that southern state politicians responded to electoral pressure with “carrots” as well as “sticks,” which is a feature of the southern political landscape that has received little mention in the literature. In support of this argument, we show that southern states increased funding for these pensions during years when Democratic gubernatorial candidates were threatened, and they distributed pensions to politically expedient counties. At the same time, the fact that southern states chose to redistribute income exclusively to Confederate veterans and widows, and not to the poor and infirm more generally, highlights the differences between the North and South. First and most importantly, offering social provision through Confederate pensions precluded southern blacks from receiving transfers. J. Morgan Kousser (1980) argues that the exclusion of blacks was a necessary precondition for political support from southern whites. ⁴ This is consistent with the Alston-Ferrie hypothesis, as most agricultural laborers in the region at the time were black. Kousser's argument is also consistent with Skocpol's view, as Skocpol argues that the ability to advocate for redistribution depends on the unification of potential beneficiaries. If race was an impassable social divide in the South, Skocpol's argument would explain why white southerners could unite behind Confederate pensions but not other forms of welfare.

We compile new data on Confederate pension legislation, spending, and individual pension applications, which we combine with county-level census data and data on gubernatorial election returns. We use these data to measure the effect of recent voting behavior, which we define as the share of the gubernatorial vote going to Republican or third party candidates, on the fraction of state expenditures allocated to Confederate pensions as well as the geographic distribution of pension funds within states. While other studies have discussed Confederate pensions in individual states, ours is the first large, cross-state analysis of these programs. ⁵ Moreover, it is the first study to use a large sample of individual pension applications to explore the geography of Confederate pension applications within states.

We use the number of Confederate pension applications filed at the county-year level as our primary measure of the allocation of pension funds within states. Because of the nature of the application records we are using, we view the number of applications filed from a county in a given year as a noisy measure of the number of applications from that county in that year that were approved, by either state or county authorities. As such, an increase in applications from a given county likely reflects the Democratic state government's desire to patronize that county. We take steps to rule out alternative explanations, such as a correlation between voting behavior and county-level economic conditions.

We find that the fraction of state expenditures on Confederate pensions significantly increased after third party or Republican gubernatorial candidates saw an increase in vote share. Within states, the number of applications from a county increased after that county experienced an increase in votes to third party or Republican candidates. Importantly, this finding is robust to controlling for county-level economic indicators, which suggests that voting patterns had a causal effect on pension outcomes. Finally, we show that, in Texas, pensions were less likely to be rejected in counties in which third party candidates had done well in the most recent election. Taken together, we feel this is strong evidence that southern Democrats implemented and augmented Confederate pensions to maintain power.

HISTORICAL BACKGROUND

Civil War Military Pensions

During the Civil War, the federal government passed the General Law of 1862, which allowed Union Army veterans and their dependents to apply for pensions if their illnesses or injuries were shown to be the result of their war experience. With the passage of the Invalid Pensions Act in 1890, a Union Army veteran could receive a pension for any illness or injury that left him unable to undertake manual labor (see Glasson 1918). By 1900, 95 percent of Union Army veterans were collecting benefits of approximately $12 per month, an amount that is roughly equivalent to 50 percent of a farmer's monthly earnings in the period. The Union Army pension had a profound effect on retirement behavior (Costa 1995, 1997), morbidity and mortality (Eli 2015; Salm 2011), and remarriage (Salisbury 2017).

Confederate veterans, however, were never allowed access to pensions from the federal government. Instead, southern states enacted their own pension systems. While the Democrats had largely regained control of state legislatures by 1876, they did not start passing pension legislation in earnest until the mid-1880s. Most existing work explains the emergence of Confederate pension laws by the fact that Confederate veterans and widows could not access Union Army pensions, taking for granted that southern states would step in to fill this gap. Other work (Ayers 1992) points to the elevated social position of Confederate veterans and widows in the South to explain why these states were willing to fund these programs.

Details of the passage of Confederate pension legislation are summarized in Table 1. States differed in terms of precise eligibility requirements; however, features common to all state programs are apparent. Pension programs typically included a means test, a residency restriction, and a remarriage prohibition for widows, although there is considerable cross-state variation in the nature of these restrictions. For example, the original pension law passed in Texas in 1899 required applicants to have been Texas residents since 1880, while North Carolina, Mississippi, and Virginia merely required applicants to be state residents at the time of application. Pension amounts differed substantially by state, ranging from a low of $15 per year in Georgia to a high of $300 per year in Tennessee. While many states initially required applicants to have been injured or widowed during the war, by the turn of the century most pensions functioned essentially as welfare for Confederate veterans and widows. Similar to the Union Army pension, southern pension programs had evolved to cover all veterans and widows in need.

Table 1 TIMELINE OF CONFEDERATE PENSION LEGISLATION

Refers to a pension law that does not require recipient to have been wounded or killed during the war.

Date of first annuity offered to Confederate veterans and widows. The Alabama legislature offered one time payments to wounded soldiers and widows as early as 1881 (with welfare-type eligibility requirements in 1891).

Amounts as of 1901. All amounts are in current dollars at the time the pension was enacted, unless otherwise stated.

From text of law amended in 1889.

First cash transfer program. A program for providing veterans with artificial limbs existed from 1877.

First cash transfer program. Several programs providing artificial limbs or land grants to veterans existed from the 1880s.

Amounts from code of 1906.

Sources:

See text and Online Appendix.

Although Confederate pensions were substantially less generous than the Union Army pension, expenditure on these programs comprised a significant fraction of state budgets (Gorman 1999; Short 2006; Ratchford and Heise 1938). Figures 1 and 2 report the number of pension applications filed in each year (where available), as well as the percentage of state expenditures allocated to pensions, separately by state. These figures indicate that these programs were widely taken up, with thousands of new applications filed from each state in most years. Spending on these programs typically peaked at 10 to 20 percent of the budget during the first two decades of the twentieth century.

Sources: See text.

Figure 1 CONFEDERATE PENSIONS BY STATE: SPENDING AND APPLICATIONS

Sources: See text.

Figure 2 CONFEDERATE PENSIONS BY STATE: SPENDING AND APPLICATIONS

Note: This figure plots the share of the gubernatorial vote from the most recent election in Texas against the fraction of pension applications filed in each year in Texas that were rejected. Sources: See text.

Figure 3 VOTING PATTERNS AND REJECTION PROBABILITIES IN TEXAS

The administration of Confederate pension programs was fairly uniform across states. Claims were evaluated first at the county level, by designated county pension boards. After being reviewed locally, claims were submitted to a state pension board, which reviewed them a second time and rendered a final judgment on the merit of each case. After approval by the state board, the state treasurer would issue a warrant for each claim on the treasury. So, while initial adjudication of pension claims was done at the county level, pensions were paid out of a central pension fund. Pension legislation typically introduced a new property tax to fund pension programs.

Consider, as an example, the pension law in Alabama. From 1899 onward, claims were assessed by a county board of examiners, appointed by the governor, and consisting of one “practicing physician of good standing in his profession” and one Confederate veteran “of good moral character” (Alabama 1907, S1998). ⁶ The county board of examiners heard applications at an office in the county seat during the first two weeks of July (ibid. S2003). During these two weeks, county boards held “office hours” (ibid. S2005), and they were obliged to “give due notice” of these office hours “by publication in some newspaper in the county or by posting at the courthouse door of the county and five other public places for three weeks” in advance (ibid. S2004). During these office hours, county boards would “subject [applicants] to an oral and physical examination” (ibid. S2010), and render a decision about the merit of each claim. Then, the county board would submit all claims to the state pension board for additional review (ibid. SS2011–2013). The state board of examiners consisted of a physician and two ex-Confederate soldiers appointed by the governor (ibid. S2000), and they met to review claims in Montgomery beginning on the second Monday in August every year (ibid. S2006). Application materials for approved claims were retained by the state auditor, which he would use to create a detailed record of pensioners (ibid. S2018); however, “all applications rejected by the board shall be returned to the county board of examiners, who shall file them with the judge of probate of the county, to be kept for future reference” (ibid. S2019). The pension was funded by “a special tax of one mill on each dollar of the taxable property of the state” (ibid. S2031).

The practice of locally adjudicating applications for pensions that were paid out of a central fund created an obvious incentive for county pension boards to approve claims of dubious merit. In principle, a claim should have been rejected if the applicant failed to meet the eligibility criteria; for instance, if the applicant owned too much property, failed the residency test, failed to prove that he served in the Confederate army, or failed to demonstrate physical disability. However, county boards routinely approved illegitimate claims, a practice of which state authorities were keenly aware. In his 1897 report to the governor, the state auditor of Alabama writes that the pension system “is being sadly abused,” noting that “applications are allowed in a great many instances that should be rejected,” and that “the Boards of Examiners, in some counties, do not appreciate their trust” (White 1897, p. 23). The same auditor writes again in 1898 that the pension law “is so open to abuse that those entrusted with its enforcement hardly know how to try to exclude those not entitled to benefits,” and he recommends that “the examination of applicants be taken from political officers” (White 1898, pp xxiv–xxv). The 1910 Alabama auditor's report similarly expresses concern that “the roll as it stands in my opinion contains the names of many who are not legally entitled to a pension” (Brandon 1910, p. 5). Similarly, the 1897 Mississippi auditor's report notes that “there are many on the [pension] rolls that are unworthy,” and that “it should be the aim of all having any duties to perform in this respect to see that these are rejected and thus to insure a larger prorata to the deserving” (Holder 1897, p. vii). Peter D. Blanck and Michael Millender (2000) also discuss the arbitrary power that elected county judges wielded in distributing pensions in Virginia.

Post-Civil War Politics in the South

By the mid-1870s, the Democratic party had largely regained control of southern politics. Voting in the South during the later nineteenth century cut primarily along racial lines, with white voters supporting Democrats and black voters supporting Republicans (Ayers 1992). The Democratic party was dominated by rural elites from the Black Belt—the portion of the South in which plantation style agriculture was common. Alston and Ferrie (1993, 1999) argue that Democratic congressmen viewed these elites as their core constituency, and acted explicitly in the best interests of this group. V. O. Key (1949) points to the role of factions within the Democratic party in providing an outlet for political competition and enabling the region to be dominated by a single political party.

Legislative efforts to disenfranchise black voters in the South during this period, including poll taxes and literacy tests, are well documented (Kousser 1974; Naidu 2012; Margo 1990). These policies began to be enacted in the region during the 1890s but did not become ubiquitous until around 1910. Prior to adopting these policies aimed at disenfranchisement, Democrats employed measures like gerrymandering, fraud, and voter intimidation to maintain power in the region, subject to some federal oversight up to the mid-1890s. While these measures significantly decreased turnout among black voters, southern elections were at least somewhat competitive before legal disenfranchisement (Naidu 2012). Importantly, Confederate veterans were often exempt from poll taxes. For example, the Georgia state legislature passed a law in 1882, “to relieve crippled and disabled Confederate soldiers from poll tax” (Georgia 1882, p. 120).

While the Democratic party claimed a majority of the southern white vote during this period, they were at risk of losing vote share to third party candidates. Populist agrarian movements were a significant presence in the South during the years following the Civil War. Beginning in Texas during the 1870s, the Farmer's Alliance had become an important political force in state legislatures by the late 1880s and early 1890s (Woodward 1951). This movement gained popularity among farmers in the face of falling agricultural prices and a perceived lack of power in their dealings with the banks and railroads (Ayers 1992; Wright 1986). This period was immediately followed by the spread of the boll weevil, which significantly reduced cotton yields, contributing to further distress among farmers (Lange, Olmstead, and Rhode 2009). As C. Vann Woodward (1951) notes, with the exception of Virginia, the majority of the populist vote came from poor, white, smallholding farmers. He quotes one characterization of this movement in Alabama as “an effort of the masses of the white to free themselves from the rule of the black-belt Democratic party of the old slave-owning type” (p. 247). Moreover, populist movements in the South during this period seemed to directly threaten Democratic votes. As Woodward explains, “the leading conservative paper of Texas described the Populists of that state as solid, native white stock ‘sober and earnest from first to last’ and estimated that 90 percent of them were ‘ex-democrats whose standing in the party was formerly as undisputed” (Woodward 1951, p. 247). Maintaining vote share among smaller white farmers at risk of voting for populist candidates would have been an important political objective for southern Democrats.

DESCRIPTION OF DATA

We combine state-level information on the passage of Confederate pension legislation, state balance sheet data, individual pension application records, and county-level census and election returns data. Data on the passage of Confederate pension laws by state legislatures is compiled from a variety of primary and secondary sources, and we obtain state balance sheet data from published annual reports to state treasurers. ⁷ This allows us to observe annual expenditures on pensions relative to total annual expenditures.

Individual pension data consists of indexes to Confederate pension applications, which are available online from southern state archives or genealogical websites such as Ancestry.com and Familysearch.org. ⁸ We have obtained these indexes for nine states: Alabama, Arkansas, Florida, Georgia, Kentucky, Louisiana, Texas, Tennessee, and Virginia. Mississippi, North Carolina, and South Carolina have existing records, but they are not indexed in machine-readable form. Information that can be gathered from these indexes varies by state. In general, the index will indicate the name of the applicant, the type of application (veteran or widow), and the county of application. In some cases, the indexes contain additional information, such as year of application, details of the soldier's military service, and the outcome of the pension application. The Online Appendix Table 1 lists the number of pension applications indexed, separately by type as well as decade. We have records of almost 300,000 pension applications in total. In general, these applications are equally divided between veterans and widows, and they are distributed across states in a way that is broadly commensurate with population. Much like spending on Confederate pensions, the number of new applications typically peaked during the first decade of the twentieth century.

We combine data on Confederate pension laws and applications with county-level census data from Michael R. Haines and Inter-university Consortium for Political and Social Research (ICPSR) (2010) and county-level data on gubernatorial election returns from ICPSR (1999). We use the county-level census data to determine characteristics of counties from which Confederate veterans and widows applied for pensions, that is, population, race composition, prevalence of farming, farm ownership structure, and crop value per acre. ⁹ We use the election returns data from 1876–1930 to determine the fraction of the gubernatorial vote that went to Democratic, Republican, and third party candidates. The “third party” designation is assigned to any party that is not classified as Democrat/Democrat equivalent or Republican/Republican equivalent. These were generally left leaning and appealed to lower income farmers and laborers. Specific parties that appear frequently include the Communists, the Greenbacks, the People's Party, the Populists, Progressives, the Prohibition Party, Socialists, and Socialist-Labor. We use gubernatorial elections as our baseline measure of voting behavior for two reasons. First, these data are available at the county level. Second, they pertain to state elections rather than federal elections, which is important because these pensions were state-level programs, so we should expect them to be responsive to state rather than federal voting patterns.

APPROACH TO MEASURING THE IMPACT OF VOTING PATTERNS ON PENSION OUTCOMES

The empirical work in this article focuses on the effect of voting patterns on Confederate pension expenditure and the distribution of pension funds within states. Because the Democrats were almost always in control of state legislatures, we view the Democratic party as the primary decision-maker. ¹⁰ Our hypothesis is that the Democrats used Confederate pensions to respond to political threats from other parties. Specifically, we expect them to have increased funding for pension programs when alternative parties gained popularity. Within states, we expect pension boards (controlled by Democrats) to have funneled pensions to counties in which alternative parties were gaining popularity. This may have occurred for two reasons: (1) Democrats may have been seeking to increase turnout among likely Democratic voters in the face of opposition; or (2) they may have used pensions to discourage Confederate veterans from supporting Republican or populist third party candidates. One confounding factor is that these other parties, particularly populist third parties, gained popularity during times of economic hardship. As such, it may be that Democrats enacted and funded pension programs when populists gained vote share because they were responding to economic hardship rather than a perceived political threat. We will address this concern directly, by showing that the estimated effect of voting patterns on pension outcomes is not sensitive to controlling for county economic indicators.

Expenditure on Pensions at the State Level

We are interested in establishing how voting patterns in gubernatorial elections affected spending on pensions at the state level. State legislatures frequently amended original pieces of pension legislation, expanding access or pension amounts; this continued into the 1920s. So, the simplest way of summarizing changes in state legislatures' behavior regarding pensions is to look at changes in the percentage of a state's budget that is spent on pensions. We construct a panel of southern states, beginning in the first year that a state introduces an initial piece of pension legislation and ending in 1922, the last year for which we have state budget data. We estimate the following:

(1)

Here, E_s,t is the fraction of state s's total expenditures in year t allocated to Confederate pensions, and O_s,t and R_s,t are measures of vote shares from the gubernatorial election prior to year t in state s going to third party (O) and Republican candidates (R), respectively. The parameter δ_t is a year fixed effect, and θ _s · t is a state-specific linear time trend. In addition to the baseline specification in which O_s,t and R_s,t enter the regression equation linearly, we allow them to affect pension expenditures quadratically. We also include lagged vote shares to capture the possibility that state legislatures took time to respond to election results.

Distribution of Pension Applications within States

Our second aim is to characterize the way voting patterns affected the distribution of pensions within states. We hypothesize that the Democrats funneled pension funds to counties in which their gubernatorial candidate was threatened. ¹¹ They may have approved more applications from these counties, processed claims from these counties more quickly, or actively solicited applications from these counties. Our data allow us to observe the number of applications filed from each county in each year. We use this as our primary measure of the distribution of pensions within states.

We believe that the number of applications filed from a county in a given year reflects effort on the part of state and local authorities to distribute pensions to that county. This is due to the nature of the pension application and review process. Claims were heard at fixed times during the year by county pension boards, who forwarded claims to the state pension board for further evaluation. In many states, pension applications were only forwarded to the state board if the county board approved them. For example, the Arkansas pension law indicates that application materials were forwarded to state boards “when said board is satisfied with the justness of the claim made by the applicant” (Arkansas 1891, Act XCI S3). Virginia, Mississippi, and North Carolina have similarly explicit provisions in their pension laws about which files were to be forwarded to state boards (Virginia 1919, Title 23; Hemingway 1917, Chapter 151; North Carolina 1889, Chapter 198). Even in states in which all applications were forwarded to the state board for secondary review, accepted and rejected pension claims were stored separately. In Alabama, for instance, rejected applications were returned to county boards, meaning that they were less likely to survive in state collections. ¹² Because collections of applications at state archives come from records of the state pension bureau, they can be presumed to consist primarily of records forwarded to or retained by state pension boards. As such, we view our count of applications at the county level as a noisy measure of the number of applications that were approved by either the county or state board.

Both state and county pension boards are likely to have approved claims for political reasons. State boards had scarce resources to allocate, and they may have prioritized claims from contested counties. County examiners had an incentive to approve as many claims as possible, even illegitimate claims; however, they faced oversight (and potential removal for incompetence) by state boards. As such, they faced a trade-off: approving dubious claims was beneficial, but it was not costless. A contested gubernatorial election may have “tipped the scales” in favor of approving more claims.

More formally, the number of applications from a given county in a given year depends on two factors. Local economic conditions—such as falling agricultural prices or output—would have affected the number of people who qualify for the pension, as these pensions were means tested. At the same time, political patronage would have affected the number of applications though the channels discussed previously. So, we can write the number of applications as follows:

The number of applications (N) is a function of current economic conditions (f(X)), political patronage (g(P)), and a random component (u). In principle, a correlation between N and gubernatorial voting patterns may work through a correlation between voting patterns and g(P) or voting patterns and f(X). Our hypothesis is that the former channel is important, and we will offer evidence to support this hypothesis.

We construct a panel of counties, beginning in the election year immediately prior to initial program passage, and ending in 1930. We use this to estimate the following:

(2)

Here, N_c,s,t is the average annual number of pension applications filed in county c of state s between election years t and t+2. ¹³ We construct this variable from our pension index data. The variable R_c,s,t is a function of Republican gubernatorial vote share in county c of state s in election year t; O_c,s,t is a function of third party vote share in the same election; φ_s,t is a state-year fixed effect; and χ _c is a county fixed effect. We include state-year fixed effects to control for any state-level factor, such as a change in the law expanding access to pensions, which might affect pension applications for all counties in that state. As such, we are identifying the effect of political factors on the within-state distribution of these applications.

Of course, this does not pin down the channel through which political variables affect the number of pension applications to come from a particular county. To argue that patronage rather than local economic conditions drives our results, we match our panel to county-level decennial census data from Haines and ICPSR (2010), which allows us to control for county-level economic indicators that may be jointly correlated with application rates and gubernatorial voting patterns. We linearly interpolate between decades to obtain county-level economic data that vary at the same frequency as our electoral data. County economic indicators include measures of crop output per acre, the ownership structure in agriculture, and agricultural land inequality. To use the prior notation, if the correlation between N and electoral variables is primarily working through a correlation between electoral variables and f(X), then controlling for salient local economic characteristics should meaningfully alter the estimated effect of electoral variables on the number of applications.

Determinants of Pension Outcomes: Texas

The only pension outcome variable that is available at the county level for multiple states is the number of pension applications filed; however, we have additional information about the outcome of pension applications from Texas. This is because we have merged two sources of data on pension applications from Texas: (1) an index available from the Texas State Archives, which contains a list of pension applications and indicates whether these applications were rejected or not; (2) an index to the collection of Texas pension files available from Ancestry.com, which contains pension application dates. The first index comes from a list of applications kept by Texas pension authorities, while the second index is a list of application records that were physically present in the archives at the time these records were digitized. We are able to locate records in the Ancestry.com index that match close to 90 percent of the records in the Texas archive index.

Having access to these two data sources allows us to do two important tests. First, we can examine directly our conjecture that our sample of pension applications is disproportionately composed of accepted pension applications. Recall that accepted and rejected pension applications were recorded in systematically different ways: in some states, rejected claims were never transmitted to the state pension board; in others, they were returned to the county upon rejection; and in others, they were retained by the state but filed in a separate collection. Our hypothesis that the number of pension applications from a given county is a noisy measure of the number of accepted applications from that county depends on accepted application records surviving in archival collections more frequently than rejected records.

We can test whether applications recorded as “rejected” in the Texas State Archive's index are less likely to be linked to the Ancestry.com index. In fact, we find this to be the case. Rejected pension records are 5 percentage points less likely to be located in the Ancestry.com pension index. This difference is highly statistically significant and robust to including county of application fixed effects. This supports our conjecture that pension application records are more likely to contain accepted files.

Our data from Texas also allow us to test whether political variables impacted the probability that a pension application is rejected. We estimate the following equation:

(3)

The variable J_i,c,t is an indicator equal to one if a claim by person i from county c who applied in year t was rejected; O_c,t and R_c,t are functions of third party and Republican vote share in the most recent gubernatorial election; δ _t is an application year fixed effect; and χ _c is a county fixed effect. We cluster standard errors at the county-year level.

EMPIRICAL FINDINGS ON THE IMPACT OF VOTING PATTERNS ON PENSION OUTCOMES

Expenditure on Pensions at the State Level

Table 2 contains estimates of Equation 1. Here, we examine the effect of state voting patterns on state expenditures on Confederate pension programs. Our results clearly indicate that the rate of spending on Confederate pensions increased following spikes in either third party or Republican vote shares. From column (1), a 10 percentage point increase in vote share to either third party or Republican candidates increased the fraction of the budget spent on pensions by almost 1 percentage point. Since this expenditure was approximately 8 percent on average, this translates into around a 12 percent increase in the fraction of state expenditures allocated to Confederate pensions, which is quite a large number. In column (2), we find no evidence that this effect is non-linear, and in column (3) we find some evidence that vote shares matter with a lag.

Table 2 EFFECT OF GUBERNATORIAL VOTE SHARES ON SPENDING ON CONFEDERATE PENSIONS AT THE STATE LEVEL

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Ordinary Least Squares (OLS) regressions of the fraction of the state budget allocated to Confederate pensions on the share of the total gubernatorial vote obtained by third party and Republican candidates in the most recent election. The unit of observation is the state and year, and the sample period is 1876–1922. The sample includes all southern states (Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, Texas, and Virginia). All regressions include state and year fixed effects and a state-specific linear time trend. Regressions are weighted by the fraction of counties with non-missing election returns data.

Sources:

See text.

Our state-level results suggest that state legislatures were responding to increases in the popularity of Republican or third party candidates when they funded Confederate pension programs. It is important to point out that, because we do not have annual data on other potentially important state-level characteristics, we are unable to pin down the precise mechanism driving these results. It may be that Democratic state legislatures felt threatened by the popularity of alternative candidates. However, it is also true that non-Democrats tended to gain popularity in times of economic distress, particularly on the farm. As mentioned earlier, these were typically populist parties. State legislatures may have been responding to this economic distress, or a desire among voters for redistributive policies, rather than a perceived political threat. As such, we view our state-level results as descriptive, and we attempt to deal more thoroughly with causality when we look at pension outcomes at the county level.

Distribution of Pension Applications within States

Table 3 contains estimates of Equation 2. Here, we are looking at the effect of voting patterns on the number of applications at the county-year level. Our county fixed effects effectively control for any permanent county characteristic that may jointly affect both application rates and voting behavior. As such, we are looking at whether a deviation in voting behavior from a county's norm induces a deviation in pension application rates from that county's norm. Moreover, because we are including state-year fixed effects, our results will not be driven by statewide trends in application rates (such as the introduction of an expansionary amendment to a pension law). Here, we are focusing on the effect of local voting behavior on the within-state allocation of pension resources.

Table 3 EFFECT OF GUBERNATORIAL VOTE SHARES ON THE DISTRIBUTION OF PENSION APPLICATIONS AT THE COUNTY LEVEL

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Panel regressions of the average annual number of pension applications filed in counties between consecutive election years on the share of the gubernatorial vote going to third party and Republican candidates in the most recent election. The unit of observation is the county-year, and the sample period is 1876–1930. States included in the sample are those with both application year and county data available: Alabama, Arkansas, Florida, Georgia, Kentucky, Texas, Virginia. States do not enter sample until they have passed an initial piece of pension legislation. All regressions contain county- and state-year fixed effects.

Sources:

See text.

Our results clearly indicate that pension application rates increased in counties that had experienced an increase in third party or Republican vote share in the most recent gubernatorial election. From column (1), a 10 percentage point increase in third party vote share relative to Democrats increases expected applications by approximately 0.6. Given that the mean number of applications to be filed in one county-year is less than 8, this represents a 7 percent increase, which is economically meaningful. ¹⁴ The coefficient on Republican vote share is much smaller and is not quite statistically significant, although it becomes significant at the 10 percent level in column (3) when we include lagged vote shares. In column (2), there is some evidence of non-linearity in the effect of third party vote share on applications.

In Table 4, we try to rule out shocks to county economic conditions as an omitted variable driving the relationship between voting patterns and pension application rates. ¹⁵ Specifically, we regress the average annual number of applications between years t and t+2 on vote shares in year t and economic indicators from year t. The economic indicators we use are the following: percent employed in agriculture, farm ownership rate, crop output per acre, and a Gini coefficient for farm land inequality (Nunn 2008). We also look at percent black, although we consider the mechanism through which this should affect application rates to be entirely consistent with our patronage story, since black voters were a likely political threat to southern Democrats prior to disenfranchisement. Here, we are primarily interested in whether or not controlling for county economic indicators alters the estimated effect of vote shares on application rates. If so, this suggests that the correlation between vote shares and application rates is driven by a joint correlation between these two variables and local economic conditions. If not, this would support our claim that political patronage drives this correlation. It is clear from Table 4 that controlling for county-level economic indicators does not meaningfully alter the estimated coefficients on vote shares. However, we do find that these county characteristics affect pension applications. For instance, the number of applications is larger in counties with a large black minority, and in predominantly agricultural counties with an intermediate level of farm ownership and land inequality. We explore interactions between county characteristics and voting patterns in a subsequent section.

Table 4 EFFECT OF GUBERNATORIAL VOTE SHARES ON THE DISTRIBUTION OF PENSION APPLICATIONS AT THE COUNTY LEVEL SENSITIVITY TO CONTROLLING FOR COUNTY CHARACTERISTICS

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Panel regressions of the average annual number of pension applications filed in counties between consecutive election years on the share of the gubernatorial vote going to third party and Republican candidates in the most recent election. The unit of observation is the county-year, and the sample period is 1876–1930. States included in the sample are those with both application year and county data available: Alabama, Arkansas, Florida, Georgia, Kentucky, Texas, Virginia. States do not enter sample until they have passed an initial piece of pension legislation. All regressions contain county and state-year fixed effects. County characteristics are obtained by linearly interpolating between county-level aggregates from published decadal censuses.

Sources:

County characteristics, Haines and ICPSR (2010); pension and electoral data, see text.

Pension Outcomes: Texas

Table 5 contains estimates of Equation 3, where we examine the effects of electoral variables on the probability of a pension claim being rejected in Texas. In columns (1)–(3), we include all years of Texas data, including 1899, which is the year in which the pension program was first introduced. In columns (4)–(6), we exclude applications filed in 1899. The idea is that there may be a learning curve associated with using a pension system for political patronage, so an effect may not have been present at the beginning of the program's life. Given the large number of applications filed during the first year the program was enacted, the outcome of these applications will have a large effect on our findings. In the first three columns, we find that a county's third party and Republican vote share typically have a negative effect on the probability of an application from that county being rejected; however, this effect is not quite significant in most specifications. In the last three columns, we omit applications from 1899, which causes the estimated effect of third party vote share to double in size and to become significant. Specifically, based on column (4), a 10 percentage point increase in third party vote share reduces the probability of having a claim rejected by 0.7 percentage points. Because the rejection probability was only 10 percent, this translates into a 7 percent reduction in the probability of rejection.

Table 5 EFFECT GUBERNATORIAL VOTE SHARES ON PROBABILITY OF PENSION REJECTION: TEXAS

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

OLS regressions of an indicator equal to one if a pension application is rejected on the share of the gubernatorial vote obtained by third party and Republican candidates in the county of application in the most recent election. The unit of observation is the pension application, and the sample period is 1899–1930. All regressions contain county fixed effects. Standard errors are clustered at the county-year level.

Sources:

See text.

In Figure 3, we plot the overall rejection probability in Texas against vote share to the Republican or third party candidate. While this is much more pronounced for third party candidates, these lines seem to inversely track one another. Especially notable is the large dip in rejection probability in 1912 that coincides with a large spike in third party vote share. There is also a smaller dip in rejection probability during the mid-1920s that coincides with a large spike in Republican vote share. It appears that Texas pension boards were more relaxed about awarding pensions in years in which Democratic gubernatorial candidates were threatened in that state.

INTERACTIONS BETWEEN POLITICAL PATRONAGE AND LOCAL CHARACTERISTICS

In this section, we explore the way in which electoral variables interact with other local characteristics to affect the disbursement of pension funds within states. These results are presented in Table 6. ¹⁶ In the top panel, we regress applications in a county on non-Democratic vote share, a county-level economic indicator, an interaction between the two, and a full set of county and state-year fixed effects. The economic indicators we look at are: (1) the fraction of the county engaged in agriculture; (2) the ownership rate among farmers; and (3) the value of crop output per acre. The positive interaction between non-Democratic vote share and percent in agriculture indicates that legislatures were more responsive to voting patterns in agricultural counties; the negative interaction between vote share and both the ownership rate and the value of crop output suggests that voting patterns mattered more in poorer counties. This gives insight into exactly which voters southern Democrats were trying to court with these programs.

Table 6 GUBERNATORIAL VOTE SHARES AND THE DISTRIBUTION OF PENSION APPLICATIONS INTERACTION EFFECTS

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Panel A contains panel regressions that replicate specifications in Table (4). We include non-Democratic vote share from the last gubernatorial election instead of including third party and Republican vote shares separately. We also include interactions between county characteristics and non-Democratic vote share. These characteristics are: mean value of farm output per acre; the fraction of farmers that are farm owners; and the fraction of households engaged in agriculture. These variables are all standardized. In panel B, we estimate a similar specification, including non-Democratic vote share, an indicator for the county having a black population greater than 20 percent, an interaction between this and non-Democratic vote share, and interactions between all three of the above variables and an indicator equal to one if the state had instituted a poll tax.

Sources:

Poll tax dates: Kousser (1974); county characteristics: Haines and ICPSR (2010); pension and electoral variables: see text.

In the bottom panel of Table 6, we look at the interaction between voting patterns, race, and voter disenfranchisement. We regress the number of applications in a county on non-Democratic vote share, an indicator equal to one if the county's population is more than 20 percent black, an interaction between non-Democrat vote share and this indicator, and interactions between each of these three variables and an indicator equal to one if the state had enacted a poll tax. As always, we include county and state-year fixed effects. ¹⁷ This exercise yields some interesting results. First, there is a large, negative interaction between the poll tax indicator and non-Democratic vote share. The interpretation is that voting behavior no longer influences a county's application rate if a poll tax has been passed in that state. Interestingly, though, the coefficient on the three-way interaction between non-Democratic vote share, the poll tax indicator, and the indicator for the county being more than 20 percent black is positive (though not quite significant). Therefore, the effect of voting patterns on application rates declined by less after a poll tax was enacted in counties with a large black population.

This sheds some interesting light on the way in which southern state governments used Confederate pensions. First, it seems that the use of Confederate pensions for patronage was a substitute for disenfranchisement. That is, the tendency to employ pensions in this way declined once the Democrats had secured a greater vote share by disenfranchising black and poor white voters. Interestingly, our results suggest that these two measures were equally, if not more, substitutable in predominantly white counties. One interpretation is that Confederate pensions were largely employed by the governing party to limit political competition among whites; repressive measures, such as voter intimidation, may have been considered sufficient to limit political competition from southern blacks.

SOUTHERN WOMEN'S GROUPS AND RELIEF FOR CONFEDERATE VETERANS

One of the focuses of Skocpol's (1992) work on Progressive Era welfare programs is the role of women's groups in advocating for aid. Women's organizations—like the National Congress of Mothers—were more united across class lines than men's organizations, which enabled them to successfully press for “maternalist” welfare programs. The glaring exception is the South, which, despite having active women's organizations, did not pass (or fund) Mothers' Pensions. Instead, influential southern women's groups were devoted to memorializing Confederate history and advocating for relief for poor veterans and widows.

Consider, for example, the United Daughters of the Confederacy (UDC), which was an organized umbrella for like-minded women's groups established in 1895 (Hyde 1959). The group's policy goals were explicit: in the minutes of the 1900 annual meeting, the Alabama division reports that, “the accepted work of the division is the care of needy Confederate veterans, the preservation of Jeff. Davis house … and the building of monuments to the Alabama dead on the four battlefields now converted into national parks” (United Daughters of the Confederacy 1901, p. 52). Membership in the UDC had exceeded 18,000 by 1900 (United Daughters of the Confederacy 1901).

The difference between the policy goals of northern and southern women's groups is instructive. One explanation is that, in the South, race formed a more significant part of a person's identity than gender, so white women's groups advocated for aid for whites rather than aid for women. Southern women reformers did not fight to win relief programs for blacks, instead putting forth spurious cultural and biological arguments for their outright exclusion. For example, they claimed that blacks were better suited for open country life than poor whites and therefore less in need of aid (Fox 2012, p. 114). Similarly, they claimed that blacks were not a “begging race” (p. 114). ¹⁸ Another explanation is the North-South difference in the foreign-born population: Cybelle Fox (2012) argues that women reformers of the North advocated for Mothers' Pensions and similar programs out of a desire to provide relief to new immigrants in order to better assimilate them (p. 110). Since new immigrants comprised a small share of the southern population, this was not a policy priority for women reformers of the South.

CONCLUSION

This article offers the first large, multi-state analysis of the introduction and dissemination of Confederate pensions in the southern United States. We show that these pensions were widely taken up and funded while states outside the region were passing and funding other types of welfare legislation. We show that the expansion of funding for Confederate pensions is significantly related to increases in the popularity of third party and Republican candidates. We also show that pensions were distributed to counties in which Democrats had lost ground to alternative parties, and we show that this cannot be explained by changing underlying economic conditions. We believe that these results indicate that southern Democrats used Confederate pensions to court the vote of Confederate veterans while withholding relief from southern blacks.

These findings offer a new perspective on the welfare state in the American South. The current literature posits that rural elites stopped southern legislatures from adopting welfare programs during the early twentieth century, and that they were successful because electoral competition had been effectively stifled. Our results suggest a more qualified version of this story, in which southern states did deploy income redistribution to curry favor with voters, albeit white voters only. Confederate pensions served the interests of a broad coalition of southern whites, including veterans, elites, and politicians: they provided relief to a cohort of Civil War veterans whose Union counterparts already had government support; they provided relief to whites and excluded blacks, which likely made the programs politically palatable in the region; and they allowed Democrats to garner votes in contested rural areas and solidify their political strength in the South into the mid-twentieth century. We believe this study offers new insight into the way in which voters and elites interact in order to shape public policy.