www.elsevier.com/locate/eeh

Available online at www.sciencedirect.com

ScienceDirectExplorations in Economic History 50 (2013) 620–633

Did the New Deal solidify the 1932 Democratic realignment?

Shawn Kantora,b,⁎, Price V. Fishbackb,c, John Joseph Wallisb,d

a Rensselaer Polytechnic Institute, United Statesb NBER, United States

c University of Arizona, United Statesd University of Maryland, United States

Received 28 October 2012Available online 27 August 2013

Abstract

The critical election of 1932 represented a turning point in the future electoral successes of the Democrats and Republicans forover three decades. This paper seeks to measure the importance of the New Deal in facilitating the Democrats' control of the federalgovernment well into the 1960s. We test whether long-differences in the county-level electoral support for Democratic presidentialcandidates after the 1930s can be attributed to New Deal interventions into local economies. We also investigate more narrowlywhether voters rewarded Roosevelt from 1932 to 1936 and from 1936 to 1940 for his efforts to stimulate depressed localeconomies. Our instrumental variable estimates indicate that increasing a county's per capita New Deal relief and public worksspending from nothing to the sample mean ($145) would have increased the long-run support for the Democratic party by 2 to 2.5percentage points. We further find that the long-run shift toward the Democratic party after 1928 was not a function of the Rooseveltlandslide victory in 1932. Roosevelt's ability to win over voters during the 1936 and 1940 elections with New Deal spending,however, did matter for the long-term.© 2013 Elsevier Inc. All rights reserved.

Jel classifications: D72; E65; N42Keywords: New Deal; Great Depression; Political economy; Voting; 1932 realignment

When I took office, the first thing we had to do wasmount an aggressive response to the worst economiccrisis we'd seen since the Great Depression, becausewe didn't want a second Great Depression… And bythe way, one of those steps was called the RecoveryAct. And I want everybody to understand here's whatit did… Now, sometimes you've got people who werecritics of what we did, but they'll show up at the ribboncutting. So I just want to make clear here what we did,

⁎ Corresponding author at: Rensselaer Polytechnic Institute, United States.E-mail address: kantos@rpi.edu (S. Kantor).

0014-4983/$ - see front matter © 2013 Elsevier Inc. All rights reserved.http://dx.doi.org/10.1016/j.eeh.2013.08.001

because people try to score political points by attackingthe Recovery Act, that's what they're attacking… relieffor laid-off workers, investment in your community—2.5 million Americans went to work today whootherwise wouldn't have gone to work.[–President Barack Obama, “Remarks by the Presidentat Ottumwa, Iowa Town Hall,” April 27, 2010 (http://

www.whitehouse.gov/the-press-office/remarks-president-ottumwa-iowa-town-hall).]

621S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

1. Introduction

The election of 1932was a turning point in the fortunesof the Democrat and Republican parties. Democrats hadwon just three of the previous ten presidential electionsand held a majority in both the House and the Senate inonly four of the previous 20 sessions of Congress.Democrats won seven of the ten presidential electionsfrom 1932 to 1968 and majorities in both the House andthe Senate in all but two of the 19Congressional elections.The NewDeal was not foreshadowed in Roosevelt's 1932campaign, and could not be the cause of the dramaticDemocratic victory that year: Hoover had won with 58%of the popular vote in 1928, while Roosevelt won with58% of the vote in 1932. Between 1932 and 1936,however, the NewDeal became the program and platformof the Democratic party. Roosevelt expanded his popularvote share to 62% in the 1936 election, and won 55% ofthe popular vote in 1940. There was nothing inevitableabout the Democrats becoming the majority party forthe next fifty years. They could have squandered theopportunity handed to them by the nation's deepestdepression, but they did not. Was there a difference in thevoters who shifted away from Hoover between 1928 and1932, and the voters who shifted toward Rooseveltbetween 1932 and 1936? Were voters who signaled theirapproval of the New Deal by shifting to the Democratsafter 1932 more likely to stay Democrats for the long-term? How much did the New Deal contribute to thepersistence of the Democratic party majorities in themid-20th century?1

The question can be given a finer point. From July1933 to June 1941 (from fiscal year 1934 to fiscal year1941), the federal government spent over $27 billion

1 In this issue Robert Fleck (2013-in this issue) also addresses theimpact of New Deal spending as well as other factors on presidentialvoting behavior in the 1930s. There are several differences in theapproaches that the two papers take. Fleck focuses on the short-runimpact in 1936 and 1940 and goes into more depth on the impact ofurbanization, farming, long-run swing voting, and the internationaldimensions of policy. We focus on the impact of spending both duringthe shorter-run of the 1930s and the longer-run impact on presidentialvoting into the 1952, 1956, and 1960 elections. Fleck focuses on totalspending, while we split the spending into two categories: AAA farmspending and public works and relief. We both use instrumentalvariable strategies: Fleck uses the amount of land in the state as hisinstrument, while we use county-specific measures of average farmsize and the presence of major rivers. The use of county-levelinstruments allows us to control for time-invariant features of eachstate in the two-stage least squares regressions in a way that Fleckcannot because he is using a state-level variable as an instrument.Generally, where there is overlap in the discussion, both papers findsimilar results.

(nominal dollars) on programs cooperatively adminis-tered by state and local governments aimed at relievingthe unemployed, building the nation's infrastructure,supporting farmers' incomes, and other projects. Totalfederal government expenditures in fiscal year 1929amounted to $3.2 billion, while New Deal spendingon cooperatively administered programs averaged$4.9 billion in 1929 dollars. We estimate the impactof that spending on electoral outcomes both in the 1930sand on long-run changes in electoral support into the1960s. The analysis helps us understand the role that anexpanding federal government may have played in theDemocrats becoming the majority party in the middle ofthe 20th century.

The Roosevelt administration funded a variety ofdifferent programs in an attempt to revive economicactivity and ameliorate the Depression's effect. The PublicWorks Administration (PWA) handed out grants to buildcivil infrastructure, while the Federal Emergency ReliefAdministration (FERA), the Civil Works Administration(CWA), and the Works Progress Administration (WPA)granted state and local governments funds to provideworkrelief and direct relief and to build and maintaininfrastructure. The New Deal launched programs thatpaid farmers to alter their land usage. New Deal agenciesloaned funds to state and local governments, banks,homeowners, farmers, and to industry in order to provideneeded liquidity. The Federal Housing Administration(FHA) sought to prop up the housing sector by insuringhome improvement and mortgage loans. We haveconstructed measures of annual federal spending inroughly 3000 counties from 1933 to 1939. We focus ontwo sets of grant programs that directly impacted localcitizens' economic well-being: relief and public worksspending and Agricultural Adjustment Administration(AAA) spending. Relief and public works spending wasused to provide employment by hiring people to buildpublic works projects or to perform public maintenanceactivities. In contrast, the AAA grants paid farmers to takeland out of production in an attempt to curtail supply sothat farm prices and farmers' incomes would rise. TheAAA potentially had perverse economic and politicaleffects because when farmers were encouraged to reduceacreage, the demand for labor fell, and farm workers weredisadvantaged. Fishback et al. (2005, 2006) have shownthat greater AAA spending in a county caused retail salesto decline and encouraged greater out-migration.

The empirical strategy has two steps. We begin byinvestigating whether voters rewarded Roosevelt from1932 to 1936 and from 1936 to 1940 for his efforts tostimulate depressed local economies. Our workinghypothesis is that local voters should have rewarded

5 See Fishback et al. (2005, Table 1) for a detailed statisticaldescription of the distribution of retail sales performance at the stateand county level.6 Our measure of New Deal spending does not encompass all federal

spending, so our analysis does not address the impact of all forms offederal expenditures. It should be noted, however, that much of theNew Deal represented an entirely new role for the federal government.For example, agricultural spending, relief spending, many forms oflending to state and local governments, and insurance of mortgageloans broke new ground for the federal government. In addition, there

622 S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

the president's efforts to help their economies emergefrom the Great Depression, all else constant. Thepredicted impact of AAA spending is more uncertainthan the effect of relief spending, however, as AAAspending benefited some segments of the farm popula-tion while others were disadvantaged. We find thatrelief spending did increase Roosevelt's vote share inthe elections of 1936 and 1940, but the effects of AAAspending were mixed.

The second step extends the analysis into the 1950sand examines the effects of the New Deal on voting in the1952, 1956, and 1960 elections. The long-run success ofthe Democratic party after 1928 was not a function of theRoosevelt landslide victory in 1932. The change in theDemocratic share of the vote between 1928 and 1932 hasa much smaller effect on the long-run shift toward theDemocrats than the increase in the Democratic voteshares in the 1936 and 1940 elections. The shift in theDemocratic share of the presidential vote between 1932and 1936 and between 1936 and 1940 is statistically andsignificantly related to the long-run success of theDemocrats.2 In the elections of 1936 and 1940, votersresponded to the benefits of the New Deal and politicalgains among those voters persisted into the 1960s. Ourdetailed empirical analysis of the 1936 and 1940 electionsshows the importance of NewDeal spending on relief andpublic works to Roosevelt's electoral successes, espe-cially in 1936.

We are aware that New Deal spending and contempo-raneous voting behavior were endogenously determined.3

We adopt an instrumental variable approach describedlater in the paper to isolate the causal role of New Dealspending on Roosevelt's electoral outcomes.

2. Geographic variation in the Great Depression andthe New Deal response

The 1930s were a decade of lost output for the U.S.economy. By 1933 both real per capita GDP and percapita retail sales had fallen to approximately two-thirdsof their 1929 peaks. In per capita terms, real retail salesreturned to its pre-Depression level by 1939, while realGDP did not return to its 1929 level until 1940.4 Modern

2 Interpreting these results requires more nuance, which we providelater in the paper.3 See Levitt and Snyder (1997) for a discussion of the endogeneity

bias in early studies to identify the political impact of governmentspending.4 See series T81 deflated by series E135 and series F4 in U.S.

Bureau of the Census (1975, pp. 210–11, 224, and 843). See alsoRomer (1992).

indices of local economic activity, such as unemploymentrates and personal income, are unavailable at the countylevel for the 1930s. As an alternative, retail sales, whichwere reported for every county in the U.S. in 1929, 1933,1935, and 1939, is a credible measure of local economicactivity. The ratio of 1939 retail sales to 1929 retail salesat the state level ranged from a low of 77% in Mississippito a high of nearly 125% in South Carolina. The NewEngland states appear to have had the most success inrecovering to their pre-Depression levels as every statehad higher real retail sales in 1939 than in 1929. In manystates there was substantially more within state variationthan there was across the states.5

The crisis of the Great Depression led the Rooseveltadministration to distribute unprecedented amounts offederal money in the form of outright grants. Thefederal government distributed $16.5 billion in grantsfrom 1933 to 1939. The New Deal grants increased thefederal government's outlays as a share of GDP fromroughly 4 to 8%. Furthermore, the federal governmentbegan spending large amounts of money where it hadspent very little before, setting the stage for a long-termstructural shift in the financial responsibilities of thenational, state, and local governments.6 The federalshare of non-military government expenditures at alllevels rose from 30% in 1932 to 46% by 1940 (Wallis,1984, pp. 141–42).

In 1940 the U.S. Office of Government Reports (OGR)produced county-level statistics on federal spending onover 30 NewDeal programs for the periodMarch 3, 1933,through July 30, 1939. We divide the New Deal grantsinto public works and relief grants; and Agricultural

were major increases in federal spending from the early 1930s onroads, public buildings, public works, and conservation. The NewDeal programs increased federal intergovernmental and direct outlayson education from $26 million in 1932 to $235 million in 1934, onhighways from $217 million to $599 million, on public welfare andemployment security from $2 million to $585 million, on housing andurban renewal from 0 in 1932 to $3 million in 1934 to $71 million in1936. Spending on the primary tasks of the federal government priorto the 1930s generally did not display the same marked jumps. SeeWallis (1985) and U.S. Bureau of Census (1975, pp., 1124–26).

623S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

Adjustment Administration benefits paid to farmers.7

Public works and relief grants had broadly similar goals ofproviding employment on public works projects and otherpublic services. Relief grants were primarily distributedunder the aegis of the Federal Emergency ReliefAdministration (FERA), the Civil Works Administration(CWA), Works Progress Administration (WPA), theSocial Security Administration categorical programs –Aid to the Blind, Aid to Dependent Children, and Old-ageAssistance – and the Farm Security Administration's“rural rehabilitation” programs. The principal goal of theseprograms was to provide immediate relief to theunemployed and low-income people. Public relief jobsranged from make-work activities to maintenance activ-ities to the building of sidewalks, post offices, schools,local roads, and other additions to local infrastructure. Thepublic works grants included expenditures by the PublicWorks Administration (PWA), Public Buildings Admin-istration (PBA), and the Public Roads Administration(PRA). These projects also employed workers, but theprograms focused less on hiring people from the reliefrolls and employed a broader class of skilled workers. Thepublic works programs also focused more on buildinglarge-scale projects like dams, roads, schools, sanitationfacilities, and other forms of civil infrastructure.8

New Deal aid to the farm sector came through theAAA's payments to farmers to remove land fromproduction. The impact of the AAA grants on localeconomic outcomes was likely smaller than the impact ofthe relief grants and potentially even negative. On onehand, farm ownersmight have received higher net incomesfrom the AAA program. Payments typically exceeded theincome farmers would have earned on the land that theytook out of production because the least productive landwas removed first. If the AAA succeeded in raising farmprices, the farmers also would have earned more on the

7 The U.S. Office of Government Reports, “County Reports,” alsoprovided information on $10.4 billion in repayable loans under avariety of programs as well as data on the Federal HousingAdministration's insurance of $2.7 billion in mortgage loans. We donot focus attention on these programs for several reasons. First, thenature of the loans and insurance were substantially different from thenonrepayable grants and it is hard to determine the true dollar size ofthe benefits that the counties received. Second, in the analysis we seekto reduce problems of endogeneity by using instrumental variables.We face difficulties in finding enough effective instruments that willallow us to simultaneously identify more than two or three New Dealvariables in a system of equations. Third, by omitting the loans andFHA insurance we reduce measurement error at the risk of increasingomitted variable bias in our estimates of the impact of New Dealgrants. We do not believe that the bias will be large from omitting theloans and FHA insured loans because these variables are largelyorthogonal to the grant variables.8 See Clarke (1996, pp. 62–68) and Schlesinger (1958, pp. 263–96).

crops they produced. In addition, the higher prices and thelimits on land usage would have encouraged farmers toraise yields on the land they kept under cultivation. On theother hand, the AAA had an adverse effect on the incomesof farm laborers, tenants, and sharecroppers. There isevidence that sharecroppers and tenants did not receivetheir full share of the AAA payments on the lands that theyhad cultivated and that some were demoted to wagelaborers.9 As land was removed from production, thedemand for farm labor likely fell, leading to declines inlaborers' incomes.10 Previous research has shown that thenet effect of greater AAA spending in a county led todiminished retail sales and out-migration.11

There was substantial variation in the extent of percapita New Deal spending across the country for the twobroad categories. Spending on relief and public workswas over $125 per person in the heavily urbanized statesin the Northeast and Midwest and well over $200 perperson in many western states. Meanwhile, relief andpublic works expenditures were below $80 per person inmany southern states. AAA expenditures were highest inagricultural regions, particularly the West North Centralregion and the Mountain West. The South receivedsubstantially higher amounts per capita than did theNortheast, but much less than the amounts spent in theWest and the West North Central.

3. Estimating the impact of New Deal spending onelectoral outcomes

There are two steps to our estimation of the politicaleffects of New Deal spending. First, we focus on theeffect of New Deal spending on the presidential vote inthe 1930s, particularly comparing the elections of 1932,1936, and 1940. Because of the probable endogeneity ofvoting and New Deal expenditures we develop a pair ofinstrumental variables. The use of instrumental variableshas a significant impact on the estimated effects,particularly for relief and public works spending. Thesecond step looks at the effect of New Deal spending andpresidential voting patterns in the 1930s on what we callthe “long-difference”: the difference between the averageDemocratic vote share for President in the 1952, 1956,and 1960 elections and the average Democratic voteshare between 1896 and 1928.

Did the New Deal have a short-term impact onRoosevelt's electoral outcomes at the county level in

9 See Depew et al. (2013-in this issue), Biles (1994, pp. 39–43),Whatley (1983), and Saloutos (1982).10 See Alston (1981).11 See Fishback et al. (2005, 2006).

13 A number of scholars have used natural resource endowments orphysical characteristics as instruments in cross-sectional analyses inpart because these factors were established long before the decisionsunder consideration in the research were made (see, e.g., Frankel andRomer, 1999; Hoxby, 2000).

624 S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

both 1936 and 1940? To answer the question weestimate the following equations:

Vi t ¼ φ1NDPRi t−4 to t þ φ2NDAi t−4 to tþ φ3Vi;1932 þ φ4Ei t−4 to t þ φ5Xi;1930 þ φ6Sþ ηi ð1Þ;

NDPRi t−4 to t ¼ λ1INSTi þ λ2Vi;1932 þ λ3Ei t−4 to tþ λ4Xi;1930 þ λ5Sþ κi ð2Þ;

NDAi t−4 to t ¼ τ1INSTi þ τ2Vi;1932 þ τ3Ei t−4 to tþ τ4Xi;1930 þ τ5Sþ μ i ð3Þ

where Vi t is the percent of the vote for Roosevelt in eachcounty i in year t (either 1936 or 1940), NDPRi t − 4 to t isreal per capita New Deal public works and relief grantsduring the four year period leading up to the election(specifically, March 1933 through December 1936 andJanuary 1937 to December 1939) and NDAi t − 4 to t is realper capita NewDeal AAA grants over the same periods.12

Vi is a measure of prior voting behavior in county i,specifically Roosevelt's vote share in 1932, before theNew Deal activity began. Including the 1932 voteeffectively makes our estimation equation a quasi-firstdifference estimation. Ei t − 4 to t is a vector of variablesmeasuring economic factors at the time of the election orthat occurred since the last election that might haveinfluenced voters' decisions in county i, such as changesin retail spending (a proxy for changes in income),unusual bouts of wetness, drought, temperature, orwhether the county experienced Dust Bowl conditions.Xi, 1930 is a vector of structural correlates, measured in1930, that might have determined how various interestgroups aligned politically; S is a vector of state dummyvariables; and INSTi is the vector of instrumentalvariables discussed below. The error terms of theequations are assumed to be independent and identicallydistributed and uncorrelated with each other.

We are concerned about possible endogeneity andomitted variable biases, including a number of concerns

12 The Office of Government Reports (1939) reported countyinformation for the period March 1933 to June 1939. The Office alsoreported state level information for each fiscal year for the publicworks, relief, and agricultural grants. We split the county-levelspending into annual fiscal year estimates by using that fiscal year'sshare of the state totals for the period March 1933 to June 1939. Wethen created calendar year estimates by taking the average of the fiscalyears. For example, the calendar year estimate from January 1 toDecember 31, 1936 is one half of the estimate for the fiscal year fromJuly 1, 1935, to June 30, 1936, plus one half of the estimate for thefiscal year from July 1, 1936, to June 30, 1937. For the calendar year1939 the estimate is equivalent to the fiscal year estimate from July 1,1938, to June 30, 1939.

with politics. The simplest is that the New Dealadministrators may have responded to voting patternswhen they made their decisions about allocating NewDeal spending across the states. Another is that they mayhave sent more money to counties with higher unem-ployment and lower income. We are not able to captureall of the economic forces in our control variables, and weknow that administrators took economic conditions intoaccount when they allocated spending, so there is likely tobe an omitted variable bias in anOLS estimation. If votersrespond to economic conditions when they vote, higherNew Deal spending will be associated with lower, nothigher, Democratic voting shares.

We constructed two instruments to address theseproblems. One is a geographic measure.13 The presenceof a major river in a county likely influenced public worksand relief spending because the potential for flooding andthe requirements for dredging and docks and other publicservices along the river provided local officials withready-made projects that they could propose to New Dealadministrators.14 More major rivers in a county meantmore public works opportunities. In the case of agricul-ture, rivers were likely to influence the types of cropschosen and, hence, the pattern of AAA spending. Not allrivers are created equal, however, and we focus on majorrivers. The measure that proved most powerful in thiscontext was a single variable measuring access to riversthat ran through 21 or more counties.15

Could rivers have influenced voting decisions? Riverscertainly influence the location of cities, farmingdecisions, and economic activity, which, in turn, mayhave influenced how people voted. However, many of theavenues by which the presence of rivers would haveinfluenced such voting – economic activity, urbanization,farm structure, state fixed effects, and homeownership –

14 As one example, Hoxby (2000) used the number of streams in anarea as an instrument for the number of school districts on the groundsthat they were natural boundaries that would have influenced thelocation of school district boundaries. As seen in the text, we usemajor rivers for different reasons.15 We experimented with several different ways of classifying majorrivers. Adding the separate variables did not change the second-stageresults appreciably, but reduced the power of the first-stage instruments,so we chose to use only the 21-or-more-county measure. We created andexperimented with several different major river categories: rivers in acounty that passed through 5 to 10 total counties, rivers that passedthrough 11 to 20 counties, 20 to 50 counties, and more than 50 counties.For 2287 counties the value of the major river variable (i.e., number ofrivers in the county that pass through 21 or more total counties) is zero.

625S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

are controlled in the second-stage vote equation. Unlikethe year to year economic fluctuations which we are notconfident of measuring, we are confident of our measuresof socio-economic characteristics. Thus, for the rivervariable to be an unsuitable instrument, it would have tohave an additional influence on the voting equation errorterm above and beyond these other control factors.

The second instrumental variable is average farm sizein a county in 1929. In an earlier paper (Fishback et al.,2003) we found that the elasticity of per capita AAAspendingwas very sensitive to farm size in 1929. Voting inthe 1930s and later could not have influenced average farmsize in 1929. There is a possibility that average farm sizebelongs in the voting equation or that it might be correlatedwith unobservables in the equation. Farm size could haveinfluenced the course of agricultural development duringthe 1930s and, thus, could have influenced how peoplevoted later. Again, however, these effects would not havebeen transitory, but reflected in the socio-economiccharacteristics that we have been able to control for: retailsales per capita, farm ownership, the share of the county'sland devoted to agriculture, and a dummy variablemeasuring whether the county experienced the DustBowl during the 1930s (see Hansen and Libecap, 2004).16

State dummy variables are included to captureunmeasured factors that were common to the countiesin the states but varied across the states: state policiesand changes in state policies over time, differences inthe cost of living across the states, policies related tostate taxation and spending, or differences in statepolitical organization, strategy, or effort. We cluster thestandard errors at the state level to account for the factthat presidential politics centers on winning majoritieswithin a specific state, not county, so political strategyacross counties within a state were likely correlated.

The second step of our analysis estimates the effectof New Deal spending and New Deal voting onlong-term electoral outcomes by regressing the long-difference between the share of votes for Democraticpresidential candidates averaged over 1952, 1956, and1960 ( Vi;1952‐1960 ) and the average share of votesfor Democratic candidates between 1896 and 1928(Vi;1896‐1928), on a vector of New Deal grant spending inthe 1930s (NDPRi and NDAi), changes in Roosevelt's

16 We further explored the validity of our instruments by regressingthe Democratic share in the 1928 election on our set of socioeconomicvariables measured in 1930 and the instruments. Neither the river noraverage farm size variable was statistically significant in thisexperiment, which lends credence to the idea that the instrumentsthemselves are unrelated to voting outcomes except perhaps throughtheir influence on New Deal spending. We thank an anonymousreferee for this suggestion (results available upon request).

support during the New Deal (ΔVi, t − (t − 4)), a vector ofstructural socioeconomic variables (Xi, 1930), and statefixed effects (S). The analysis is conducted at the countylevel (subscript i). Specifically, we estimate the followingequation:

Vi;1952‐1960–Vi;1896‐1928 ¼ β1ΔVi;t− t−4ð Þþ β2NDPRi þ β3NDAiþ β4Xi;1930 þ β5 Sþ εi ð4Þ

where εi is the error term.We chose 1960 as the terminal year because political

scientists debate whether 1964 or 1968 were criticalelections themselves. ΔVi, t − (t − 4) is a vector of threevariablesmeasuring the change in Roosevelt's share of thepopular vote in county i in election year t (1932, 1936, and1940) relative to the Democratic vote share in the priorelection. We include these variables to determine whetherand when votes for Roosevelt were instrumental inshifting long-term voter sentiment toward the Democraticparty. The shift in the vote between 1928 and 1932reflected Hoover's failure rather than Roosevelt's policies,while the shifts in votes between 1932 and 1936, and 1936and 1940, reflect two different regimes under Roosevelt. IfRoosevelt's policy really mattered for long-run Demo-cratic success, we would expect to see the 1936 and 1940vote shifts play an important role, even after controllingfor New Deal spending that we measure directly.

The results from the first-stage regression are shown inTable 1. The table shows the estimates for relief and AAAspending separately for the period 1933 to 1936 and 1937to 1940, as they will be used separately in the estimates ofthe 1936 and 1940 votes. The specific t-tests for theindividual instruments in the first-stage show that thepresence of large rivers led to statistically significantlyhigher New Deal relief and public works spending asexpected, but had no impact on AAA spending.Meanwhile, average farm size had a statistically signifi-cant impact on AAA spending but no statisticallysignificant impact on relief and public works spending.Thus, the instruments are separately identifying the impactof the two types of spending in the way we anticipated.The river variable is the key instrument that is being usedto identify the impact of public works and relief and thefarm size variable is the primary instrument identifying theimpact of the AAA grants. The F-tests for the coefficientsof the excluded instruments for each equation are shownin the bottom of Table 2 for the 1936 regressions and atthe bottom of Table 4 for the 1940 regressions. All of theF-statistics reject the hypothesis that the coefficients of thetwo instruments are simultaneously equal to zero in thetwo first-stage regressions at the 1% level. Stock and

Table 1First-stage coefficients from 1936 & 1940 2SLS regressions, column (6).

Corresponding to Table 2 (1936), col. (6) Corresponding to Table 3 (1940), col. (6)

Relief & public works(1933–1936)

AAA(1933–1936)

Relief & public works(1937–1939)

AAA(1937–1939)

Roosevelt's share of the vote in 1932 0.7321(1.87)

0.7118(3.24)

0.4118(1.64)

0.2536(2.60)

Excluded instrumentsNumber of rivers in the county flowing through21 or more counties

20.61(3.37)

6.029(1.34)

17.25(3.10)

0.3744(0.16)

Average farm size in 1929 −0.0034(−0.04)

1.465(6.84)

−0.1045(−1.57)

0.7225(3.29)

Contemporaneous economic variablesPercentage change in real per capita retail sales1933–1935 or 1935–1939

8.458(0.45)

7.568(0.71)

1.757(0.15)

−24.12(−2.74)

Real per capita retail sales, 1935 or 1939 −0.0013(−0.01)

−0.0348(−1.40)

−0.0165(−0.66)

−0.0293(−2.18)

Months of excess or severe drought,1933–1936 or 1937–1940

0.8039(1.41)

0.1221(0.14)

−0.0483(−0.12)

0.0410(0.12)

Months of excess or severe wetness,1933–1936 or 1937–1940

8.308(1.42)

1.640(1.11)

4.177(1.24)

−0.1949(−0.41)

Average monthly temperature, 1933–1936 or1937–1940 relative to 1920s average

−1194.0(−1.37)

722.5(1.34)

−284.0(−0.54)

479.8(2.21)

“Dustbowl” County 54.43(2.91)

308.1(3.15)

33.31(2.27)

240.9(7.37)

Structural socioeconomic variablesPct. of population black, 1930 −0.1495

(−0.45)0.2876(1.01)

−0.3681(−1.75)

0.3448(3.02)

Pct. of population living in urban area, 1930 −0.3725(−0.88)

−0.9043(−3.64)

0.5139(2.54)

−0.2856(−3.18)

Pct. of population manufacturing workers, 1929 −1.227(−0.74)

−2.183(−3.14)

−0.8653(−1.71)

−0.7005(−2.66)

Pct. of population foreign born, 1930 0.5869(0.36)

1.090(1.15)

1.205(1.00)

0.8046(2.09)

Pct. of population illiterate, 1930 −2.573(−2.27)

0.8334(1.27)

−1.897(−2.87)

0.4364(0.77)

Pct. of population belonging to religiousorganizations, 1926

−0.7563(−1.34)

−0.0257(−0.23)

−0.2471(−1.48)

−0.2019(−2.44)

Pct. of county's land in farms, 1929 −1.132(−6.59)

−0.0453(−0.20)

−0.6783(−3.48)

0.0293(0.26)

Percentage of farm operated by tenants, 1929 −0.6319(−1.71)

0.0520(0.16)

−0.7655(−2.83)

−0.3798(−1.12)

Percentage of households owning homes, 1930 −0.2929(−0.45)

0.0253(0.06)

−0.1602(−0.49)

−0.1936(−0.81)

Percentage of households owning radios, 1930 0.1186(0.17)

1.962(3.06)

−0.3531(−0.68)

1.388(2.89)

Tax returns filed per capita 1184.0(1.24)

−71.21(−0.19)

−412.9(−1.71)

−730.4(−3.00)

Constant 1407.0(1.64)

−820.9(−1.48)

474.4(0.92)

−496.5(−2.23)

State fixed effects Yes Yes Yes YesR2 0.310 0.748 0.396 0.668Adjusted R2 0.294 0.742 0.382 0.660N 2899 2899 2899 2899

Notes & sources: See Tables 2, 3, and 4 for sources and summary statistics of the included instruments variables. t-Statistics are in parentheses. Themean (standard deviation) of the river instrument is 0.226 (0.456) and 69.81 (69.11) for the average farm size variable. See data appendix inFishback et al. (2005) for sources.

626 S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

Table 2OLS and 2SLS estimates of Roosevelt's percentage of the vote, 1936.

Variables Mean(std. dev.)

(1) OLS (2) 2SLS (3) OLS (4) 2SLS (5) OLS (6) 2SLS

Dependent variableRoosevelt's share of the

vote in 193666.22 (18.12)

Independent variablesPer capita AAA spending,

1933–1936100.9 (171.7) 0.0016 (1.06) 0.0040 (1.41) 0.0061 (6.98) 0.0072 (4.41) 0.0062 (6.40) 0.0076 (2.98)

Per capita public works and reliefspending, 1933–1936

145.1 (169.4) 0.0015 (1.02) 0.0661 (2.41) 0.0010 (0.93) 0.0282 (2.06) 0.0010 (0.74) 0.0369 (2.16)

Roosevelt's share of the votein 1932

67.89 (18.21) 0.8656 (20.8) 0.8530 (17.9) 0.8177 (22.1) 0.7939 (20.0) 0.8160 (22.3) 0.7874 (18.4)

Constant 7.082 (2.51) 4.414 (0.94) 13.42 (3.39) 3.236 (0.67) −9.719 (−0.37) −63.7 (−1.45)Contemporaneous economic

variablesNo No No No Yes Yes

Structural socioeconomic variables No No Yes Yes Yes YesState fixed effects Yes Yes Yes Yes Yes YesF-statistic: excluded instruments in

AAA equation (p-value)18.8 (0.000) 14.2 (0.000) 25.2 (0.000)

F-statistic: excluded instruments inrelief & public works equation(p-value)

6.92 (0.002) 6.74 (0.003) 5.69 (0.006)

Kleibergen–Paap F-statistic 5.87 6.88 5.07R2 0.89 0.62 0.92 0.88 0.93 0.85Adjusted R2 0.89 0.92 0.93N 2901 2901 2901 2901 2899 2899

Notes & sources: The instrumental variables that are used are the number of rivers in the county flowing through 21 or more counties and averagefarm size in 1929 (see Table 1). t-Statistics are in parentheses. Standard errors are clustered at the state level. See Table 4 for the coefficients of thecontemporary and structural socioeconomic variables from columns (5) and (6). See Table 1 for sources.

627S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

Yogo (2002) constructed critical values for tests of weakinstrument strength when there are multiple instrumentsand Kleibergen and Paap (2006) developed the F-statisticfor the test when using robust standard errors. For 1936,the Kleibergen–Paap F-statistic ranges from 5.07 to 6.88.The Stock–Yogo critical values for weak instrument testsare 7.03 for 10% weak instrument bias and 4.58 for 15%.Thus, we can reject the hypothesis of weak instrumentbias if we are willing to accept weak instrument bias of upto 15%. Similarly, for the 1940 regression specifications2, 3 and 4, the Kleibergen–Paap F-statistic ranges from4.88 to 6.45 and we can again reject the hypothesis ofweak instrument bias if we are willing to accept up to 15%weak instrument bias.

4. Effects of New Deal spending on voting in 1936and 1940

There are a number of reasons why New Dealspending might not have been directly correlated withDemocratic voting in the 1930s. First, the South wassolidly Democratic and had been since the late 19thcentury. There was little that Roosevelt and New Deal

administrators could gain from making large grants tothe South and incrementally increasing the Democraticvoting share; they were going to win those states in anycase. In addition, southern Democrats often opposedincreasing spending for relief and public works projectsin the South that might have made black citizens moreindependent of white political patronage (Alston andFerrie, 1999). As a result, per capita New Deal spendingin the South was lower than any other region of thecountry despite the fact that the Democratic share of thevote was higher than in any region of the country (see,e.g., Arrington, 1970; Reading, 1973; Wright, 1974).

Our county level data and state fixed effects controlfor the pure Southern effect, but similar effects couldexist within other states. Democrats had a strategicreason to channel spending toward swing states andcounties where a small shift in voters' decisions couldmove a county, state, or region from the Republican toDemocratic column. This strategic spending dovetailed,in part, with New Deal efforts to relieve the unem-ployed who were concentrated in urban industrial areas.The heavily industrialized belt of states running westfrom New York to Illinois had been the states where

17 This statement is supported further by regressions in which weestimated the impact of New Deal spending after 1936 on voting in1940, while holding constant Roosevelt's vote share in 1936. Thoseregressions, which are available from the authors, show very smalleffects of the New Deal spending.

628 S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

Democrats and Republicans closely contested electionsin the late 19th century and early 20th century. Theseswing states were more likely to receive large relief andpublic works grants.

Table 2 shows the results from OLS and 2SLSestimations of the effect of New Deal spending on theDemocratic share of the presidential vote in 1936,including and excluding various economic and socioeco-nomic controls and both OLS and 2SLS results. All of theequations are run as quasi-first differences in the sensethat Roosevelt's 1932 vote share is included as a controlvariable so the coefficients represent the marginal impactof the New Deal variables on the 1936 vote relative to the1932 outcome in each respective county. All of theequations include state fixed effects and standard errorsare clustered at the state level.

In Table 2 the coefficients of the New Deal spendingvariables are always positive. As we expected, theimpact of New Deal spending on electoral outcomes inthe 1930s is much stronger when the IV analysis isperformed. In the estimates, the 2SLS coefficients arealways larger than the OLS coefficients, for both theAAA and the public works and relief variables. TheOLS and 2SLS coefficients on AAA spending are notmuch different, but the coefficients on the relief andpublic works spending are 30 to 40 times bigger in theIV than in the OLS. If we focus on the 2SLS estimateswith all correlates (column (6)), increasing relief andpublic works spending in a county from zero to thesample mean of $145 would have increased Roosevelt'sshare of the vote by 5.4 percentage points relative to his1932 share. The $145 increase is a reasonablecounterfactual since relief and public works spendingin 1929 was almost zero (there were some highwaygrants) and a one standard deviation increase in reliefand public works grants would have been $169. Anincrease in AAA spending from zero to the sample mean($101) would have increased the Roosevelt vote shareby a modest 0.8 percentage point. The smaller AAAeffect is consistent with the conflicting effects of theAAA for farm owners versus farm workers and tenants.

It appears that the long-run effects of New Dealspending on presidential voting were concentrated in theyears from 1933 to 1936. Table 3 presents estimates forthe 1940 election. The first two columns present OLS and2SLS estimates of Roosevelt's vote share in 1940 usingNew Deal spending from 1937 to 1939. The only othercontrols are Roosevelt's vote share in 1932, whicheffectively makes the estimation a quasi-first-difference,and state fixed effects. We control for the 1932 electionreturns because, as we showed above, the New Deal hadan impact on the 1936 election results so including 1936

would potentially dampen the true effect of the NewDeal.Columns (3) and (4) add structural socioeconomicvariables to the equation using the 1937 to 1939 spendingdata, and columns (5) and (6) add contemporaneouseconomic variables to them. Finally, columns (7) and (8),respectively, show OLS and 2SLS estimates using theNew Deal spending patterns over the entire New Dealperiod, 1933 to 1939.

The central difference between New Deal spendingbefore and after 1936 is the degree to which stategovernments determined the level of New Deal grantallocations to counties within the states. Before 1936,most relief funds were allocated by the Federal Emer-gency Relief Administration (FERA). FERA fundslegally became the property of the states when theywere granted and the federal relief administration haddifficulty controlling the administration and allocation ofrelief funds within the states. In 1936 and after, the largestrelief program was the Works Progress Administration(WPA). The federal government had much more directcontrol over the allocation of WPA funds than FERAfunds. As Wallis et al. (2006) show, the allocation ofWPA funds was significantly more responsive toeconomic factors like unemployment and lower incomesthan FERA allocations, and significantly less likely to beinfluenced by political factors. In light of the difference inpre- and post-1936 methods of allocating funds, perhapsit is not surprising that the effect of NewDeal spending onvoting behavior would be different in the two periods.

As was the case in the 1936 voting model, the 2SLScoefficient of the relief and public works variable is largerthan the OLS estimate. The coefficient on the 1937 to1939 spending on the 1940 vote is smaller than in the1936 analysis and is not statistically significant in 2SLSspecifications (4) and (6) when additional correlates areadded. If a county were to receive an average amount ofNew Deal relief and public works spending in the 1937 to1939 period ($132 per capita), then Roosevelt's voteshare in 1940 would have increased by 2.6 percentagepoints (132 × 0.0196) relative to 1932. The AAAcoefficients for the 1940 voting in Table 3 are evensmaller than the modest effects found for the 1936 votesin Table 2. Thus, the main effect of the New Deal hadbecome manifest by 1936 and New Deal spending after1936 seems to have had only marginal effects onRoosevelt's electoral support after 1936.17

Table 3OLS and 2SLS estimates of Roosevelt's percentage of the vote, 1940.

Mean(std. dev.)

(1) OLS (2) 2SLS (3) OLS (4) 2SLS (5) OLS (6) 2SLS (7) OLS (8) 2SLS

Dependent variableRoosevelt vote share, 1940 60.09 (20.46)

Independent variablesPer capita AAA spending, 1937–1939 54.82 (89.12) −0.0048

(−1.06)−0.0014(−0.327)

0.0022(1.42)

0.0005(0.13)

0.0040(2.44)

0.0016(0.24)

Per capita public works andrelief spending, 1937–1939

132.0 (105.6) 0.0079(2.14)

0.0446(1.40)

0.0048(2.08)

0.0144(0.78)

0.0056(2.67)

0.0196(0.92)

Per capita AAA spending, 1933–1939 155.7 (249.9) 0.0004(0.46)

0.0001(0.04)

Per capita public works andrelief spending, 1933–1939

277.1 (259.6) 0.0015(1.06)

0.0088(0.97)

Roosevelt's share of the vote in 1932 67.89 (18.21) 0.8106(15.2)

0.8109(15.2)

0.7901(15.8)

0.7868(14.5)

0.7896(15.7)

0.7848(15.3)

0.7914(15.7)

0.7841(15.1)

Constant 4.273(1.20)

8.226(1.51)

12.89(2.26)

10.45(1.59)

−46.26(−1.35)

−55.92(−1.73)

−49.84(−1.42)

−70.03(1.79)

Contemporaneous economic variables No No Yes Yes Yes YesStructural socioeconomic variables No No Yes Yes Yes Yes Yes YesState fixed effects Yes Yes Yes Yes Yes Yes Yes YesF-statistic: excluded instruments in AAA

equation (p-value)7.08(0.002)

8.22(0.001)

6.07(0.005)

20.2(0.000)

F-statistic: excluded instruments in relief& public works equation (p-value)

12.6(0.000)

6.98(0.002)

6.61(0.003)

7.94(0.001)

Kleibergen–Paap F-statistic 4.26 6.22 4.88 6.45R2 0.88 0.85 0.90 0.90 0.90 0.90 0.88 0.90Adjusted R2 0.88 0.90 0.90 0.88N 2901 2901 2901 2901 2899 2899 2899 2899

Notes & sources: The instrumental variables that are used are the number of rivers in the county flowing through 21 or more counties and averagefarm size in 1929 (see Table 1). t-Statistics are in parentheses. Standard errors are clustered at the state level. See Table 4 for the coefficients to thecontemporary and structural socioeconomic variables from columns (5) and (6). See Table 1 for sources.

629S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

Finally, Table 4 reports the estimated coefficients ofthe structural and contemporaneous economic variablesfrom specifications (5) and (6) for the 1936 regressionsin Table 2 and the 1940 regressions in Table 3. Theresults of the contemporaneous economic variables donot indicate that Roosevelt particularly benefited orsuffered as a result of economic fluctuations at the locallevel. One strong effect, though, is that counties withgreater per capita retail sales growth in the inter-electionperiod were more likely to side with Roosevelt in the1936 election, holding support in 1932 constant.Generally speaking, however, we find little evidencethat negative climate shocks in the form of the droughts,extreme wetness, or the horrors of the Dust Bowl weretied to Roosevelt.

The variables measuring the structural socioeconomiccharacteristics of the counties showed that Rooseveltpicked up support in 1936 and 1940, relative to 1932,from counties with larger black populations, more urbanresidents, moremanufacturing workers, and a larger shareof foreign-born residents (in 1936). Support was lower in

counties with greater religious participation (in 1940),greater homeownership, and in counties with a largerfraction of their land in farms. Fleck (2013-in this issue)goes into much greater depth on the impact of these typesof factors and the impact of prior swing voting on thevoting for Roosevelt in the 1930s.

5. Long-term results

To estimate the effect of the New Deal on thelong-term electoral support for the Democrats from thepre- to post-New Deal era, we calculated the averageDemocratic share of the presidential vote in electionsfrom 1952 to 1960 and subtracted the average Demo-cratic share of the vote in elections from 1896 to 1928.The regression estimates are presented in Table 5. TheOLS specification in column (1) includes only theNew Deal variables and state fixed effects; specification(2) only includes the marginal vote changes during the1930s and state fixed effects; specification (3) includesboth sets of variables; and, finally, specification (4) adds

Table 4Coefficients of contemporaneous economic and structural socioeconomic variables from 1936 and 1940 OLS and 2SLS regressions.

From Table 2 (1936) From Table 3 (1940)

Mean (std. dev.) (5) (6) Mean (std. dev.) (5) (6)

Contemporaneous economic variablesPercentage change in real percapita retail sales 1933–1935or 1935–1939

0.2477 (0.1804) 2.533 (3.26) 2.202 (2.32) 0.2044 (0.1812) 1.156 (1.01) 1.080 (0.97)

Real per capita retail sales, 1935or 1939

435.1 (228.7) 0.0044 (3.36) 0.0045 (1.25) 530.7 (273.0) −0.0007 (−0.36) −0.0004 (−0.18)

Months of excess or severedrought, 1933–1936 or1937–1940

11.87 (10.33) −0.0171 (−0.58) −0.0487 (−1.34) 6.979 (10.56) −0.0643 (−0.94) −0.0654 (−0.95)

Months of excess or severewetness, 1933–1936 or1937–1940

1.194 (2.508) 0.0225 (0.41) −0.2810 (−1.10) 1.353 (2.583) −0.1190 (−1.30) −0.1828 (−1.38)

Average monthly temperature,1933–1936 or 1937–1940relative to 1920s average

1.014 (0.0168) 20.61 (0.81) 62.40 (1.46) 1.008 (0.0140) 59.21 (1.71) 66.99 (1.94)

“Dustbowl” County 0.0203 (0.1348) −0.7844 (−0.82) −3.683 (−1.66) 0.0203 (0.1348) −2.200 (−2.28) −1.681 (−0.70)

Structural socioeconomic variablesPct. of population black, 1930 11.20 (18.44) 0.1209 (4.83) 0.1249 (4.72) 11.20 (18.44) 0.0820 (3.52) 0.0876 (3.47)Pct. of population living in urbanarea, 1930

20.47 (24.42) 0.0482 (5.97) 0.0633 (3.32) 20.47 (24.42) 0.0538 (3.49) 0.0435 (2.72)

Pct. of population manufacturingworkers, 1929

3.336 (3.389) 0.3353 (6.48) 0.3795 (5.21) 3.336 (3.389) 0.2104 (4.39) 0.2182 (3.89)

Pct. of population foreign born,1930

4.664 (5.805) 0.3349 (3.18) 0.3202 (2.73) 4.664 (5.805) 0.1560 (0.92) 0.1428 (0.93)

Pct. of population illiterate, 1930 5.190 (5.504) −0.0301 (−0.51) 0.0623 (0.76) 5.190 (5.504) −0.0356 (−0.51) −0.0048 (−0.06)Pct. of population belonging toreligious organizations, 1926

45.70 (18.17) −0.0174 (−1.04) 0.0082 (0.30) 45.70 (18.17) −0.0374 (−1.67) −0.0353 (−1.63)

Pct. of county's land in farms,1929

64.39 (26.93) −0.0346 (−3.02) 0.0040 (0.71) 64.39 (26.93) −0.0615 (−4.19) −0.0504 (−2.29)

Percentage of farm operated bytenants, 1929

31.73 (15.94) −0.0136 (−0.77) 0.0113 (0.56) 31.73 (15.94) −0.0337 (−1.47) −0.0243 (−0.91)

Percentage of households owninghomes, 1930

51.86 (12.99) −0.0961 (−3.23) −0.0804 (−2.15) 51.86 (12.99) −0.1380 (−2.98) −0.1374 (−2.80)

Percentage of households owningradios, 1930

34.50 (17.65) 0.0144 (0.42) 0.0066 (0.17) 34.50 (17.65) 0.0518 (1.06) 0.0642 (1.25)

Tax returns filed per capita 0.0118 (0.0130) −33.88 (−1.56) −77.45 (−1.40) 0.0118 (0.0130) 25.66 (0.94) 28.80 (1.03)

Notes & sources: The coefficients were drawn from the full regression of voting returns in the respective years and included the variables reportedhere and in Tables 2 and 3. t-Statistics are in parentheses.

630 S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

structural socioeconomic variables describing counties in1930.

The results in column (2) provide the simplest test ofthe hypothesis we set out to ask in the paper: Did theNew Deal have an effect on the long-run shift to theDemocratic party? The key comparison is between thecoefficient on the change in the Democratic sharebetween the elections of 1928 and 1932, which ispositive but statistically insignificant, and the coeffi-cients on the change in the Democratic share betweenthe elections of 1932 and 1936, and between theelections of 1936 and 1940. The coefficients on the laterelections are positive, of much greater magnitude, and

statistically significant. The results in the fullestspecification (4) show that a 1 percentage point shiftin the share voting for Roosevelt between 1932 and1936 translated into a statistically significant 0.39percentage point increase in Democratic support overthe long-term that we analyze. Similarly, a 1 percentagepoint shift in the Democratic vote share from 1936 to1940 contributed a statistically significant 0.38 percent-age point increase in long-term Democratic voting. Thecoefficient on the change in the 1928 to 1932 vote shareis a statistically insignificant .08. In the simplest terms,a one percentage point shift toward the Democratsbetween the elections of 1932 and 1936 had four times

Table 5OLS estimates of long-difference change in Democratic support, 1896–1960.

Variables Mean (std. dev.) (1) (2) (3) (4)

Dependent variableLong-difference in Democratic vote a −5.234 (18.56)

Independent variablesPer capita AAA spending. 1933–1939 155.7 (249.9) −0.0026 (−1.64) −0.0015 (−0.68) 0.0003 (0.18)Per capita public works and relief spending,

1933–1939277.1 (259.6) 0.0026 (1.60) 0.0018 (1.60) 0.0012 (1.56)

ΔDemocratic vote share 1932–1928 23.93 (11.59) 0.0425 (0.48) 0.0466 (0.57) 0.0843 (1.28)ΔDemocratic vote share 1936–1932 −1.665 (7.647) 0.3517 (3.44) 0.3493 (3.52) 0.3875 (6.07)ΔDemocratic vote share 1940–1936 −6.117 (6.505) 0.4217 (2.95) 0.4064 (2.60) 0.3815 (3.08)Pct. of population black, 1930 11.19 (18.44) −0.2714 (−3.41)Pct. of population living in urban area, 1930 20.46 (24.42) 0.0204 (1.58)Pct. of population manufacturing workers, 1929 4.582 (5.034) 0.1329 (1.42)Pct. of population foreign born, 1930 4.663 (5.804) 0.8676 (7.43)Pct. of population illiterate, 1930 5.197 (5.520) 0.0651 (0.50)Pct. of population belonging to religious

organizations, 192645.68 (18.19) 0.0272 (1.12)

Pct. of county's land in farms, 1929 64.38 (26.95) −0.0153 (−0.91)Percentage of farm operated by tenants, 1929 31.25 (15.35) 0.1608 (3.20)Percentage of households owning homes, 1930 50.84 (12.22) −0.0915 (−1.44)Percentage of households owning radios, 1930 51.93 (18.38) −0.2952 (−3.58)Tax returns filed per capita 0.0171 (0.0166) −78.01 (−2.88)Constant −5.554 (−11.0) −3.086 (−1.88) −3.569 (−2.27) 9.019 (2.00)State fixed effects Yes Yes Yes YesR2 0.74 0.76 0.76 0.81Adjusted R2 0.74 0.76 0.76 0.80N 2901 2901 2901 2901 2901

Notes & sources: t-statistics are in parentheses. Standard errors are clustered at the state level. Presidential voting in 1952, 1956, and 1960 wascollected from ICPSR file number 8611.a The dependent variable is the long-difference between the average vote for the Democratic presidential candidate in 1952, 1956, and 1960, and

the average vote for the Democratic presidential candidates from 1896 to 1928, inclusive. See Eq. (4) in the text. The summary statistics reportedabove correspond to unweighted measures. Weighting by 1930 population results in a mean of 0.910 and a standard deviation of 18.24.

631S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

the effect on the Democratic share in the 1950s as didthe same magnitude shift between 1928 and 1932.

This conclusion requires the nuanced interpretation wementioned earlier. There was an enormous shift towardsthe Democrats between the 1928 and 1932 election.Hoover wonwith 58% of the vote in 1928, Roosevelt with58% in 1932 (remember that the means of the variables inthe table are the unweighted average of the change invotes at the county level). But of that 16 percentage pointshift, a one percentage point shift produced a statisticallyinsignificant .08 percentage point increase in thelong-term Democratic share. Roosevelt's share of the1936 votes was 62%, an increase of only 4 percentagepoints. But each percentage point increase in the 1936Democratic share produced a .39 percentage pointincrease in the long-term Democratic voting share.While the shift towards Democratic votes in 1932 wasvery large, the smaller shifts that occurred in 1936 and1940 were much more persistent from a statisticalperspective. The shift in votes toward the Democrats

after the NewDeal was in place seems to have had a moremeaningful long-term effect. It was Roosevelt's ability towin or maintain electoral support between 1932 and 1936and between 1936 and 1940 that helped to cement theDemocratic coalition.

The effect of New Deal spending on the long-termsuccess of the Democratic party worked through thechange in the share voting for Roosevelt 1932, 1936, and1940 elections. Using coefficients from specification (6) inTable 2, increasing the relief and public works spending ina county from zero to an amount equal to the sample meanof $145 would have increased Roosevelt's share of thevote by 5.4 percentage points in 1936 relative to his 1932landslide performance. After multiplying this change bythe coefficient of 0.39 for the change in the vote from 1932to 1936 from Table 5 specification (4), the indirect effectadds 2.1 percentage points to the support for Democraticpresidents in the 1950s. A similar procedure using thecoefficients from Table 3 suggests that the public worksand relief spending had an indirect effect through the

18 See Fishback et al. (2005, 2006, 2007, 2010), Fishback andKachanovskaya (2011), Barreca et al. (2012), and Depew et al. (2013-in this issue).

632 S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

change in the vote for Roosevelt from 1936 to 1940 andwould have added an additional 0.6 percentage point inlong-term support for Democrats. In all, New Dealspending may have raised the Democratic voting shareover the long-term by somewhere in the neighborhood of2 to 2.5 percentage points.

In terms of estimating whether New Deal spendingdirectly affected the long-difference in the Democraticvote, the specifications reported in columns (3) and (4) inTable 5 are over-controlled in the sense that the changingvotes share in the 1930 elections are included asexplanatory variables. It appears that the effect of NewDeal spending exerted an effect on the long-difference inthe Democratic voting shares through its effect on theelection results in 1936 and 1940. The direct effect of NewDeal spending is both small and statistically insignificantin this experiment.

The full specification in column (4) indicates the clearstructural voting changes taking place after the 1932election that also strengthened the Democratic majority.Counties with more manufacturing workers, the foreignborn, and a greater share of tenant farmers all shiftedsignificantly toward the Democrats after the 1932 criticalelection. In contrast, counties that were better off in 1930tended to shift towards the Republicans. Specifically,counties with a greater share of tax returns filed, greaterhomeownership, and greater radio ownership in 1930shifted away from the Democrats. The statistically strongnegative coefficient of the black population variableseems counter to conventional wisdom that black voterssupported Democratic presidential candidates. The key tounderstanding this result is that counties with large blackpopulations in 1930 were in the South, which votedoverwhelmingly Democratic in the early 1930s. As whitesouthern conservative voters began moving out of theDemocratic and into the Republican party, a shift that wasnot consummated until the 1980s, the Democratic shareof the southern vote fell. Black voters enfranchised by theVoting and Civil Rights Acts would not appear untilelections later in the 1960s.

6. Conclusions

The New Deal programs that most directly affectedthe lives of unemployed Americans during the GreatDepression were the emergency spending and publicworks programs, such as the FERA, CWA, WPA, PRA,PWA, and the farm programs. The New Deal distributedlarge sums of money to state and local governments toprovide employment and relief and to build a wide arrayof public works. The New Deal paid farmers to changetheir production patterns in an attempt to raise commodity

prices. These grants represented a substantial and novelchange in the federal government's role in the economy.Recent studies have shown that the public works andrelief programs had salutary effects in raising personalincomes and retail purchases, and contributed to areduction in crime and several types of death rates. TheAAA farm programs aided farm owners, but reducedeconomic opportunities for share tenants, sharecroppers,and unskilled farm workers.18

Was Roosevelt rewarded for his efforts to improveeconomic conditions? Yes, especially with regard tospending on public works and relief. Drawing togetherthe implications of our long- and short-term analyses ofvoter support for Democratic presidential candidatespaints the following picture. The election of 1932reflected an adverse judgment on Herbert Hoover andproduced a weakly persistent shift in voting patternstowards the Democrats. In contrast, the shift inDemocratic votes between the 1932 and 1936 and the1936 and 1940 elections helped to secure a permanentchange in the electorate that lasted into the 1960s. NewDeal policies could not have affected the 1932 vote, butour results show that the change in voter behaviorbetween 1932 and 1936 was a direct response to theNew Deal. The results suggest that an average increasein NewDeal relief and public works spending resulted ina 5.4 percentage point increase in the 1936 Democraticvoting share, and a smaller amount in 1940. Theestimated persistence of this shift suggests that NewDeal spending increased long-term Democratic supportby 2 to 2.5 percentage points. Thus, it appears thatRoosevelt's early, decisive actions created long-lastingpositive benefits for the Democratic party. The effects ofAAA spending were minimal because support from thefarmers who benefitted was partially offset by opposi-tion from those who saw their opportunities decline.Finally, on net, between 1936 and 1940 additionalspending on public works and relief and the AAA addedlittle to the support Roosevelt had already gained by1936.

The New Deal did contribute to the ascendancy of theDemocratic party after the 1930s. We find significanteffects of New Deal spending for relief and public worksprograms in the shift towards the Democratic party after1932. Democratic gains made in the elections after 1932had larger and more persistent effects than the gains madein the election of 1932. The New Deal did play an

633S. Kantor et al. / Explorations in Economic History 50 (2013) 620–633

important role in consolidating Democratic gains for atleast two decades.

Acknowledgments

The authors have benefitted from helpful comments andsuggestions from Jamie Carson, William Collins (editor),Robert Fleck, Gavin Wright, and an anonymous referee,and from conference participants at the 2010 UC MercedConference on the Politics of Federal Spending and the2012 NBER Conferences on the Microeconomics of theNewDeal. Fishback's and Kantor's work on the NewDealhas been supported by the National Science FoundationGrants SBR-9708098, SES-0080324, SES-0214395,SES-0617942, SES-0617972, and SES-0921732. We aresolely responsible for the views expressed in the article.

References

Alston, L.J., 1981. Tenure choice in southern agriculture. Explorationsin Economic History 18 (3), 211–232.

Alston, L.J., Ferrie, J., 1999. Southern Paternalism and the AmericanWelfare State. Cambridge University Press, New York.

Arrington, L.J., 1970. Western agriculture and the New Deal.Agricultural History 44 (4), 337–353.

Barreca, A.I., Fishback, P.V., Kantor, S., 2012. Agricultural policy,migration, and malaria in the United States in the 1930s.Explorations in Economic History 49 (4), 381–398.

Biles, R., 1994. The South and the New Deal. University of KentuckyPress, Lexington, KY.

Clarke, J.N., 1996. Roosevelt's Warrior: Harold L. Ickes and the NewDeal. Johns Hopkins University Press, Baltimore.

Depew, B., Fishback, P.V., Rhode, P., 2013. New deal or no deal inthe Cotton South: The effect of the AAA on the agricultural laborstructure. Explorations in Economic History 50 (4), 469–489(in this issue).

Fishback, P.V., Kachanovskaya, V., 2011. In search of the multiplier forfederal spending in the states during the Great Depression. NationalBureau of Economic Research Working Paper No. 16561.

Fishback, P.V., Kantor, S., Wallis, J.J., 2003. Can the New Deal's threeR's be rehabilitated? A program-by-program, county-by-countyanalysis. Explorations in Economic History 40 (3), 278–307.

Fishback, P.V., Horrace, W.C., Kantor, S., 2005. Did New Deal grantprograms stimulate local economies? A study of federal grants andretail sales during the Great Depression. The Journal of EconomicHistory 65 (1), 36–71.

Fishback, P.V., Horrace, W.C., Kantor, S., 2006. The impact of NewDeal expenditures on mobility during the Great Depression.Explorations in Economic History 43 (2), 179–222.

Fishback, P.V., Haines, M.R., Kantor, S., 2007. Births, deaths, andNew Deal relief during the Great Depression. The Review ofEconomics and Statistics 89 (1), 1–14.

Fishback, P.V., Johnson, R.S., Kantor, S., 2010. Striking at the rootsof crime: the impact of social welfare spending on crime duringthe Great Depression. Journal of Law and Economics 53 (4),715–740.

Fleck, R., 2013. Why did the electorate swing between parties duringthe Great Depression. Explorations in Economic History 50 (4),602–622 (in this issue).

Frankel, J.A., Romer, D., 1999. Does trade cause growth? AmericanEconomic Review 89 (3), 379–399.

Hansen, Z.K., Libecap, G.D., 2004. Small farms, externalities, and theDust Bowl of the 1930s. Journal of Political Economy 112 (3),665–694.

Hoxby, C.M., 2000. Does competition among public schools benefitstudents and taxpayers? American Economic Review 90 (5),1209–1238.

Kleibergen, F., Paap, R., 2006. Generalized reduced rank tests using thesingular value decomposition. Journal of Econometrics 133 (1),97–126.

Levitt, S.D., Snyder Jr., J.M., 1997. The impact of federal spending onHouse election outcomes. Journal of Political Economy 105 (1),30–53.

Reading, D.C., 1973. New Deal activity and the states, 1933 to 1939.Journal of Economic History 33 (4), 792–810.

Romer, C.D., 1992. What ended the Great Depression? The Journal ofEconomic History 52 (4), 757–784.

Saloutos, T., 1982. The American Farmer and the New Deal. IowaState University Press, Ames, IA.

Schlesinger Jr., A., 1958. The Age of Roosevelt: The Coming of theNew Deal. Houghton-Mifflin, Boston.

Stock, J., Yogo, M., 2002. Testing for weak instruments in linear IVregression. National Bureau of Economic Research TechnicalWorking Paper No. 284. National Bureau of Research, Cambridge,Massachusetts (October).

U.S. Census Bureau, 1975. Historical Statistics of the United States:From Colonial Times to 1970. Government Printing Office,Washington, DC.

Wallis, J.J., 1984. The birth of the old federalism: financing the NewDeal, 1932–1940. Journal of Economic History 44 (1), 139–159.

Wallis, J.J., 1985. Why 1933? The origins and timing of nationalgovernment growth, 1933–1940. Research in Economic History 4,1–51.

Wallis, J.J., Fishback, P.V., Kantor, S., 2006. Politics, relief, andreform: Roosevelt's efforts to control corruption and manipulationduring the New Deal. In: Glaeser, E., Goldin, C. (Eds.),Corruption and Reform. University of Chicago Press, Chicago.

Whatley, W.C., 1983. Labor for the picking: the New Deal in theSouth. Journal of Economic History 43 (4), 905–929.

Wright, G., 1974. The political economy of New Deal spending: aneconometric analysis. The Review of Economics and Statistics 56(1), 30–38.