econ referee report 3

NBER WORKING PAPER SERIES

MUNICIPAL HOUSEKEEPING: THE IMPACT OF WOMEN'S SUFFRAGE ON PUBLIC EDUCATION

Celeste K. Carruthers Marianne H. Wanamaker

Working Paper 20864 http://www.nber.org/papers/w20864

NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue

Cambridge, MA 02138 January 2015

Ye Gu, Ahiteme Houndonugbo, Andrew Moore, James England, III, and Nicholas Busko provided outstanding research assistance. We thank Martha Bailey, Don Bruce, Larry Kenny, Matthew Murray, Michael Price, Melissa Thomasson, Nicholas Sanders, Christian Vossler, Robert Whaples, and anonymous referees for thoughtful comments and suggestions, as well as seminar participants at Appalachian State University and the 2012 Appalachian Spring Conference in World History and Economics. We are also grateful to Larry Kenny for sharing data. Seed funding for this project was provided by the University of Tennessee Office of Research. Additional support includes a grant from the Spencer Foundation, grant number 201200064, and a grant from the University of Kentucky Center for Poverty Research via the U.S. Department of Health and Human Services, Office of the Assistant Secretary for Planning and Evaluation, grant number 5 U01 PE000002-06. The opinions and conclusions expressed herein are solely those of the authors and should not be construed as representing the opinion or policy of the Spencer Foundation, the U.S. Department of Health and Human Services, any agency of the Federal Government, or the National Bureau of Economic Research. All errors are our own.

NBER working papers are circulated for discussion and comment purposes. They have not been peer- reviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications.

© 2015 by Celeste K. Carruthers and Marianne H. Wanamaker. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source.

Municipal Housekeeping: The Impact of Women's Suffrage on Public Education Celeste K. Carruthers and Marianne H. Wanamaker NBER Working Paper No. 20864 January 2015 JEL No. H75,I24,N32

ABSTRACT

Gains in 20th century real wages and reductions in the black-white wage gap have been linked to the mid-century ascent of school quality. With a new dataset uniquely appropriate to identifying the impact of female voter enfranchisement on education spending, we attribute up to one-third of the 1920-1940 rise in public school expenditures to the Nineteenth Amendment. Yet the continued disenfranchisement of black southerners meant white school gains far outpaced those for blacks. As a result, women’s suffrage exacerbated racial inequality in education expenditures and substantially delayed relative gains in black human capital observed later in the century.

Celeste K. Carruthers Department of Economics Center for Business and Economic Research 702 Stokely Management Center The University of Tennessee Knoxville, TN 37996-0570 [email protected]

Marianne H. Wanamaker Department of Economics University of Tennessee 524 Stokely Management Center Knoxville, TN 37996 and NBER [email protected]

1 Introduction

“The men have been carelessly indifferent to much of this civic housekeeping, as they

have always been indifferent to details of the household... The very multifariousness

and complexity of a city government demand the help of minds accustomed to detail

and variety of work, to a sense of obligation for the health and welfare of young

children and to a responsibility for the cleanliness and comfort of others.”

- Jane Addams1

At the dawn of the twentieth century, the United States held a position of distinction in the

provision of public education. Among its Western Hemisphere peers, the U.S. exhibited the high-

est “common school” (grade 1-8) enrollment rates and was showing early leadership in the race

towards mass secondary education. Indeed, the country had enjoyed a substantial and persistent

mass education advantage since the middle of the previous century, aided by the country’s commit-

ment to a set of “egalitarian principles” (Goldin, 2001). These principles included “public funding,

openness, gender neutrality, local (and also state) control, separation of church and state, and an

academic curriculum,” and they drove the United States to world leadership in education provision

by 1900. As the twentieth century progressed, the United States strengthened its leadership posi-

tion in the provision of public education and brought secondary education to the masses. By the

dawn of the second world war, the median 19-year-old was a secondary school graduate (Goldin,

1998).

The human capital consequences of gains in school resources and quality are somewhat con-

troversial for the latter part of the 20th century (Hanushek, 1996) but less-so for earlier cohorts of

Americans. Card & Krueger (1992a) and Card & Krueger (1992b) document an associated upward

trend in the rate of return to schooling for white and black Americans. And because black school

quality was rising more quickly than white after 1930, they attribute a substantial portion of the

narrowing black-white wage gap to corresponding improvements in relative black human capital.

The available literature cites the continued application of American egalitarian principles and

strong labor market demand for an educated workforce to explain the steady advance of public

1Quoted in Harper (1922), p. 178.

2

education provision (Goldin & Katz, 2009). Yet, as we show in Figure 1, the growth of education

provision was not constant over the course of the century. The commitment of states and local

school districts to the funding of public education exhibits a marked uptick around 1920, the same

year that the Nineteenth Amendment to the U.S. Constitution guaranteed full voting rights to all

adult females. Given the timing of these changes, is there a role for universal suffrage in explaining

the renewed commitment of school districts to public education finance? In addition to marking the

beginning of accelerated growth in overall school spending, relative black school quality measures

also dipped to unprecedented lows in this same period before rising again in the 1930s and 40s.2

Is there an explanation for this shift in relative resources that is related to the expanded electorate

under the Nineteenth Amendment?

An extensive empirical literature documents a greater propensity of women to support the pro-

vision of public goods, to hold other-regarding preferences, to foster the expansion of government

to benefit child welfare and, in some ways, to hold Goldin’s “egalitarian principles” closer to heart.3

Standard models of electoral competition indicate that policy makers will respond to shifting pref-

erences of their electoral base by altering public service allocations and their own voting behavior.

The testable implication is that the enfranchisement of women would have resulted in greater edu-

cation expenditures following the ratification of the Nineteenth Amendment.4 Other literature has

identified a measurable impact of expanded suffrage on government provision of other services.5

Interestingly, this literature has found no increase in education spending in the wake of suffrage

despite the fact that education was one of the fastest-growing elements of state and local expen-

ditures in this period and despite evidence from a variety of literatures that women have stronger

preferences than men for child welfare (Lott & Kenny (1999); Miller (2008)). We hypothesize that

2See Margo (1990) Table 2.5, pp 20-21.

3See, inter alia, Eckel & Grossman (1996); Croson & Gneezy (2009); Li et al. (2011); Doepke

& Tertilt (2014).

4An important caveat to this expectation, however, is the possibility that the impact of suffrage

on public spending was fleeting as policymakers became less wary of the female vote (Moehling

& Thomasson, 2012).

5Husted & Kenny (1997); Lott & Kenny (1999); Miller (2008).

3

a muted or negligible response to suffrage at the state level belied a strong local response. Control

of schools was highly decentralized in the early 20th century, and the majority of education funds

were local outlays resulting from taxes administered by counties and school districts.6

A second implication, at least for the segregated South, is that white spending stood to gain

more from women’s preferences than did black. Severe disenfranchisement through poll taxes,

literacy tests, and voter intimidation meant that although the Southern electorate became more

female after 1920, it did not become any less white until later decades. Because schools were seg-

regated and education expenditures were race-specific, a testable implication is that the expansion

of voting rights to women affected expenditures on white schools differently from black schools.

To our knowledge, we are the first to examine this question in the literature. For a later period,

Cascio & Washington (2013) show that the Voting Rights Act, which extended the de jure fran-

chise to black southerners, resulted in higher education expenditures in counties with higher black

population concentrations.

We use a new county-level panel dataset of annual education expenditures for three Southern

U.S. states - Alabama, Georgia, and South Carolina - to test the notions that suffrage elevated

public resources for education and that benefits accrued more heavily to white schools. These data

are unique in that they are disaggregated by county and race, and we know of no other long-running

source of education spending at this level of detail.7 Alabama, Georgia, and South Carolina are

not among the twenty-nine states that granted full voting rights to adult females prior to 1920.

Rather, the counties in this panel were compelled to extend the voting franchise by the passage of

the Nineteenth Amendment, which none of the three states ratified until after 1950.

This single point of intertemporal variation in suffrage reduces the threat of policy endogeneity

6Nationwide, local sources contributed 83 percent of school revenues for the 1919-1920 school

year, versus 44 percent for 2004-2005 (Snyder et al., 2008).

7We focus on Southern states in order to test all model predictions in Section 2, including

differential implications by race. Other Southern states are excluded from the analysis either due

to the lack of consistent expenditure reporting over the time period in question or because spending

data are not disaggregated by race. See Appendix B for evidence that these three states did not

diverge significantly from the rest of the nation in their response to women’s suffrage.

4

(from, for instance, unobserved progressivism that led states to extend suffrage rights and increase

education spending at the same time) but presents the challenge of identifying variation in women’s

right to vote. We therefore exploit spatial variation in the “dosage” of suffrage, namely cross-

county variation in white female population shares and estimated female voter shares as measures

of the relative power, perceived or actual, of women in the democratic process.

Consistent with expectations, a higher dose of suffrage is associated with higher education

spending after 1920. Each percentage point increase in the white female population share increased

per-pupil spending by 0.7 percent, indicating that up to one-third of 1920-1940 expenditure gains

are attributable to women’s suffrage. The results further indicate that expanded suffrage had a

significant, positive impact on both black and white school expenditures, but that white school

spending gains far outpaced those for black schools. After having stabilized in the latter part of

the 1910-1920 decade, the ratio of black to white per-capita spending fell by 19 percent in the

years following suffrage. Our estimates indicate that all of this relative decline can be attributed to

the Nineteenth Amendment, and we cautiously propose that the ratio of black to white education

expenditures would have been substantially higher in its absence.

The question of whether mass enfranchisement affects the provision of education has bearing

for modern-day developed and developing countries, where the decisive voter is less proximate to

decision-makers and where the returns to expansion of public education are steep (Duflo, 2001).

Improving public schools in the United States had long-term impacts on wages and inequality, and

our findings imply that women’s suffrage was partly responsible.

2 Historical and Theoretical Foundation

This study is motivated by the observation that per capita education expenditures in the United

States demonstrate a marked uptick around 1920. The accelerated growth of education funding

may be the result of numerous causes: the impact of World War I and the ensuing recession, rising

living standards and incomes of the “roaring twenties,” a modernizing work force and a rising

demand by employers for formal human capital, changes in compulsory schooling requirements,

or expanded “free tuition” legislation.8 Our study examines the role of an expanded electorate in 8See Goldin & Katz (2009) for a more thorough discussion of the social, economic, and political

landscape that contributed to the growth of public education in the early 20th century.

5

explaining the expenditure increase.

A causal role for female suffrage is supported by two distinct lines of economic research. First,

models of electoral competition indicate that policymakers act in accordance with the views of

decisive or “swing” voters in the electorate.9 The Nineteenth Amendment doubled the size of the

electorate in some states; if the new voters also exhibited a greater preference for spending on

public education, we should observe an acceleration in expenditures as a result.10

Second, a series of empirical results document a greater preference of women for goods that

enhance child welfare and for the provision of public goods in general. In the intra-household con-

9The early, seminal literature on competitive political economy indicates that politicians adopt

policy platforms matching the preferences of voters at the median of an issue spectrum (Bowen,

1943; Black, 1948; Baumgardner, 1993). If women are more public-goods loving than men, and if

there is variability within the male electorate on public goods preferences, the decisive voter along

the spectrum of preferences regarding education provision shifts decidedly towards more education

funding after women’s suffrage, even if the decisive voter is still male. A similar outcome obtains

in models of probabilistic voting (Lindbeck & Weibull, 1987, 1993; Persson & Tabellini, 2000).

In this case, voters choose candidates based on their policy platforms and on individual or group-

specific preferences over candidates which are independent of policy and imperfectly observed

by candidates. The equilibrium outcome of the model is that candidates adopt platforms which

represent a weighted social welfare function where the weights reflect both the size of a particular

group and the expected responsiveness of groups/individuals to policy changes in terms of votes. In

either case, and under relatively weak assumptions, expanding the electorate at the scale realized by

the Nineteenth Amendment would have shifted platforms of vote-maximizing candidates towards

the preferences of newly enfranchised voters.

10A critical view of electoral competition models might emphasize the role of political activism

outside of voting per se as a driver of political behavior and public expenditures. Women were

certainly active in the political process and pressed their agendas prior to being granted suffrage

(Schuyler, 2006). Nevertheless, to the extent that newly-acquired voting rights reflected a more

potent voice in the political process, we expect a change in the allocation of public monies to more

closely reflect female preferences.

6

text, a number of studies have shown an increased propensity of women to invest in the health and

welfare of their own children, relative to their male counterparts.11 As a result, welfare outcomes

for children in the household also tend to rise with the mother’s financial resources.12

In addition to an increased propensity to invest financial resources in their own children, women

also appear to prefer higher quantities of public goods in general and goods benefitting children

(not necessarily their own) in particular.13 Doepke & Tertilt (2009) propose that the expansion of

women’s legal rights in the nineteenth century, prior to women’s suffrage, resulted in increased

investments in children’s education. In terms of the Nineteenth amendment, Moehling & Thomas-

son (2012) attribute state participation in the Sheppard-Towner maternal education program to

women’s suffrage, although the effect seems to wane over time. Lott & Kenny (1999) credit the

enfranchisement of women with an increase in overall government expenditures and revenue af-

ter 1920. At the state level, however, they find no significant impact on specific components of

government expenditure including social services and education. Miller (2008) demonstrates that

suffrage and the increased voting power of women resulted in a sizable increase in local public

health spending and a decrease in child mortality rates, but no change in state educational spend-

ing.14 Neither Miller (2008) nor Lott & Kenny (1999) examined the impact of suffrage on local

educational spending, which we contend would have been more sensitive to women’s enfranchise-

ment given the dominant role of local districts in determining the allocation of resources to public

schools. 11See Doepke & Tertilt (2014) for a summary of empirical findings.

12Anthropometric status, nutrition, and child survival rights have all been shown to increase with

the mother’s income share. See Atkin (2009) and Duflo (2003) for the anthropometric results,

Rubalcava et al. (2009) for nutritional status, and Thomas (1990) for child survival results. See

Wolpin (1993) for a more complete summary of the literature on health outcomes of children.

13See, inter alia, Eckel & Grossman (1996); Croson & Gneezy (2009); Li et al. (2011); Doepke

& Tertilt (2014).

14See also Husted & Kenny (1997) for evidence that enfranchisement of the poor via the elimi-

nation of poll taxes and literacy tests increased public welfare spending.

7

3 Data

3.1 Education Statistics

We utilize a newly transcribed dataset of county-level black and white public school statistics be-

tween 1910 and 1940 for three Southern states: Alabama, Georgia, and South Carolina. Focusing

on Southern states allows us to test for impacts of suffrage on spending inequality between races,

and these three states were chosen for their consistency in reporting expenditure values for black

and white schools separately over the critical years 1910-1940. Each state’s department of edu-

cation or equivalent office published an annual or biennial report containing statistics on revenues

and expenditures. The data and data collection process are described in detail in Carruthers &

Wanamaker (2013).

Local school districts were the locus of control over schooling expenditures in this era both

because the majority of revenues were locally-sourced and because state contributions to schools

were subject to the spending discretion of local districts. Snyder et al. (2008) report the nationwide

contribution of revenue receipts from federal, state and local sources, and we present these data in

Table 1. Nationwide, federal spending is minimal throughout the period of interest. Local control

waned over time; the 1919-1920 school year saw 83.2% of revenues emanating from local sources,

a number which had fallen to 68% by 1939-1940.15 State appropriations filled the gap left by the

relative decline of local school revenues between 1920 and 1940. Critically, however, education

expenditures reported in state department of education reports include all spending from state

and federal transfers, since local school districts were the clearinghouse for all public education

support.

For the states in our sample, school districts are sub-county constructs. But education expen-

ditures are not consistently published at the sub-county level, and we have no ability to measure

female voter power at the school district level. As a result, we perform our analysis at the county

level and maintain that the southern county is the best available unit of analysis given data con-

straints.16 Alabama and Georgia report statistics by district, which we then aggregate to the county

15Importantly, we do not observe a discernible break in the local funding share of total receipts

between 1919/20 and 1929/30.

16The number of distinct school districts per county varied over time; in 1920 the ratio was

8

level. The South Carolina data were aggregated to the county level before being published. Echo-

ing the previous literature, Appendix B shows that the state-level education spending response was

small or negligible for these three states, implying that any change in education provision after the

Nineteenth Amendment would have been driven by local decisions.

Transcribed education data are assembled into a county-by-race panel for 1910-1940 describ-

ing the school finances of each county.17 This panel is matched to additional county-level variables

from industrial and agricultural censuses taken in 1910, 1920, 1930, and 1940 (Center, 2011).18

Relevant statistics from these reports include crop value per capita, the percent of land devoted

to agriculture, and manufacturing employment and earnings. Annual measures are interpolated

between census years to fully populate the panel.19 Philanthropic activity directed toward black

schools was an important factor in both black and white school spending (Carruthers & Wana-

maker, 2013), and we match each county and year with the number of new Rosenwald classrooms

built therein.20 We additionally control for the presence of secret ballots, which pertain to Georgia

1.73 for Alabama, 1.56 for Georgia and 41.2 for South Carolina where individual townships were

granted the power to levy school taxes in addition to county taxes. The sample average in 1920 is

8.3.

17Georgia schools data are reported biannually between 1930 and 1940, and we interpolate lin-

early between reporting years.

18County boundaries changed over time, and Georgia continued to form new counties throughout

the period in question. We rely on county boundary change data from Horan & Hargis (1995) and

aggregate up to the “supercounty” level to ensure consistent boundaries over time. This brings the

number of counties from 272 to 235.

19Because the interpolation process itself may generate discrete changes in 1920, we test the

robustness of our results to a set of controls interacted with year fixed effects in Appendix C.

20Benefactor Julius Rosenwald challenged localities with matching grants for black school con-

struction. Over 5,000 rural schools for black students were built by the Rosenwald Founda-

tion between 1916 and 1933, and each is documented in the online catalog at Fisk University:

http://rosenwald.fisk.edu/.

9

in 1922 and later years.21

3.2 Measuring Voter Power

The conceptual framework outlined above generates testable implications regarding the dosage

of suffrage treatment across local areas. We use three dosage measures: the percentage of the

voting population white and female in 1920 when the amendment was enacted; the percentage of

the voting population white and female in each year (interpolated between census years); and the

estimated density of female voters as a percentage of the voting-age population in 1920 (an early

turnout proxy).

To accurately size the male and female voting-age population, we require more granular pop-

ulation statistics than those available in published census volumes. We populate decennial 1910-

1940 age-by-race-by-gender cells for each county in the three-state sample using data from the

genealogy website Ancestry.com.22 The 1920 ratio of white females to the voting-age population

is the first proxy above. The second is calculated by interpolating the number of white females and

the size of the voting-age population between census years. For the third (female voter percent-

age), we match female population data to total voter counts for the 1920 general election (Clubb

et al., 2006), the first where all U.S. women could participate. We use Bayesian methods with

informative priors to infer 1920 voter turnout rates by gender for each county. See Appendix A for

details on this procedure.23 We then use the estimated female turnout to calculate female turnout

as a percent of the voting-age population.

21The expected impact of secret ballots on education expenditures is ambiguous, but as the re-

action of political systems to voters or potential voters is the effect of interest, we control for this

variation in voting structures which may have impacted female voter uptake.

22Ancestry.com has transcribed the full manuscripts of U.S. Census returns through 1940 from

the originals housed at the National Archives. The data is indexed and searchable, facilitating

tabulations by county and demographic characteristics. Outside of limitations to enumeration in

the original census year and the readability of census manuscripts, there are no known limitations

to the Ancestry.com data.

23Estimates of race and gender-specific turnout proved too noisy for this exercise.

10

Each of these proxies has its merits. Fixed measures of voter power in 1920 are less likely

to be endogenous to education expenditure trends than a population share that changes over time,

but relying on a dosage proxy from a point in time increases measurement error in other years.

Population percentages represent potential female voter power and are less likely to be endogenous

to education expenditures than voter percentage would be. But if policymakers responded to female

voter turnout at the polls more so than potential voters, the voter turnout measure is a better proxy

for suffrage dose. The share of the population consisting of voting females is a distilled form of the

population share proxy, and one that allows the effect of suffrage to operate through the political

process per se. The choice of suffrage dosage matters little for our overall conclusions, and results

for all three proxies are reported in Section 4.

Given the central role female population percentages take in our analysis, it is appropriate to

question which counties had higher shares of white voting-age females in and around 1920. Table 2

describes the correlation between 1920 white female population shares and other observable county

features as of 1920.24 White females are conditionally more prevalent in counties with lower black-

white ratios, higher shares of land devoted to agriculture, fewer adults employed in manufacturing,

and lower crop values per capita. Overall, these observable county-level covariates explain 75%

of the intra-state variation in 1920 white female population shares. Principal results to follow

control for these covariates (which are time-varying in our specifications, rather than being fixed

in 1920) and essentially test whether the remaining (25%) variation in female population shares

is associated with differentially higher or lower education spending after the 19th Amendment,

relative to before.25

24The regression is PctFc,1920 = βXc,1920 + ϵct where the dependent variable is expressed in

0-100 percentages and Xc,1920 is a vector of covariates from the bottom of Table 3, excluding

variables with no cross-sectional variation in 1920 (secret ballots and Rosenwald classrooms).

25Given the economic and statistical significance of black-white population ratios, robustness

checks test whether results are sensitive to a more flexible quadratic control for this variable. Point

estimates are nearly equivalent in sign and significance. See Appendix C.

11

4 Empirical Strategy and Results

Table 3 lists descriptive statistics for school spending and other school quality outcomes, suffrage

treatment measures, and Census controls. The gap between white and black spending is striking.

Between 1910 and 1940 in these three states, black students were allocated 24 cents for every

dollar directed toward white students. This gap narrowed after 1940, particularly in the wake of

civil rights legislation more than two decades after the close of our panel.

We consider 1921 (that is, the 1920-1921 school year) to be the first post-suffrage reporting

year for these states. Financial reports that form the basis of these data referred to academic years

(e.g., August 1919 - July 1920). The Nineteenth Amendment was officially in place as of August

1920, well after funds were spent for the 1919-1920 school year. Though there may have been

anticipatory effects on spending immediately prior to the Nineteenth Amendment, we are most

interested in the change in school spending trends after policymakers faced a new electorate.26

The stylized facts relevant to this application are presented in Figures 1 and 2. The top panel of

Figure 1 plots per-capita school spending by year for all of the United States, where a substantial

increase in expenditures is evident in the years immediately following 1920. The same metric for

the transcribed three-state sample of local school data is located in the bottom panel of Figure 1.

The trajectory of spending in these states mimics the nationwide change with a sharp upward shift

in spending trends after 1920.

In both panels, a noticeable dip in expenditures is apparent during the war years (1914-1918).

These reductions are likely a result of reductions in municipal tax revenues, but there is no reliable

data on local tax receipts at the county level for this period. If post-1920 gains are just a recovery

from this reduction and unassociated with suffrage, the results should show no differential response

in counties with more voting power among women. On the other hand, if reductions in spending

around World War I are somehow correlated with the proxies for suffrage used in the empirical

26If there are anticipatory effects, they will serve to bias our results toward zero. Note that any

long-term expenditure changes would have been funded out of changes in state and local taxes

which themselves would have taken some time to come into effect. In the near term, local leaders

who were cognizant of the political economy implications of suffrage could have redirected funds

from other public service areas to benefit education.

12

results below, the interpretation of those results is muddled given the possibility for “catch-up”

spending in the hardest-hit municipalities.27 We return to this issue in Section 4.1 and Appendix

Section D.

The spending trends highlighted in Figure 1 differed by race. Figure 2 shows that the spending

trajectory shifts in 1921 were steeper in white schools. In panel A of Figure 2, it is apparent that

post-1920 growth in black expenditures per pupil exhibited a much slower increase. The same is

true for specific school resources. Teachers per pupil (panel B) and average teacher salary (panel

C) show sharp shifts for whites after 1920, but the increases are more muted for blacks, especially

in the case of teacher salaries.28

Importantly, panel D of Figure 2 indicates that these expenditure and resource shifts did not

coincide with changes in enrollment per capita. Thus the post-1920 changes in spending per pupil,

where we will focus our analysis, are not likely driven by any sudden, demand-side changes in the

size of the student population. We do not rule out that subsequent enrollment may have responded

to higher school quality, but simply highlight no discrete shift in these metrics at the 1921 juncture.

4.1 Difference-in-Difference Identification

The apparent shift in post-1920 resources cannot be attributed to suffrage per se if there are other

events impacting spending in the years immediately following 1920. To identify the contribution

of women’s suffrage specifically to this expenditure growth, we utilize a difference-in-difference

estimator where treatment is the suffrage dosage proxy interacted with an indicator, POSTt, equal

to 1 in all years after 1920. If women’s suffrage led to higher education expenditures, we should

27Male casualties may have been a source of cross-county variation in the gender ratio, but war-

induced variation in our suffrage dose proxies were probably very small. Only 0.19 percent of the

voting-age population were killed in the war, and cross-county variation in war casualties likely

account for a small share of the 9.9-point standard deviation of female population shares.

28The 1919 marker for white teachers per 1,000 pupils is affected by one outlier in Georgia. That

year, the state reported atypically low white enrollment figures for Gordon County. This probable

reporting error attenuates results for white teachers and spending per pupil downward.

13

observe a larger post-1920 effect in counties where proxies for female voter power are higher.29

A difference-in-difference estimator of the treatment effect of women’s suffrage is as follows:

Yct = δ0 + δ1POSTt ∗ DFct + δ2D F ct + θc + θt + βXct + εct, (1)

where Yct is a measure of public educational spending for county c at time t (in 1925 dollars), DFct

is a measure of female voter power, θc is a county fixed effect, θt a year fixed effect, and εct is an

error term.30 Xct is a matrix of time-varying county-level observable characteristics summarized

at the bottom of Table 3. Economic controls (value of crops per capita, percent of land devoted to

agriculture, average annual income in manufacturing, share of adults working in manufacturing),

population controls (black-white population ratio, cubic function of total county population, adult

female population share) and other variables that may have affected school spending (presence of a

secret ballot, number of newly constructed Rosenwald classrooms) are included in all estimates.31

29A number of other econometric tools are available here. The most flexible of these is a modified

event study estimator where year fixed effects are interacted with dosage proxies to generate year-

by-year estimates of the impact of suffrage dosage on spending. This strategy, however, makes

hefty demands on our sample of 235 counties. We present the results of this analysis in Appendix

D and note that the results are consistent with those obtained in other specifications below but are

imprecisely estimated as would be expected given our limited sample size. In addition, we have

estimated a more flexible functional form for the post-1920 impact which identifies the impact of

the suffrage dosage on the level of resources and on both the linear and quadratic growth trends.

Those results, available upon request, imply similar effects to the ones identified here.

30The presence of year fixed effects precludes identification of a coefficient on POSTt as a

stand-alone regressor.

31Agricultural economic activity was a close substitute to schooling in rural areas and also drove

incomes, but Southern economies were shifting to a more industrial emphasis over this time period.

We thus include measures of both agricultural and manufacturing variables. Outside private funds,

proxied by Rosenwald classrooms, may have crowded out local investment in school spending or

may have been crowded in by it and, if so, are endogenous. Our results are not sensitive to including

this control. The Appendix discusses robustness checks with additional controls for county-wide

literacy. Although literacy is an important proxy for human capital within a given area, we exclude

14

We estimate δ1 over a 20-year post-Amendment horizon, and robust standard errors are clustered

within counties in all estimates.

Of course, voter density and population shares are not randomly assigned across counties and

may be correlated with potential confounders, including the size of the school-age population,

constituent preferences for education, and migration in response to or anticipation of better school

resources. In this context, any such unobserved, county-specific and time-invariant variable that

is correlated with both a county’s (proxied) female voter power and the school spending measure

will be absorbed in the county fixed effect. Similarly, any post-1921 impact that is common across

counties will be absorbed by year fixed effects. Identification comes from the differential effect of

female voter power on education spending in the years following 1920 relative to the pre-suffrage

years in the panel.

The remaining concern and primary threat to our identification strategy is the possibility that

unobserved and heterogeneous trends in omitted variables are more prevalent in high-dosage areas,

and that these omitted variables affect the educational spending in ways we falsely attribute to suf-

frage. In robustness checks discussed in Appendix C, inferences are qualitatively unchanged when

we undertake additional steps to recognize heterogeneous trends in the analysis. Additionally,

balancing tests show that the correlation between pre-suffrage spending trends and our suffrage

dosage metrics are largely insignificant.

Finally, the interpretation of δ1 hinges on the assertion that the identified impact of the dosage

proxies in the post-1920 years is attributable to suffrage per se. Alternative explanations which

are consistent with a differential impact after 1920 for counties with higher female voter power

are difficult to come by. To our minds, the leading contender is that δ1 is measuring a differen-

tial rebound from World War I spending reductions in these counties which is unassociated with

suffrage. If higher female populations somehow reduced spending during World War I more than

other counties, δ1 may simply reflect post-1920 rebounds from these spending reductions. Con-

trary to this hypothesis, however, we observe no difference in wartime spending based on suffrage

dosage. These results and further discussion are located in Appendix D.

this control from our main analysis. Available literacy data pertain to the population age ten and

older, which may be endogenous to improving school quality.

15

4.1.1 Baseline Results

Results corresponding to each of three proxies, which are measured in percentage points (0-100),

are reported in Tables 4 - 6, respectively. The top row of each table gives the δ1 coefficient under

a variety of functional forms. Column 1 in each table includes year fixed effects and Xct controls

interpolated between census years. Column 2 replaces year fixed effects with state-year fixed

effects, and Column 3 in each table drops the Xct variables which were interpolated between census

years (all variables except secret ballot legislation and Rosenwald school controls) out of concern

that post-1920 estimated impacts are spuriously driven by changes in the slope of the interpolated

covariates as of the 1920 census. The impact on overall spending is largely impervious to these

changes in functional form.

We turn first to results for the impact of suffrage via a higher share of white females in 1920,

which varies across counties but not time (Table 4). This dosage proxy, measured at a point in

time on the eve of women’s suffrage, is least subject to endogenous time-varying omitted variables

(e.g., mobility) but most subject to error in measuring the time-varying power of new voters. Each

percentage point increase in the white female population in 1920 (per Columns 1 - 3 of Table 4) is

associated with an increase in education expenditures per pupil of between 0.7 and 0.9 log points

between 1920 and 1940, indicating that the overall treatment effect of increasing female voting

power from null to the typical dosage was 19 to 26 log points.32

Table 5 lists coefficient estimates for POST ∗ DFct with the annual population share of white

females standing in for DFct, a value that varies both over time and across counties. This proxy is

a more accurate measure of women’s potential voting bloc in a given year, but also more likely to

be endogenous. Nevertheless, the first row of Column 1 - 3 point estimates are in broad agreement

with those from Table 4, and the implied suffrage treatment effect is equivalent to that in Table 4,

19 to 26 log points.33

Our third proxy for women’s voter power is the share of the 1920 population who we estimate

to have been female voters (Table 6). Standard errors are bootstrapped to reflect estimation of the

32Obtained by multiplying each coefficient by the average dosage value, 29.85, from Table 3.

33Obtained by multiplying each coefficient by the average dosage value, 30.21, from Table 3.

16

proxy variable.34 Incrementally higher female turnout is associated with expenditure impacts of

between 1 and 2 log points, and the implied suffrage treatment effect is somewhat smaller at 7

to 13 log points.35 We note that treatment effects from this third dosage are highly susceptible to

attenuation bias from measurement error since female turnout is estimated (see Appendix A). Still,

we view the results in Table 6 as confirmation that women’s proximity to the political process itself

was responsible for the observed impact of female population shares observed in Tables 4 and 5.

Our results are comparable to estimates of suffrage treatment effects on public spending overall

in this era. Using a similar methodology, Lott & Kenny (1999) estimate that typical turnout gains

from suffrage increased state-level spending by 14 percent immediately and 21 percent after 25

years (p. 1176).

The size of the suffrage treatment effect also indicates that the Nineteenth Amendment played

a dominant role in the growth of public school expenditures after 1920. Average expenditure (per

pupil) in the three Southern states in our sample increased by approximately 78 log points between

the pre-suffrage years of 1910-1920 and the post-suffrage years through 1940. Our estimates

indicate that between 24 and 33 percent of that increase emanated from an expanded electorate.

4.2 Testing the Interracial Prediction

In this section, we test whether white schools benefitted differentially from suffrage relative to

their black counterparts. Segregated schools in this era resulted in fully separable school budgets,

allowing us to test the interracial prediction directly by replacing Yct in Equation 1 with black

expenditures per (black) pupil, white expenditures per (white) pupil, and the ratio of black to white

per pupil school expenditures. The ratio allows us to gauge the impact of suffrage on this common

measure of relative school quality.36

34Bootstrapped standard errors are computed by estimating the model 100 times for 125 ran-

domly sampled counties (with replacement).

35Suffrage “treatment” for this proxy implies going from 0 to an average of 7 percent of the

electorate as female voters. See Table 3.

36Because enrollment changes were minimal following 1920 (see Figure 2), changes in the black-

white funding ratio are driven by changes in expenditures.

17

Results across Tables 4 - 5 (Columns 1-3, Rows 3-4) indicate that female suffrage exacerbated

gaps in black and white school quality by raising the level of white school spending more than

black. Point estimates for black spending gains, while sometimes statistically significant, range

from 0.27 to 0.50 log points per white female share while the same metrics for white spending

range from 0.54 to 0.70 log points. Although the coefficients differ in Table 6, the implied suffrage

effect for both black and white education spending is the same.

The possibility that black spending increased at all following the enfranchisement of predom-

inantly white women evokes Myrdal’s (1944) paradox: given widespread disenfranchisement of

blacks in the South, why were they provided any public services? And then why were blacks

provided more public education resources following the Nineteenth Amendment? The answer lies

beyond the scope of this paper and the data at hand, but three possibilities are worthy of note. First,

Margo (1991) proposed that Tiebout sorting by black families led localities to compete for black

labor by supporting black schools; perhaps new women voters were more attuned to the importance

of black labor. Second, education leaders may have held the perception that new women voters

were more altruistic toward the welfare of black families and schools. Last, the observed gains in

black spending may have been driven by omitted factors affecting both white and black school sys-

tems. Still, even if no post-1921 changes in black spending are attributable to suffrage, and if the

trajectory of black spending is an adequate counterfactual to the trajectory of white spending, the

implication is that women’s suffrage was nevertheless responsible for a large share of the overall

rise in white education spending after and including 1921.

The finding that white schools benefitted far more from suffrage than did black schools suggests

that the ratio of black to white school expenditures, the so-called education “gap,” suffered at the

hands of the Nineteenth Amendment. Indeed, we estimate in the second rows of Tables 4 - 5 that

the relative quality of black schools fell substantially and significantly following women’s suffrage.

The mean value of black to white spending per pupil, multiplied by 100, was 23.8 over this period

(see Table 3), and we estimate that each percentage point increase in the white female population

percentage decreased this ratio by 0.52 to 0.57 (Tables 4 - 5).37 Taking the more conservative point

37We note that this ratio is very noisy, however, as it is derived from four series of transcribed

data: white and black spending, and white and black enrollment. R2 statistics indicate that Equa-

tion 1 estimates of the black-white ratio have much less overall explanatory power than estimates

18

estimates from Table 5, the treatment effect of suffrage on the black-white expenditure ratio was

between 15 and 16 cents per dollar of white school spending. To scale this effect, we note that the

average pre-suffrage ratio of black to white per pupil expenditures was 27.2 and the post-suffrage

ratio in the same fell to 22.0. Results based on voter turnout (Table 6) echo these conclusions.

On their face, then, our estimates indicate that the entirety of the post-1920 reduction in relative

black spending can be attributed to women’s suffrage. The size of the coefficients also suggest that,

in the absence of women’s suffrage, the black/white ratio would have risen substantially after 1920

rather than exhibit such a drastic fall. For blacks in the South, female enfranchisement was a mixed

blessing, likely bringing additional resources to their schools but also lowering the quality of their

schools relative to their white counterparts.

4.3 Duration of Impact

Next we turn to the question of whether the estimated effects were fleeting or persistent over

time. The expected timing of any spending response to voter enfranchisement is ambiguous in this

context. The Nineteenth Amendment was never reversed, yet Moehling & Thomasson (2012) find

evidence that the impact of suffrage on public health expenditures waned quickly as policymakers

became accustomed to the female vote.

At the same time, the influx of voters following 1920 was unprecedented and it took years, if

not decades, for the full impact of female enfranchisement to form (see, for example, short-term

versus long-term estimates by Lott & Kenny (1999)). This was especially true in the South where

female voter uptake lagged the rest of the nation but grew over time. Our own estimates of female

voter turnout indicate an increase from 20.2% in 1920 to 29.0% in 1940. (See Appendix A, Figure

3). This gradual increase in female voter participation may have resulted in changes in behavior

relative to the counterfactual well into the 1920s and 1930s as elected officials continued to gauge

the preferences of the new electorate. In addition, because education expenditures were subject to

a public budgeting process with significant lags, female voter preferences may have taken several

years to find their way to expenditure outcomes. Consistent with this last conjecture, in Appendix

Section E we show that spending growth was steeper after suffrage in counties with more local

for other outcomes.

19

control of school budgets.

To map out the spending changes over time, we modify our difference-in-difference estimator

in Equation 1 to include 5-year time dummies:

Yct = δ0 + δaT1921,1925 ∗ DFct + δbT1926,1930 ∗ D F ct + δcT1931,1935 ∗ D

F ct + δdT1936,1940 ∗ D

F ct

+δ2D F ct + θc + θt + βXct + εct, (2)

where each Tt0,t1 is an indicator for t ∈ [t0, t1]. This functional form allows us to measure the

differential impact of suffrage over four time horizons: 1921-1925, 1926-1930, 1931-1935, and

1936-1940, each relative to the omitted window of 1910-1920.

The coefficients δa −δd are reported in Columns 4 - 7 of Tables 4 - 6. Each column corresponds

to the coefficient on the year group dummy interacted with a dosage proxy. For each outcome of

interest, findings unambiguously indicate that the impact of suffrage began slowly in the early

1920s and accelerated over time with no evidence of tapering by 1940. Looking first to Table

4, the estimated impact on spending per pupil accelerated from an insignificant 0.27 log points

per population share between 1921 and 1925 to 1.4 log points per population share between 1936

and 1940 - a result that is echoed by Table 5. Results using voter share (Table 6) as the proxy

are smaller, but reflect the same trajectory. In this formulation, black spending gains are rarely

statistically significant, but white spending follows the same increasing pattern. As a result, the

impact on the ratio of black to white expenditures deepens over time, although it appears to level

off after 1935 and is estimated to reverse in Table 6. These estimates are consistent with those

from an event study estimator discussed in Appendix D.

Thus we conclude that women’s suffrage had a substantial impact on school spending that

grew over the course of the 1920s and 1930s. As a byproduct of this finding, we conclude that

women’s suffrage contributed to the growing black-white school quality gap at least through 1940

and perhaps longer.

4.4 The Impact on School Resources

Our final question is whether spending gains manifested as discernible changes in segregated

school resources that voters would have been aware of and that might have directly affected stu-

dent success. We focus on black and white teachers per pupil and average black and white teacher

20

salaries. Trends in these resources are depicted in Figure 2, where we see a noticeable upward shift

in school resources after the 19th Amendment, much more so for white school resources than for

black. We estimate Equation 1 for these four outcomes. The percent of the voting-age population

who were white and female in a given year (corresponding to Table 5) serves as the dosage proxy.

Though this time-varying dosage is more subject to endogenous mobility, it is less subject to mea-

surement error at points well removed from 1920, which is important for estimating the impact of

suffrage on resources that are measured with less consistency than spending.

Table 7 lists results for teachers per pupil and teacher salaries. In the first two rows, the suf-

frage dosage proxy is correlated with positive and significant increases in white teacher salaries,

but substantial reductions in the same for black teachers. By these estimates, suffrage brought an

increase in white teacher salaries of between 9 and 15 percent.38 At the same time, reductions in

black salaries were on the order of 45% under a conservative point estimate of -1.5 log points. Al-

though the estimates are imprecisely measured, we find increases in both black and white teaching

forces of between 2 and 3 teachers per 1000 enrolled students.39 This is a 5 to 10 percentage point

increase above the values listed in Table 3.

Although the fiscal impact of suffrage is the main emphasis of our analysis, these results indi-

cate that local administrators may have increased white school quality by increasing the number of

teachers and paying them more and partially offset those costs by reducing the pay to black school

teachers. The fact that the number of black school teachers increased commensurate with white is

somewhat surprising, but given the large reduction in black teacher salaries, it is consistent with an

overall negative effect on black school spending.

38Computed as the product of point estimates and mean values for this dosage proxy, e.g., 0.003×

30.21.

39Note that the exclusion of one outlier 1919 entry for white teachers per pupil (one county in

Georgia reported very little white enrollment in that year, likely in error) increases the sign and

significance of that outcome and suggests an impact of 4 to 5 additional white teachers per 1,000

white pupils.

21

5 Conclusion

A steady rise in school resources over the course of the 20th century is a much-celebrated feature of

the United States’ education system, and an associated rise in the relative quality of black schools

after 1940 has been linked to reductions in the black-white wage gap for workers later in the

century. In the three Southeastern states we examine, over the years 1921 - 1940, real county-

level per pupil school spending increased more than twofold over average spending from 1910

to 1920. The estimates in this paper attribute up to one-third of this increase to female voter

enfranchisement via the Nineteenth Amendment. Spending gains were higher in counties with

higher female representation in the electorate around 1921 and later, and higher for white schools

than for black schools.

Our findings are consistent with economic models of intra-household bargaining and concep-

tual expectations about the localized political economy response to asymmetric voter enfranchise-

ment. The reaction of public finance allocations to the extension of women’s voting rights provides

strong support for the idea that suffrage shifted and increased the pivotal voter’s preferences for

public education. The adoption of women’s suffrage in the United States and the subsequent im-

pact of suffrage on public education represents a historic episode that should shape expectations for

the relationship between women’s rights and human capital accumulation in modern developing

countries. As women gain electoral power, public resources for education improve.

At the same time, these findings are an important caveat to the equalizing impacts of voter en-

franchisement noted elsewhere in the literature. Cascio & Washington (2013) show that the Voting

Rights Act of 1965 led to a sizable, significant increase in the ratio of black to white education

expenditures. We find that, forty-five years earlier, women’s suffrage resulted in the reverse. With

evidence from the 19th century, Acemoglu & Robinson (2000) propose that franchise expansion

in Europe led to a reduction in income inequality after 1870. In contrast, and although the black-

white wage gap cannot be directly measured prior to 1940, we conjecture that selective franchise

expansion in the United States exacerbated racial income inequality by limiting the relative ascent

of black human capital acquisition.

22

References

Acemoglu, D., & Robinson, J. A. (2000). Why Did the West Extend the Franchise? Democ-

racy, Inequality, and Growth in Historical Perspective. Quarterly Journal of Economics, 4(115),

1167–1199.

Atkin, D. (2009). Working for the Future: Female Factory Work and Child Health in Mexico.

Working Paper.

Baumgardner, J. R. (1993). Tests of Median Voter and Political Support Maximization Models:

the Case of Federal/State Welfare Programs. Public Finance Review, 21(1), 48–83.

Black, D. (1948). On the Rationale of Group Decision Making. Journal of Political Economy,

56(1), 23–34.

Bowen, H. R. (1943). The Interpretation of Voting in the Allocation of Economic Resources.

Quarterly Journal of Economics, 58(1), 27–48.

Card, D., & Krueger, A. B. (1992a). Does School Quality Matter? Returns to Education and the

Characteristics of Public Schools in the United States. The Journal of Political Economy, 100,

1–40.

Card, D., & Krueger, A. B. (1992b). School Quality and Black-White Relative Earnings: A Direct

Assessment. The Quarterly Journal of Economics, 107.

Carruthers, C. K., & Wanamaker, M. H. (2013). Closing the gap? The Effect of Private Philan-

thropy on the Provision of African-American Schooling in the U.S. South. The Journal of Public

Economics, 101, 53–67.

Carter, S. B., Gartner, S. S., Haines, M. R., Olmstead, A. L., Sutch, R., & Wright, G. (Eds.) (2006).

Historical Statistics of the United States: Earliest Times to the Present. Cambridge University

Press.

Cascio, E. U., & Washington, E. L. (2013). Valuing the Vote: The Redistribution of Voting Rights

and State Funds Following the Voting Rights Act of 1965. Quarterly Journal of Economics,

129(1), 379–433.

23

Center, M. P. (2011). National Historical Geographic Information System: Version 2.0. University

of Minnesota.

Clubb, J. M., Flanigan, W. H., & Zingale, N. H. (2006). Electoral Data for Counties in

the United States: Presidential and Congressional Races, 1840-1972. ICPSR08611-v1.

Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor]

doi:10.3886/ICPSR08611.v1.

Corder, J. K., & Wolbrecht, C. (2004). Using Prior Information to Aid Ecological Inference: A

Bayesian Approach. In G. King, O. Rosen, & M. A. Tanner (Eds.) Ecological Inference: New

Methodological Strategies, (pp. 144–161). Cambridge University Press.

Croson, R., & Gneezy, U. (2009). Gender differences in preferences. Journal of Economic Liter-

ature, 47(2), 1–27.

Doepke, M., & Tertilt, M. (2009). Women’s liberation: What’s in it for men? Quarterly Jounral

of Economics, 124(4), 1541–1591.

Doepke, M., & Tertilt, M. (2014). Does Female Empowerment Promote Economic Development?

Working Paper.

Duflo, E. (2001). Schooling and Labor Market Consequences of School Construction in Indonesia:

Evidence from an Unusual Policy Experiment. American Economic Review, 91(4), 795 – 813.

Duflo, E. (2003). Grandmothers and Granddaughters: Old-Age Pensions and Intrahousehold Al-

location in South Africa. World Bank Economic Review, 17(1), 1–25.

Eckel, C. C., & Grossman, P. J. (1996). Altruism in Anonymous Dictator Games. Games and

Economic Behavior, 16, 181–191.

Goldin, C. (1998). Americaâçs graduation from high school: The evolution and spread of sec-

ondary schooling in the twentieth century. Journal of Economic History, 58(2), 354–75.

Goldin, C. (2001). The Human Capital Century and American Leadership: Virtues of the Past.

Journal of Economic History, 61(2), 263–92.

24

Goldin, C., & Katz, L. F. (2009). Why the United Stated Led in Education: Lessons from Sec-

ondary School Expansion, 1910 to 1940. In D. Eltis, F. D. Lewis, & K. L. Sokoloff (Eds.)

Human Capital and Institutions: A Long-Run View. Cambridge University Press.

Haines, Michael R. and Inter-university Consortium for Political and Social Research (2010). His-

torical, Demographic, Economic, and Social Data: The United States, 1790-2002. ICPSR02896-

v3.

Hanushek, E. (1996). School resources and student performance. In G. Burtless (Ed.) Does Money

Matter? The Effect of School Resources on Student Achievement and Adult Success. Washington:

Brookings Institution Press.

Harper, I. H. (Ed.) (1922). The History of Woman Suffrage: 1900-1920, vol. 5. New York, NY:

National American Woman Suffrage Association.

Horan, P. M., & Hargis, P. G. (1995). County Longitudinal Template, 1840-1990. ICPSR06576-v1.

Hoynes, H. W., & Schanzenbach, D. W. (2009). Consumption responses to in-kind transfers: Evi-

dence from the introduction of the food stamp program. American Econoimc Journal: Applied

Economics, 1, 109–139.

Husted, T. A., & Kenny, L. W. (1997). The Effect of the Expansion of the Voting Franchise on the

Size of Government. Journal of Political Economy, 105(1), 54–82.

Li, S. X., Eckel, C. C., Grossman, P. J., & Brown, T. L. (2011). Giving to Government: Voluntary

Taxation in the Lab. Journal of Public Economics, 95(11-12), 1190–1201.

Lindbeck, A., & Weibull, J. W. (1987). Balanced-budget redistribution as the outcome of political

competition. Public Choice, 52, 273–297.

Lindbeck, A., & Weibull, J. W. (1993). A Model of Political Equilibrium in a Representative

Democracy. Journal of Public Economics, 51, 195–209.

Lott, J. R., & Kenny, L. W. (1999). Did Women’s Suffrage Change the Size and Scope of Govern-

ment? Journal of Political Economy, 107, 1163–1198.

25

Margo, R. A. (1990). Race and Schooling in the South, 1880-1950: An Economic History.

Chicago: The University of Chicago Press.

Margo, R. A. (1991). Segregated Schools and the Mobility Hypothesis: A Model of Local Gov-

ernment Discrimination. Quarterly Journal of Economics, 106, 61–73.

Miller, G. (2008). Women’s Suffrage, Political Responsiveness, and Child Survival in American

History. Quarterly Journal of Economics, 123(3), 1287 – 1327.

Moehling, C. M., & Thomasson, M. A. (2012). The Political Economy of Saving Mothers and Ba-

bies: The Politics of State Participation in the Sheppard-Towner Program. Journal of Economic

History, 72(1), 75–103.

Myrdal, G. (1944). An American Dilemma: The Negro Problem and Modern Democracy. New

York: Harper and Row.

Persson, T., & Tabellini, G. (2000). Political Economics: Explaining Economic Policy. Cambridge,

MA: The MIT Press.

Rubalcava, L., Turuel, G., & Thomas, D. (2009). Investments, Time Preferences, and Public

Transfers Paid to Women. Economic Development and Cultural Change, 57(3), 507–538.

Schuyler, L. G. (2006). The Weight of Their Votes: Southern Women and Political Leverage in the

1920s. Chapel Hill, NC: The University of North Carolina Press.

Snyder, T. D., Dillow, S. A., & Hoffman, C. M. (2008). Digest of Education Statistics, 2007.

NCES 2008022. National Center for Education Statistics.

Thomas, D. (1990). Intra-Household Resource Allocation: An Inferential Approach. The Journal

of Human Resources, 25(4), 635–664.

Wolpin, K. I. (1993). Determinants and Consequences of the Mortality and Health of Infants and

Children. In Handbook of Population and Family Economics, Volume 1, (pp. 483–557). Elsevier.

26

A Voter Turnout Data

This study makes use of voter turnout estimates by gender to estimate the impact of females’

electoral participation on school spending and resource trends. Precise data on the gender of 1920

voters are not available at the county level. Instead, we have county-level counts of total votes

for every general election during this period (Clubb et al., 2006). We use Bayesian methods with

informative priors to infer gender vote shares for each county, relying on total turnout statistics as

well as demographic data from the U.S. Census.

For each county i, we observe the number of voters Vi, the adult population Ni, and the gender

composition xji, where j = 1,2 indicates female and male groups, respectively.

We start by modeling the number of voters Vi as a draw from a binomial distribution:

Vi ∼ bin (zi,Ni) , (3)

where zi is the probability that an individual votes. That probability in turn is the weighted sum of

the group-specific probabilities. It becomes:

zi = x1ip1i + x2ip2i, (4)

where p1i and p2i represent the probability that a female and male votes, respectively. Our goal is

to find the posterior distributions that characterize pji for j = 1,2 for the 1920 election.

We follow Corder & Wolbrecht (2004) and approximate the logistic transformations θji of

the group specific probabilities pji with normal distributions. That is, pji = exp(θji)

1+exp(θji) and θji ∼

norm (µj,τj), where µj is the mean of the distribution and τj is its precision. We use uninformative

normal distributions as priors on µ1 (female groups). For µ2 (male groups), we choose normal

prior distributions that reflect the turnouts observed for the 1912 and 1916 presidential election,

for which women did not vote. We assume gamma distributions for the priors on the precisions τj

for all groups.

Equations 3 and 4, along with the specified prior distributions complete the model. The model

is solved through numerical simulations. Specifically, we use Metropolis-Hasting algorithms and

Markov Chain Monte Carlo (MCMC) techniques. The posterior distributions of p1i and p2i are ob-

tained based on the data (Vi, Ni and xji) and MCMC draws from the prior distributions of µj and τj.

The point estimates of the group specific turnouts are obtained by sampling from their respective

27

posterior distributions. For each MCMC draw, we also compute the mean of pji across counties.

The results yield the aggregate posterior distributions of pj. The aggregate point estimates of pj

are obtained by sampling from the aggregate posterior distributions.

Our estimates indicate that female voter turnout rose over time. Figure 3 gives the estimated

value of p1 between 1920 and 1940. Female turnout rates increased steadily through 1936 before

falling somewhat in the 1940 election.

B Extension: State Spending After Suffrage

The analysis in this paper is limited to the provision of state and local education resources in three

Southern states. The sample is thus limited in its ability to address the responses of counties other

than those located in the three states included in our panel.

Whether the three states in the panel are representative of the nation cannot be tested directly

with county-level data from other states. We can, however, estimate the impact of female suffrage

on state level education spending for these three states compared to the nation at large. We em-

ploy the same state-level finance data and socioeconomic control variables used by Lott & Kenny

(1999) and Miller (2008) to replicate previous work estimating the effect of suffrage on state-level

educational spending.40 We then go on to test whether the three sampled states reacted differently

than the rest of the nation.

To situate results with those of Lott & Kenny (1999) and Miller (2008), we estimate a dynamic

fixed effects specification informed by both studies:

ln(Yst) = α + SUFFRAGEstβ1 + SUFFRAGEst ∗ THREESTATEsβ2

+Xstψ + θs + α(t) + θsα(t) + θt + εst, (5)

where ln(Yst) is the natural log of states’ per-capita education expenditures, SUFFRAGEst is a

measure of suffrage treatment, THREESTATEs is an indicator equal to one for Alabama, Geor-

40We thank Larry Kenny for graciously providing these data and associated documentation. State

expenditure and revenue data from 1870-1915 were originally provided to Lott and Kenny by John

Wallis.

28

gia, and South Carolina, and Xst is a matrix of socioeconomic controls.41 The parameter α(t)

controls for linear trends common to all states, θs is a state fixed effect, and θsα(t) controls for

state-specific trends. Specifications with and without time trend controls (i.e., (α(t) + θsα(t))) are

estimated. Equation 5 additionally controls for year fixed effects (θt) since cross-state variation in

the timing of women’s suffrage is not collinear with any one year fixed effect. There are two mea-

sures of suffrage treatment. The first follows Lott & Kenny (1999) and defines SUFFRAGEst,

that is, the “dosage” of suffrage, to be the product of the number of years since suffrage was imple-

mented and the share of adults who are female in a given year. Second, following Miller (2008),

SUFFRAGEst is defined to be a binary indicator equal to one in states and years with women’s

suffrage.42 Table 8 summarizes suffrage and summary statistics for the nation and the three-state

sample. Clearly, the three Southern states exhibited lower spending than the rest of the country,

and they had less exposure to suffrage rights prior to the 1920 Nineteenth Amendment. Although

the South may not be representative in terms of the levels of spending outcomes, Equation 5 tests

whether they were fundamentally different in terms of the impact of suffrage on the growth of

spending after suffrage.

Equation 5 is estimated with and without an interaction between the SUFFRAGEst variable

and the THREESTATES indicator. Results are reported in Table 9. Columns 1 and 3 report

the estimated coefficients for two different proxies of SUFFRAGEst without the interaction for

our sample states. The Lott & Kenny (1999) replication in Column 1 measures no statistically

significant increase in education expenditures while the Column 3 specification attributes to fe-

male suffrage a marginally significant 14.3 percent gain in educational spending, relative to linear

41Controls include the presence of a literacy test, secret ballot indicator, number of motor vehicle

registrations, log population density, fraction of the population rural, fraction of the population

black, fraction of the population older than 65, fraction of women working, fraction of the popula-

tion illiterate, fraction of the labor force in manufacturing, fraction of the population foreign-born,

and the average real manufacturing wage.

42These suffrage dosage measures are inappropriate to the exercise in the main text as all states

in our three-state sample were forced to grant women the right to vote by federal legislation in

1920.

29

state-specific time trends. The second coefficients of Columns 2 and 4 measure the difference in

post-suffrage spending in the three-state subsample, relative to the baseline impact in other states.

Neither interaction is statistically significant, indicating that state educational spending in these

three states responded no differently to suffrage than the rest of the nation.

C Robustness and Falsification Checks

The primary difference-in-difference identification strategy estimates the change in log per-pupil

education spending following 1921, with fixed effects to control for unobserved heterogeneity

within counties and years. In this section we present results from five robustness checks of this

empirical approach before turning to pre-treatment “effects” of the suffrage dosage.

C.1 Robustness Checks

Results for the five variations are listed in Table 10. Column 1 repeats baseline results from Table

5. Coefficients represent estimates of the impact of 19th Amendment dosage (POST ∗ DFct) on

outcomes listed in the leftmost column. Given the strong relationship between female population

shares and the overall black-white population ratio (see Table 2 and related discussion), in Column

2 we modify the Xct vector of controls to control for a quadratic rather than linear function of

the black-white population ratio. Point estimates, standard errors, and statistical significance are

nearly unchanged relative to the baseline model.

Next, we modify dependent variables to measure per-adult (voting age) spending outcomes

rather than per-pupil outcomes. Column 3 lists results for POST ∗ DFct coefficients when Equa-

tion 1 is estimated for spending per adult of voting age. The dependent variable denominator is

necessarily interpolated between census years, increasing measurement error. Results from our

preferred model, reported in the main text, utilize pupil counts that are available each year of the

panel. Column 3 coefficient estimates depart quantitatively from baseline results, but our inter-

pretation of the impact of the 19th Amendment is broadly consistent with inference drawn from

baseline results. Higher dosages of women’s suffrage from higher female population shares lead

to significantly higher education spending, benefitting white students more so than black students.

Column 4 lists results when we estimate a regression adjustment model for per-pupil outcomes,

30

which allows the impact of Xct covariates to change in the new regime (e.g., by letting the impact of

crop value to vary across pre-suffrage and post-suffrage years). Specifically, regression adjustment

ensures that the γ coefficients are not also incorporating the effects of control variables whose

impact may have changed in 1920. The regression adjustment estimating equation is as follows:

ln(Yct) = γ0 + γ1POSTt + γ2POSTtD F ct + Xctβ + POSTt(Xct − X̄)Ψ + θc + εct, (6)

where X̄ is a vector of means and other variables defined as before. In expectation, the average

treatment effect evaluated at the sample mean is (γ1 − γ0) + γ2DFct. Table 10 lists γ̂2 estimates

in Column 4. The impact of dosage on per-pupil spending overall is very similar with regression

adjustment, although the impact of dosage on white per-pupil spending is estimated to be much

larger than in the baseline model, leading to a much more negative impact on the ratio of black to

white per-pupil spending. Our baseline results, in that sense, are conservative.

The principal threat to our basic difference-in-difference identification strategy is the possibility

that areas with higher dosages of women’s suffrage – that is, areas with higher female population

shares – possess an inherent trend toward higher education spending regardless of women having

the right to vote. If so, a difference-in-difference estimator could detect significant but biased shifts

after 1920 in these locations.

One strategy to address this threat is to control for interactions between a linear time trend and

pre-“treatment” (pre-suffrage) observable variables, like so:

ln(Yct) = δ0 + δ1POST ∗ DFct + δ2D F ct + θc + θt + β1Xct + β2t ∗ X

1920 c εct, (7)

where X1920c is the vector of observed economic and population covariates as of 1920 and t is a

time trend.43 In addition to observables included in the main analysis (summarized at the bottom

of Table 3) we control for county-wide literacy rates among the population over ten years of age.44

If there are different time trends in high-dosage locations, and these trends are correlated with an

43See Hoynes & Schanzenbach (2009) and Carruthers & Wanamaker (2013) for other non-

experimental applications of this method. It is not feasible to replace X1920ct with D F ct in Equation

7 as the differential trend by dosage is precisely the object of interest.

44Haines, Michael R. and Inter-university Consortium for Political and Social Research (2010).

31

observable in X1920c , they will be accounted for in the t ∗ X1920c term in Equation 7. The parameter

δ1 is then estimated net of these trends.

Column 5 of Table 10 lists coefficient estimates for δ1 after controlling for these trends. The

overall impact of suffrage dosage on per-pupil spending is mitigated but still positive and statisti-

cally significant, and the ratio of nominal per-pupil black to white spending falls by roughly the

same percent as in baseline results.

Lastly, Column 6 lists results from the main Equation 1 specification with the addition of

controls for county-wide literacy. Literacy is a sensible proxy for countywide human capital, but

we omit this from the list of controls in our main analysis for two reasons. First, available literacy

data cover all persons over ten years of age, which may be endogenous to improving school quality.

Second, we only observe literacy as of 1910, 1920, and 1930, so data for the last ten years of the

panel must be extrapolated. Nevertheless, controlling for this imperfect measure of literacy has

almost almost no bearing on point estimates. Column 6 results are nearly identical to baseline

results in Column 1.

C.2 Pre-treatment and Falsification Tests

It remains possible that heterogeneous, unobserved, and endogenous trends are present and induced

increases in education spending in high-dose counties, even in the absence of women’s suffrage.

Though we cannot test this possibility directly, we can use pre-suffrage spending and population

data to (1) assess the importance of our key dosage proxy in determining trends in educational

spending prior to suffrage and (2) project spending outcomes as if the 19th Amendment never

happened. Specifically, we estimate the following for school years 1910 up to and including 1920,

the last fiscal year before district leaders faced an expanded electorate:

ln(Yct) = δ0 + δ1trend + δ2D F ct + δ3trend ∗ D

F ct + θc + βXct + εct, (8)

where DFct is the share of the voting-age population who are white females in a given year t (anal-

ogous to Table 5), trend is a linear time trend, θc is a county fixed effect, and Xct includes control

variables defined above (with the exception of secret ballot laws and Rosenwald school controls,

which were not present until after 1920, and with the addition of countywide literacy rates interpo-

lated between 1910 and 1920). Table 11 lists point estimates for δ1, δ2, and δ3 coefficients for our

32

four dependent variables of interest.

The first row of Table 11 results shows that pre-suffrage conditional trends in school spend-

ing were downward for black schools and insignificantly sloped for white schools. The second

row of estimates indicates that counties with higher shares of white voting-age females tended to

realize lower black and white school spending, although the log-sum of black and white spend-

ing exhibited no significant association with female population shares. Most important are results

for the interaction trend ∗ DFct. If high-dose counties were on a differentially upward spending

trajectory prior to suffrage, a coincidental acceleration of that trend around 1921 would manifest

as a spurious “effect” of suffrage dosage. However, prior to 1921, we observe no significant log

spending trends in high-dose counties. Point estimates for log spending outcomes are very small

and insignificant. Interestingly, the nominal ratio of black to white spending was on a downward

path in incrementally more female counties, although the 0.09 point estimate is a modest fraction

of our main result for this outcome (-0.57 per Table 5, Column 1).

To underscore the point that pre-suffrage conditions do not foretell post-suffrage outcomes,

Figure 4 plots actual spending outcomes (dots) against projections (lines) from predicted values of

Equation 8. Point estimates from the 1910-1920 model are fit to observed right-hand-side variables

from 1921 and later. Solid lines trace average predictions (unweighted) across counties. The

pre-suffrage model does a very poor job of fitting 1921 and later outcomes, greatly understating

the path of each outcome’s realized time series.45 Which is to say that – based on observable

county characteristics, prevailing trends, and county fixed effects – post-suffrage school spending

significantly exceeded expectations.

To summarize robustness and falsification exercises, we find that results from the principal

identification strategy are robust – with important caveats regarding economic significance – to de-

nominator changes in the dependent variable and two alternative treatments of time-varying trends

in the effect of other control variables. Moreover, pre-suffrage estimates of spending outcomes

suggest that counties with more females were not on differential spending trajectories prior to the

19th Amendment. Overall, the evidence indicates that the 19th Amendment is responsible in part

45Allowing for higher-ordered time trends exacerbates the poor performance of the pre-suffrage

model.

33

for post-1920 gains in educational spending, and that white students benefitted more than black

students.

D Event Study Estimates

Other econometric tools are available to identify the effect in question in this paper. In particular,

the post-1921 differential impact of women’s suffrage dosage can be identified using an event study

estimator. The event study estimator is conceptually similar to a difference-in-difference estimator,

but the treatment effect is estimated in each year after the event in question. Pre-treatment effects

are estimated as a falsification test.

In this context, the event study estimator traces the impact of suffrage dose over time by re-

placing the POSTt ∗ DFct in Equation 1 with a set of interactions between year fixed effects and

DFct. For dosage, we utilize time-varying white female population shares (i.e., the dosage proxy

from Table 5 results) and estimate the following:

Yct = δ0 + 1940∑

κ=1910

δκ1(t = κ) ∗ DFcκ + δ2D F ct + θc + θt + βXct + εct, (9)

where variables are defined as for Equation 1 above. Each δκ measures the suffrage treatment

effect in that year. County fixed effects and year fixed effects account for variation in outcomes

that are constant across counties or years.

We present the point estimates for δ1910 − δ1940 when Yct represents per pupil expenditures in

Panel A of Figure 5. Estimates are noisy, but the treatment effect of suffrage hovers around zero for

1910-1920 before beginning a steady upward climb. The treatment effect is consistently positive

and frequently statistically significant after 1925, and reflects the increasingly important role of

suffrage over time (also documented in Tables 4 - 6).

Estimates of the suffrage treatment effect on white expenditures (Panel B of 5) follow a pattern

similar to that for overall expenditures with statistically significant treatment effects in the 1930s.

The annual treatment effect on black expenditures increases more slowly after 1921 and tapers off

more quickly in the 1930s (Panel C). Point estimates in each case match those in the main results,

which reflect and average effect across 1921-1940.

Pre-1920 estimates serve as an important falsification test. Namely, as Figure 4 makes clear,

school expenditures suffered substantially during the World War I era. If these reductions are some-

34

how related to the suffrage proxies incorporated in our analysis, the resulting estimates, although

unbiased, are measuring not the impact of suffrage but, rather, the impact of having a particular

population structure in the World War I years as captured by the proxies. Clearly this is not the

effect of interest.

Contrary to what we might expect if the World War I impact was correlated with suffrage

dosage proxies, we observe no significant “effect” of suffrage in the pre-1920 years. The point

estimates hew close to zero and show no particular deviation in the war years. This is consistent

with the interpretation that the relationship between spending and the suffrage dosage proxies is

attributable to suffrage per se.

In short, event study estimates, although noisier than our main results, reflect a similar pattern

of suffrage treatment effects on school expenditures in these three southern states.

E Duration of Impact - Extension

In Section 4.3 of the main text, we discuss results showing that the impacts of suffrage were

larger in later years of the panel. This begs the question of why female enfranchisement should

have taken so much time to impact local public education, since the potential electorate shifted

sharply beginning with the 1920-1921 school year. Even if the new electorate proved to be a

strong and credible threat, policymakers may have been constrained in their ability to quickly shift

large amounts of public expenditures toward school budgets.

We posit that counties where local revenues (from property, poll, and miscellaneous local taxes)

account for a larger share of school expenditures just prior to suffrage would have been able to more

flexibly respond to enfranchisement in the years following. With this idea in mind, we estimate a

differenced version of Equation 1 and replace the dosage proxy with the ratio of locally-sourced

school revenues to total school expenditures (as in other parts of the study, by race) to better

understand the relationship between local control and growth in spending items of interest.

The local tax data on hand are not ideal for testing the overall revenue response to suffrage,46

46We do not observe the total value of all local tax collections, but only the portion appearing in

school budgets. Furthermore, local tax revenue data are sporadically and inconsistently reported

in the state reports we have transcribed.

35

but nevertheless, serve as valuable (albeit noisy) measures of the tax base readily available to public

schools.

Specifically, we estimate the following:

∆Yct = δ̃0 + δ̃1∆D F c + δ̃2∆POST ∗ D

F c + β∆Xct + εct, (10)

where DFc is the (0-100) percent of school expenditures accounted for by local revenues and

other variables are defined as before. Point estimates for δ̃1 and δ̃2 are in Table 12. The first

coefficient of interest, δ̃1, estimates the pre-suffrage relationship between spending growth and

incremental increases in the local control of school spending. Results for δ̃1 in Column 1 indicate

that counties with more local control of school spending were generally on a declining spending

path prior to suffrage. Column 2 lists point estimates for δ̃2, a gage of whether incrementally

greater shares of local funds are associated with deviations from underlying Column 1 trends.

Indeed, growth in per-pupil spending (total, black, and white) appears to be steeper after suffrage

in counties with more local control of school budgets. Thus, the new electorate may have taken

some time to be fully reflected in school resources because local leaders were constrained by

limited discretionary funds.

36

Tables and Figures

FIGURE 1: Trends in per capita public educational expenditures (1982-1984 dollars)

Nationwide

4 0

6 0

8 0

1 0

0 1

2 0

1 4

0 P

e r

ca p

ita p

u b

lic e

d u

ca tio

n a

l e xp

e n

d itu

re s

1910 1920 1930 1940 year

Three Southern States

5 0

1 0

0 1

5 0

2 0

0 2

5 0

P e

r p

u p

il p

u b

lic e

d u

ca tio

n a

l e xp

e n

d itu

re s

1910 1920 1930 1940 year

Source: Nationwide: Authors’ calculations, Carter et al. (2006). Three Southern States: Authors’ calculations and annual reports of states’ Department of Education or equiv- alent office in AL, GA, and SC.

37

TABLE 1: Distribution of Revenue Receipts for U.S.

Schools, 1919/1920-1969/70

Federal State Local

1919-20 0.3 16.5 83.2

1929-30 0.4 16.9 82.7

1939-40 1.8 30.3 68.0

1949-50 2.9 39.8 57.3

1959-60 4.4 39.1 56.5

1969-70 8.0 39.9 52.1

Notes: Percent of total revenue re- ceipts for United States public schools emanating from each fiscal source. Source: Snyder et al. (2008)

FIGURE 2: Trends in per pupil expenditures, teachers per pupil, average teacher salaries, and enrollment per capita, 1910-1940

A. Real school expenditures per pupil B. Teachers per pupil

0 2 0

4 0

6 0

1910 1920 1930 1940 year

White Black

2 0

2 5

3 0

3 5

1910 1920 1930 1940 year

White Black

C. Real average teacher salary D. Enrollment per capita (000s)

2 0 0

4 0 0

6 0 0

8 0 0

1 0 0 0

1910 1920 1930 1940 year

White Black

.4 5

.5 .5

5 .6

1910 1920 1930 1940 year

White Black

Source: Authors’ calculations and numerous annual reports of states’ Department of Education or equivalent office. County-level data are averaged across years without weights. Salary and expenditure statistics are in inflation-adjusted 1925 dollars.

38

TABLE 2: Correlates of 1920 white female population shares

Observable county characteristic Coefficient (st. err.)

Total Population (000s) -0.950 (0.79)

Crop Value Per Capita -0.0106* (0.0053)

Percent of Land Devoted to Agriculture (0-100) 0.0459** (0.022)

Black-white Ratio (x100) -7.97*** (0.68)

Average Annual Manufacturing Earnings 0.000114 (0.0027)

Percent of Adults in Manufacturing (0-100) -0.160** (0.063)

Observations (counties) 236 R2 0.75

Source: Authors’ calculations and numerous annual reports of three Southern states’ Department of Education or equivalent office. The ta- ble lists results of a simple regression of the percent of counties’ 1920 population who are white females (0-100 percent) against other observ- able features of counties in 1920, restricted to counties with observable expenditures per pupil in 1920. Robust standard errors are in parenthe- ses below each coefficient. *** indicates statistical significance at 99% confidence (with respect to zero), ** at 95%, and * at 90%.

FIGURE 3: Estimates of female voter turnout rates, 1920-1940

20.2

26.1 26.1

29.4 30.7

29

0 1

0 2

0 3

0 F

e m

a le

t u

rn o

u t

ra te

e st

im a

te s

1920 1924 1928 1932 1936 1940

Source: Nationwide: Authors’ calculations from Clubb et al. (2006) for AL, GA, LA, SC, NC.

39

TABLE 3: County schooling resources and population summary statistics, 1910-1940

(1) (2)

Mean Standard deviation

Dependent variables ln(per-pupil educational expenditures) 2.94 (0.57) ln(per-pupil white educational expenditures) 3.40 (0.65) ln(per-pupil black educational expenditures) 1.67 (0.66) Ratio of black to white per-pupil expenditures (x100) 23.83 (26.13) ln(ave. white teacher salary) 6.43 (0.44) ln(ave. black teacher salary) 5.24 (0.75) White teachers per 1000 pupils 31.39 (13.36) Black teachers per 1000 pupils 21.98 (9.48)

Suffrage dosage measures Percent of electorate white female 30.21 (9.85) Percent of 1920 electorate white female 29.85 (9.27) Percent of 1920 electorate female voters 7.00 (4.07)

Socioeconomic control variables Population (in thousands) 30.51 (35.84) Crop value per capita 115.31 (55.20) Percent of land devoted to agriculture 65.0 (15.8) Black-white population ratio 1.08 (1.03) Secret ballot (Georgia 1922 and later) 0.33 (0.47) Average annual manufacturing earnings 635.82 (189.01) Percent of adults in manufacturing 6.71 (6.62) New Rosenwald classrooms (non-zero from 1921-1933) 0.36 (1.60)

n (county-years) 7,148

Source: Authors’ calculations and numerous annual reports of three Southern states’ De- partment of Education or equivalent office.

40

T A

B L

E 4

: E

st im

at ed

ch an

ge s

in lo

ca le

du ca

ti on

al sp

en di

ng pe

r pu

pi l,

af te

r th

e N

in et

ee nt

h A

m en

dm en

t D

os ag

e P

ro xy

: P

er ce

nt W

hi te

F em

al e

P op

ul at

io n

in 19

20

20 -Y

ea r

W in

do w

(t hr

ou gh

19 40

) 19

21 -1

92 5

19 26

-1 93

0 19

31 -1

93 5

19 36

-1 94

0

(1 )

(2 )

(3 )

(4 )

(5 )

(6 )

(7 )

C oe

ffi ci

en to

n P O S T t *D

F c t

ln (A

ll S

pe nd

in g

pe r

P up

il )

0. 00

69 **

* 0.

00 86

** *

0. 00

65 **

* 0.

00 27

0. 00

69 **

* 0.

01 0*

** 0.

01 4*

** (0

.0 01

7) (0

.0 01

7) (0

.0 01

5) (0

.0 01

9) (0

.0 02

0) (0

.0 02

0) (0

.0 02

2)

R at

io of

B la

ck to

W hi

te S

pe nd

in g

P er

P up

il -0

.5 7*

** -0

.5 6*

** -0

.6 1*

** -0

.4 6*

* -0

.5 3*

* -0

.7 1*

** -0

.7 3*

** (0

.1 6)

(0 .1

6) (0

.1 3)

(0 .1

8) (0

.2 1)

(0 .1

6) (0

.1 6)

ln (W

hi te

S pe

nd in

g pe

r P

up il )

0. 00

54 **

* 0.

00 70

** *

0. 00

73 **

* 0.

00 10

0. 00

51 **

0. 00

81 **

* 0.

01 4*

** (0

.0 01

8) (0

.0 01

8) (0

.0 01

6) (0

.0 01

9) (0

.0 02

2) (0

.0 02

2) (0

.0 02

4)

ln (B

la ck

S pe

nd in

g pe

r P

up il )

0. 00

27 0.

00 43

** 0.

00 04

3 0.

00 22

0. 00

63 **

0. 00

18 -0

.0 01

5 (0

.0 02

1) (0

.0 01

9) (0

.0 02

0) (0

.0 02

3) (0

.0 02

7) (0

.0 03

0) (0

.0 03

0)

Y ea

r F

ix ed

E ff

ec ts

Y N

Y Y

Y Y

Y S

ta te

-Y ea

r F

ix ed

E ff

ec ts

N Y

N N

N N

N In

te rp

ol at

ed C

on tr

ol s

Y Y

N Y

Y Y

Y O

bs er

va ti

on s

7, 14

7 7,

14 7

7, 14

7 7,

14 7

7, 14

7 7,

14 7

7, 14

7 N

um be

r of

co un

ti es

23 5

23 5

23 5

23 5

23 5

23 5

23 5

N ot

es :

C oe

ffi ci

en te

st im

at es

of E

qu at

io ns

1 (C

ol um

ns 1-

3) an

d 2

(C ol

um ns

4- 7)

w it

h pe

r pu

pi l

ex pe

nd it

ur es

(i n

na tu

ra ll

og s)

or th

e ra

ti o

of bl

ac k

to w

hi te

pe r-

pu pi

le xp

en di

tu re

s as

th e

de pe

nd en

tv ar

ia bl

e. C

lu st

er -r

ob us

t( by

co un

ty )

st an

da rd

er ro

rs ar

e in

pa re

nt he

se s

be lo

w ea

ch co

ef fi

ci en

t. **

* in

di ca

te s

st at

is ti

ca ls

ig ni

fi ca

nc e

at 99

% co

nfi de

nc e

(w it

h re

sp ec

tt o

ze ro

), **

at 95

% ,a

nd *

at 90

% .

41

T A

B L

E 5

: E

st im

at ed

ch an

ge s

in lo

ca le

du ca

ti on

al sp

en di

ng pe

r pu

pi l,

af te

r th

e N

in et

ee nt

h A

m en

dm en

t D

os ag

e P

ro xy

: P

er ce

nt W

hi te

F em

al e

P op

ul at

io n

in E

ac h

Y ea

r

20 -Y

ea r

W in

do w

(t hr

ou gh

19 40

) 19

21 -1

92 5

19 26

-1 93

0 19

31 -1

93 5

19 36

-1 94

0

(1 )

(2 )

(3 )

(4 )

(5 )

(6 )

(7 )

C oe

ffi ci

en to

n P O S T t *D

F c t

ln (A

ll S

pe nd

in g

pe r

P up

il )

0. 00

70 **

* 0.

00 85

** *

0. 00

63 **

* 0.

00 26

0. 00

72 **

* 0.

01 0*

** 0.

01 4*

** (0

.0 01

6) (0

.0 01

6) (0

.0 01

4) (0

.0 01

8) (0

.0 02

0) (0

.0 01

9) (0

.0 02

0)

R at

io of

B la

ck to

W hi

te S

pe nd

in g

P er

P up

il -0

.5 3*

** -0

.5 2*

** -0

.5 4*

** -0

.4 2*

** -0

.5 0*

** -0

.6 7*

** -0

.6 8*

** (0

.1 6)

(0 .1

6) (0

.1 2)

(0 .1

7) (0

.2 0)

(0 .1

6) (0

.1 6)

ln (W

hi te

S pe

nd in

g pe

r P

up il )

0. 00

55 **

* 0.

00 70

** *

0. 00

67 **

* 0.

00 08

4 0.

00 56

** *

0. 00

83 **

* 0.

01 3*

** (0

.0 01

8) (0

.0 01

8) (0

.0 01

5) (0

.0 01

9) (0

.0 02

2) (0

.0 02

2) (0

.0 02

3)

ln (B

la ck

S pe

nd in

g pe

r P

up il )

0. 00

32 0.

00 50

** *

0. 00

10 0.

00 27

0. 00

67 **

* 0.

00 24

-0 .0

01 0

(0 .0

02 0)

(0 .0

01 9)

(0 .0

01 8)

(0 .0

02 2)

(0 .0

02 6)

(0 .0

02 8)

(0 .0

02 7)

Y ea

r F

ix ed

E ff

ec ts

Y N

Y Y

Y Y

Y S

ta te

-Y ea

r F

ix ed

E ff

ec ts

N Y

N N

N N

N In

te rp

ol at

ed C

on tr

ol s

Y Y

N Y

Y Y

Y O

bs er

va ti

on s

7, 14

7 7,

14 7

7, 14

7 7,

14 7

7, 14

7 7,

14 7

7, 14

7 N

um be

r of

co un

ti es

23 5

23 5

23 5

23 5

23 5

23 5

23 5

N ot

es :

C oe

ffi ci

en te

st im

at es

of E

qu at

io ns

1 (C

ol um

ns 1-

3) an

d 2

(C ol

um ns

4- 7)

w it

h pe

r pu

pi l

ex pe

nd it

ur es

(i n

na tu

ra ll

og s)

or th

e ra

ti o

of bl

ac k

to w

hi te

pe r-

pu pi

le xp

en di

tu re

s as

th e

de pe

nd en

tv ar

ia bl

e. C

lu st

er -r

ob us

t( by

co un

ty )

st an

da rd

er ro

rs ar

e in

pa re

nt he

se s

be lo

w ea

ch co

ef fi

ci en

t. **

* in

di ca

te s

st at

is ti

ca ls

ig ni

fi ca

nc e

at 99

% co

nfi de

nc e

(w it

h re

sp ec

tt o

ze ro

), **

at 95

% ,a

nd *

at 90

% .

42

T A

B L

E 6

: E

st im

at ed

ch an

ge s

in lo

ca le

du ca

ti on

al sp

en di

ng pe

r pu

pi l,

af te

r th

e N

in et

ee nt

h A

m en

dm en

t D

os ag

e P

ro xy

: F

em al

e V

ot er

P er

ce nt

ag e

in 19

20

20 -Y

ea r

W in

do w

(t hr

ou gh

19 40

) 19

21 -1

92 5

19 26

-1 93

0 19

31 -1

93 5

19 36

-1 94

0

(1 )

(2 )

(3 )

(4 )

(5 )

(6 )

(7 )

C oe

ffi ci

en to

n P O S T t *D

F c t

ln (A

ll S

pe nd

in g

pe r

P up

il )

0. 00

99 **

0. 01

82 **

* 0.

01 15

** -0

.0 01

5 0.

01 15

** 0.

01 61

** *

0. 02

21 **

* (0

.0 04

7) (0

.0 05

1) (0

.0 04

5) (0

.0 06

2) (0

.0 05

8) (0

.0 05

1) (0

.0 05

6)

R at

io of

B la

ck to

W hi

te S

pe nd

in g

P er

P up

il -1

.2 9*

* -1

.3 4*

* -1

.5 1*

** -1

.0 7*

-1 .1

7 -1

.6 8*

** -1

.4 6*

** (0

.5 5)

(0 .5

7) (0

.4 9)

(0 .5

9) (0

.7 5)

(0 .5

3) (0

.5 4)

ln (W

hi te

S pe

nd in

g pe

r P

up il )

0. 00

84 0.

01 47

** 0.

01 41

** *

-0 .0

04 0

0. 00

83 0.

01 43

** 0.

02 52

** *

(0 .0

05 2)

(0 .0

05 7)

(0 .0

05 1)

(0 .0

05 2)

(0 .0

06 3)

(0 .0

06 9)

(0 .0

07 3)

ln (B

la ck

S pe

nd in

g pe

r P

up il )

-0 .0

07 0

0. 00

45 -0

.0 10

2* -0

.0 09

4 0.

00 27

-0 .0

11 0

-0 .0

11 7

(0 .0

07 0)

(0 .0

06 5)

(0 .0

06 0)

(0 .0

07 5)

(0 .0

11 5)

(0 .0

09 9)

(0 .0

09 9)

Y ea

r F

ix ed

E ff

ec ts

Y N

Y Y

Y Y

Y S

ta te

-Y ea

r F

ix ed

E ff

ec ts

N Y

N N

N N

N In

te rp

ol at

ed C

on tr

ol s

Y Y

N Y

Y Y

Y O

bs er

va ti

on s

7, 14

7 7,

14 7

7, 14

7 7,

14 7

7, 14

7 7,

14 7

7, 14

7 N

um be

r of

co un

ti es

23 5

23 5

23 5

23 5

23 5

23 5

23 5

N ot

es :

C oe

ffi ci

en t

es ti

m at

es of

E qu

at io

ns 1

w it

h pe

r pu

pi l

ex pe

nd it

ur es

(i n

na tu

ra l

lo gs

) or

th e

ra ti

o of

bl ac

k to

w hi

te pe

r- pu

pi l

ex pe

nd it

ur es

as th

e de

pe nd

en t

va ri

ab le

. B

oo ts

tr ap

pe d

st an

da rd

er ro

rs ar

e in

pa re

nt he

se s.

** *

in di

ca te

s st

at is

ti ca

l si

gn ifi

ca nc

e at

99 %

co nfi

de nc

e (w

it h

re sp

ec t

to ze

ro ),

** at

95 %

,a nd

* at

90 %

.

43

TABLE 7: Estimated changes in teacher salaries and teachers per pupil after the Nineteenth Amendment

Dosage Proxy: Percent White Female Population in Each Year

Observations Counties 20-Year Window (through 1940)

(1) (2) (3)

Coefficient on POSTt*DFct

ln(ave. white teacher salary) 7,115 235 0.003* 0.004*** 0.005*** (0.002) (0.002) (0.001)

ln(ave. black teacher salary) 7,115 235 -0.025*** -0.015*** -0.018*** (0.006) (0.005) (0.004)

white teachers per 1000 pupils 7,145 235 0.086 0.087 0.010 (0.079) (0.082) (0.063)

black teachers per 1000 pupils 7,145 235 0.076 0.087 0.002 (0.071) (0.072) (0.054)

Year Fixed Effects Y N Y State-Year Fixed Effects N Y N Interpolated Controls Y Y N

Notes: Coefficient estimates of Equations 1 with teachers per pupil, or the log of inflation-adjusted average teacher salary as the dependent variable. Cluster-robust (by county) standard errors are in parentheses below each coefficient. *** indicates statistical significance at 99% confidence (with respect to zero), ** at 95%, and * at 90%.

TABLE 8: Spending and suffrage summary statistics, three-state sample versus all states (1870 - 1940)

(1) (2)

Variable All states Three-state subsample

ln(total expenditures) 2.816 2.052 (0.963) (0.962)

ln(education expenditures) 1.312 0.790 (1.212) (1.313)

Y EARSINCEst∗Percent Female 0.200 0.166 (0.238) (0.238)

Female suffrage (0,1) 0.415 0.329 (0.493) (0.471)

n (state-years) 1,882 122

Source: Authors’ calculations and Lott & Kenny (1999). Cluster-robust (by state) standard deviations are in parentheses. Y EARSINCEst is equal to the number of since elapsed since full suffrage. pctfem is the share of the adult population that is female. SUFFst is an indicator for full suffrage in state s in year t, which is achieved for all states by 1920.

44

TABLE 9: Estimated changes in state education expenditures after women’s suffrage, three-state sample versus all states

(1) (2) (3) (4)

Suffrage proxy Y EARSINCEst*Percent female SUFFst

Time trend controls No No Yes Yes

SUFFRAGEst -0.132 -0.121 0.143* 0.150* (0.178) (0.179) (0.086) (0.087)

SUFFRAGEst ∗ THREESTATEs -0.220 -0.160 (0.383) (0.230)

Observations 1,827 1,827 1,827 1,827 Adjusted R-squared 0.71 0.71 0.78 0.78

Notes: Selected coefficient estimates from Equation 5 for state-level public expenditures. Y EARSINCEst is equal to the number of since elapsed since full suffrage. SUFFst is an indi- cator for full suffrage in state s in year t, which is achieved for all states by 1920. Time trend controls (α(t) + θs ∗ α(t) in Equation 5) are excluded in Columns 1 and 2. Cluster-robust (by state) standard errors are in parentheses below each coefficient. Additional controls include socioeconomic variables listed in the text, state fixed effects, and year fixed effects. *** indicates statistical significance at 99% confidence (with respect to zero), ** at 95%, and * at 90%.

45

TABLE 10: Estimated changes in local educational spending after the Nineteenth Amendment: Robustness Checks

Dosage Proxy: Female Voter Percentage in Each Year

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

Model Quadratic black-white Spending Regression t ∗ X1920ct Literacy

Baseline ratio control per adult adjustment interactions controls

ln(All Spending) 0.0070*** 0.0069*** 0.0040** 0.0078*** 0.0049*** 0.0070*** (0.0016) (0.0016) (0.0017) (0.0029) (0.0019) (0.0016)

Ratio of Black to White Spending -0.53*** -0.54*** -1.83* -0.82*** -0.49*** -0.53*** (0.16) (0.15) (1.12) (0.25) (0.19) (0.16)

ln(White Spending) 0.0055*** 0.0057*** 0.0036* 0.0103*** 0.0052** 0.0054*** (0.0018) (0.0018) (0.0019) 0.0035) (0.0021) (0.0018)

ln(Black Spending) 0.0032 0.0029 -0.0052* 0.0044 0.0069** 0.0031 (0.0020) (0.0020) (0.0029) (0.0043) (0.0029) (0.0020)

Year Fixed Effects Y Y Y Y Y Y State-Year Fixed Effects N N N N N N Interpolated Controls Y Y Y Y Y Y Observations 7,134 7,134 7,134 7,134 7,134 7,134 Number of Counties 232 232 232 232 232 232

Notes: Baseline results for the impact of suffrage dosage on per-pupil spending and the black-white per-pupil ratio (POSTtDFct from Equation 1) are in Column 1. Column 2 adds the square of each county’s interpolated black-white ratio to the list of controls. Column 3 substitutes per-pupil dependent variables for per-adult of voting age dependent variables. Column 4 lists POSTtDFct coefficient estimates from regression adjustment (Equation 6). Column 5 lists POSTtD

F ct coefficient estimates from Equation 7, which modifies the basic difference-in-difference model by including

controls for interactions between a linear time trend and 1920 covariates. Column 6 adds over-10 literacy rates to the lost of controls. *** indicates statistical significance at 99% confidence (with respect to zero), ** at 95%, and * at 90%.

46

TABLE 11: Spending Trends and Female Population Shares Prior to Suffrage

(1) (2) (3) (4)

Ratio of All Black to White White Black

Outcome Spending Spending Spending Spending

trend -0.0087* -1.0189** 0.0033 -0.0214*** (0.0051) (0.5093) (0.0060) (0.0066)

DFct -0.0096 -0.3471 -0.0153** -0.0264*** (0.0066) (0.5325) (0.0070) (0.0067)

trend ∗ DFct -0.0003 -0.0843** 0.0003 -0.0003 (0.0003) (0.0401) (0.0003) (0.0005)

Observations 2,540 2,540 2,540 2,540 Number of Counties 234 234 234 234

Notes: Selected results from Equation 8, a pre-suffrage model of school ex- penditures and population characteristics. *** indicates statistical significance at 99% confidence (with respect to zero), ** at 95%, and * at 90%.

TABLE 12: Equation 10 Results: Local Control and Growth in School Spending After Suffrage

(1) (2)

Coefficient on DFct POSTt*D F ct

ln(All Spending per Pupil) -4.6E-04*** 7.5E-04*** (07.0E-05) (7.0E-05)

Ratio of Black to White Spending per Pupil 2.4E-03 1.4E-03 (0.009) (0.013)

ln(White Spending per Pupil) -4.1E-04*** 7.0E-04*** (1.1E-04) (1.0E-04)

ln(Black Spending per Pupil) -5.9E-04*** 1.1E-03*** (9.0E-05) (1.1E-04)

Observations 6,775 Number of Counties 227

Notes: The table lists results for the impact of local control (ratio of locally- sourced school revenues to total school expenditures, by race) on growth in per- pupil spending and the black-white per-pupil ratio. Point estimates for DFc in Equation 10), the pre-suffrage trend in counties with incrementally more local control are in Column 1. Point estimates for POSTtDFc , representing the post-suffrage deviation from underlying trends, are in Column 2. *** indicates statistical significance at 99% confidence (with respect to zero), ** at 95%, and * at 90%.

47

FIGURE 4: Actual Spending Outcomes Versus Projected Outcomes Based on Pre-Suffrage Estimates

Real school expenditures per pupil

2 2 .5

3 3 .5

1910 1920 1930 1940 year

Actual Spending Projected Spending

Real white school expenditures per pupil

2 .5

3 3 .5

4

1910 1920 1930 1940 year

Actual Spending Projected Spending

Real black school expenditures per pupil

1 1 .5

2 2 .5

1910 1920 1930 1940 year

Actual Spending Projected Spending

Source: Authors’ calculations and numerous annual reports of states’ Department of Education or equivalent office. County-level data are averaged across years without weights. Figures plot actual spending outcomes (dots) against predicted values (lines) from the pre-suffrage spending model (Equation 8).

48

FIGURE 5: Event Study Coefficients and 95% Confidence Intervals

A. Real school expenditures per pupil

− .0

1 0

.0 1

.0 2

10 20 30 40 year

B. Real white school expenditures per pupil

− .0

1 0

.0 1

.0 2

10 20 30 40 year

C. Real black school expenditures per pupil

− .0

1 0

.0 1

.0 2

10 20 30 40 year

Source: Authors’ calculations and numerous annual reports of states’ Department of Education or equivalent office. Estimates from Equation 9 with 95 percent confidence intervals.

49