Full text of Working Papers (Federal Reserve Bank of Chicago) : The Impact of Rosenwald Schools on Black Achievement, Working Paper 2009-26

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Federal Reserve Bank of Chicago

The Impact of Rosenwald Schools on
Black Achievement
Daniel Aaronson and Bhashkar Mazumder

REVISED
September, 2011
WP 2009-26

THE IMPACT OF ROSENWALD SCHOOLS ON BLACK ACHIEVEMENT

Daniel Aaronson
Federal Reserve Bank of Chicago

Bhashkar Mazumder
Federal Reserve Bank of Chicago

September, 2011
 
Abstract
The Black-White gap in schooling among Southern-born men narrowed sharply between the World Wars.
From 1914 to 1931, nearly 5,000 schools were constructed as part of the Rosenwald Rural Schools
Initiative. Using Census data and World War II records, we find that the Rosenwald program accounts for
a sizable portion of the educational gains of rural Southern Blacks. We find significant effects on school
attendance, literacy, years of schooling, cognitive test scores, and Northern migration. The gains are
highest in the most disadvantaged counties, suggesting that schooling treatments have the largest impact
among those with limited access to education.

 
 
 
 
We thank Jesse Smith and Beth Howse of Fisk University for helping us obtain the Rosenwald data,
making the archives available to us, and answering our many questions; the Minnesota Population Center
and Joe Ferrie for making available an early version of the 1930 5 percent IPUMS sample; Joe Ferrie for
sharing his discovery of the AGCT test score data; David Benson, Jon Davis, Shani Schechter, and Zach
Seeskin for their valuable research assistance; Derek Neal and Bob Margo for very helpful conversations;
and seminar participants at a number of universities and conferences for their comments. The views
expressed in this paper are not necessarily those of the Federal Reserve Bank of Chicago or the Federal
Reserve System.

1

I. Introduction
Since the path breaking work of Schultz (1961) and Becker (1964), it has been well-recognized
that investment in human capital is a primary vehicle for economic development. Indeed, over the last 50
years, developing countries have made the improvement of basic education a principal goal in their effort
to enhance living standards. Yet, despite substantial progress, lack of access to schools and poor school
quality remains a persistent problem (UN Millennium Project 2005). For example, in many developing
countries, schools often lack basic infrastructure like sanitation (UNICEF 2011) and educational
equipment like textbooks, blackboards and desks (Glewwe and Kremer 2006) and rates of teacher
absenteeism are often very high (Banerjee and Duflo, 2006). Some of these problems can be traced to
ineffective or corrupt governments (Lewis and Pettersson 2009), exacerbated by a lack of political
influence among the rural poor.

There is also some uncertainty about whether providing more

educational inputs improves learning and other long-term outcomes (Glewwe and Kremer 2006). In this
paper, we present new evidence based on a large American schooling intervention from nearly a century
ago that may shed light on current efforts to improve school access and school quality in developing
countries today.
At the turn of the 20th century, the educational opportunities available to rural Blacks living in the
segregated American South were quite similar to what is available to the rural poor in many countries
today: inadequate school buildings, classrooms, and equipment. Moreover, White-run public institutions
were not held accountable for these failures since Blacks lacked political representation.

As a

consequence, Blacks born in the South between 1880 and 1910 completed three fewer years of schooling
than their White counterparts. While both groups made absolute gains, Blacks experienced no relative
progress over this thirty year period.

2

However, for cohorts born during a relatively short period between the World Wars, the Southern
racial education gap improved dramatically (see Figure 1).1 Within a generation, the racial gap in the
South declined to well under a year and was comparable in size to the racial gap in the North, which
remained roughly unchanged for cohorts born between 1880 and 1940.
The Rosenwald Rural Schools Initiative was a program explicitly designed to narrow racial
schooling gaps in the South during this period of Black relative progress.2 The Initiative was the result of
a collaboration between Booker T. Washington, the principal of the Tuskegee Institute in Alabama, and
Chicago businessman and philanthropist Julius Rosenwald. The two men developed a matching grant
program that, between 1913 and 1931, facilitated the construction of almost 5,000 schoolhouses for
Southern rural Black children. By the time the program ended, we estimate that approximately 36 percent
of Southern rural Blacks of school-age could have attended a Rosenwald school.
In addition to making schooling more accessible, the program represented a sea change in the
quality of schools. The buildings were constructed based on modern designs that ensured adequate
lighting, ventilation and sanitation. Classrooms were required to be fully equipped with books, chairs,
desks, blackboards and other materials to ensure an adequate learning environment. A number of other
initiatives -- including minimum teacher salaries, newly built teacher homes, and training programs often
in concert with other philanthropic efforts like the Jeannes Fund3 -- were introduced to recruit and prepare
teachers. While historical data on these dimensions of school quality unfortunately do not exist,4 we can
examine the overall impact of these various programs on long-term measures of human capital.

1

For research on the Black-White schooling and income gap during the 20th century, see Smith (1984), Smith and
Welch (1989), Margo (1990), Donohue and Heckman (1991), Collins and Margo (2006), and Neal (2006).
2
Donohue, Heckman, and Todd (2002) discuss the role of Northern Philanthropists, and Rosenwald in particular, on
Black school attendance in the early 20th century. Card and Krueger (1992) study the effects of improvements in
school quality experienced by Blacks from 1915 to 1966, but do not explicitly look at the effects of the Rosenwald
schools. For historical descriptions of the Rosenwald Rural Schools Initiative, see McCormick (1934), Embree
(1936), Ascoli (2006), and Hoffschwelle (2006).
3
The Jeannes Fund, established in 1907, primarily funded supervisors who helped train teachers in Black rural and
urban Southern schools. Underscoring the close link to the Rosenwald School Initiative, Booker T. Washington
served on the Jeannes Fund Board.
4
As we discuss later, available measures that have been used in prior research to proxy for school quality (e.g.
school expenditures, teacher counts) will generally not capture the school quality improvements of the Rosenwald
schools and can lead to misleading inferences regarding the effects of the program.

3

We do this by linking newly uncovered records of the location and the timing of the Rosenwald
schools to large samples drawn from the Decennial Censuses and World War II enlistment records. Our
main finding is that rural Black students with access to Rosenwald schools completed over a full year
more education than rural Black students with no access to Rosenwald schools, a magnitude that, in the
aggregate, explains close to 40 percent of the observed Black-White convergence in educational
attainment in the South for cohorts born between 1910 and 1925. We find similarly large effects on
literacy rates and evidence that exposure to the schools increased rates of migration to the North among
rural Blacks, potentially fueling further economic gains. Accounting for the full cost of building and
maintaining the schools, we estimate that the additional human capital acquisition generated by the
Rosenwald schools implies an internal real rate of return of about 7 to 9 percent.
Our estimates also provide insight into the relative importance of school quality versus school
access in explaining the effects of the program. We estimate that the program increased the rate of school
attendance by roughly 5 percentage points, well below the 36 percent of students who could have
attended a Rosenwald school in 1932. Some simple calculations suggest that about one-third of the
overall improvement in years of completed schooling is due to an increase in the fraction of students
attending school. The remaining two-thirds is due to gains in schooling among students who would have
attended school even in the absence of the Rosenwald program. This suggests that the bulk of the gain in
human capital investment is likely due to improvements in school quality which would have led many
students to stay in school longer than they would have in a pre-Rosenwald school.5
A unique aspect of our analysis is that we use a measure of cognitive ability, the Army General
Classification Test (AGCT), a precursor to the modern Armed Forces Qualifying Test (AFQT). We find
that childhood access to the schools raised the AGCT scores of rural Blacks by about 0.2 to 0.45 standard
deviations. By comparison, the Black-White test score gap among World War II enlistees was about 1.1
5

It is difficult to attribute exactly how much of the gain in completed schooling is due to school quality. Some of
the gain in years of schooling among students who would have attended school in the absence of the Rosenwald
program could have been due to lower costs of schooling rather than to improvements in school quality. On the
other hand, improvements in school quality may have also encouraged some children who would not have otherwise
attended school, to enroll in a Rosenwald school.

4

standard deviations.

Further, we find that the test score effect is eliminated when we control for

educational attainment, suggesting that the pathway for the improvement in cognitive skills is through
schooling. This finding is consistent with other studies that show that racial gaps in test scores are not
immutable and can be influenced by interventions (Neal and Johnson 1996; Hansen, Heckman, and
Mullen 2004; Cascio and Lewis 2006; Chay, Guryan, and Mazumder 2009).
Across all outcomes, we find that the gains are largest in counties with large fractions of former
slaves. This suggests that schooling interventions targeted at historically deprived communities may
yield especially large returns, consistent with a view in the development literature that introducing
schools in disadvantaged rural areas disproportionately benefits students facing the highest cost of
attending school and where rudimentary school inputs are lacking (e.g. Duflo 2001; Glewwe and Kremer
2006).
Identifying causal effects of the Rosenwald Initiative presents an empirical challenge because
school location decisions were not random. In order to receive matching grants, local citizens were
required to donate substantial resources for school construction. Consequently, counties with greater
demand for educational resources may have more likely received a grant. It is plausible that students in
those high-demand counties may have experienced better outcomes even in the absence of the Rosenwald
program. However, we find no empirical evidence to support this conjecture. We show that Black school
attendance rates in 1910, and the change in those rates between 1900 and 1910, were similar between
counties that received a school and those that did not, and that the prevalence of Rosenwald schools
across counties was not systematically related to observable measures of Black socioeconomic conditions
prior to the Rosenwald Fund’s creation.
Nevertheless, we use several empirical strategies to address any concerns about the selective
location of schools. Our research design exploits both the geographic and temporal variation in the
location of schools, allowing comparisons of cohorts born within the same county by using county fixed
effects. Repeated cross-sections of Census data for some outcomes allow us to control for county-specific
time trends. We also take advantage of the explicit targeting of the program to rural Blacks allowing us to
5

use rural Whites and urban Blacks as control groups in a difference-in-difference framework. As an
additional robustness check, we conduct a separate exercise where we only exploit variation arising from
idiosyncratic factors that influenced the location decisions of some of the earliest schools built in
Alabama that preceded the large-scale rollout of the program. The estimates from this exercise are very
similar to what we find with our full sample. This strategy also addresses concerns regarding the possible
selective migration of families with a strong preference for education to the Rosenwald counties. Finally,
we directly address concerns about selective migration by recalculating Rosenwald exposure rates based
on county of birth rather than on county of residence for a sample of WWII enlistees and find similar
results.
The next section provides background on the Rosenwald schools and sketches out our analytical
framework. Section III describes the data. Section IV outlines our empirical strategy. Section V presents
our Census results on school attendance and literacy. Section VI discusses our results on adult outcomes,
including educational attainment and test scores using the World War II enlistment data and migration
using the 1940 Census. Section VII describes how these results differ across pre-Rosenwald community
characteristics. Section VIII explains how we compute an internal rate of return for the Rosenwald
Initiative. Brief conclusions are offered in section IX.
II. The Rosenwald Rural Schools Initiative
A.

Historical Background
The Rosenwald Rural Schools Initiative arose from the collaboration of Booker T. Washington,

the principal of the Tuskegee Institute in Alabama, and Julius Rosenwald, a Chicago area businessman.
Frustration with the disbursement of funds by local county education boards and the general inadequacy
of Black schools6 led Washington to seek out Northern philanthropists, including but not limited to

6

For accounts of early 20th century Black schooling, see Bond (1934), McCormick (1934), Myrdal (1944), Margo
(1990), and Hoffschwelle (2006). One contemporary description appears in the South Carolina Superintendent of
Education’s 1917-18 school report: “The school buildings are in the most instances wretched, the terms short, and
salaries low, practically no equipment, and the preparation and fitness of the teachers generally very inferior.” The

6

Rosenwald, to build new schools for rural Blacks. After the opening of the first six schools near
Tuskegee around 1913-14, Rosenwald agreed to partly fund up to 100 additional schools, primarily in
Alabama. Thereafter, the program spread quickly and, by the end of the decade, 716 schools were built in
11 states. Figure 2a displays a map of the number of Rosenwald schools by county as of the 1919-20
school year, the first year in which we know the complete spatial distribution of school buildings. Early
schools were primarily clustered in Alabama, as well as in parts of Louisiana, Tennessee, Kentucky,
North Carolina and Virginia. Three final states – Florida, Oklahoma, and Texas – were approved for
funding in 1920, boosting the final list of eligible states to the 14 shown in Figure 2a.
Rosenwald school construction accelerated in the 1920s, growing by 18 percent per year. Figures
2b and 2c illustrate this expansion by registering the number of schools by county as of 1925 and 1932.
When the initiative closed in 1932,7 76 percent of counties with rural Black children had a Rosenwald
school and 92 percent of rural Black children in the 14 states lived in a county with at least one
Rosenwald school. Nevertheless, in most counties the number of Rosenwald schools was insufficient to
serve all potential students. Figure 2d displays our estimate of the fraction of school-age rural Black
children that could be accommodated by Rosenwald schools by county as of school year 1931-32. These
calculations take the ratio of the count of Rosenwald teachers multiplied by an assumed student teacher
ratio of 45, which was typical of the time, to the count of 7 to 17 year old rural Blacks derived from the
100 percent Census manuscripts. We estimate that roughly 36 percent of the Southern rural Black schoolage population, and 25 percent of all Southern Black school-age children, could have attended Rosenwald
schools by 1930.8

report also describes overcrowded classrooms, lack of blackboards, seats, and windows, and “in many
cases…superintendents (who) did not even know the location of many African-American schools” (Brannon 1919
as described in Weathers 2008). Racial funding inequities are also described bluntly in Washington’s letters to
Rosenwald and catalogued more systematically in Johnson (1941) and Margo (1990). For example, the data
reported in Johnson (1941) indicates that the average county spent 39 cents on Black school salaries per capita for
every dollar on White school salaries per capita in 1930.
7
The Fund voluntarily closed within a year of Rosenwald’s death in 1932. Its closure was hastened by a significant
fall in Sears’ stock value, which made up two-thirds of the Fund’s assets prior to the market crash (McCormick
1934).
8
Fund documents, starting with the name of the Initiative, are clear that the program was intended for rural
communities. For example, a well-publicized experiment to build five urban high schools in the late 1920s was

7

The program’s varied timing and geographic development provides the basis for our main
identification strategy. Figure 3 plots the mean and the inter-quartile range of Rosenwald rural coverage
across counties over the 1919 to 1930 period, highlighting the substantial amount of cross-county
variation. The temporal coverage of Rosenwald schools varied substantially within counties as well. For
example, although Oklahoma was among the last states to be funded by Rosenwald, by 1930 it had the
second highest share of rural Black coverage. In contrast, although Alabama hosted the first schools, by
1930 its coverage was among the lowest.
B.

A Framework for Understanding the Effect of the Rosenwald Schools
The canonical model of the school investment decision (Becker 1967; Card 1995) provides the

conceptual basis for understanding how the Rosenwald schools improved student outcomes. In this
framework, the optimal amount of schooling for individual i is determined by equating the marginal
economic return (in terms of future wages) of an additional year of schooling, Bi, to the marginal cost, Ci.
The marginal return to a year of school is a positive function of the quality Q of school s (Bi = Bi(Qs).
The marginal cost is a function of foregone contemporaneous earnings and the direct cost of attending
school.
Now imagine a population that does not have access to the socially efficient number of schools
either because of political oppression or other institutional restrictions. One clear prediction of the model
is that by increasing the Qs that pertains to this group, individuals will choose to spend more time in
school and thus ultimately improve their long-term skills. The Rosenwald Initiative aimed to improve Qs
among rural Blacks along several important dimensions. First, the program subsidized the construction of
modern facilities that were conducive to learning. This included building designs that provided for
adequate lighting, ventilation, and sanitation. The program also required adequate provision of school

short-lived and very small-scale. That said, a small number of counties with sizable cities, notably Shelby County in
Tennessee which contains Memphis, appear to have a high degree of Rosenwald coverage. In some of the analysis
to follow, we make sharp restrictions on the definition of rural and urban that ensures that no urban (living in a town
or incorporated place with over 2,500 residents) students are included. See Sections III.C. Elsewhere, when we rely
on self-reports of rural status, any effect from an urban building will be picked up on urban Blacks and lead us to
sometimes underestimate the effect of Rosenwald on our main treatment group, rural Blacks. See Section IV.

8

equipment (e.g. desks and blackboards)9 and directly-funded school libraries that ensured supplies of
books. All of these key features were lacking in the previous schools that served rural Blacks. Second,
the Rosenwald program made an effort to improve teacher quality by first and foremost improving the
physical environment for teaching,10 but also providing for teacher training schools and programs,11
incentivizing communities to raise teacher salary standards, and in some cases, subsidizing the
construction of teacher homes. Third, the program actively sought to increase the length of the school
term.
The model also indicates that a second channel by which the intervention may have increased
educational attainment among rural Blacks is by reducing the marginal cost, Ci, of attending school. The
construction of new schools made access easier for those who lived far from existing school buildings.12
This is particularly true for high school instruction, which was virtually nonexistent prior to the Fund’s
involvement at this level starting in 1926.13 The reduction in costs of schooling has two implications.
First, among those families for whom the costs of schooling their children were so high that they would

9

“School equipment received the same careful scrutiny to ensure that the building could have the greatest impact on
its occupants. Blackboards along three walls served the teacher for instruction and students for practice assignments.
Modern patent desks replaced the rough wooden slabs, pews, and benches typical of many other Black schools.
Often African American community members found it difficult to pay for patent desks in addition to their
contribution to the building and asked to be relieved of this burden. White school officials would have preferred to
transfer used furnishings from White schools over to Black ones. However, the Rosenwald Fund remained firm and
refused to make final payment on buildings that did not meet its standards for the exterior or interior.” From
http://www.preservationnation.org/travel-and-sites/sites/southern-region/rosenwald-schools/development-ofrosenwald-plans/community-school-plans.html.
10
Numerous letters from key participants make clear the difficult conditions hampering teaching and the importance
of the buildings in improving this situation. A poignant example is Booker T. Washington’s June 12, 1912 letter to
Rosenwald that concludes: “Many of the places in the South where the school are now taught are as bad as stables,
and it is impossible for the teacher to do efficient work in such places.”
11
This included county training schools, summer training programs, and actively working with other philanthropies,
most notably the Jeannes Fund, to help train teachers. See, for example, Reed (2004) who discusses the training
standards that applied to Rosenwald instructors at one particular school in North Carolina.
12
Indeed, in 1929, the Fund began offering three year grants to support bus services, conditional on term length and
minimum teacher salary standards (Hoffschwelle 2006). Reducing distance from school may also increase
attendance of teachers (Kremer, Chaudhury, Rogers, Muralidharan, and Hammer 2005; Chaudhury, Hammer,
Kremer, Muralidharan, and Rogers 2006; Das, Dercon, Habyarimana, and Krishnan 2007) and thereby work on the
school quality margin.
13
Although the high school movement was well underway in the North by the mid-1920s (Goldin and Katz 1999),
public high schools for Southern Blacks were scarce. Alabama and South Carolina contained no four-year
accredited Black public high schools, and Florida had only two that offered any high school instruction. However,
by 1932, Rosenwald Fund pamphlets claimed that roughly 10 percent of their schools offered at least two years of
high school instruction (Donohue, Heckman, and Todd 2002).

9

have opted for no schooling in the absence of the program, the presence of a new school nearby may have
led them to obtain some schooling. In addition to this effect at the “extensive margin”, the lower costs of
schooling among those who were already enrolled in school could have led them to obtain more
schooling.
Finally, during this era, school finances were determined by county school boards that were
dominated by Whites. If this politically dominant group could implicitly tax the Rosenwald subsidy, it is
possible that White students in Rosenwald counties may have gained from gifts intended for rural Black
schools. Indeed, the diversion of fungible financial resources to White schools was an endemic problem
in the South during this period (Bond 1934; Margo 1990; Ascoli 2006). Moreover, anecdotes suggest that
the success of the Rosenwald schools led to additional demand for schooling among Whites, which could
have accelerated the diversion of resources.14 A possible implication is that the Rosenwald program may
have also benefitted rural Whites, an issue which we return to later when we discuss our differencing
strategy in section IV and our results in section V.
Similarly, given the fungibility of money and the institutional environment of the time, it is likely
that some of the public financial resources that were targeted to the Rosenwald schools “crowded out”
funding that would have otherwise gone to existing Black schools. To counteract both the possibility of
crowd out as well as the diversion of funds to White schools, the Rosenwald Fund deliberately focused on
providing physical resources (e.g. better buildings, equipment, and teachers) that could not be easily
expropriated. This implies that traditional measures that have been used to proxy for school quality in
previous research (e.g. school expenditures, counts of teachers), would not only fail to capture
improvements in actual school quality but may even lead to counter-intuitive conclusions regarding the
effects of the Rosenwald schools on racial gaps in educational resources.
C.

Matching Grants and School Location Selection

14

See e.g. McCormick (1934). In a 1919 letter to Rosenwald, Robert Moton, the Principal of Tuskegee wrote: “Let
me repeat that there is no movement in America that is doing more, not only in providing larger and more
satisfactory school equipment for the Negro race, but doing equally as much in stimulating White people towards
making more adequate provisions for the education of their own children, so that you are not only helping Negroes
but Whites as well.”

10

The strategies pursued by the Rosenwald Fund reflected the experiences and visions of both
Washington and Rosenwald. Washington felt strongly that in the wake of slavery, the Black community
could develop only through self-reliance. Rosenwald, meanwhile, was a strong proponent of using
matching grants to foster community support; indeed, he had employed the concept of “buy-in” for his
previous philanthropic efforts. Consequently, the Fund was unambiguous in its requirement that it would
provide financial backing conditional on local support, succinctly summarized by the Fund’s refrain:
“Help only where help was wanted, when an equal or greater amount of help was forthcoming locally,
and where local political organizations co-operated” (McCormick 1934; Hoffschwelle 2006).15 Local
Blacks and state and county governments provided the majority of the funding, particularly after
construction was complete (see Table A1).16 Over time, matching became even more critical, with the
Rosenwald share of contributions falling from around 25 percent for the earliest schools to the 10 to 15
percent range during the last five years of the program.17
This funding mechanism suggests that individuals from communities that were particularly open
to improving Black schools, and thus were able to convince the Fund to invest in their community, might
have experienced better outcomes even in the absence of the Rosenwald program. We briefly provide
several pieces of evidence regarding the possibility of selective school location. In Figure 4, we show
that the initial counties where Rosenwald schools were built by 1919 had slightly lower levels and weaker
trends in Black rural school attendance in the pre-Rosenwald period (1900 to 1910) than those counties

15

Letters available at the Fund archives at Fisk University provide a few examples of communities that struggled to
obtain the required resources and were therefore denied funding.
16
Table A1does not include nonmonetary donations of time, materials, and land from local citizens. Comprehensive
records of in-kind donations do not exist. Anecdotal evidence from the archives suggests that in-kind donations
were particularly important for the earliest schools.
17
Although the Rosenwald Fund ultimately only covered a small share of the building expenses, it played a crucial
role in providing the prestige and credibility to garner the necessary financial and nonfinancial support of local
White and Black communities. For example, the Fund hired canvassers to explain available opportunities and guide
local Black leaders through the fundraising process (Hoffschwelle 2006). The Fund consulted with and, to varying
degrees, gained the support of White government officials who acted as the state agents for Black schools.
Rosenwald money also likely helped buy local White acquiescence, including county education board approval for
maintaining schools post-construction (Donohue, Heckman, and Todd 2002).

11

that never received a school.18 Thus, with respect to a key outcome, we find no evidence of positive
selection for the earliest schools. Nevertheless, this concern also motivates our analysis in section VI
where we isolate idiosyncratic variation in the location of some of the earliest schools built in Alabama.
In Appendix A, we investigate the extent to which pre-Rosenwald county characteristics affected
school location decisions.

We find that pre-existing Black socio-economic conditions (e.g. school

attendance, literacy, occupational status) and trends in these conditions cannot predict the timing or
intensity of initial Rosenwald school locations in a statistically or economically significant way (Table
A4).19 However, we do find some evidence that counties with higher levels of White literacy, irrespective
of White occupational structure, were more likely to build an initial school in the program’s early years.
This result is consistent with Washington’s strategy, perhaps continued by the Fund after his untimely
death in 1915, of avoiding areas that might lead to White backlash.20 In any event, we remove any bias
that is generated by the matching grant strategy or selection on White characteristics, by using county
fixed effects and county fixed effects interacted with Census year. The latter approach further limits our
variation to differences in exposure across birth cohorts within counties in a given year.
III. Data
A.

Rosenwald Schools
Through an agreement with the caretaker of the Rosenwald Fund’s archives -- Fisk University in

Nashville, Tennessee -- we received digital versions of the index cards used to track the Fund’s 4,972
construction projects. These cards are the only complete database of the individual Rosenwald schools.
Each card contains a description of a school, teacher home, or industrial shop, or some combination

18

If we go back even further to 1880, the counties which built the initial Rosenwald schools had virtually identical
levels of Black rural school attendance as the counties that never obtained a Rosenwald school.
19
Black occupational status has a marginally statistically significant effect in some specifications involving the
schools built during the 1920s but the size of the effect is not qualitatively large.
20
Neither Washington nor the Rosenwald Fund challenged segregation, which almost surely increased White
support for the schools. The view within the Fund echoed Washington’s well-known belief that education and
economic needs had to be addressed first —a strategy that led to deep conflicts with other activists, notably W.E.B
Du Bois and the NAACP.

12

thereof. Information is limited to the location (state and county), year of construction, school name,
number of teachers (or home/shop rooms), number of acres of land, insurance valuation, and construction
cost.

Cost is broken down by four possible funding sources: the Rosenwald Fund, local Black

individuals, local White individuals, and local public governments. Room additions, as well as complete
destructions due to fire or weather, are recorded in handwriting ex-post although it is difficult to know
how complete these adjustments ultimately are. However, for completeness, we include all recorded
additions and rebuilds in the relevant year that they take place. Appendix Table A1 provides basic
statistics about all Rosenwald school construction projects. Our analysis uses a database that includes
4,932 schools with the capacity to hold 13,746 teachers in 888 counties.21
B.

Census (1900-1930)
We pool cross-sectional samples drawn from the 1900 to 1930 decennial Censuses using the

Integrated Public Use Microdata Series (or IPUMS, see Ruggles, Alexander, Genadek, Goeken,
Schroeder, and Sobek 2010). In particular, we use the 1 percent sample for 1900, the 1.4 percent sample
for 1910, the 1 percent sample for 1920, and an early version of the 1930 5 percent sample.22 These data
are linked to the Rosenwald schools by county of residence and birth year. Importantly, we can also
distinguish between those living in rural or urban areas within a county.23
The two relevant outcomes available in these Censuses are school attendance and literacy.
School attendance refers to whether an individual attended or enrolled in school between September 1 and
the Census date (June 1 in 1900, April 15 in 1910, January 1 in 1920, and April 1 in 1930). For this
measure, we construct a pooled sample of over 640,000 Black and White children between the ages of 7
21

Official Rosenwald Fund records tally 4,977 schools in 883 counties. Our database starts with 4,972 index cards.
We delete 36 of these cards because the project did not involve a schoolhouse (22 cases), contained missing
information on cost or teachers (10 cases), or was never built (4 cases). Additionally, we drop the four Missouri
projects. We also exclude county training schools because of uncertainty as to whether they housed students or were
used for their original purpose to train teachers. However, our results are impervious to the inclusion of county
training schools.
22
Since the 1910 Census oversamples certain groups, we utilize sample weights in our main estimates.
23
It may be the case that the Rosenwald Fund’s vision of a rural community differs from the technical Census
definition of less than 2,500 people, adding attenuation bias to our estimates. However, in internal documents, the
Fund often used the Census definition for data organization and evaluation.

13

and 17. Literacy, which is asked of individuals 10 and older, refers to the ability to both read and write in
any language. Collins and Margo (2006) find that this measure acts as a proxy for completing 1 to 3
years of schooling. They also show that literacy rises with age. To abstract from literacy effects due to
schooling at young ages, we restrict the sample to those who are at least 15. We also restrict the sample
to those under 23 to avoid spurious correlation arising from the possibility that adults with high literacy
moved into Rosenwald counties for their children’s schooling.24 This sample of 15 to 22 year olds totals
over 430,000 persons.
Appendix Table A2 presents descriptive statistics of these samples. Of note, the rural BlackWhite school attendance gap was 21 percentage points in 1910 but narrowed to 9 percentage points by
1930. In urban areas, the Black-White attendance gap fell from 13 to 7 percentage points. The table also
illustrates the striking racial differences in measures of family background such as parent literacy and
home ownership. For example, as late as 1930, the Black-White gap in father’s literacy was about 20
percentage points. With this detailed individual-level data, we are able to control for such factors in our
empirical analysis.
C.

World War II Enlistment Records
We also draw from records of US Army World War II enlistees, available from the National

Archives and Records Administration.

Our overall sample includes roughly 1.8 million men born

between 1910 and 1928 and living in Rosenwald states when they joined the Army between 1940 and
1946.25 We use age and county of residence at enlistment to link soldiers to their potential access to
Rosenwald schools. Unlike the Census, however, we do not know rural status. Instead, we use the 1910
through 1930 Censuses to classify counties based on their rural composition. Since the Census uses a
population cutoff of 2,500 for defining whether a city or incorporated place is urban, we classify a county
as rural if it had no individuals living in places with a population greater than 2,500 in the 1910 through
24

We find generally weaker effects when the age range is expanded to 15 to 30 year olds suggesting that selective
migration is not a significant concern. Sections VI, VII.D.3, and VII.E provide more evidence on this latter point.
25
Records are available for enlistees starting in 1938 but the samples are very small. See Feyrer, Politi, and Weil
(2008) for more description of the data.

14

1930 Censuses. This ensures that we do not include any urban students in our treatment group. We
define a county as urban if more than half of the county’s residents from 1910 through 1930 lived in cities
containing greater than 2,500 people. It is possible therefore that some treated individuals may be
contained in our control groups. These restrictions produce a sample of nearly 1 million men with nearly
equal numbers living in our definition of rural and urban counties.26
A key advantage of this data, relative to the early Censuses, is that it provides measures of human
capital during adulthood. This includes completed years of schooling beyond grammar school27 from
which we also calculate indicators for attending and completing high school. A unique feature of the data
is that, for a brief period in 1943, we have scores from the Army General Classification Test (AGCT)
used to determine military occupation. Since these test scores were thought lost to history, we describe
them in Appendix B. In addition, for 1941 and 1942, there are data on height, which we use as a validity
check since height is unlikely to be impacted by access to schools after age 5 (Martorell, Schroeder,
Rivera, and Kaplowitz 1995; Behrman and Hoddinott 2005).
A clear concern with the enlistment records is that there may be selection into who was inducted
into the Army. We discuss a variety of ways in which we address potential selection bias in the next
section. In one such exercise, we use a smaller subsample of about 17,000 WWII enlistees, for whom we
have determined the county of birth by matching individuals’ name, state of birth, and year of birth to the
Social Security Death Master File (DMF). We obtained the place of birth (county or city) from a match
provided to us from the Social Security Administration (SSA) using their NUMIDENT file.28

26

Our primary classification excludes counties with a rural share between 50 and 100 percent. We discuss the
implications of using alternative rural and urban thresholds in section VII.A.
27
For those who did not complete grammar school, we impute years of schooling based on race, birth year, and state
economic area from the 1940 Census and exclude individuals who had completed fewer than four years of
schooling, the military’s requirement in 1941-1942 (Perrott 1946). Our results are not sensitive to small changes in
imputation methods.
28
We drew three samples from the WWII data to match to SSA data: Blacks living in Rosenwald counties,
individuals who enlisted in 1941 and 1942 for whom we have height, and individuals for whom we have AGCT
scores. There are many sources of possible selection with this sample: 1) Individuals must die between 1965 and
2007; 2) they must enter the Social Security System and own a Social Security Number; 3) they must be uniquely
identified; and 4) they must provide a clearly recognized county or city name for their place of birth that can be
matched to the Rosenwald database.

15

Summary statistics of the WWII enlistee samples are presented in Appendix Table A3. The
Black-White gap in years of schooling for the full sample is about 2.2 years. The racial difference in
AGCT scores is about 25 points which is equivalent to 1.1 standard deviations based on the test score
distribution for the full sample.
IV. Empirical Strategy
A.

Regression Model
For our analysis using the Census data, the main specification is of the following form:
(1)           

 

where yibct is school attendance or literacy for individual i born in year b living in county c in Census year
t, female, black, rural and blackrural are indicators of being in one of those demographic categories, Xibct
is a vector of family background characteristics including mother’s literacy, father’s literacy, father’s
occupational status and father’s home ownership, age is interacted with state s and Census year t, county
represents county fixed effects, and εibct is an error term. Standard errors are clustered at the county
level.29
Our first measure of Rosenwald treatment (“ROSE”), Rct, is an indicator of whether a Rosenwald
school was present in an individual’s county c as of Census year t. In our tables, we refer to this measure
as “Rosenwald presence.” Rct is limited in at least two ways: it fails to distinguish differences in
Rosenwald exposure between birth cohorts within a county30 and it does not adjust for the breadth of
Rosenwald coverage within a county. Despite these drawbacks, we start with Rct because it provides a
straightforward approach that is easy to interpret.
29

Clustering at the state level, as in Bester, Conley, and Hansen (2010), has minimal impact on our inferences.
For example, a 13 year old in 1930 living in a county that first opened a Rosenwald school in 1928 has only two
years of exposure but is treated the same as a 13 year old living in a county that built a Rosenwald school in 1924
and has 6 years of exposure.

30

16

Our second, more comprehensive measure of “Rosenwald exposure,” Ebc, estimates the average
Rosenwald coverage that each student born in year b and living in county c experienced over ages 7 to
13.31 Specifically for individuals 13 or older as of a Census,

∑

N

, where Tct is the

number of Rosenwald teachers in county c in year t and Nct is the number of rural Blacks between the
ages of 7 and 17 in the county in each year. For the numerator of our coverage measure, we assume a
class size of 45 students per teacher, a standard at the time (Johnson 1941) and an assumption used by the
Rosenwald Fund in internal and published documents. The denominator is computed from the digitized
100 percent 1920 and 1930 Census manuscript files available through ancestry.com and interpolated for
1919 and 1921 through 1929. For individuals under 13 as of a Census, we average the coverage rates up
to and including the year of the Census. That is, for 10 year olds:

∑

N

. Since Ebc takes

on values between 0 and 1, the γ coefficients can be interpreted as the effect of going from no Rosenwald
exposure in one’s county to complete exposure (i.e. every rural school-age Black child could have
attended a Rosenwald school).32

However, when considering economic magnitudes, it may be

particularly revealing to describe the effects at the mean level of exposure in order to account for
population-wide changes in racial gaps over particular time periods or cohorts.
For our analysis using the World War II data, the main specification is of the following form:
 

(2)

3

   

 

 

There are a few differences in (2) compared to the Census-specific specification (1). We rely exclusively
on Ebc as our measure of ROSE since most of the WWII enlistees attended school in the late 1920s and
1930s when the vast majority of counties with Black children had at least one Rosenwald school.
31

Since we cannot identify which schools built after 1926 were high schools, we confine our analysis to the effects
of exposure during the ages of 7 to 13. However, our results are robust to defining exposure over the ages of 7 to
17.
32
In some cases, the exposure measure exceeds 1; in such cases, we topcode values at 1. Rosenwald coverage rates
exceed 1 for 6 counties in 1919, 48 counties in 1925 and 109 counties in 1930.

17

However, even with county fixed effects, there remains significant cross-cohort variation in the timing of
the construction of schools to exploit. Xc includes the same average family background characteristics as
(1) but, because these measures do not exist for individuals in the WWII data, it is measured at the county
level using the Census year when a cohort was between the ages of 0 and 9. As described in section III.C,
rural is also defined at the county level. To control for age, we include a set of age dummies for each
race. We also include a set of race-specific time dummies, enlqtr, to control for the 28 quarters (indexed
as q) from 1940 Q1 to 1946 Q4 in which men could have enlisted. Finally, the sample is composed
entirely of men and therefore the gender indicator is dropped.
In both (1) and (2), we interact our Rosenwald measures with race and rural status to take
advantage of the explicit targeting of the treatment to rural Blacks while allowing the other groups (e.g.
rural Whites) to potentially serve as controls. Consequently, the γs enable us to construct “differenced”
estimates. For example, to calculate the effect of complete exposure versus no exposure on Black rural
children we would sum γ0, γ1, γ2, and γ3. The sum of γ2 and γ3 provides the difference in effects between
rural and urban Blacks (labeled “Black, Rural-Urban” in the tables). Similarly, the sum of γ1 and γ3
estimates the effect on the Black-White gap in rural areas (“Black-White Rural”). Finally, γ3 taken alone,
provides an estimate of the “triple difference,” which describes the effect of Rosenwald exposure on the
Black-White gap in rural school attendance relative to the Black-White gap in urban school attendance.
The advantage of differencing across groups is that common factors that may be correlated with
Rosenwald school exposure are removed. In particular, suppose that in the absence of the interactions,
the error term is structured as

ict

= ωct + ωct,race + ωct,rural+ ωct,race,rural. By interacting rural with ROSE,

any correlation between ROSE and the error component ωct,rural is absorbed. This design accounts for
factors, such as a positive shock to the rural economy, that benefited both rural Whites and rural Blacks
and were also correlated with the formation of the schools. Similarly, ωct,race can account for race-specific
factors that happened to be coincident with the construction of Rosenwald schools that would also affect
urban Blacks.

Finally, we can difference out both race and rural status, thereby eliminating the

correlation between ROSE and ωct,race and ωct,rural. In this case, our identification rests on the assumption
18

that there are no unobservable factors, ωct,race,rural, that are correlated with ROSE that only affect rural
Blacks. Put differently, any alternative explanation for our findings must utilize variation that had no
effect on rural Whites or urban Blacks.
Each of our four key estimators -- Rural Black; Black, Rural-Urban; Black-White Rural; triple
difference -- requires different assumptions about how unobservables vary by demographic group.
Moreover, the possibility that Rosenwald resources were diverted to White schools is not only
theoretically plausible (see Section II.B) but visible in the actions and words of Fund leaders (see footnote
13). Consequently, any positive effect on Whites would imply Black-White estimators may understate
the gains for rural Blacks. Additionally, for the Census specifications that rely on self-reported rural
status, if Rosenwald schools were built in areas that exceeded 2,500 residents, our comparison of rural
and urban Blacks would understate the gains for rural Blacks. Therefore, we consider all four estimators,
including the simple rural Black estimator, in assessing the range of effects of Rosenwald on rural Black
schoolchildren. In practice, the estimates are typically tightly bunched, a consequence of the control
groups revealing little systematic selection favoring rural communities or Black children, the Fund’s
building being concentrated in areas deemed rural by the Census, and diversion of key resources to White
schools being of minor importance.
In addition to differencing, we exploit variation in Rosenwald school coverage either over time
(for Rct) or across cohorts (for Ebc) to control for unobserved county characteristics by using county fixed
effects. With the exposure measure, Ebc, we can also specify separate county fixed effects for each
Census year (i.e. add county

yearct to equation 1) to address any long-term (e.g. 10 year) time trends

that are county-specific. This is because even within a particular county in a particular Census year, there
is sufficient variation in Rosenwald exposure across birth cohorts due to the timing of school
construction. This variation allows us to overcome threats to identification that arise from the possibility
that Rosenwald schools were built in counties with particular characteristics at a point in time (see
Appendix A) or that were exhibiting certain trends over long periods of time. Moreover, this framework
accounts for concurrent policy changes at the state or national level, such as the introduction and
19

expansion of compulsory schooling and child labor laws, as well as more general trends such as
improvements in health (e.g. disease eradication) or the lessening in importance of “intergenerational
drag” from slavery (Margo 1990).33 Coupled with our differencing strategy, the fixed effects approach
implies that any alternative explanation for our findings must be rely on within-county variation that only
affected rural Blacks and happened to coincide with the timing of school construction for cohorts who
were of school-age at the time the schools were built.
To address remaining concerns related to unobservables, and in particular, to their potential
impact on initial nonrandom residential household location decisions, we also provide three additional
sets of results: a) family fixed effects34, b) estimates based on the initial Alabama schools, and c)
estimates based on exposure in one’s location at birth rather than contemporaneous residential location.
In the latter two cases, residential location precedes large-scale rollout of the Rosenwald program and, by
construction, is unlikely to be tainted by endogenous migration decisions.
B.

Selection in World War II Data
The World War II sample introduces an additional concern about potential bias from nonrandom

induction into the Army. Prospective enlistees were screened on mental and physical characteristics.
Moreover, these criteria were lowered when manpower needs rose after the US entered the war (Lew
1944). Since our sample includes both men who were drafted through lotteries as well as volunteers,
mean characteristics such as years of education may differ from the overall population.
Of particular concern for our purposes is the extent to which selection into the military might bias
our parameter estimates of the effects of Rosenwald exposure. A positive bias could arise, for example, if
counties where Rosenwald schools were most effective at improving educational outcomes happened to
have other characteristics that led to greater recruitment and higher induction rates of rural Blacks.
Alternatively, as manpower needs became the preeminent concern for the military, and standards were
33

We note that Lleras-Muney (2002) finds no impact of compulsory schooling and child labor laws on Black
education. Likewise, the eradication of hookworm disease (Bleakley 2007) predates our cohorts, and primarily
impacted Whites living in coastal areas (Coelho and McGuire 2006; Keller, Leathers, and Densen 1940).
34
We can only use family fixed effects for studying effects of school attendance with the 1900-30 Census samples.

20

loosened for more recent cohorts, recruitment might have become more intense in counties where
Rosenwald schools were less effective, possibly leading to a negative bias.
A useful way to gauge whether selection into the sample may have been related to the success of
the schools is to compare the probability of selection with Rosenwald exposure. If the measures are
correlated, selection would be of greater concern. We estimate the probability of selection by taking the
ratio of the number of actual male enlistees by county, birth year, and race, to an estimate of the
corresponding population of each group. For the numerator, we use counts from the population of male
WWII enlistees; the denominator is derived from digitized records of the complete 1930 Census
manuscript files available at ancestry.com. We use 1930 as a baseline because the 1940 data are not yet
available. There are a total of 26,448 cells for each race (1,392 counties × 19 cohorts).35
Figure 5 shows the scatter plot of the probability of selection for Blacks against Rosenwald
exposure Ebc, along with a regression line for a subsample of 20,494 observations for counties that had a
positive number of Black residents in the 1910 Census and non-missing values for exposure and
probability of selection. Although visually there appears to be very little relationship between the
variables, the regression coefficient is positive, 0.036 (0.009) and statistically significant (standard errors
are clustered at the county level). However, once we include county fixed effects and cohort dummies,
the coefficient is reduced to 0.004 and is no longer quantitatively important or statistically significant.36
Therefore, our first approach to dealing with selection is to incorporate county fixed effects, age
dummies interacted with race, and quarter of enlistment dummies interacted with race in our
specifications. The use of county fixed effects should eliminate any selection bias that is associated with
time invariant county level characteristics that may be common to the selection of both Whites and
Blacks.

The age-by-race dummies and quarter of enlistment-by-race dummies eliminate selection

35

The estimated probabilities are greater than 1 for 124 cells for Whites and 363 cells for Blacks. In these cases, we
topcode the values at 1.
36
The coefficient is further reduced to -0.0006 (0.007) if we exclude 298 cells where the probability of selection is
0.6 or greater. Such high probabilities of selection are likely to be implausible, particularly since the denominator
contains both males and females. The estimate is reduced to 0.002 (0.007) if we simply exclude 184 cells where the
probability of selection is estimated to be 1 or more.

21

operating at the national level that affect some ages or time periods differently by race. For example, if
military recruiters began to target Blacks of all ages in greater numbers across all counties as the War
heated up in late 1942 and 1943, this form of selection would be absorbed by these indicators.
These controls, however, do not account for county-specific factors that might have changed over
time, across birth cohorts, or differed by race. To build on the last example, the increasing demand for
soldiers as the War intensified may have led recruiters to look for volunteers among 18 year old Blacks
living in counties that had Rosenwald high schools. This particular form of selection involves the
“interaction” of county-by-cohort-by-race and therefore is not absorbed by our fixed effects, though it is
worth noting that our earlier exercise found no evidence suggesting that selection at this level was a
problem. In any case, we cannot sweep out effects at this level of interaction through indicator variables
and maintain identification since our exposure measure, Ebc, is defined at the county-by-cohort level.
Therefore, our second strategy is to directly account for the probability of selection at the countyby-cohort-by-race level using Inverse Probability Weighting (IPW). This approach involves weighting
our regressions by the inverse of the probability of enlistment for each observation based on their cell.
Define p to be the true likelihood that a given individual will enlist, and ̂ to be an estimate of that
likelihood. Then, weighting the regression equations by wi =   removes any selection bias, as long as the
observables used to estimate the probabilities account for all sample selection within cells (e.g. Hirano,
Imbens, and Ridder 2003; Wooldridge 2002; Chay, Guryan, and Mazumder 2009). Note that unlike
studies that use a selection equation (e.g. propensity score) with a sample to estimate ̂ , we have the
universe of World War II enlistees and the full population counts from the 1930 Census.
Any sources of selection that operate within cells defined by county-by-cohort-by-race would not
be addressed with our IPW strategy. For example, suppose it were the case that among a set of rural
Black men attending Rosenwald schools who were the same age and lived in the same county, only those
who were the most successful students chose to volunteer for the Army. The use of weights derived from
probabilities estimated at the county-by-cohort-by-race level would not address this source of bias since

22

the probabilities differed within cells. Likewise, any other characteristics operating below the level of
county-by-cohort-by-race that influenced selection also would not be addressed by IPW.
To address the possibility that individual preferences for military service could be correlated with
the effectiveness of the Rosenwald schools, our third strategy confines the sample to individuals who
were drafted in the year in which they first became age-eligible for the draft. When conscription was first
instituted in October 1940, all men between the ages of 21 and 35 were required to register. In November
1942, the draft age was reduced to 18. Therefore, we construct a sample that for each year includes only
draftees who were either of the minimum draft age or one year older (to account for the fact that we do
not know the exact date of birth but only the year of birth). Thus the sample includes 21 and 22 years old
who were drafted in 1940 and 1941, 20 and 21 year olds drafted in 1942, and 18 and 19 year olds drafted
during 1943 to 1946.37 Since there is unambiguous scope for 20 to 22 year old draftees to have
volunteered at younger ages, we also report results solely for the small subsample of 18 and 19 year old
draftees during 1943 to 1946.38
Finally, we describe results on similar education outcomes using the 1940 Census, which does not
contain the potential selection problems related to military enlistment but requires using a more blunt
measure of Rosenwald exposure.
V. Census Results on School Attendance and Literacy
A.

School Attendance
Table 1 provides results for school attendance using the indicator of county Rosenwald presence,

Rct. The top panel displays the γ coefficients from equation (1). Below that, we report a series of
estimates based on combinations of the γs: 1) the implied effects for each of the four demographic groups

37

According to our data, 95 percent of draftees were at least 21 in 1941. In 1942, 9.2 percent of draftees were 20
and 90 percent were 21 or older. During this time, few 18 to 20 year olds volunteered. After the draft age was
lowered to 18, the age distribution shifted markedly. Between 1943 and 1946, 25 to 30 percent of draftees were 18
or 19. Note that draftees made up 75 percent of the military during the War, peaking at 92 percent in 1943.
38
For that sample, we cannot include county fixed effects since there is no variation within county in Ebc for the
relevant cohorts.

23

– Black rural, White rural, Black urban, and White urban, 2) difference-in-difference estimates between
Black rural and Black urban, White rural and White urban, Black rural and White rural, and Black urban
and White urban; and 3) the triple differenced estimate of Black-White rural less Black-White urban. We
start with a minimal specification in column (1) that includes only year effects to show the basic patterns
in the data. Naturally, other important factors, like family background characteristics, are likely to affect
school attendance. In Column (2), we add controls for gender, age, parents’ literacy, father’s occupation
score, father’s homeownership, state fixed effects, and the White literacy rate in the county in 1910. In
column (3), we omit these controls but add county fixed effects.

Column (4) includes both the

demographics and the county fixed effects.
Beginning with the most parsimonious specification, we find an economically and statistically
significant effect on rural Blacks; the presence of a Rosenwald school in one’s county boosted school
attendance among potentially eligible children by 8.9 (0.7) percentage points. By comparison, there is no
effect on either rural Whites or urban Whites. There is a smaller, but statistically significant, effect on
urban Blacks that could reflect the possibility that some Blacks classified as urban by the Census attended
Rosenwald schools. Alternatively, the effect on urban Blacks may reflect that there were higher school
attendance rates among all Blacks living in Rosenwald counties. When we difference out the common
effect of being Black (Black, Rural-Urban) or the common effect of living in a rural county (Black-White,
rural), the Rosenwald effect remains economically large and highly significant. Finally, differencing the
rural Black-White effect and the urban Black-White effect, we find that a Rosenwald school is estimated
to raise school attendance by 6.7 (1.1) percentage points.39
Concentrating on the most complete specification in column (4), the four different estimates of
the treatment effects on rural Blacks (Black rural, Black rural – Black urban, Black rural – White rural,
and the triple difference) range from 4 to 7 percentage points and are all statistically significant at the 1
percent level. The magnitudes of these estimates are economically important. For brevity, we focus on
what the effects of the program imply for the Black-White gap among rural students across our outcomes
39

We get extremely similar results if we use those attending school and not working as an outcome.

24

in the various samples, but in principle we can calculate similar statistics for the rural – urban gap among
Blacks or the triple difference. Over the 1910 to 1930 period, the Black-White rural difference in school
attendance fell by about 11.5 percentage points from 1910 to 1930. Our estimates suggest that Rosenwald
schools can account for 5.8 percentage points, or 50 percent, of this decline.40
Table 2 reports school attendance results using the more refined Rosenwald exposure measure,
Ebc. Column (1) begins with a sparse specification and shows large and significant effects of complete
exposure on rural Blacks using any of our four estimates. We find that adding county fixed effects
(column 2) and further adding age interactions by state and year (column 3) has relatively little effect on
our four key estimators. For example, the simple un-differenced estimate of “Black Rural” is about 0.12
in all three cases.
However, when we include county fixed effects by Census year (column 4), we estimate a large
effect on urban Whites (γ0). As we discussed in section II.B, this might reflect the possible diversion of
public funds to White schools. If on the other hand, this reflects some form of positive selection
regarding the placement of schools in counties that would have experienced school attendance gains even
without the schools, our differenced estimators may remove this bias. Indeed, we find that our three
differenced estimates yield statistically significant effects that are only slightly lower in magnitude than
we find in column (3).41 As we will show later, we do not find statistically significant positive effects for
Whites in any of our other results.
Overall, moving from the blunt Rosenwald treatment measure shown in Table 1 to the more
nuanced measure in Table 2 lowers the implied aggregate effects of the program.

For example,

specification (4), which includes the baseline controls along with county-by-year fixed effects and age
interactions by state and Census year, suggests that going from no exposure to Rosenwald schools (Ebc=0)

40

We take the effect of 6.3 percentage points on Black-White Rural and scale this down to 5.8 percentage points
since 92 percent of rural Black school-age children in 1930 were living in a county with a Rosenwald school.
41
We also find that the results are similar if we use age dummies rather than a linear term, include county-specific
age (cohort) trends, interact the state- and Census year-specific age trends with race or rural status, use alternative
IPUMS samples, use alternative weighting methods, or construct our exposure measure using both urban and rural
Blacks.

25

to the mean level of Rosenwald exposure for rural Blacks in 1930 (Ebc=0.27)42 raised school attendance of
rural Blacks relative to rural Whites by about 3.1 percentage points. That estimate accounts for 27
percent of the reduction in the gap between 1910 and 1930.
Finally, in columns (5) and (6), we allow for family fixed effects (within Census year). As was
the case with column (4), we find a positive effect on urban Whites but also find that the differenced point
estimates are similar to what we find in other specifications. However, the standard errors rise sharply.
Consequently, only the effects on the Black-White rural gap remain statistically significant at reasonable
confidence levels. Nevertheless, by moving to a comparison of siblings who may have had different
levels of exposure to schools simply because of the timing of their birth relative to the construction of the
schools further narrows the scope for alternative explanations of our findings.
It is important to note that relative to the full capacity of the Rosenwald schools, the gains in
school attendance were relatively modest, suggesting that most of the students who were attending
Rosenwald schools would have attended another school in the absence of the program. To be concrete, in
1932, approximately 36 percent or 650,000 of the 1.8 million school-age rural Blacks were in Rosenwald
schools. If we assume that Rosenwald schools increased school attendance by roughly 5 percentage
points, which is within the range of our estimates, then 560,000 (1.8 million x (1 – 0.05)) of the 650,000
Rosenwald students would have attended school in the absence of the program. While Rosenwald
schools may have “crowded out” other schools, the program nonetheless appears to have been beneficial
in increasing attendance and, as we document below, other human capital measures. This success was
likely a consequence of improvements in both quantity, via lowering the cost of attending school, and
quality relative to the documented inadequacy of the pre-Rosenwald rural Black schools.
B

Literacy
Table 3 reports results for literacy using our Census sample of 15 to 22 year olds. Using our

preferred specification for Rct (column 2), we find that the presence of a Rosenwald school in the county
raises rural Black literacy rates by 9.3 (0.6) percentage points. As with school attendance, we estimate no
42

Exposure for all Blacks (rural and urban) was 0.28 in 1930 (see Table A2).

26

effect for rural Whites and a small positive effect on urban Blacks. This leads to difference-in-difference
estimates of 7.2 (0.8) percentage points for the rural-urban Black difference and 9.3 (0.5) percentage
points for the Black-White rural difference.

The triple difference is somewhat lower at 5.3 (0.9)

percentage points. However, some of this attenuation is due to an estimated decline in literacy rates
among urban Whites in Rosenwald counties. This may be driven by sampling error rather than by a true
decline in literacy since literacy rates were already close to 100 percent among urban Whites by 1910.
In columns (3) and (4), we turn to the exposure measure. Although we again find quantitatively
large and statistically significant effects for rural Blacks, we now find larger negative effects for urban
Whites and small but statistically significant negative effects for rural Whites. These results are difficult
to reconcile with the framework from section II.B that suggested that, if anything, the effects on Whites
could be positive. Nevertheless, using the differenced estimators, our preferred specification in column
(4) shows that complete exposure to Rosenwald schools improved Black literacy relative to Whites in
rural areas by just under 25 percentage points. Similarly, the effect on literacy for rural Blacks relative to
urban Blacks is nearly 18 percentage points. The estimated effect of complete exposure on the difference
between the Black-White rural gap and the Black-White urban gap is to narrow this difference by 16.5
percentage points. Focusing on the aggregate implications of the program, we estimate that access to the
Rosenwald schools accounted for 55 percent of the closing of the Black-White rural gap in literacy among
15 to 22 year olds from 1910 to 1930.
C.

Earliest Schools Built in Alabama
Thus far, we rely on fixed effects and differences between treatment and control groups to

address any potential selection driven by local demand for education. This section presents an alternative
approach based on the location of the earliest schools.
The Rosenwald Fund archival records supply clues that the initial schools were heavily clustered
in specific geographic areas (see Figure 2a) for idiosyncratic reasons that were largely unrelated to

27

economic or educational circumstances.43 We focus on the first schools in Alabama where the evidence
for exogenous school location is most transparent and compelling. In particular, Booker T. Washington
found himself located in Alabama due to happenstance 44 and archival records suggest that he sought to
build the initial schools there in order to quickly and efficiently develop a model for the future rollout of
the program:
“At present, it is thought wise to confine the schoolhouse building to the State of Alabama with
the view of getting experience that will enable us to render the best service for the least money
and in the shortest time possible.” 45
Moreover, we know from future construction activity (e.g. Figures 2b-d), school expenditure data
(Johnson 1941), and anecdotes46 that there is little to suggest that Alabama’s underlying demand for
Black schooling was high relative to the rest of the South. Therefore, we estimate the effects of the
Alabama schools built between 1913 and 1920 on school attendance rates of 7 to 17 year olds as of the
1920 Census, using data drawn from the 1900 to 1920 Censuses.
Specifically, we compare the effects of the program on children who lived in Alabama counties
along the state border to a control group consisting of children who lived in contiguous counties on the

43

The cluster of schools in Eastern Virginia and North Carolina could be explained by their proximity to Virginia’s
Hampton Institute where Washington was trained. The clusters in Louisiana, North Carolina, and Tennessee may
have been related to the presence of certain individuals, particularly the state agents for Black schools or county
officials who happened to be sympathetic to the Rosenwald program. Washington specifically considered exploiting
such friendly contacts in a June 1912 letter to Rosenwald:
“The wisest plan would be…to get…a half dozen county superintendents and county boards who are in
thorough sympathy with the plan, get them to work in their county, and in this way it would soon attract the
attention of other county officials…”
During 1918 and 1919, when the Fund began to divert more resources out of Alabama, strong letters of interest were
received from the Boards of Education of Louisiana, North Carolina, and Tennessee. Appendix A also highlights
that the initial school sites were not strongly related to observable characteristics of the counties.
44
Washington was born into slavery in Virginia where he became trained as a teacher at the Hampton Institute.
Local citizens in Tuskegee contacted the Hampton Institute in Virginia in search of a founding director who they
expected would be White. The principal of the Hampton Institute recommended Washington and this was the basis
of his location in Alabama.
45
Source: “Plan for Erection of Rural Schoolhouses,” date and author unknown, Fisk archives.
46
Bond (1969) describes how Alabama’s school superintendent noted in 1911 that local school boards were averse
to building or repairing Black schools, even with funds remaining after all work had been completed on White
schools.

28

other side of the border (see Figure A1 for a map of the counties included).47 Because there were few
large urban areas along the border and sample sizes are limited, we have insufficient power to estimate
differences by rural status. Further, due to lack of power, we only use Rct, the indicator of Rosenwald
presence in one’s county. Since Rct does not take into account the intensity of treatment -- the fact that 12
year olds in 1920 who were treated will have had, on average, more years of exposure to the schools than
7 year olds in 1920-- and since we do not know the year that schools were built prior to 1919, we also
estimate the effects for a slightly older sample of 9 to 17 year olds. Our regression specification is the
following:
(3)

 
_

where γ1, the estimated effect on Blacks in Alabama relative to Whites, is our coefficient of interest.
Table 4 reports the results.

We find that when we combine all four (Georgia, Florida,

Mississippi, and Tennessee) borders, the presence of a Rosenwald school increases the likelihood of
school attendance of Blacks relative to Whites by 5.1 (3.2) percentage points. The effect rises to 7.6 (3.5)
percentage points among 9 to 17 year olds. These estimates are quite similar to those reported in Tables 1
and 2.48
Another potential threat to the validity of our inferences is that the presence of a Rosenwald
school in a county might have prompted families who placed a high value on their children’s schooling to
migrate to or stay in these counties.49 A possible implication of such selective migration may be to
overstate the effects of the program since outcomes for these children might be higher even in the absence
of the program. It is difficult to directly assess the potential magnitude of this bias since the rate of
47

In general, the Alabama counties had lower levels of schooling and poorer socioeconomic outcomes than the
adjacent counties outside of Alabama. However, we use county fixed effects to sweep out that source of variation.
48
Rosenwald exposure was around 10 percent in Alabama around 1919. Therefore, the point estimates suggest that
over half of the Rosenwald students in Alabama would not have attended another school in the absence of the
Initiative, suggesting far less crowd-out than during the remainder of the program’s existence.
49
Emmett Scott of the Tuskegee Institute in 1918 noted that the presence of the schools appeared to be valued by
families who stayed in the South: “Of the rural Black people who choose to remain in the South, many will tell you
that they are content because they have a good school for their children to attend, a friendlier understanding with
their White neighbors, and a brighter outlook because of the Rosenwald rural school.” (McCormick 1934).

29

migration across geographic areas is generally unavailable prior to the 1940 Census. However, the
earliest schools are also largely purged of location choices since the rapid building of the pilot schools in
Alabama preceded the large-scale rollout of the program throughout the South. It is unlikely that
migration would have responded quickly to the pilot program. We provide further evidence on selective
migration in section VI.D.
VI. Results on Adult Outcomes from World War II Enlistment Records and the 1940 Census
A.

Years of Schooling
Table 5 reports a variety of results using the WWII enlistment records. The first two columns

examine the effects of Rosenwald exposure on completed years of schooling, with the columns
differentiated by whether inverse probability weighting (IPW) corrects for nonrandom sample selection.
Using the preferred IPW specification (column 2), we find that complete exposure raised Black rural
schooling levels by 1.2 years and had no meaningful effect on the other three groups, including urban
Blacks. Consequently, complete exposure to Rosenwald narrowed the Black-White rural difference, the
Black rural-urban difference, and the triple difference by 1.2 to 1.4 years.50 Based on these estimates,
Rosenwald exposure accounted for nearly 40 percent of the narrowing of the Black-White gap in
completed schooling for cohorts born between 1910 and 1925.51
These results are extremely similar to what we obtain when we use the 1940 Census, which does
not contain potential selection problems related to military enlistment. Unfortunately, county geocodes
are not currently available in the 1940 IPUMS and therefore we must rely on state economic areas
50

As explained in section III.C, our sharper classification of rural and urban counties eliminates those with rural
share greater than or equal to 50 percent but less than 100 percent. For this sample, we find complete Rosenwald
exposure effects on Blacks of: 0.74 (0.15) on years of education; 0.11 (0.02) on some high school; 0.04 (0.02) on
completed high school and 5.57 (1.97) on AGCT scores. In all cases, we find no significant effects for Whites. In
short, the results are always statistically significant but around 50 to 60 percent of the effect sizes estimated for the
Black-White rural differences, consistent with the attenuation we expect in mixed rural-urban counties.
51
The Black-White gap for Southern born men is about 3 years for the pre-Rosenwald cohorts born between 1905
and1909 (see Figure 1) for whom Ebc averaged only 0.01 and closes to a gap of 1.8 years for the 1925 birth cohort
for whom Ebc = 0.36. If we use Table 6’s average estimated effect from full Rosenwald exposure of 1.3 years, the
effect at the mean is 1.3 × (0.36 – 0.01) = 0.46. Therefore, Rosenwald explains about 38 percent (0.46/(3.0-1.8)) of
the closing of the gap across these cohorts.

30

(SEAs), aggregations of contiguous counties with similar economic characteristics developed by the
Census Bureau. We also cannot distinguish between rural and urban areas within an SEA. Nevertheless,
by using a cohort similar in age to WWII enlistees, specifically those aged 18 to 25 in 1940, we find that
complete exposure raised completed years of schooling of Blacks relative to Whites by 1.36 (0.18) years.
Our earlier findings on school attendance imply that about one-third of the gain in completed
schooling can be attributed to students who would not have attended any school in the absence of the
program. Specifically, in section V.A we estimated that the Rosenwald program caused approximately
90,000 additional students to attend schools during the early 1930s. If we further assume that these
students gained, on average, an additional 3 years of schooling, this would imply that increases in school
attendance contributed about 0.4 years to the overall increase of 1.2 to 1.4 years of completed schooling
that we estimate in Table 5.52 The remainder of the estimated increase in completed schooling, therefore,
can be attributed to greater time spent in school among those students who would have attended a nonRosenwald school even in the absence of the program. We think that this latter gain most likely reflects
the sharp improvement in school quality that the Rosenwald schools represented.
B.

High School Matriculation and Completion
In columns (3) and (4), we focus on the program’s effect on attending and completing high school

(hereafter, we only show the IPW results). The estimates suggest that full exposure to Rosenwald schools
is associated with a 17.0 percentage point increase in the probability of attending some high school
amongst rural Blacks. Again, we estimate no significant effects for urban and rural Whites and urban
Blacks. The gain relative to urban Blacks is 13.3 percentage points and the gain relative to rural Whites is
18.6 percentage points. The triple difference estimate is 20.4 percentage points and, like all of the key
estimates, highly significant.

The aggregate effect of the program would be to raise high school

attendance of rural Blacks by between 4.6 to 7.1 percentage points for cohorts born in 1925 compared to

52

We use three years in our calculation, because this was the average gain in education among Southern Blacks born
in 1925-29 relative to 1905-09. Our calculation is as follows: 90,000/650,000 × 3 + 560,000/650,000 × 0 = 0.42.

31

those born just prior to 1910. Raising high school attendance leads to a notable increase in high school
graduation. Among rural Blacks, complete exposure raises the probability of high school completion by
8.3 percentage points and by 8 to 9 percentage points relative to our control groups. This translates into an
aggregate increase in high school completion across cohorts of rural Blacks of about 3 percentage points.
C.

AGCT Scores
Column (5) presents the results on AGCT scores. Recall that we are relegated to using a vastly

smaller sample of individuals who enlisted over a short period in 1943 so estimates are less precise and
could be more susceptible to selection. Nevertheless, we again find quantitatively large and in some cases
statistically significant effects across our estimates of rural Blacks ranging from 5 to 10 points. For
example, the triple difference estimate is 8.0 (4.0) points. Given that the standard deviation of AGCT
scores for the full sample of Blacks and Whites is 23.5 points, our estimates suggest that full exposure
improved Black test scores by between 0.2 to 0.45 standard deviations and, in the aggregate, would have
led to an improvement of between 0.08 to 0.15 standard deviations for cohorts born in 1925 compared to
those born just prior to 1910.
Since the publication of Herrnstein and Murray (1994), several studies show that environmental
factors influence the Black-White test score gap (Neal and Johnson 1996, Hansen, Heckman, and Mullen
2004, Cascio and Lewis 2001, Chay, Guryan, and Mazumder 2009). Our results confirm that a very
straightforward intervention, the provision of schools, has a sizable effect on test scores.

Further

confirming that schooling influences these scores, we also find that the Rosenwald test score effect
disappears when we include educational attainment as a covariate in the regressions (column 6).
We note some concern over the surprising marginally significant negative effects on rural and
urban Whites. However, as we discuss below, these effects are largely eliminated when we turn to the
“young draftee” sample where selection is less of a concern.
D.

Further robustness checks

32

1. Height
One potential validity check is to measure the Rosenwald effect on outcomes for which we expect
additional schooling to have little, and perhaps no, influence. A good candidate is height, since effective
interventions on height are believed to be confined mostly to the early life period, well before children
enter school (Martorell, Schroeder, Rivera, and Kaplowitz 1995; Behrman and Hoddinott 2005). Since
the Rosenwald Initiative targeted children beyond the early life period and moreover was not designed to
treat childhood nutrition or health,53 we expect Rosenwald exposure to have no impact on height. That is
indeed what we report in column (7). These results do not suggest any obvious remaining confounding
factor.
2. Young Draftees
As we discuss in section IV.B, there may be concern that our initial strategies of using a wide
array of fixed effects and inverse probability weighting may not sufficiently remove all potential sources
of selection, especially those operating at the individual level where unobserved preferences may matter.
Therefore, to address this potential source of bias, we re-estimate our regressions for two outcomes,
education and AGCT scores, using the sample of young draftees described earlier. The results are shown
in columns (8) and (9). In short, we find the same general pattern of results as in the full sample.
Complete exposure to Rosenwald raised rural Black schooling levels by over 1.5 years and narrowed the
Black-White rural gap, the Black rural-urban gap, and the triple difference by between 0.8 to 1.3 years.
For test scores, the samples are considerably smaller since they include only 18-19 year old
draftees who enlisted over a 10 week period in 1943 and do not include county fixed effects. We find that
the simple undifferenced “Black Rural” estimate is now slightly larger at 8.4 percentage points using this
specification compared to column (5) and significant at the 10 percent level. We also find that the
negative effect on urban Whites is eliminated and the negative effect on rural Whites is greatly reduced.
53

While there is no historical documentation of health initiatives in the primary or secondary Rosenwald schools,
Julius Rosenwald had interest in health initiatives (Ascoli 2006). The Rosenwald school plans embraced some
school hygiene issues, including lighting, ventilation, and bathrooms (see www.preservationnation.org/travel-andsites/sites/southern-region/rosenwald-schools).

33

Our differenced estimates, however, indicate very similar-sized magnitudes as with our larger sample.
The triple difference estimate suggests that full exposure increased test scores by 7.8 percentage points
(5.7), nearly identical to the 8.0 point estimate for the full test score sample.
We also estimate the regressions (not shown) on just the subsample of 18 to 19 year old draftees
who enlisted from 1943 through 1946 for whom there was virtually no scope for volunteering prior to
their draft. Once more, we find similar results. For example, our four key estimates of the education
effect on rural Blacks range from 0.8 to 1.5 and are highly statistically significant. Similarly for attending
high school, our estimates range from 0.13 to 0.22 and for completing high school range from 0.02 to
0.07. Finally, the positive AGCT score effects already described in the previous paragraph are estimated
using this sample.
3. Exposure Based on County of Birth
To further address concerns about selective migration, we reran our WWII analysis on a select
subsample of WWII enlistees where a source of exposure determined prior to the time of school
attendance, namely exposure based on one’s county of birth, was available. The regression specification
is similar to equation (2):
(4)

 

   
 

but

 

is computed for county of birth rather than county of enlistment and counties are not

differentiated by their rural status. The results are shown in Table 6. For each outcome (years of
completed education, some high school, completed high school, and AGCT score), we display a first
column of results that use county of enlistment and the full sample and a second that is based on
enlistment county but restricts the sample to those we know county of birth. The third column reports
results that use county of birth exposure. We find statistically indistinguishable effects whether we use
county of enlistment or county of birth, regardless of the outcome. Consequently, we again find no
support for concern that our results are contaminated by selective migration.
34

E.

South-to-North Migration
Although we find little evidence of parents moving to counties with higher levels of Rosenwald

exposure, their children’s improved human capital could offer greater opportunity to relocate to superior
labor markets, which at the time chiefly meant Northern cities (e.g. Bowles 1970, Margo 1990, Card and
Krueger 1992). To study this question, we use the 1940 Census, the first to ask about migration, to run
the following regression:
(5)

 

,

,
,

The outcome

,

,

.

is either South-to-North or South-to-South migration, computed from a comparison

of the individual’s 1940 residential state economic area (SEA) to her residential SEA five years prior.
The sample consists of residents of Rosenwald states in 1935. The regression covariates include a female
indicator, a Black indicator, an indicator for Black and female, and a quadratic in age interacted with race
and 1935 state of residence.
Results for both migration measures are presented by age cohort in Table 7. We find that
complete Rosenwald exposure increased the likelihood of Northern migration among 17 to 21 year old
Blacks relative to 17 to 21 year old Whites by 2.5 (1.2) percentage points (column 2). By contrast, we
find no Black-White difference on within-South migration for this same age cohort (column 5). We also
find no Black-White difference within the school-age population or within those individuals aged 22 to
30, who were likely finished with school and potentially raising their own children. Indeed, the only
group of Southerners in the late 1930s that responded with their feet to higher Rosenwald exposure was
Blacks finishing school, and they increased their propensity to move North by over 60 percent, from 1.4
to 2.3 percent at a Rosenwald exposure rate typical for this cohort.

35

While we are certainly not the first to link education to the Great Migration, we believe our
results are the first to use a potentially exogenous source of educational improvement to make a causal
claim about the importance of human capital to Black Northern migration at that critical time.54
VII. Heterogeneous Effects
A.

By initial school conditions
Given the extremely poor educational conditions facing many rural Blacks prior to the Rosenwald

program and the inability to secure financing for schools through existing institutional arrangements, it
seems plausible that the introduction of the Rosenwald program disproportionately benefited those
students with high costs of schooling and with especially high marginal rates of return. We explore this
possibility by re-estimating our models using samples that are stratified by county rates of Black school
attendance in the period prior to the initiation of the Rosenwald program.
The first two columns of Panel A of Table 8 report these results.55 Regardless of which of the
three differenced estimators we report, the effects are statistically and economically larger for those
residing in counties where the 1910 Black attendance rate was at or below the median. Indeed, when we
split the sample into quartiles (unreported), the effects monotonically decline with higher initial Black
school attendance. By contrast, there is no consistent pattern when we stratify by pre-Rosenwald White
school attendance rates (columns 3 and 4). These results suggest that the program’s influence was largest
where the opportunity for Blacks to invest in schooling was lacking and therefore where there was
substantial room for progress.
In Panel B, we show comparable results for education and AGCT scores in the World War II data
stratified by 1920 Black school attendance rates.56 For years of completed education, we again see
sharply higher estimates in the counties that were at or below the median. For AGCT scores, we find

54

See also Duflo (2004) for a similar result following the Indonesian school construction program.
For this exercise, we use our baseline specification from column (3) of Table 2, where the full sample effect of
Rosenwald exposure on rural Blacks is 0.124, the Black rural-urban gap is 0.070, and the Black-White rural gap is
0.127. The triple difference estimate is 0.076. The coefficients are reported in Appendix Table A5.
56
We used 1920 for stratification since over 80 percent of the WWII sample entered school in 1920 or later.
55

36

considerable effects in the bottom half of the Black school attendance distribution and no effect for those
in the top half of the distribution.
B.

By Black Population Share and Past Slave Share
Researchers starting with Bond (1934) argue that Blacks historically fared worse in counties

where they were more numerous because Whites were successfully able to divert educational resources
by taking advantage of the political exclusion of Blacks (see also Margo 1990 and Card and Krueger
1996). These conditions arose in areas where there were large populations of slaves due to agricultural
conditions favoring certain crops (Fogel and Engerman 1974). The Rosenwald program might have
potentially overcame this exclusionary system by providing actual physical infrastructure that could not
be diverted. Therefore, we next examine how the effects of the schools varied over various direct
measures that reflect these historically deprived communities.
We first split the sample based on the Black share of the 1910 county population (Panel A,
columns 5 and 6). We find that the effects were significantly larger in the counties with high Black
population shares. In columns 7 and 8, we stratify by the share of 1890 county land that was used to
cultivate one of four labor intensive crops -- cotton, cane sugar, rice and tobacco -- using data from Chay
and Munshi (2011).

Here, we find larger point estimates in the counties that had high shares of

plantations, though the differences are not as stark as with high Black population shares. Finally, we
stratify by the slave share of the population in 1860 for counties that had at least some slaves using data
from Haines (2010). Columns 9 and 10 report sharply higher effects in those counties that had the most
slaves prior to the Civil War. In columns 11 and 12, we further stratify by whether a county was in the
top quartile in both plantation land and slave share. Again, the effects are particularly large for the
counties in which rural Blacks were most likely to have been historically disadvantaged.

37

Overall, the evidence in Table 8 strongly suggests that students schooled in the most
disadvantaged communities, as measured by conditions prone to slavery or by pre-Rosenwald levels of
Black schooling, benefited the most from access to the Rosenwald schools.57
VIII. Real Internal Rate of Return
In order to gauge the success of the program as a public policy investment, we calculate its real
internal rate of return. Since this requires many simplifying assumptions, the results of this exercise
should be viewed with some caution. The precise details of the calculation are provided in Appendix C.
We assume that the schools were gradually phased out by 195058 and therefore only cohorts of rural
Blacks born in Rosenwald states between 1906 through 1942 received benefits from the program. The
size of each cohort is estimated from the 1910 to 1950 Censuses.
We calculate the stream of yearly benefits over the 1926 to 2002 period by aggregating the labor
earnings experienced by all cohorts in that year. For each cohort, we estimate the implied effects of the
program on completed years of schooling based on their childhood exposure to the schools and our Table
5 estimates. We assume that the return to a year of education was either 5 or 7 percent and that these
returns were earned when individuals were between the ages of 20 and 60.59 The typical lifecycle
earnings profile to which we apply these returns is computed for Southern–born Blacks from the 1940 to
1980 Censuses. We calculate the stream of costs incurred between 1913 and 1950 using the total costs of
Rosenwald school construction and our estimates of teacher salaries, other school maintenance costs, and
the foregone contemporaneous earnings that Rosenwald students would have received during the

57

We have also run the results separately by age and gender. There appears to be some evidence, though not
uniformly consistent, that the impact of Rosenwald is larger for older students. This could reflect the possibility that
the effects have a cumulative effect over time. We find that the program’s effects are statistically similar for males
and females.
58
Since we have no information about the timing of school closures, we assume that the buildings were used for
about 20 years. There are some anecdotes of schools that remained open well into the 1950s.
59
The estimated return to a year of education for Southern Blacks in the 1940 and 1950 Census is 5 percent. We
suspect that the returns may have been even larger for those exposed to Rosenwald schools. This is based on an
analysis of the effects of the program on the earnings of Blacks who remained in the South in the 1940 and 1950
Census in an earlier draft of this paper where estimated returns were in excess of 10 percent.

38

additional 1.2 years that they were in school. Earnings and costs are deflated to 1925 dollars using the
Consumer Price Index.
We estimate that the real internal rate of return was between 7 and 9 percent. By way of
comparison, the average real yield on 8 to 12 year US treasury bonds was 5 percent during the 1919 to
1932 period.60 Moreover, this calculation does not include other potential benefits such as improvements
in family planning (Aaronson, Lange, and Mazumder 2011), health or intergenerational linkages.61
IX. Conclusions
At the turn of the twentieth century, the education infrastructure available to American Southern
Blacks, particularly those living in rural areas, resembled the conditions faced by some rural communities
in developing nations today. For example, rates of literacy and school enrollment in some developing
countries are as low or even lower today than they were among rural Blacks in 1910. But over a
moderately short period between the World Wars, the Southern racial education gap declined markedly.
While no single explanation likely accounts for this rapid convergence, we show that the Rosenwald
Rural Schools Initiative is a significant contributor, explaining about 40 percent of the narrowing of the
racial education gap among the cohorts that we study. Moreover, the program stimulated migration to
better labor market opportunities in the North. In sum, the Rosenwald Initiative highlights the large
productivity gains that can arise when substantial improvements to school quality and access are
introduced to relatively deprived environments. This conclusion is accentuated by the especially large
gains measured in communities that were contending with the worst pre-Rosenwald educational
conditions.
60

While the Rosenwald Initiative ended in 1932, many schools remained open well beyond. Therefore, a more
appropriate comparison could be to a longer period. For example, the average real yield between 1919 and 1939,
still well before WWII interest rate caps were in place, is under 3 percent. We take nominal yields (8 year U.S.
bonds from 1919 to 1925 and 12 year U.S. bonds thereafter) from the National Bureau of Economic Research’s
macro database (series 13033). Inflation expectations do not exist for this period. Instead, we use the average 8 or
12 year realization of future inflation rates, computed from the NBER macro database’s Consumer Price Index
(series m04128).
61
Aaronson and Mazumder (2008) estimate the intergenerational income elasticity to be roughly 0.4 to 0.5 for the
children of Rosenwald era students (including White and Black, North and South, born between 1930 and 1950).

39

Left somewhat unresolved are the channels by which the Rosenwald program improved student
outcomes. We find significant gains in school attendance suggesting progress along the quantity of
schools margin. Yet we speculate that the bulk of the gains in human capital may be attributable to
improvements in the quality of the schools that rural Black students attended. However, systematic data
do not exist for many of the relevant dimensions of school quality (e.g. teacher quality, curriculum,
physical infrastructure).

Further, traditional measures such as school spending may not adequately

capture school quality, especially in light of the persistent institutional inequities that accompanied
segregated schools.
More generally, the gains in human capital acquisition due to school improvements likely had
implications for economic development in the 20th century South, as well as for the US economy in
general. While beyond the scope of the current paper, we view this link as an important future research
question and the Rosenwald program as a useful contributor towards understanding the causal
relationship between human capital acquisition and economic progress.
In assessing the overall importance of the Rosenwald schools, one may reasonably argue that
racial convergence in educational standards with the North was inevitable. Indeed, by the time of Brown
vs. Board of Education in 1954, common measures of Black-White educational resource gaps had mostly
been closed (Card and Krueger 1992, Donohue, Heckman, and Todd 2002). Yet many observers -notably Booker T. Washington, but modern researchers such as Margo (1990) and Donohue, Heckman,
and Todd (2002) as well -- point to the fundamental funding inequities driven partly by institutional
discrimination, and likely exacerbated by liquidity constraints, to argue that major investments in Black
schools required outside intervention in the early part of the 20th century. The racial convergence that
occurs in relatively short-order after the introduction of the Rosenwald program seems to validate the
view that some prodding was necessary. Subsequent progress in Black educational attainment and
cognitive skill development has likewise been significantly aided by a series of private and public
interventions, including but not limited to NAACP litigation (Donohue, Heckman, and Todd 2002),

40

desegregation of schools and hospitals (Welch and Light 1987, Guryan 2004, Chay, Guryan, and
Mazumder 2009), and civil rights legislation (Donahue and Heckman 1991).
Our results may also inform the historical literature concerning the nature of Black economic
progress in the first half of the 20th century. Margo (1990) presents a framework in which a combination
of human capital (supply side), institutional factors (demand side), and “intergenerational drag” all played
a role in keeping the relative earnings of adult Black men flat from 1900 to 1940. We show that an
expansive schooling intervention in the South did in fact have a sizable effect on the level of human
capital of Blacks born after 1910. Our study suggests that the relative importance of the supply of human
capital may have played a more consequential role than previously thought in accounting for early 20th
century Black economic progress.

41

References
Aaronson, Daniel, Fabian Lange and Bhashkar Mazumder. 2011. “The Essential Complementarity of the Quality
and Quantity of Children.” Working paper, Federal Reserve Bank of Chicago.
Aaronson, Daniel and Bhashkar Mazumder. 2008. “Intergenerational Economic Mobility in the US, 1940 to 2000 .”
Journal of Human Resources 43(1): 139-172.
Ascoli, Peter. 2006. Julius Rosenwald: The Man Who Built Sears, Roebuck and Advanced the Cause of Black
Education in the American South. Bloomington, IN: Indiana University Press.
Banerjee, Abhijit and Esther Duflo. 2006. “Addressing Absence.” Journal of Economic Perspectives 20(1): 117132.
Becker, Gary. 1964. Human Capital: A Theoretical and Empirical Analysis, with Special Reference to Education.
Chicago, IL: University of Chicago Press.
Becker, Gary. 1967. Human Capital and the Personal Distribution of Income: An Analytical Approach. Ann Arbor,
MI: University of Michigan Press.
Behrman, Jere and John Hoddinott. 2005. “Programme Evaluation with Unobserved Heterogeneity and Selective
Implementation: The Mexican PROGRESA Impact on Child Nutrition.” Oxford Bulletin of Economics and Statistics
67(4): 547-569.
Bester, C. Alan, Timothy Conley, and Christian Hansen. 2010. “Inference with Dependent Data Using Cluster
Covariance Estimators.” Journal of Econometrics, forthcoming.
Brannon, J. 1919. “Report of State Agent for Negro Schools.” in Annual Report of the State Superintendent of
Education of the State of South Carolina, 1920, Volume 2. Columbia, SC: South Carolina Department of Education.
131.
Bleakley, Hoyt. 2007. “Disease and Development: Evidence from Hookworm Eradication in the American South.”
Quarterly Journal of Economics 122(1): 73-117.
Bond, Horace Mann. 1934. The Education of the Negro in the American Social Order. New York, NY: Octagon
Press.
Bond, Horace Mann. 1969. Negro Education in Alabama: A Study in Cotton and Steel. New York, NY: Octagon
Press.
Bowles, Samuel. 1970. “Migration as Investment: Empirical Tests of the Human Capital Approach to Geographic
Mobility.” Review of Economics and Statistics 52(4): 356-362.
Card, David and Alan Krueger. 1992. “School Quality and Black-White Relative Earnings: A Direct Assessment.”
Quarterly Journal of Economics 107(1): 151-200.
Card, David and Alan Krueger. 1996. “School Resources and Student Outcomes: An Overview of the Literature and
New Evidence from North and South Carolina.” Journal of Economic Perspectives 10(4): 31-50.
Card, David. 1995. “Earnings, Schooling, and Ability Revisited.” Research in Labor Economics, Volume 14: 23-48.
Cascio, Elizabeth and Ethan Lewis. 2006. “Schooling and the Armed Forces Qualifying Test: Evidence from
School-Entry Laws.” Journal of Human Resources 41(2): 294-318.

42

Chaudhury, Nazmul, Jeffrey Hammer, Michael Kremer, Karthik Muralidharan, and Halsey Rogers. 2006. “Missing
in Action: Teacher and Health Worker Absence in Developing Countries.” Journal of Economic Perspectives 20(1):
91-116.
Chay, Kenneth, Jonathan Guryan and Bhashkar Mazumder. 2009. “Birth Cohort and the Black-White Achievement
Gap: The Role of Health Soon After Birth.” Working paper, Federal Reserve Bank of Chicago.
Chay, Kenneth and Kaivan Munshi. 2011. “Slavery's Legacy: Black Mobilization in the Postbellum South”. Brown
University typescript.
Coelho, Philip and Robert McGuire. 2006. "Racial Differences in Disease Susceptibilities: Intestinal Worm
Infections in the Early Twentieth-Century American South." Social History of Medicine 19(3): 461-482.
Collins, William and Robert Margo. 2006. “Historical Perspectives on Racial Differences in Schooling in the United
States.” In Handbook of the Economics of Education: Volume 1, edited by E. Hanushek and F. Welch. New York:
North-Holland. 107-154.
Craig, Lee A. To Sow One Acre More: Childbearing and Farm Productivity in the Antebellum North. Baltimore:
Johns Hopkins University Press, 1993.
Das, Jishnu, Stefan Dercon, James Habyarimana, and Pramila Krishnan. 2007. “Teacher Shocks and Student
Learning: Evidence from Zambia.” Journal of Human Resources 42(4): 820-862.
Donohue, John and James Heckman, 1991. “Continuous versus Episodic Change: The Impact of Civil Rights Policy
on the Economic Status of Blacks.” Journal of Economic Literature 29(4): 1603-1643.
Donohue, John, James Heckman, and Petra Todd. 2002. “The Schooling of Southern Blacks: The Roles of Legal
Activism and Private Philanthropy, 1910-1960.” Quarterly Journal of Economics 117(1): 225-268.
Duflo, Esther. 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, Esther. 2004. “The Medium Run Effects of Educational Expansion: Evidence from a Large School
Construction Program in Indonesia.” Journal of Development Economics 74(1): 163-197.
Embree, Edwin. 1936. Julius Rosenwald Fund: Review of Two Decades, 1917-1936. Chicago, IL: The Julius
Rosenwald Fund.
Ferrie, Joseph, Karen Rolf, and Werner Troesken. 2009. “… Healthy, Wealthy, and Wise? Physical, Economic and
Cognitive Effects of Early Life Conditions on Later Life Outcomes in the U.S., 1915-2005." Working paper,
Northwestern University.
Feyrer, James, Dimintra Politi, and David Weil. 2008. “The Economic Effects of Micronutrient Deficiency:
Evidence from Salt Iodization in the United States.” Working paper, Dartmouth College.
Fogel, Robert and Stanley Engerman. 1974. Time on the Cross. Boston, MA: Little Brown.
Glewwe, Paul and Michael Kremer. 2006. “Schools, Teachers, and Education Outcomes in Developing Countries.”
in Handbook of the Economics of Education: Volume 2, edited by Eric Hanushek and Finis Welch. New York:
Elsevier. 945-1017.
Goldin, Claudia and Lawrence Katz. 1999. “Human Capital and Social Capital: The Rise of Secondary Schooling in
America, 1910 to 1940.” Journal of Interdisciplinary History 29(4): 683-723.
Guryan, Jonathan. 2004. “Desegregation and Black Dropout Rates.” American Economic Review 94(4): 919-943.

43

Haines, Michael. 2010. “Historical, Demographic, Economic, and Social Data: The United States, 1790-2002.”
ICPSR02896-v3 data file. Ann Arbor, MI: Inter-university Consortium for Political and Social Research.
Hansen, Karsten, James Heckman, and Kathleen Mullen. 2004. “The Effect of Schooling and Ability on
Achievement Test Scores.” Journal of Econometrics. 121(1-2): 39-98.
Herrnstein, Richard and Charles Murray. 1994. The Bell Curve: Intelligence and Class Structure in American Life.
New York: Free Press.
Hoffschwelle, Mary. 2006. The Rosenwald Schools of the American South. Gainesville, FL: University Press of
Florida.
Hirano, Keisuke, Guido Imbens, and Geert Ridder. 2003. “Efficient Estimation of Average Treatment Effects Using
the Estimated Propensity Score.” Econometrica 71(4), 1161-1189.
Johnson, Charles. 1941. Statistical Atlas of Southern Counties: Listing and Analysis of Socio-Economic Indices of
1104 Southern Counties. Chapel Hill: University of North Carolina Press.
Karpinos, Bernard. 1958. “Height and Weight of Selective Service Registrants Processed for Military Service
During World War II.” Human Biology 30(4): 292-321.
Keller, Alvin, W. S. Leathers and Paul Densen, 1940. "The Results of Recent Studies of Hookworm in Eight
Southern States." American Journal of Tropical Medicine s1-20(4): 493-509.
Kremer, Michael, Nazmul Chaudhury, Halsey Rogers, Karthik Muralidharan, and Jeffrey Hammer. 2005. “Teacher
Absences in India: A Snapshot.” Journal of the European Economic Association 3(2-3): 658-667.
Lew, Edward. 1944. “Interpreting the Statistics of Medical Examinations of Selectees.” Journal of the American
Statistical Association 39(227): 345-356.
Lewis, Maureen and Gunilla Pettersson. 2009. “Governance in Education: Raising Performance.” Working paper,
World Bank.
Lleras-Muney, Adriana. 2002. “Were Compulsory Attendance and Child Labor Laws Effective? An Analysis from
1915 to 1939.” Journal of Law and Economics 45(2): 401-435.
Margo, Robert. 1990. Race and Schooling in the South, 1880-1950. Chicago, IL: University of Chicago Press.
Martorell, Reynaldo, Dirk Schroeder, Juan Rivera, and Haley Kaplowitz. 1995. “Patterns of linear growth in rural
Guatemalan adolescents and children.” Journal of Nutrition 125(4): 1060S-1067S.
McCormick, Scott. 1934. “The Julius Rosenwald Fund.” The Journal of Negro Education 3(4): 605-626.
Myrdal, Gunnar. 1944. An American Dilemma: The Negro Problem and Modern Democracy. New York: Harper
and Row.
Neal, Derek. 2006. “Why has Black-White Convergence Stopped?” In Handbook of the Economics of Education:
Volume 1, edited by Eric Hanushek and Finis Welch. New York: Elsevier. 511-576.
Neal, Derek and William Johnson. 1996. “The Role of Premarket Factors in Black-White Wage Differences.”
Journal of Political Economy 104(5): 869-895.
Perrott, J. 1946. “Selective Service Rejection Statistics and Some of Their Implications.” American Journal of
Public Health 36: 336-342.

44

Reed, Betty. 2004. The Brevard Rosenwald School: Black Education and Community Building in a Southern
Applachian Town, 1920-1966. Jefferson, NC: McFarland and Company.
Ruggles, Steven, J. Trent Alexander, Katie Genadek, Ronald Goeken, Matthew B. Schroeder, and Matthew Sobek.
2010. Integrated Public Use Microdata Series: Version 5.0 [Machine-readable database]. Minneapolis: University
of Minnesota.
Schultz, Theodore W. 1961. “Investment in Human Capital.” American Economic Review 51(1): 1-17.
Smith, James. 1984. "Race and Human Capital." American Economic Review 74(4): 685-698.
Smith, James and Finis Welch. 1989. "Black Economic Progress after Myrdal." Journal of Economic Literature
27(2): 519-564.
Staff, Personnel Research Section, Adjutant General's Office. 1947. “The Army General Classification Test, With
Special Reference to the Construction and Standardization of Forms 1a and 1b.” Journal of Educational Psychology.
38: 385–420.
UNICEF. 2011. “WASH in Schools, Call to Action.”
http://www.unicef.org/wash/schools/washinschools_53108.html
UN Millennium Project. 2005. Toward Universal Primary Education: Investments, Incentives and Institutions.
London: Earthscan.
Weathers, Lindsay. 2008. “The Rosenwald School Building Program in South Carolina, 1917-1932.” submission to
the National Register of Historic Places, National Park Services, U.S. Department of the Interior.
Welch, Finis and Audrey Light. 1987. “New Evidence on School Desegregation.” U. S. Commission on Civil Rights
Clearinghouse, Publication 92.
Wooldridge, Jeffrey. 2002. “Inverse Probability Weighted M-Estimators for Sample Selection, Attrition, and
Stratification.” Portuguese Economic Journal 1(2): 117-139.
www.preservationnation.org/travel-and-sites/sites/southern-region/rosenwald-schools.2008.

45

Table 1:  School Attendance Effects of Rosenwald School Presence in County
(1)
0.011
[0.007]

(2)
0.014
[0.006]**

(3)
0.001
[0.007]

(4)
0.010
[0.007]

γ1

0.024
[0.010]**

0.017
[0.008]**

0.034
[0.009]***

0.022
[0.008]***

γ2

‐0.013
[0.007]*

‐0.012
[0.006]**

0.004
[0.006]

‐0.001
[0.005]

γ3

0.067
[0.011]***

0.047
[0.010]***

0.055
[0.010]***

0.041
[0.010]***

0.089
[0.007]***

Differences (Rose minus no Rose)
0.066
0.094
[0.007]***
[0.007]***

0.072
[0.007]***

γ0

Black Rural
(γ 0 + γ 1 + γ 2 + γ 3)
White Rural
(γ 0 + γ 2)

‐0.002
[0.004]

0.002
[0.004]

0.005
[0.006]

0.008
[0.005]

Black Urban
(γ 0 + γ 1)

0.034
[0.008]***

0.031
[0.008]***

0.036
[0.009]***

0.032
[0.009]***

White Urban
(γ 0)

0.011
[0.007]

0.014
[0.006]**

0.001
[0.007]

0.010
[0.007]

Difference In Difference
0.035
0.059
[0.009]***
[0.009]***

Black, Rural Urban
Black, Rural‐Urban
(γ 2 + γ 3)

0.054
[0.009]***

0.040
[0.009]***

White, Rural‐Urban
(γ 2)

‐0.013
[0.007]*

‐0.012
[0.006]**

0.004
[0.006]

‐0.001
[0.005]

Black‐White Rural
(γ 1 + γ 3)

0.091
[0.006]***

0.065
[0.006]***

0.089
[0.006]***

0.063
[0.006]***

Black‐White Urban
(γ 1)

0.024
[0.010]**

0.017
[0.008]**

0.034
[0.009]***

0.022
[0.008]***

Triple Difference
B‐W Rural ‐ B‐W Urban
(γ 3)
Controls
County Fixed Effects
N

0.067
[0.011]***

0.047
[0.010]***

0.055
[0.010]***

0.041
[0.010]***

N
N

Y
N

N
Y

Y
Y

650167

650167

650167

650167

Notes: Samples include children between the ages of 7 and 17 in the 1900, 1910, 1920 and 1930 IPUMs.  
Dependent variable is school attendance.  Columns 1 and 3 only include year dummies.  The controls in 
columns 2 and 4 include year dummies, age, female dummy, father's and  mother's literacy, father's 
occupational score and father's home ownership.  Column 2 also includes state fixed effects and county 
White literacy rate in 1910.  Estimates use Census sampling weights.  Standard errors clustered on county 
are shown in brackets.  
* significant at 10%; ** significant at 5%; *** significant at 1% 

Table 2:  School Attendance Effects of Rosenwald Exposure
(1)
0.005
[0.010]

(2)
‐0.002
[0.012]

(3)
0.002
[0.015]

(4)
0.138
[0.021]***

(5)
0.147
[0.028]***

(6)
0.137
[0.028]***

γ1

0.045
[0.013]***

0.055
[0.013]***

0.052
[0.013]***

0.05
[0.013]***

0.029
[0.077]

0.030
[0.075]

γ2

‐0.025
[0.009]***

‐0.016
[0.009]*

‐0.006
[0.011]

‐0.008
[0.009]

0.006
[0.033]

‐0.001
[0.033]

γ3

0.093
[0.016]***

0.081
[0.016]***

0.076
[0.017]***

0.065
[0.016]***

0.057
[0.089]

0.043
[0.088]

Black Rural
(γ 0 + γ 1 + γ 2 + γ 3)

0.119
[0.012]***

0.119
[0.014]***

0.239
[0.042]***

0.209
[0.042]***

White Rural
(γ 0 + γ 2)

‐0.020
[0.006]***

‐0.017
[0.008]**

‐0.003
[0.009]

0.13
[0.019]***

0.153
[0.018]***

0.136
[0.018]***

Black Urban
(γ 0 + γ 1)

0.051
[0.016]***

0.053
[0.019]***

0.054
[0.022]**

0.188
[0.024]***

0.177
[0.071]**

0.166
[0.070]**

White Urban
(γ 0)

0.005
[0.010]

‐0.002
[0.012]

0.002
[0.015]

0.138
[0.021]***

0.147
[0.028]***

0.137
[0.028]***

Black, Rural‐Urban
(γ 2 + γ 3)

0.069
[0.017]***

0.066
[0.018]***

0.062
[0.083]

0.042
[0.082]

White, Rural‐Urban
Whit R l U b
(γ 2)

‐0.025
0 025
[0.009]***

‐0.016
0 016
[0.009]*

‐0.006
0 006
[0.011]

‐0.008
0 008
[0.009]

0 006
0.006
[0.033]

‐0.001
0 001
[0.033]

Black‐White Rural
(γ 1 + γ 3)

0.139
[0.011]***

0.136
[0.012]***

0.127
[0.012]***

0.115
[0.011]***

0.086
[0.045]*

0.073
[0.046]

Black‐White Urban
(γ 1)

0.045
[0.013]***

0.055
[0.013]***

0.052
[0.013]***

0.050
[0.013]***

0.029
[0.077]

0.030
[0.075]

γ0

Difference (Effect of Complete Exposure) 
0.124
0.245
[0.014]***
[0.022]***

Difference In Difference
0.070
0.056
[0.021]***
[0.015]***

Triple Difference
B‐W Rural ‐ B‐W Urban
(γ 3)
Baseline Controls
Age‐St.‐Yr
County F.E
County by Year F.E.
Family F.E
Birth Order
N

0.093
[0.016]***

0.081
[0.016]***

0.076
[0.017]***

0.065
[0.016]***

0.057
[0.089]

0.043
[0.088]

Y
N
N
N
N
N

Y
N
Y
N
N
N

Y
Y
Y
N
N
N

Y
Y
N
Y
N
N

N
N
N
N
Y
N

N
N
N
N
Y
Y

643284

643284

643284

643284

482346

482346

Notes: Samples include children between the ages of 7 and 17 in the 1910, 1920 and 1930 IPUMs.  Dependent variable is 
school attendance.  Estimates show the effect of complete exposure (exposure = 1) to Rosenwald schools between the 
ages of 7 and 13 relative to no exposure (exposure=0).  The controls include year dummies, age, female dummy, father's 
and  mother's literacy, county White literacy rate in 1910 (column 1 only), father's occupational score and father's home 
ownership and state dummies  (column 1 only).  Estimates use Census sampling weights.  Standard errors clustered on 
county are shown in brackets except for columns 5 and 6 which cluster on families.  
* significant at 10%; ** significant at 5%; *** significant at 1% 

Table 3:  Literacy Effects of Rosenwald School Presence in County, or Rosenwald School Exposure
(1)

(2)
Rosenwald Presence
‐0.030
‐0.018
[0.005]***
[0.005]***

(4)
Rosenwald Exposure
‐0.058
‐0.051
[0.009]***
[0.011]***

γ1

0.052
[0.008]***

0.039
[0.007]***

0.086
[0.017]***

0.083
[0.016]***

γ2

0.020
[0.005]***

0.018
[0.004]***

0.029
[0.008]***

0.012
[0.008]

γ3

0.064
[0.010]***

0.053
[0.009]***

0.182
[0.022]***

0.165
[0.020]***

γ0

Black Rural
(γ 0 + γ 1 + γ 2 + γ 3)

 Differences (Rose minus no Rose)
0.106
0.093
[0.007]***
[0.006]***

(3)

Difference (Effect of Exposure) 
0.239
0.209
[0.017]***
[0.015]***

White Rural
(γ 0 + γ 2)

‐0.010
[0.003]***

0.000
[0.003]

‐0.029
[0.008]***

‐0.039
[0.008]***

Black Urban
(γ 0 + γ 1)

0.022
[0.008]***

0.021
[0.007]***

0.028
[0.014]*

0.032
[0.011]***

White Urban
(γ 0)

‐0.03
[0.005]***

‐0.018
[0.005]***

‐0.058
[0.009]***

‐0.051
[0.011]***

Black, Rural‐Urban
(γ 2 + γ 3)

 Difference In Difference
0.084
0.071
[0.010]***
[0.008]***

Difference In Difference
0.211
0.177
[0.020]***
[0.017]***

White, Rural‐Urban
(γ 2)

0.020
[0.005]***

0.018
[0.004]***

0.029
[0.008]***

0.012
[0.008]

Black‐White Rural
(γ 1 + γ 3)

0.116
[0.006]***

0.092
[0.005]***

0.268
[0.018]***

0.248
[0.016]***

Black‐White Urban
(γ 1)

0.052
[0.008]***

0.039
[0.007]***

0.086
[0.017]***

0.083
[0.016]***

 Triple Difference
B‐W Rural ‐ B‐W Urban
(γ 3)
Controls
County F.E.
County by Year F.E.
N

 Triple Difference

0.064
[0.010]***

0.053
[0.009]***

0.182
[0.022]***

0.165
[0.020]***

N
N
N
431976

Y
Y
N
431976

Y
Y
N
425115

Y
N
Y
425115

Notes: Samples includes individuals between the ages of 15 and 22 in the 1900, 1910, 1920 and 1930 IPUMs.  
Dependent variable is literacy.  Estimates in columns 1 and 2 show the effect of the presence of a Rosenwald school in 
one's county as of the Census year.  Estimates in columns 3 and 4 show the effect of complete exposure (exposure = 1) 
to Rosenwald schools between the ages of 7 and 13 relative to no exposure (exposure=0).  The sample sizes are lower 
in columns 3 and 4 because there are a few Rosenwald counties for which we cannot calculate exposure.  The controls 
include year dummies, age, female dummy, father's and  mother's literacy, father's occupational score and father's 
home ownership.  Specifications without county fixed effects also include state fixed effects and county White literacy 
rate in 1910.  Estimates use Census sampling weights.  Standard errors clustered on county are shown in brackets. 
* significant at 10%; ** significant at 5%; *** significant at 1% 

Table 4:  School Attendance Effects of Rosenwald County Presence Using Alabama Borders

Panel A:  7 to 17 year olds
(1)
AL/GA
Border
γ0
(White)

(2)
AL/FL
Border

(3)
AL/MS
Border

(4)
AL/TN
Border

(5)
All
Borders

0.044
[0.034]

‐0.044
[0.034]

‐0.023
[0.041]

0.012
[0.024]

0.096
[0.050]*
4878

γ1
(Black‐White)
N

0.096
[0.085]
0.031
[0.065]
1505

0.032
[0.056]
4188

0.100
[0.100]
2391

0.051
[0.032]
12438

Panel B:  9 to 17 year olds
(1)
AL/GA
Border
γ0
(White)
γ1
(Black‐White)
N

(2)
AL/FL
Border

(3)
AL/MS
Border

(4)
AL/TN
Border

(5)
All
Borders

0.053
[0.037]

0.088
[0.069]

‐0.055
[0.027]*

‐0.040
[0.051]

0.007
[0.025]

0.117
[0.042]***
3905

‐0.014
[0.106]
1194

0.070
[0.059]
3340

0.105
[0.126]
1909

0.076
[0.035]**
9929

Notes: The sample uses 7 to 17 year olds or 9 to 17 year olds in the 1900, 1910 and 1920 Censuses living in
any of the counties on either side of Alabama's borders to estimate the effect of the presence of a Rosenwald
school in an Alabama county by 1920 on White school attendance (γ0) and the Black‐White difference in
school attendance (γ1) by specific border and for all borders pooled together. Note that certain Alabama
counties may appear in more than one border comparison. For this reason, the sample in column 5 is not
equal to the sum of the samples in columns 1 to 4. The controls include year dummies, age, female dummy,
father's and mother's literacy, father's occupational score, father's home ownership and county fixed effects.
Regressions also control for the presence of a Rosenwald school in non‐Alabama counties interacted with
being black. Estimates use Census sampling weights. Standard errors clustered on county are shown in
brackets.
* significant at 10%; ** significant at 5%; *** significant at 1% 

Table 5:  Effects of Rosenwald Exposure on Outcomes in World War II Data
"Young Draftees"
(8)
(9)
AGCT
Education scores
1.077
0.025
[0.635]
[1.372]

All ages and years, Volunteers and Draftees 
(2)
(3)
(4)
(5)
(6)
Some
Complete
AGCT
AGCT
Education
H.S.
H.S.
scores
incl. Ed.
0.061
0.048
0.056
0.003
‐2.275
‐1.769
[0.131]
[0.119]
[0.019]
[0.018]
[1.198]* [1.054]*

Height
‐0.027
[0.139]

γ1

‐0.017
[0.298]

‐0.131
[0.256]

‐0.017
[0.043]

‐0.007
[0.028]

2.008
[1.971]

‐0.328
[2.796]

0.033
[0.110]

‐0.271
[0.312]

2.775
[3.134]

γ2

‐0.146
[0.164]

‐0.100
[0.157]

‐0.071
[0.024]

‐0.004
[0.022]

‐2.867
[3.170]

‐0.010
[2.512]

‐0.051
[0.177]

‐0.561
[0.831]

‐2.258
[1.831]

‐1.986
[3.941]

‐0.191
[0.175]

1.335
[0.411]***

7.832
[5.714]

(1)

γ0

γ3

1.186
1.377
0.204
0.090
8.033
[0.367]*** [0.339]*** [0.056]*** [0.036]*** [4.006]**

(7)

Difference (Effect of Complete Exposure) 
Black Rural
1.084
1.193
0.171
0.083
4.899
‐4.094
‐0.235
(γ 0 + γ 1 + γ 2 + γ 3) [0.232]*** [0.228]*** [0.039]*** [0.024]*** [4.156]
[3.352]
[0.155]

1.580
8.374
[0.600]*** [4.615]*

White Rural
(γ 0 + γ 2)

‐0.085
[0.097]

‐0.053
[0.102]

‐0.015
[0.015]

0.000
[0.014]

‐5.142
[2.935]*

‐1.779
[2.282]

‐0.078
[0.109]

0.516
[0.535]

‐2.232
[1.213]*

Black Urban
(γ 0 + γ 1)

0.044
[0.279]

‐0.083
[0.244]

0.038
[0.043]

‐0.003
[0.024]

‐0.267
[2.243]

‐2.097
[2.549]

0.007
[0.158]

0.806
[0.692]

2.800
[2.744]

White Urban
(γ 0)

0.061
[0.131]

0.048
[0.119]

0.056
[0.019]

0.003
[0.018]

‐2.275
[1.198]*

‐1.769
[1.054]*

‐0.027
[0.139]

1.077
[0.635]

0.025
[1.372]

Difference In Difference
1.040
1.276
0.133
0.086
5.166
‐1.996
[0.362]*** [0.334]*** [0.058]*** [0.034]*** [4.723]
[4.230]

‐0.242
[0.222]

0.774
[0.915]

5.574
[5.369]

‐0.010
[2.512]

‐0.051
[0.177]

‐0.561
[0.831]

‐2.258
[1.831]

‐2.314
[2.747]

‐0.158
[0.137]

‐0.328
[2.796]

0.033
[0.110]

‐0.271
[0.312]

2.775
[3.134]

‐1.986
[3.941]

‐0.191
[0.175]

1.335
[0.411]***

7.832
[5.714]

Black, Rur‐Urb
(γ 2 + γ 3)
White, Rur‐Urb
(γ 2)
B‐W Rural
(γ 1 + γ 3)
B‐W Urban
(γ 1)

‐0.146
[0.164]

‐0.100
[0.157]

‐0.071
[0.024]

‐0.004
[0.022]

‐2.867
[3.170]

1.169
1.246
0.186
0.083
10.041
[0.215]*** [0.221]*** [0.037]*** [0.022]*** [3.487]***
‐0.017
[0.298]

‐0.131
[0.256]

‐0.017
[0.043]

‐0.007
[0.028]

2.008
[1.971]

1.064
10.606
[0.267]*** [4.779]**

Triple Difference
BW Rur ‐ BW Urb
(γ 3)
County F.E
Inverse Prob. Wts
N

1.186
1.377
0.204
0.090
8.033
[0.367]*** [0.339]*** [0.056]*** [0.036]*** [4.006]**
Y
N

Y
Y

Y
Y

Y
Y

Y
Y

Y
Y

Y
Y

Y
Y

N
Y

980020

980020

980020

980020

50239

50239

464698

196930

18693

Notes: Sample is drawn from World War II enlistment records and includes men born between 1910 and 1928 who enlisted 
between 1940 and 1946 and who lived in either entirely rural or predominantly urban counties based on the 1910‐1930 
Census (see text for details).   Estimates show the effect of complete exposure (exposure = 1) to Rosenwald schools 
between the ages of 7 and 13 relative to no exposure (exposure=0).  The controls include quarter of enlistment  dummies 
interacted with race (except for columns 5, 6 and 9), age dummies interacted with race, and county fixed effects.  Columns 
2 through 9 use the inverse of the probability of being in the military by race, county and year of birth.  Standard errors 
clustered by county are shown in brackets.  
* significant at 10%; ** significant at 5%; *** significant at 1% 

Table 6:  Effects of Using Rosenwald Exposure Based on County of Birth
(1)

  
γ0
(White)
γ1
(Black‐White)
County Exp. Measure
Sample type
N

γ0
(White)
γ1
(Black‐White)
Cty. Exp. Measure
Sample type
N

(2)
(3)
Education
‐0.033
‐0.337
‐0.466
[0.052]
[0.281]
[0.355]
0.534
0.951
0.668
[0.188]*** [0.260]*** [0.209]***

(5)
(6)
Some High School
0.016
0.006
‐0.071
[0.008]*
[0.045]
[0.059]
0.071
0.145
0.094
[0.027]*** [0.042]*** [0.032]***

Residence
Full
2035356

Residence
Full
2035356

Residence
Restricted
17550

Birth
Restricted
17371

(7)
(8)
(9)
Completed High School
‐0.001
‐0.018
‐0.038
[0.009]
[0.040]
[0.051]
0.031
0.069
0.089
[0.019]*
[0.036]* [0.031]***
Residence
Full
2035356

Residence
Restricted
17550

Birth
Restricted
17371

(4)

Residence
Restricted
17550

(10)

(11)
AGCT Score
‐0.223
‐2.110
[0.738]
[7.339]
4.91
7.448
[1.653]*** [3.619]**
Residence
Full
92971

Residence
Restricted
4129

Birth
Restricted
17371
(12)
1.702
[7.949]
4.991
[3.408]
Birth
Restricted
4071

Notes: Sample uses the World War II enlistment records and includes men born between 1910 and 1928 
who enlisted between 1940 and 1946.  The "restricted" samples use only a subset of men who were residing 
in Rosenwald counties who also could be matched to SSA death records, and who provided SSA a place of 
birth that is easily matched to a county.   
* significant at 10%; ** significant at 5%; *** significant at 1% 

Table 7:  Effects of Rosenwald Exposure on Migration (1940 Census)

(1)
(2)
(3)
South to North Migration
Age in 1940
8 to 16 
17 to 21
22 to 30

(4)
(5)
(6)
South to South Migration
Age in 1940
8 to 16 
17 to 21
22 to 30

γ0
(White)

0.005
[0.006]

0.000
[0.007]

0.009
[0.010]

0.013
[0.013]

0.014
[0.015]

0.002
[0.011]

γ1
(Black‐White)

0.004
[0.007]

0.025
[0.012]**

‐0.014
[0.013]

0.008
[0.018]

0.000
[0.019]

0.011
[0.025]

N

68044

35750

54521

68044

35750

54521

Mean for Blacks

0.008

0.014

0.022

0.035

0.052

0.070

Notes: The sample uses individuals between the ages of 8 and 30 in the 1940 IPUMS who lived in a 
Rosenwald state in 1935.  Controls in Panel B include a female dummy, female*Black dummy, a 
quadratic in age interacted with state interacted with race.  Standard errors clustered on state 
economic area are shown in brackets.    
* significant at 10%; ** significant at 5%; *** significant at 1% 

Table 8:  Results Based on Stratified Samples
Panel A:  Census School Attendance Results Stratified by…
1910 Black
1910 White
1910 Black
1890 Plantation
1860
School Attendance
School Attendance
Population Share
Share of Land
Slave Share (if >0)
at/below
above
at/below
above
at/below
above
at/below
above
at/below
above
Median
Median
Median
Median
Median
Median
Median
Median
Median
Median
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
Black, Rur‐Urb
0.122
0.013
0.091
0.058
0.014
0.090
0.042
0.086
0.019
0.084
[0.025]*** [0.019] [0.032]*** [0.019]*** [0.027] [0.025]*** [0.029] [0.027]*** [0.025] [0.021]***
(γ 2 + γ 3)
B‐W Rural
(γ 1 + γ 3)

0.207
0.058
0.119
0.148
0.067
0.138
0.108
0.130
0.063
0.147
[0.019]*** [0.015]*** [0.016]*** [0.016]*** [0.022]*** [0.013]*** [0.018]*** [0.015]*** [0.021]*** [0.015]***

B‐W Rural ‐ 
0.146
B‐W Urban (γ 3) [0.028]***
N

255887

0.015
[0.020]
296191

0.072
0.087
[0.025]*** [0.021]***
337151

296440

0.027
[0.028]
281651

0.081
0.063
0.084
[0.020]*** [0.029]** [0.021]***
352418

203511

333198

0.044
[0.028]

0.078
[0.023]***

226559

288338

Top quart.
Slave Share excluding
AND
those in 
Plantation
(11)
(11)
(12)
0.082
0.048
[0.042]* [0.020]**

Age
7 to 10
11 to 13 14 to 17
(13)
(14)
(15)
0.073
0.123
0.045
[0.017]*** [0.018]*** [0.030]

Male
Female
(16)
(17)
0.060
0.077
[0.026]** [0.020]***

B‐W Rur.
(γ 1 + γ 3)

0.224
0.096
[0.037]*** [0.015]***

0.096
0.129
0.176
[0.013]*** [0.014]*** [0.019]***

0.126
0.129
[0.014]*** [0.013]***

B‐W Rural ‐ 
B‐W Urban (γ 3)
N

0.104
0.061
[0.061]* [0.019]***
83940
398919

0.047
0.102
0.141
[0.019]** [0.019]*** [0.033]***
250214
170990
222059

0.075
0.075
[0.022]*** [0.019]***
324141
319122

Black, Rur‐Urb
(γ 2 + γ 3)

Sex

Panel B:  World War II Education and AGCT Results Stratified by 1920 Black School Attendance…
Education
AGCT Scores
at/below
above
at/below
above
Median
Median
Median
Median
(1)
(2)
(3)
(4)
Black, Rur‐Urb
1.567
0.767
15.499
‐10.577
(γ 2 + γ 3)
[0.498]*** [0.423]**
[8.598]*
[6.44]*
B‐W Rur.
(γ 1 + γ 3)

1.584
0.953
[0.336]*** [0.314]***

20.889
[6.137]***

‐0.751
[4.163]

B‐W Rural ‐ 
1.997
0.894
B‐W Urban (γ 3) [0.552]*** [0.394]**
N
306693
521279

22.677
[7.341]***
15053

‐3.073
[4.596]
27983

Notes: For Panel A the samples include children between the ages of 7 and 17 in the 1900, 1910, 1920 and 1930 IPUMs.  
The dependent variable is school attendance.  Sample sizes vary across columns due to different rates of non‐missing
values of the variable used for stratification.  The specification corresponds to column (3) in Table 2.  Estimates use Census 
sampling weights.  For Panel B, the samples are drawn from World War II enlistment records and includes men born 
between 1910 and 1928 who enlisted between 1940 and 1946 who lived in either entirely rural or predominantly urban 
counties (see text for details) based on the 1910‐1930 Census.  The specification corresponds to that shown in Table 6 and 
includes inverse probability weights of being in the military by race, county and year of birth. Standard errors clustered on 
county are shown in brackets.  
* significant at 10%; ** significant at 5%; *** significant at 1% 

Figure 1: Black‐White Gap in Education by Birth Cohort, 
vs. Timing of Rosenwald School Construction 
6000

‐0.50

,,~--+-­
,

5000

‐1.00
4000
‐1.50
3000
‐2.00
2000
‐2.50
1000

‐3.00

--~

‐3.50

, ,,i
0

Years of Birth/Years of School Construction
Born in Rosenwald States

Born Outside Rosenwald States

-·-

# of Rosenwald Schools

Numb
ber of Rosenwald Schools

Black‐White Gap in Years of Schooling

0.00

Figure 2a: Number of Rosenwald Schools, as of 1920

# of Schools in County

*

Montgomery, AL
1 (154)
2 (57)

3 (26)
4(15)
5 to 9 (30)

10 to 14(4)
15 to 19 (3)
-

20to29(0)
30to39(0)
40to59(0)

0

60 and above (0)
100
200

Figure 2b: Number of Rosenwald Schools, as of 1925

# of Sch ools in County

*

Montgomery, AL
1 (222)
2 (101)
3 (83)

4 (50)
5 to 9 (133)
10 to 14 (41)
15 to 19 (12)
20 to29(6)
30 to39(2)

-

40to59(1)

-

60 and above (O)
100
200

0

Miles

Figure 2c: Number of Rosenwald Schools, as of 1932

# of Schools in County

*

Montgomery, AL
1 (237)
2 (118)
3 (112)
4(69)
5 to 9 (185)
10 to 14(69)
15 to 19 (36)

-

20 to 29 (30)
30to39(5)
40to59(1)

0

60 to 1000 (1)
100
200

Figwe 2d: Share of Black Rural School Age Children in Rosenwald Schools, as of 1932

Coverage by County
Mo ntgomery, AL
c = 0 (508)
0.0 < c < 0.1 (86)
0.1 <= c < 0.2 (130)
0.2 <= c < 0.3 (112)

*

0.3 <= c < 0.4
0.4 <= c < 0.5
0.5 <= c < 0.6
0.6 <= c < 0.7

(114)
(78)
(72)
(38)

-

0.7 <= c < 0.8 (39)

-

0.0 <- < < 0.9 (30)
c >= 0.9 (150)

-

0

100
Miles

200

 

Figure 3:  Distribution of Rosenwald Share of  Rural Black School Age 
Children Across Counties
0.50
0 50

Mean and Interquartile Range

0.45
0.40
0.35
0.30
0.25
0.20

Mean exposure for 
the 1910‐30 school 
attendance sample

0.15
0 15
0.10
0.05
0.00
1919

1920

1921

1922

1923

1924

Year

1925

1926

1927

1928

1929

1930

Figure 4: Black Rural School Attendance Rates, 1900‐1930
(ages 10‐13)
0.90

0.85

School A
Attendance Rate

0.80

0.75
Rosenwald County by 1919

0.70

Never Rosenwald
0.65
0 65
Rosenwald County between 1920 and 
1931
0.60

0.55

0.50

0.45
1900

1910

1920

1930

Figure 5: Probability of Black Enlistment vs. Rosenwald Exposure
Each Observation is a county‐birth year (N = 20494)
1.2

Probability of Selection into Army

1

0.8

0.6

0.4

0.2
y = 0.0359x + 0.1018
R² = 0.0051

0
0

0.2

0.4

0.6
Rosenwald Exposure

0.8

1

1.2

Appendix A: School Location Selection
This appendix describes a set of county-level regressions that show the association between preRosenwald county characteristics and school location decisions. The regressions take the following form:
(6)

where ROSEc is defined as (a) whether a Rosenwald school was built in county c by 1919, (b) the share of
the school age population covered by Rosenwald schools in c by 1919, and (c) the analogously defined
Rosenwald coverage rate in c by 1931. The pre-Rosenwald county characteristics, countyc, come from
the 1900 and 1910 Census and include race-specific measures of educational and economic status such as
school enrollment, literacy, and occupational status.1 We define these variables to be either as of 1910 or
to be the change between 1900 and 1910.2 The regressions also incorporate state fixed effects, industry
share controls, plantation land as of 1890, and political participation measures as of 1880. The plantation
land and political participation measures were generously supplied by Kenneth Chay and Kaivan Munshi
(2011).3
Table A4 shows the results. In the first three columns, we regress the various Rosenwald
measures on the 1910 level of county characteristics. We find no evidence that 1910 Black socioeconomic characteristics can predict the location of the earliest schools. We also find no statistically
significant, or economically large, effects of these characteristics on the final rates of coverage in 1931.
Moreover, using changes in county characteristics between 1900 and 1910, rather than 1910 levels,
columns (4) to (6) report no evidence that Black socio-economic progress can predict the location of the
Rosenwald schools.4 This provides some comfort that our key outcomes when measured prior to the

1

The occupational status measures are provided by IPUMS and are based on 1950 levels of income and education
by occupation.
2
We have also run specifications that allow us to incorporate both levels and changes. The broad contours of the
results are similar to what we report in Table A4.
3
When we include this data we drop 172 counties from the sample. We find virtually identical results if we include
these counties and drop the plantation land and political participation measures.
4
In columns 4 through 6 we have also included the 1910 level of the rural black population. When we add 1900
data we drop 5 counties due to missing data.

Rosenwald period, do not appear to be significantly correlated with the location of where schools were
built, especially in the 1910s, suggesting limited scope for the notion that our results could be due to
reverse causality.
That said, one intriguing finding is that a 10 percentage point increase in a county’s 1910 White
literacy rate is associated with a 4.5 percentage point increase in the probability that a Rosenwald school
is built in the county by 1919. It is not clear what this relationship reflects. From letters in the
Rosenwald archives, we know that Washington believed that the program had important racial
implications, and in a variety of ways sought to minimize White backlash as much as possible. These
results seem consistent with that strategy. Moreover, the results are not eliminated after controlling for
lynchings and political participation using the Chay and Munshi data. Alternatively, areas with higher
White literacy may have been more prosperous and had a higher demand for more skilled Black labor.
However, neither the level nor the change in industrial composition of the White workforce has a
statistically or economically significant impact on school location and the White literacy results are highly
robust to these controls.
Although we believe that school selection was fairly idiosyncratic prior to 1919, as time passed,
there is perhaps some suggestive evidence that schools were concentrated in areas with better socioeconomic characteristics. This can be seen from a comparison of columns (2) to (3). Column (2) reports
regressions of the Rosenwald coverage rate in 1919 on pre-Rosenwald Census characteristics. Again, we
find little evidence that Black or White observables matter with the exception of White literacy. But by
1931 (column 3), there is some marginally statistically significant evidence that Rosenwald coverage was
higher in counties with higher Black occupational status in 1910. Further, the point estimates on Black
school attendance and literacy in column (3) are positive and slightly higher than they are in column (2),
though they remain statistically and economically insignificant.

The positive effects of Black

socioeconomic characteristics are also not robust to looking at changes in these measures between 1900
and 1910 (columns 5 and 6).

In any case, our econometric strategy (use of county fixed effects and county-by-year fixed
effects) is robust to sources of bias at the county-level. The fact that we often find precisely estimated
“zero” effects for our control groups once we include county fixed effects, is reassuring on this point as
well. Moreover, the results in Table A4 provide further support for using the location of pre-1919
Alabama schools and state and county of birth (see sections VI and VII.D) to identify causal Rosenwald
effects.

Appendix B: Army General Classification Test (AGCT) scores
Although the enlistment records database does not appear to contain test scores, Joe Ferrie
discovered ,through National Archives documentation, that a May 1943 Army training manual (TM-12305, May 1, 1943) instructed punch card operators to input AGCT scores into the weight field (Ferrie,
Rolf, and Troesken 2009). Specifically the instructions read “Weight (AGCT will be punched in this
field) 76-78.” An examination of the data confirms that for a period from March 1943 to May 1943, the
weight field was occupied by test scores. Specifically, we plotted the mean and standard deviation of the
data contained in the “weight” field for a 40 week period in 1943 for all enlistees across all large
enlistment cities. In New York City, for example, it is apparent that the mean value of weight abruptly
changes from around 150 to 100 starting in March 1943. The mean stays at around 100 for the following
10 weeks and thereafter becomes noisy.
Based on an evaluation of the means and standard deviations of the weekly data in the weight
field in the period beginning in March 1943, we were able to classify about 98,000 of the weight
observations for men in the Rosenwald states as actually representing test scores. We also confirmed that
our data replicates the distributions of weight and tests scores from previous historical studies using other
samples of World War II enlistees. Figure A2 plots separate kernel density estimates for weight and test
scores and compares this to data from previous historical studies (Staff, Personnel Research Section, The
Adjutant General’s Office 1947; Karpinos 1958). We find that AGCT scores have a lower fat tail and
peak at around 110 while weight peaks at around 140 pounds, consistent with these other historical
sources.
Finally, we note that prior to March 1943 the correlation between the data in the weight field and
completed schooling was only about 0.06. For the sample in which we are convinced the data contains
test scores, the correlation with schooling is roughly 0.60.

Appendix C: Details of the Internal Rate of Return Calculation
Cohort Size: We want to calculate benefits for cohorts of rural Southern Blacks who survived to age 7
and could have attended a Rosenwald school. We estimate the number of rural Blacks born in Rosenwald
states who were 7 years old in the 1910, 1920, 1930, 1940 and 1950 Census and use this as our estimate
of the size of the 1903, 1913, 1923, 1933 and 1943 cohorts. For example, we estimate the size of the
1913 cohort to be about 161,000. The size of other cohorts is a linear extrapolation of these counts.
Unfortunately, in 1940 and 1950 we do not know rural status. Therefore, to get the count of rural Blacks
in those years, we extrapolate forward the rate of the decline in the rural share of the Southern Black
population from 1910 to 1930.
Earnings Stream: To calculate a base earnings level on which to apply the return to education, we
estimate the lifecycle earnings for a representative cohort born in 1920. In particular, we calculate the
mean earnings of Blacks born in Rosenwald states in 1920 in the 1940, 1950, 1960, 1970 and 1980
Censuses and then interpolate earnings at other ages. It is not possible to estimate the earnings stream for
younger cohorts because earnings was not asked in the Census prior to 1940. Earnings are deflated to
1925 dollars using the CPI-U. We also subtract 10 percent of the value of earnings under the assumption
that rural-born Blacks had lower earnings than urban-born Blacks (“rural penalty”). This suggests that for
rural Blacks born in the South in 1920, total real earnings from age 20 to 60 was about $33,000 in 1925
dollars.5
Costs: We take the total construction costs of primary buildings, replacement buildings, rehabilitated
buildings, teacher homes and industrial shops from the Rosenwald database and apply them to the school
year in which they were built. For teacher salaries, we use data from Margo (1990) on salaries of Black
teachers in 7 southern states in 1910 and 1936. We convert these to 1925 dollars and add a 10 percent
premium on the conservative assumption that Rosenwald teachers were paid more on average. We then
interpolate teacher salaries for the other years based on these values. We use the Rosenwald database to
estimate the number of teachers based on the number of classrooms. We assume that maintenance costs
were about 20 percent of the total of variable costs (teacher salaries being the other variable cost).
Finally, we assume that the foregone contemporaneous earnings of attending an additional 1.2 years of
school (see Table 5, column 2) was $40 in 1925 dollars.6

5

 Ideally we would like our earnings trajectory to only include individuals who did not attend a Rosenwald school 
and who presumably had a lower earnings base.  Unfortunately, there is no way to address this directly.  Among 
our sensitivity checks we have increased the rural penalty to 20 percent.  This can be viewed as an alternative way 
of lowering the baseline level of earnings for our calculation. 
6
 We estimate that the foregone contemporaneous earnings of an additional 1.2 school years was approximately 
$30 to $35 in 1925 dollars.  We choose $40 to be conservative.  We come to this estimate in three independent 
ways.  First, we use the 1940 Census and estimate weekly earnings, by age, of 14 to 17 year old Blacks in 
Rosenwald states. Since the majority of working Black children in the early 20th century rural South were employed 
on the farm, we restrict the sample to those who report their industry as agriculture.  Since earnings was not asked 
of those younger than 14, we interpolate the weekly earnings of 7 to 13 year olds and then calculate an average 
for 7 to 17 year olds.  We then calculate earnings over 26 weeks (assuming a 6 month school term length) for each 
age.  We further inflate this by a factor of 1.2 to account for our effect on “years” of education and finally convert 
to 1925 dollars using the CPI.  This method yields an estimate of foregone earnings of $31.  Our second approach 
also uses the 1940 Census. By age, we calculate the difference in annual earnings between Southern Blacks who 

Baseline Estimates: If we assume that the effect of Rosenwald exposure is to increase schooling by 1.2
years, the low end of our point estimates, and that the return to education is 5 percent, we find that the
internal rate of return (IRR) is 7.2 percent. If, instead, we assume an effect of 1.4 years on completed
schooling, the upper end of our point estimates, and a rate of return of 7 percent, the estimated IRR rises
to 9.3 percent.
Sensitivity Analysis: We experiment with varying the “rural penalty” from 0 to 20 percent. If we increase
the rural penalty to 20 percent, our estimates fall to between 6.7 and 8.8 percent. Reducing the rural
penalty to 0 increases our estimates to between 7.6 and 9.8 percent. We also experiment with increasing
the share of maintenance costs as a fraction of variable costs from 20 to 33 percent. This modification
lowers the range of estimates to between 6.7 and 8.7 percent. Increasing the amount of the construction
costs has little effect. For example, raising such costs by 20 percent to better account for the donation of
land, labor, and materials and the possibility that some capital costs are not well recorded by the Fund’s
index cards, lowers the IRR by only two tenths of a percentage point. If we assume that there were no
foregone contemporaneous earnings among Rosenwald students, the range of our IRR estimates increases
to 7.9 to 10.0 percent. On the other hand, if the foregone contemporaneous earnings was 50 percent
higher than the baseline assumption ($60), our range of estimates falls to 6.9 to 9.0 percent. The IRR
estimate appears to be most sensitive to the assumed rate of return of a year of schooling. For example,
raising the rate of return from 5 to 9 percent raises the IRR by about 2.5 percentage points. If we use our
most conservative set of assumptions in combination, the IRR is estimated to be 5.8 percent.

work in agriculture but do not attend school and Southern Blacks who work in agriculture while simultaneously 
attending school.  That difference could be attributed to the lost earnings implied by school attendance. Again, we 
multiply this amount by 1.2 years of schooling and deflate to 1925 dollars using the CPI.  This method results in an 
estimate of foregone earnings of $34.  Finally, we convert the estimates of the contribution to farm income by 
children from Craig (1993) to 1925 dollars. His estimates, which are based on Northern farmers from the 1860 
Census, suggest that forgone earnings was about $31.   

Table A1:  Descriptive Statistics about Rosenwald School Projects by School

Mean
School Building Details (n=4,968)
Rosenwald school
Rosenwald teacher homes 
Rosenwald shops
Country Training Schools
Fraction of schools with additions
Fraction of schools rebuilt
Fraction of schools burned

5,343
882
221
3,425
815

Min

Max

0.993
0.044
0.035
0.049
0.062
0.007
0.016

Cost of Rosenwald Schools  (n=4,932)
Real Cost of original schools
  borne by local Blacks
  borne by local Whites
  borne by local government
  borne by Rosenwald Fund

Std dev

7,854
1,021
1,226
7,193
569

583
0
0
0
58

169,761
16,528
39,375
163,473
7,859

8,176

583

169,761

Fraction of schools with positive contributions from:
   local Blacks
0.92
   local Whites
0.29
   local government
0.96
   Rosenwald Fund
1.00
Real cost, including changes
   to schools (additions/rebuilds)

5,604

Notes:
Excludes 4 Missouri projects.  Counts Rosenwald schools with known number of teachers and date of construction.  
Costs deflated to 1925 dollars.

Table A2:  Summary statistics of 1900‐1930 IPUMS samples
School Enrollment Sample of 7 to 17 year olds (N= 650167)
1900
1910
Blacks
Whites
Blacks
Whites
School Enrollment
All Ages
0.38
0.59
0.60
0.80
Age 7 to 10
0.36
0.55
0.61
0.82
Age 11 to 14
0.49
0.72
0.69
0.87
Age 15 to 17
0.28
0.46
0.45
0.66
Male
0.37
0.58
0.57
0.80
Female
0.40
0.60
0.64
0.81
Rural
0.37
0.58
0.59
0.80
Urban
0.48
0.61
0.68
0.81
Family Characteristics
Father literate
0.39
0.85
0.53
0.87
Mother literate
0.30
0.81
0.49
0.87
Father Occ. Score
15.37
18.60
15.01
19.81
Father Owned home
0.26
0.58
0.27
0.57
Rosenwald Measures
Presence in County
0.00
0.00
0.00
0.00
Exposure (ages 7 to 13)
0.00
0.00
0.00
0.00
Geography
Rural
0.87
0.85
0.86
0.82
City Population
9101
19922
9963
20733
Alabama
0.11
0.07
0.10
0.06
Arkansas
0.05
0.07
0.05
0.06
Florida
0.03
0.02
0.03
0.02
Georgia
0.13
0.08
0.15
0.08
Kentucky
0.03
0.12
0.02
0.11
Louisiana
0.08
0.05
0.08
0.05
Maryland
0.02
0.06
0.03
0.05
Mississippi
0.12
0.04
0.12
0.04
North Carolina
0.08
0.08
0.09
0.08
Oklahoma
0.01
0.04
0.01
0.08
South Carolina
0.11
0.04
0.12
0.04
Tennessee
0.06
0.10
0.05
0.09
Texas
0.09
0.17
0.09
0.17
Virginia
0.08
0.07
0.07
0.07
Number of observations
21082
39887
28390
71358
Literacy Sample of 15 to 22 year olds (N = 431976)
Literacy
All ages
0.61
0.90
0.71
0.94
Age 15 to 17
0.60
0.89
0.72
0.94
Age 18 to 22
0.61
0.91
0.71
0.93
Rural
0.57
0.89
0.68
0.92
Urban
0.79
0.97
0.85
0.98
Rosenwald Measures
Presence in County
0.00
0.00
0.00
0.00
Exposure (ages7 to 13)
0.00
0.00
0.00
0.00
Number of observations
13056
24620
19146
51027

Blacks

1920
Whites

Blacks

1930
Whites

0.74
0.78
0.81
0.55
0.72
0.76
0.72
0.82

0.85
0.91
0.92
0.65
0.85
0.86
0.85
0.87

0.75
0.81
0.85
0.52
0.73
0.77
0.74
0.79

0.84
0.90
0.93
0.63
0.83
0.84
0.83
0.86

0.63
0.66
15.47
0.27

0.90
0.91
20.34
0.53

0.73
0.80
15.70
0.27

0.93
0.95
20.63
0.47

0.49
0.02

0.29
0.02

0.91
0.28

0.73
0.30

0.82
17021
0.11
0.06
0.03
0.14
0.02
0.08
0.02
0.11
0.10
0.01
0.12
0.04
0.08
0.07
17680

0.77
30530
0.07
0.06
0.03
0.08
0.10
0.05
0.04
0.04
0.08
0.09
0.04
0.09
0.17
0.07
52188

0.78
26542
0.10
0.05
0.04
0.13
0.02
0.08
0.02
0.11
0.11
0.02
0.10
0.05
0.09
0.08
115146

0.74
35335
0.07
0.06
0.04
0.07
0.09
0.05
0.05
0.04
0.10
0.08
0.04
0.08
0.16
0.07
304436

0.80
0.82
0.79
0.77
0.90

0.96
0.96
0.96
0.95
0.98

0.88
0.89
0.86
0.85
0.94

0.98
0.98
0.98
0.97
0.99

0.46
0.00
11998

0.29
0.00
33623

0.91
0.13
78477

0.73
0.13
200029

Table A3:  Summary statistics of WWII enlisted men sample
All Counties
Mean

Blacks
s.d.

N

Outcomes
Education
Some H.S.
Completed H.S.
AGCT Score
Height

7.4
0.33
0.12
65.1
68.1

2.7
0.47
0.32
16.4
3.2

400168
400168
400168
11816
176352

Demographics
Age
Drafted
Enlisted, 1940
Enlisted, 1941
Enlisted, 1942
Enlisted, 1943
Enlisted, 1944
Enlisted, 1945
Enlisted, 1946
Enlist Prob.

23.2
0.86
0.01
0.09
0.35
0.26
0.10
0.13
0.06
0.20

4.0
0.35
0.08
0.28
0.48
0.44
0.30
0.33
0.24
0.20

Rosenwald Exposure
All cohorts
0.33
born <1915
0.18
born 1916‐22
0.35
born 1923‐28
0.39
Geography
% Rural
Alabama
Arkansas
Florida
Georgia
Kentucky
Louisiana
Maryland
Mississippi
No. Carolina
Oklahoma
So. Carolina
Tennessee
Texas
Virginia

0.71
0.10
0.03
0.07
0.11
0.02
0.06
0.03
0.14
0.11
0.02
0.08
0.07
0.10
0.06

Mean
9.7
0.62
0.36
90.8
68.7

Rural Counties
Whites
s.d.

N

Mean

Blacks
s.d.

N

Mean

Whites
s.d.

N

2.9 1440294
0.49 1440294
0.48 1440294
23.0
75545
4.4 688865

7.1
0.28
0.09
65.5
68.0

2.6
0.45
0.29
16.8
3.2

103906
103906
103906
3476
48387

9.2
0.546
0.301
87.24
68.53

2.852
0.498
0.459
23.07
4.037

397726
397726
397726
21050
193329

400168 22.97
400168 0.64
400168 0.06
400168 0.12
400168 0.36
400168 0.19
400168 0.11
400168 0.12
400168 0.05
400168 0.22

4.07
0.48
0.24
0.32
0.48
0.39
0.31
0.32
0.21
0.18

1440294
1440294
1440294
1440294
1440294
1440294
1440294
1440294
1440294
1440294

23.0
0.88
0.01
0.11
0.36
0.24
0.11
0.13
0.06
0.29

3.9
0.33
0.07
0.31
0.48
0.43
0.31
0.33
0.23
0.34

103906
103906
103906
103906
103906
103906
103906
103906
103906
103906

22.86
0.699
0.048
0.137
0.349
0.185
0.112
0.12
0.049
0.276

3.944
0.459
0.214
0.344
0.477
0.388
0.316
0.325
0.216
0.294

397726
397726
397726
397726
397726
397726
397726
397726
397726
397726

0.30
0.23
0.30
0.31

400168
75810
205351
119007

0.33
0.17
0.35
0.40

0.34 1440294
0.23 274574
0.35 748228
0.37 417492

0.24
0.14
0.25
0.27

0.22
0.15
0.23
0.24

103906
17546
55133
31227

0.251
0.145
0.264
0.289

0.292
0.216
0.293
0.313

397726
70703
209531
117492

0.28
0.30
0.16
0.25
0.31
0.15
0.24
0.16
0.34
0.32
0.14
0.28
0.25
0.31
0.24

400168
400168
400168
400168
400168
400168
400168
400168
400168
400168
400168
400168
400168
400168
400168

0.72
0.08
0.03
0.04
0.09
0.08
0.03
0.04
0.05
0.11
0.07
0.05
0.11
0.17
0.05

0.28
0.27
0.18
0.20
0.29
0.28
0.16
0.19
0.22
0.31
0.25
0.21
0.31
0.38
0.23

1.00
0.09
0.01
0.03
0.10
0.01
0.05
0.07
0.18
0.09
0.00
0.06
0.03
0.06
0.21

0.00
0.29
0.12
0.16
0.30
0.12
0.22
0.26
0.38
0.28
0.07
0.23
0.18
0.23
0.41

103906
103906
103906
103906
103906
103906
103906
103906
103906
103906
103906
103906
103906
103906
103906

1.00
0.074
0.03
0.022
0.089
0.113
0.024
0.082
0.067
0.077
0.031
0.019
0.108
0.109
0.156

0.00
0.261
0.172
0.146
0.285
0.316
0.154
0.274
0.251
0.266
0.173
0.138
0.31
0.311
0.363

397726
397726
397726
397726
397726
397726
397726
397726
397726
397726
397726
397726
397726
397726
397726

1440294
1440294
1440294
1440294
1440294
1440294
1440294
1440294
1440294
1440294
1440294
1440294
1440294
1440294
1440294

Table A3:  Summary statistics of WWII enlisted men sample (continued)
Urban Counties
Mean

Blacks
s.d.

N

Mean

Whites
s.d.

N

Mean

Non‐Rural, Non‐Urban counties
Blacks
Whites
s.d.
N
Mean s.d.

N

Outcomes
Education
Some H.S.
Completed H.S.
AGCT Score
Height

8.1
0.41
0.16
66.6
68.1

2.8
0.49
0.37
16.3
3.3

112957
112957
112957
4725
45678

10.5
0.74
0.46
97.3
68.6

2.7
0.44
0.50
21.9
4.9

372180
372180
372180
21109
181464

7.2
0.30
0.11
63.0
68.2

2.7
0.46
0.31
15.8
3.1

183305
9.5 2.894
183305 0.60 0.49
183305 0.347 0.476
3615 88.86 22.75
82287 68.76 4.25

670388
670388
670388
33386
314072

Demographics
Age
Drafted
Enlisted, 1940
Enlisted, 1941
Enlisted, 1942
Enlisted, 1943
Enlisted, 1944
Enlisted, 1945
Enlisted, 1946
Enlist Prob.

23.7
0.84
0.01
0.08
0.32
0.30
0.10
0.12
0.06
0.22

4.3
0.36
0.09
0.27
0.47
0.46
0.30
0.33
0.23
0.10

112957 23.32
112957 0.59
112957 0.06
112957 0.12
112957 0.37
112957 0.21
112957 0.10
112957 0.10
112957 0.04
112957 0.24

4.26
0.49
0.24
0.32
0.48
0.40
0.30
0.31
0.20
0.10

372180
372180
372180
372180
372180
372180
372180
372180
372180
372180

23.0
0.86
0.01
0.08
0.37
0.25
0.10
0.13
0.06
0.14

3.9
0.35
0.08
0.27
0.48
0.43
0.30
0.34
0.24
0.11

183305
183305
183305
183305
183305
183305
183305
183305
183305
183305

22.83
0.637
0.068
0.108
0.361
0.183
0.11
0.12
0.049
0.174

4.023
0.481
0.251
0.31
0.48
0.387
0.313
0.326
0.217
0.092

670388
670388
670388
670388
670388
670388
670388
670388
670388
670388

Rosenwald Exposure
All cohorts
0.38
born <1915
0.22
born 1916‐22
0.41
born 1923‐28
0.46

0.37
0.29
0.38
0.39

112957
27214
54358
31385

0.36
0.17
0.39
0.46

0.39
0.24
0.39
0.41

372180
82696
190232
99252

0.35
0.18
0.37
0.41

0.27
0.19
0.27
0.28

183305
31050
95860
56395

0.362
0.174
0.383
0.439

0.339
0.227
0.337
0.356

670388
121175
348465
200748

Geography
% Rural
Alabama
Arkansas
Florida
Georgia
Kentucky
Louisiana
Maryland
Mississippi
No. Carolina
Oklahoma
So. Carolina
Tennessee
Texas
Virginia

0.10
0.36
0.11
0.33
0.37
0.18
0.19
0.07
0.22
0.27
0.13
0.21
0.34
0.35
0.08

112957
112957
112957
112957
112957
112957
112957
112957
112957
112957
112957
112957
112957
112957
112957

0.31
0.09
0.02
0.09
0.10
0.09
0.01
0.02
0.02
0.07
0.05
0.02
0.13
0.25
0.01

0.10
0.29
0.14
0.29
0.30
0.29
0.12
0.15
0.15
0.26
0.22
0.14
0.34
0.43
0.11

372180
372180
372180
372180
372180
372180
372180
372180
372180
372180
372180
372180
372180
372180
372180

0.80
0.08
0.04
0.05
0.08
0.02
0.08
0.01
0.17
0.15
0.03
0.12
0.04
0.11
0.02

0.12
0.27
0.20
0.22
0.27
0.14
0.27
0.12
0.38
0.36
0.16
0.33
0.21
0.31
0.13

183305
183305
183305
183305
183305
183305
183305
183305
183305
183305
183305
183305
183305
183305
183305

0.79
0.081
0.043
0.029
0.083
0.062
0.036
0.02
0.054
0.144
0.098
0.077
0.091
0.165
0.018

0.12
0.273
0.203
0.167
0.275
0.241
0.187
0.141
0.226
0.351
0.297
0.267
0.287
0.371
0.132

670388
670388
670388
670388
670388
670388
670388
670388
670388
670388
670388
670388
670388
670388
670388

0.32
0.15
0.01
0.12
0.16
0.03
0.04
0.01
0.05
0.08
0.02
0.05
0.13
0.14
0.01

Table A4:  Determinants of Location of Rosenwald Schools Using Pre‐Rosenwald County Characteristics

Rural Black Population
Black School Enrollment
Black Literacy
Black Occ. Status
Black Occ. Ed. Score
White School Enrollment
White Literacy
White Occ. Status
White Occ. Ed. Score 
% Teachers
Fraction Repub. Voters
(1880)
Ever elect, Black St. Rep.
(before 1880)
Ever elect, Black St. Sen.
(before 1880)
Plantation Share of Land 
(1880)
N

Indep. Vars use 1910 levels
(1)
(2)
(3)
School
Coverage
Coverage
by 1919
1919
1931
0.017
0.000
‐0.004
[0.004]***
[0.001]
[0.003]
‐0.003
‐0.003
0.052
[0.064]
[0.021]
[0.048]
0.031
0.022
0.060
[0.084]
[0.027]
[0.063]
0.006
0.002
0.007
[0.005]
[0.002]
[0.004]*
0.004
0.001
‐0.007
[0.006]
[0.002]
[0.004]*
0.064
0.047
0.099
[0.122]
[0.039]
[0.091]
0.447
0.101
0.443
[0.213]**
[0.069]
[0.159]***
‐0.014
‐0.004
0.004
[0.008]*
[0.003]*
[0.006]
0.012
0.004
0.003
[0.007]*
[0.002]
[0.005]
1.881
‐0.006
0.561
[1.481]
[0.478]
[1.109]
‐0.054
‐0.018
‐0.068
[0.091]
[0.029]
[0.068]
‐0.017
‐0.006
‐0.051
[0.045]
[0.014]
[0.033]
‐0.067
0.005
‐0.015
[0.060]
[0.019]
[0.045]
‐0.073
‐0.041
0.149
[0.314]
[0.101]
[0.234]
873
868
868

Indep. Vars use Change 1900‐10
(4)
(5)
(6)
School
Coverage
Coverage
by 1919
1919
1931
1.414
4.444
8.744
[5.698]
[1.820]**
[4.208]**
0.019
0.015
0.045
[0.043]
[0.014]
[0.032]
0.006
0.021
‐0.065
[0.068]
[0.022]
[0.050]
‐0.001
0.001
0.004
[0.004]
[0.001]
[0.003]
0.000
‐0.001
‐0.005
[0.003]
[0.001]
[0.002]**
0.119
0.036
0.178
[0.067]*
[0.022]
[0.050]***
‐0.031
0.010
0.086
[0.163]
[0.052]
[0.120]
‐0.005
‐0.001
0.005
[0.006]
[0.002]
[0.004]
0.002
0.000
‐0.004
[0.004]
[0.001]
[0.003]
1.147
0.393
2.127
[0.923]
[0.294]
[0.681]***
‐0.072
‐0.018
‐0.056
[0.091]
[0.029]
[0.067]
‐0.018
‐0.003
‐0.039
[0.046]
[0.015]
[0.034]
‐0.080
0.008
0.002
[0.060]
[0.019]
[0.045]
0.152
‐0.015
0.107
[0.290]
[0.093]
[0.214]
873
868
868

Notes:  Sample includes counties with rural blacks.  All regressions also include indicators for missing any of the following black school 
white literacy.  Regressions also include indicators for state and control for the fraction of workers in agriculture, nondurable manufact
Columns 4 through 6 also control for the 1910 level of the rural Black population. Standard errors are in brackets.  * significant at 10%;

Table A5:  Coefficients from Regressions Based on Stratified Samples
Panel A:  Census School Attendance Results Stratified by…
1910 Black
1910 White
1910 Black
School Attendance
School Attendance
Population Share
at/below
above
at/below
above
at/below
above
Median
Median
Median
Median
Median
Median
(1)
(2)
(3)
(4)
(5)
(6)
‐0.015
0.025
‐0.023
0.026
0.028
‐0.028
γ0
[0.021]
[0.013]*
[0.025] [0.012]** [0.011]** [0.026]
γ1

0.060
0.044
0.048
0.062
0.040
0.057
0.044
0.046
[0.023]*** [0.015]*** [0.019]** [0.017]*** [0.017]** [0.016]*** [0.022]** [0.017]***

γ2

‐0.024
[0.016]

γ3

0.146
[0.028]***
255887

N

γ0
γ1

‐0.001
[0.010]

0.020
[0.018]

‐0.028
[0.010]***

‐0.013
[0.009]

0.008
[0.017]

‐0.022
[0.012]*

0.003
[0.019]

1860
Slave Share (if >0)
at/below
above
Median
Median
(9)
(10)
0.025
‐0.005
[0.017]
[0.018]
0.019
[0.018]

0.069
[0.019]***

‐0.025
[0.013]*

0.006
[0.014]

0.015
0.072
0.087
0.027
0.081
0.063
0.084
0.044
0.078
[0.020] [0.025]*** [0.021]*** [0.028] [0.020]*** [0.029]** [0.021]*** [0.028] [0.023]***
296191
337151
296440
281651
352418
203511
333198
226559
288338

Top quart.
Slave Share excluding
AND
those in 
Plantation
(11)
(11)
(12)
0.012
0.010
[0.043]
[0.014]

Age
7 to 10
11 to 13 14 to 17
(13)
(14)
(15)
‐0.032
‐0.043
0.049
[0.011]*** [0.011]*** [0.023]**

0.121
0.034
[0.046]*** [0.016]**
[0 046]*** [0 016]**

0.050
0.028
[0.015]*** [0.014]**
[0 015]*** [0 014]**

γ2

‐0.022
[0.045]

γ3

0.104
0.061
[0.061]* [0.019]***
83940
398919

N

1890 Plantation
Share of Land
at/below
above
Median
Median
(7)
(8)
0.025
‐0.018
[0.014]*
[0.027]

‐0.013
[0.010]

0.035
[0.028]
[0 028]

0.026
0.022
‐0.096
[0.009]*** [0.008]** [0.019]***
0.047
0.102
0.141
[0.019]** [0.019]*** [0.033]***
250214
170990
222059

Sex
Male
(16)
0.015
[0.016]

0.052
0.054
[0.019]*** [0.013]***
[0 019]*** [0 013]***
‐0.015
[0.012]

‐0.413
[0.437]

0.059
[0.233]

‐1.788
[3.815]

2.322
[2.065]

γ2

‐0.43
[0.380]

‐0.127
[0.218]

‐7.177
[6.367]

‐7.504
[5.557]

22.677
[7.341]***
15053

‐3.073
[4.596]
27983

γ3
N

1.997
0.894
[0.552]*** [0.394]**
306693
521279

0.002
[0.012]

0.075
0.075
[0.022]*** [0.019]***
324141
319122

Panel B:  World War II Education and AGCT Results Stratified by 1920 Black School Attendance…
Education
AGCT Scores
at/below
above
at/below
above
Median
Median
Median
Median
(1)
(2)
(3)
(4)
γ0
0.155
‐0.046
5.52
‐3.566
[0.302]
[0.119]
[3.114]* [1.390]**
γ1

Female
(17)
‐0.01
[0.015]

Notes: See notes accompanying Table 8 for details. Standard errors, clustered on county, are shown in brackets.  
* significant at 10%; ** significant at 5%; *** significant at 1% 

Figure A1
Alabama Border Counties Used for Table 4A

Legend
Included county

0

50
Miles

100

 

Figure A2:  Comparison of  AGCT  and  Weight Data to  Historical Sources
Panel A:  Kerenel Density of AGCT scores and Weight 

Panel B:  AGCT scores from Staff, Personnel Research Section, The Adjutant General’s Office (1947)

Panel C:  Descriptive statistics for weight for WWI and WWII enlistees from Karpinos (1958)

Working Paper Series
A series of research studies on regional economic issues relating to the Seventh Federal
Reserve District, and on financial and economic topics.
U.S. Corporate and Bank Insolvency Regimes: An Economic Comparison and Evaluation
Robert R. Bliss and George G. Kaufman

WP-06-01

Redistribution, Taxes, and the Median Voter
Marco Bassetto and Jess Benhabib

WP-06-02

Identification of Search Models with Initial Condition Problems
Gadi Barlevy and H. N. Nagaraja

WP-06-03

Tax Riots
Marco Bassetto and Christopher Phelan

WP-06-04

The Tradeoff between Mortgage Prepayments and Tax-Deferred Retirement Savings
Gene Amromin, Jennifer Huang,and Clemens Sialm

WP-06-05

Why are safeguards needed in a trade agreement?
Meredith A. Crowley

WP-06-06

Taxation, Entrepreneurship, and Wealth
Marco Cagetti and Mariacristina De Nardi

WP-06-07

A New Social Compact: How University Engagement Can Fuel Innovation
Laura Melle, Larry Isaak, and Richard Mattoon

WP-06-08

Mergers and Risk
Craig H. Furfine and Richard J. Rosen

WP-06-09

Two Flaws in Business Cycle Accounting
Lawrence J. Christiano and Joshua M. Davis

WP-06-10

Do Consumers Choose the Right Credit Contracts?
Sumit Agarwal, Souphala Chomsisengphet, Chunlin Liu, and Nicholas S. Souleles

WP-06-11

Chronicles of a Deflation Unforetold
François R. Velde

WP-06-12

Female Offenders Use of Social Welfare Programs Before and After Jail and Prison:
Does Prison Cause Welfare Dependency?
Kristin F. Butcher and Robert J. LaLonde
Eat or Be Eaten: A Theory of Mergers and Firm Size
Gary Gorton, Matthias Kahl, and Richard Rosen

WP-06-13

WP-06-14

1

Working Paper Series (continued)
Do Bonds Span Volatility Risk in the U.S. Treasury Market?
A Specification Test for Affine Term Structure Models
Torben G. Andersen and Luca Benzoni

WP-06-15

Transforming Payment Choices by Doubling Fees on the Illinois Tollway
Gene Amromin, Carrie Jankowski, and Richard D. Porter

WP-06-16

How Did the 2003 Dividend Tax Cut Affect Stock Prices?
Gene Amromin, Paul Harrison, and Steven Sharpe

WP-06-17

Will Writing and Bequest Motives: Early 20th Century Irish Evidence
Leslie McGranahan

WP-06-18

How Professional Forecasters View Shocks to GDP
Spencer D. Krane

WP-06-19

Evolving Agglomeration in the U.S. auto supplier industry
Thomas Klier and Daniel P. McMillen

WP-06-20

Mortality, Mass-Layoffs, and Career Outcomes: An Analysis using Administrative Data
Daniel Sullivan and Till von Wachter

WP-06-21

The Agreement on Subsidies and Countervailing Measures:
Tying One’s Hand through the WTO.
Meredith A. Crowley

WP-06-22

How Did Schooling Laws Improve Long-Term Health and Lower Mortality?
Bhashkar Mazumder

WP-06-23

Manufacturing Plants’ Use of Temporary Workers: An Analysis Using Census Micro Data
Yukako Ono and Daniel Sullivan

WP-06-24

What Can We Learn about Financial Access from U.S. Immigrants?
Una Okonkwo Osili and Anna Paulson

WP-06-25

Bank Imputed Interest Rates: Unbiased Estimates of Offered Rates?
Evren Ors and Tara Rice

WP-06-26

Welfare Implications of the Transition to High Household Debt
Jeffrey R. Campbell and Zvi Hercowitz

WP-06-27

Last-In First-Out Oligopoly Dynamics
Jaap H. Abbring and Jeffrey R. Campbell

WP-06-28

Oligopoly Dynamics with Barriers to Entry
Jaap H. Abbring and Jeffrey R. Campbell

WP-06-29

Risk Taking and the Quality of Informal Insurance: Gambling and Remittances in Thailand
Douglas L. Miller and Anna L. Paulson

WP-07-01

2

Working Paper Series (continued)
Fast Micro and Slow Macro: Can Aggregation Explain the Persistence of Inflation?
Filippo Altissimo, Benoît Mojon, and Paolo Zaffaroni

WP-07-02

Assessing a Decade of Interstate Bank Branching
Christian Johnson and Tara Rice

WP-07-03

Debit Card and Cash Usage: A Cross-Country Analysis
Gene Amromin and Sujit Chakravorti

WP-07-04

The Age of Reason: Financial Decisions Over the Lifecycle
Sumit Agarwal, John C. Driscoll, Xavier Gabaix, and David Laibson

WP-07-05

Information Acquisition in Financial Markets: a Correction
Gadi Barlevy and Pietro Veronesi

WP-07-06

Monetary Policy, Output Composition and the Great Moderation
Benoît Mojon

WP-07-07

Estate Taxation, Entrepreneurship, and Wealth
Marco Cagetti and Mariacristina De Nardi

WP-07-08

Conflict of Interest and Certification in the U.S. IPO Market
Luca Benzoni and Carola Schenone

WP-07-09

The Reaction of Consumer Spending and Debt to Tax Rebates –
Evidence from Consumer Credit Data
Sumit Agarwal, Chunlin Liu, and Nicholas S. Souleles

WP-07-10

Portfolio Choice over the Life-Cycle when the Stock and Labor Markets are Cointegrated
Luca Benzoni, Pierre Collin-Dufresne, and Robert S. Goldstein

WP-07-11

Nonparametric Analysis of Intergenerational Income Mobility
with Application to the United States
Debopam Bhattacharya and Bhashkar Mazumder

WP-07-12

How the Credit Channel Works: Differentiating the Bank Lending Channel
and the Balance Sheet Channel
Lamont K. Black and Richard J. Rosen

WP-07-13

Labor Market Transitions and Self-Employment
Ellen R. Rissman

WP-07-14

First-Time Home Buyers and Residential Investment Volatility
Jonas D.M. Fisher and Martin Gervais

WP-07-15

Establishments Dynamics and Matching Frictions in Classical Competitive Equilibrium
Marcelo Veracierto

WP-07-16

Technology’s Edge: The Educational Benefits of Computer-Aided Instruction
Lisa Barrow, Lisa Markman, and Cecilia Elena Rouse

WP-07-17

3

Working Paper Series (continued)
The Widow’s Offering: Inheritance, Family Structure, and the Charitable Gifts of Women
Leslie McGranahan
Demand Volatility and the Lag between the Growth of Temporary
and Permanent Employment
Sainan Jin, Yukako Ono, and Qinghua Zhang

WP-07-18

WP-07-19

A Conversation with 590 Nascent Entrepreneurs
Jeffrey R. Campbell and Mariacristina De Nardi

WP-07-20

Cyclical Dumping and US Antidumping Protection: 1980-2001
Meredith A. Crowley

WP-07-21

Health Capital and the Prenatal Environment:
The Effect of Maternal Fasting During Pregnancy
Douglas Almond and Bhashkar Mazumder

WP-07-22

The Spending and Debt Response to Minimum Wage Hikes
Daniel Aaronson, Sumit Agarwal, and Eric French

WP-07-23

The Impact of Mexican Immigrants on U.S. Wage Structure
Maude Toussaint-Comeau

WP-07-24

A Leverage-based Model of Speculative Bubbles
Gadi Barlevy

WP-08-01

Displacement, Asymmetric Information and Heterogeneous Human Capital
Luojia Hu and Christopher Taber

WP-08-02

BankCaR (Bank Capital-at-Risk): A credit risk model for US commercial bank charge-offs
Jon Frye and Eduard Pelz

WP-08-03

Bank Lending, Financing Constraints and SME Investment
Santiago Carbó-Valverde, Francisco Rodríguez-Fernández, and Gregory F. Udell

WP-08-04

Global Inflation
Matteo Ciccarelli and Benoît Mojon

WP-08-05

Scale and the Origins of Structural Change
Francisco J. Buera and Joseph P. Kaboski

WP-08-06

Inventories, Lumpy Trade, and Large Devaluations
George Alessandria, Joseph P. Kaboski, and Virgiliu Midrigan

WP-08-07

School Vouchers and Student Achievement: Recent Evidence, Remaining Questions
Cecilia Elena Rouse and Lisa Barrow

WP-08-08

4

Working Paper Series (continued)
Does It Pay to Read Your Junk Mail? Evidence of the Effect of Advertising on
Home Equity Credit Choices
Sumit Agarwal and Brent W. Ambrose

WP-08-09

The Choice between Arm’s-Length and Relationship Debt: Evidence from eLoans
Sumit Agarwal and Robert Hauswald

WP-08-10

Consumer Choice and Merchant Acceptance of Payment Media
Wilko Bolt and Sujit Chakravorti

WP-08-11

Investment Shocks and Business Cycles
Alejandro Justiniano, Giorgio E. Primiceri, and Andrea Tambalotti

WP-08-12

New Vehicle Characteristics and the Cost of the
Corporate Average Fuel Economy Standard
Thomas Klier and Joshua Linn

WP-08-13

Realized Volatility
Torben G. Andersen and Luca Benzoni

WP-08-14

Revenue Bubbles and Structural Deficits: What’s a state to do?
Richard Mattoon and Leslie McGranahan

WP-08-15

The role of lenders in the home price boom
Richard J. Rosen

WP-08-16

Bank Crises and Investor Confidence
Una Okonkwo Osili and Anna Paulson

WP-08-17

Life Expectancy and Old Age Savings
Mariacristina De Nardi, Eric French, and John Bailey Jones

WP-08-18

Remittance Behavior among New U.S. Immigrants
Katherine Meckel

WP-08-19

Birth Cohort and the Black-White Achievement Gap:
The Roles of Access and Health Soon After Birth
Kenneth Y. Chay, Jonathan Guryan, and Bhashkar Mazumder

WP-08-20

Public Investment and Budget Rules for State vs. Local Governments
Marco Bassetto

WP-08-21

Why Has Home Ownership Fallen Among the Young?
Jonas D.M. Fisher and Martin Gervais

WP-09-01

Why do the Elderly Save? The Role of Medical Expenses
Mariacristina De Nardi, Eric French, and John Bailey Jones

WP-09-02

Using Stock Returns to Identify Government Spending Shocks
Jonas D.M. Fisher and Ryan Peters

WP-09-03

5

Working Paper Series (continued)
Stochastic Volatility
Torben G. Andersen and Luca Benzoni

WP-09-04

The Effect of Disability Insurance Receipt on Labor Supply
Eric French and Jae Song

WP-09-05

CEO Overconfidence and Dividend Policy
Sanjay Deshmukh, Anand M. Goel, and Keith M. Howe

WP-09-06

Do Financial Counseling Mandates Improve Mortgage Choice and Performance?
Evidence from a Legislative Experiment
Sumit Agarwal,Gene Amromin, Itzhak Ben-David, Souphala Chomsisengphet,
and Douglas D. Evanoff

WP-09-07

Perverse Incentives at the Banks? Evidence from a Natural Experiment
Sumit Agarwal and Faye H. Wang

WP-09-08

Pay for Percentile
Gadi Barlevy and Derek Neal

WP-09-09

The Life and Times of Nicolas Dutot
François R. Velde

WP-09-10

Regulating Two-Sided Markets: An Empirical Investigation
Santiago Carbó Valverde, Sujit Chakravorti, and Francisco Rodriguez Fernandez

WP-09-11

The Case of the Undying Debt
François R. Velde

WP-09-12

Paying for Performance: The Education Impacts of a Community College Scholarship
Program for Low-income Adults
Lisa Barrow, Lashawn Richburg-Hayes, Cecilia Elena Rouse, and Thomas Brock
Establishments Dynamics, Vacancies and Unemployment: A Neoclassical Synthesis
Marcelo Veracierto

WP-09-13

WP-09-14

The Price of Gasoline and the Demand for Fuel Economy:
Evidence from Monthly New Vehicles Sales Data
Thomas Klier and Joshua Linn

WP-09-15

Estimation of a Transformation Model with Truncation,
Interval Observation and Time-Varying Covariates
Bo E. Honoré and Luojia Hu

WP-09-16

Self-Enforcing Trade Agreements: Evidence from Antidumping Policy
Chad P. Bown and Meredith A. Crowley

WP-09-17

Too much right can make a wrong: Setting the stage for the financial crisis
Richard J. Rosen

WP-09-18

Can Structural Small Open Economy Models Account
for the Influence of Foreign Disturbances?
Alejandro Justiniano and Bruce Preston

WP-09-19

6

Working Paper Series (continued)
Liquidity Constraints of the Middle Class
Jeffrey R. Campbell and Zvi Hercowitz

WP-09-20

Monetary Policy and Uncertainty in an Empirical Small Open Economy Model
Alejandro Justiniano and Bruce Preston

WP-09-21

Firm boundaries and buyer-supplier match in market transaction:
IT system procurement of U.S. credit unions
Yukako Ono and Junichi Suzuki
Health and the Savings of Insured Versus Uninsured, Working-Age Households in the U.S.
Maude Toussaint-Comeau and Jonathan Hartley

WP-09-22

WP-09-23

The Economics of “Radiator Springs:” Industry Dynamics, Sunk Costs, and
Spatial Demand Shifts
Jeffrey R. Campbell and Thomas N. Hubbard

WP-09-24

On the Relationship between Mobility, Population Growth, and
Capital Spending in the United States
Marco Bassetto and Leslie McGranahan

WP-09-25

The Impact of Rosenwald Schools on Black Achievement
Daniel Aaronson and Bhashkar Mazumder

WP-09-26

7