The Effects of the Great Depression on Educational Attainment
Department of Economics
3203 SE Woodstock Blvd.
Portland, OR 97202-8199
This paper examines the relationship between the Great Depression and the educational
attainment of young adults in the 1930s, taking advantage of the state-level variation in
employment as individuals were turning a critical age. In general, there were negligible effects
of the Great Depression’s severity on average years of schooling beyond the cohort and state-
specific effects. Regional differences in availability of appropriate schools seem to matter for
the substitution effect to operate to increase the years of schooling during the recession.
Furthermore, at the top end of educational attainment, the income effect seems to outweigh the
substitution effect as the severity of the Great Depression is associated with a large drop in white
male’s college attendance. In sum, the Great Depression may have increased the average
educational attainment, but the net effects seem small. More importantly, it appears to have
compressed the distribution of educational attainment among white males.
JEL Classification Codes: I20, N32
Keywords: the Great Depression, educational attainment, schooling, human capital investment
I thank participants at 2007 SOLE Annual meeting in Chicago and RAND Labor and Population
Group Brown Bag Lunch for helpful comments. I especially would like to thank Leah Platt
Boustan, Jim Hosek and Arie Kaptyen for helpful comments. All errors are mine.
This paper examines the relationship between the Great Depression and the educational
attainment of young adults who were growing up during the 1930s. The era of the Great
Depression is one of the most tumultuous periods in American economic history. Output
declined by more than 40 percent between 1929 and 1932, and the unemployment rate exceeded
20 percent in 1932 and 1933. Unemployment was severe as the majority of those who were
unemployed experienced a spell of unemployment longer than a year in certain parts of the
country (Margo 1993).
Tremendous efforts have been devoted to the analysis of the Great Depression. Most of
studies on the Great Depression have been concerned with its causes, its propagation mechanism
and immediate effects (e.g., Friedman and Schwartz 1971, Bernanke 1983, Hamilton 1987, and
Romer 1990 to name a few). Despite the great interest, however, there has been relatively little
research on the long-term consequences of the Great Depression for those who grew up during
this period.1 This paper tries to fill the gap by investigating the effect of the Great Depression on
the level of educational attainment of those who grew up during this tumultuous time in the
history of the United States.
Generally, the theoretical effects of recessions on human capital investment are
ambiguous as economic downturns could affect educational attainment in several ways.
Recessions may affect the budget constraints of households through unemployment and income
losses. This could lead to the affected individuals leaving school earlier than would otherwise be
optimal (income effect). Furthermore, in the absence of a perfect loan market, economic
downturns could increase the proportion of individuals who are liquidity constrained, leading
further to interruption of schooling. When a breadearner of a family becomes unemployed,
1 One of a few exceptions is Fishback, Haines and Kantor (2007) who analyze the
mortality and fertility rates between 1929 and 1940.
children of a certain age may give up schooling and look for a job to help the family (added-
worker effect). On the other hand, recessions could lower the opportunity costs of attending
school, thus increasing the schooling of the affected cohorts (substitution effect). Recent studies
generally find that the substitution effect dominates the income effect, thus schooling seems to
increase during recessions.
In addition to the standard factors, the Great Depression could have affected schooling
decisions even more because of its severity and length. Two features of the Great Depression
stand out. First, many banks went under during the Great Depression and in the absence of
deposit insurance, many people lost their bank savings after massive bank failures. As a result,
some had to quit schooling, as they were unable to withdraw their deposits in time for tuition
payments. Second, because of the deflation spiral during this period, the real interest rate shot up
between 1930 and 1933.2 As prices started to rise after 1933, the real interest rate turned
negative and stayed negative until 1938. As the higher interest rate exerts a similar effect as
lowering the return on education on decision to invest in human capital, the return on education
fluctuated widely during the Great Depression.
As in many sectors that were affected by the Great Depression, the education sector
underwent drastic changes during the early 1930s. Funding was cut. School building
construction was halted. School days were shortened. In some states, pay for teachers and
administrators were in arrears. As budgets were cut and demand for vocational education
increases, curricula were modified to emphasize more on job-related skills and less on art, music,
2 Whether or not the deflation was anticipated is subject to controversy. See Cecchetti
(1992) and Hamilton (1992).
Effects of recessions on schooling have been analyzed for a number of developing
countries. Thomas, Beegle, Frankenberg, Sikoki, Strauss and Teruel (2004), using the
longitudinal data, analyze the effect of Indonesia’s financial crisis in 1997. Schady (2004), using
three waves of Peruvian household survey data, examines the effects of the 1989-91 recession on
education among school-age children. Both studies find that economic downturns tend to
increase schooling of school-age children and young adults as the opportunity cost of schooling
diminishes and employment opportunities dry up.
When exogenous events that affect the entire economy are catastrophic, such as wars and
revolutions, there is often an unambiguous impact on education of particular cohorts. For
example, Ichino and Winter-Ebmer (2004) examine the effects of WWII, by comparing
educational attainment and earnings of Austrians and Germans who grew up during WWII to
those in Sweden and Switzerland that did not participate in WWII. They find that Austrians and
Germans who were born between 1930 and 1939 have lower average education by about 16 to
26 percent of a year compared to the cohorts born before and after them. This translates into 1.5
to 4.2 percent loss of subsequent earnings. On the other hand, they do not find declines in years
of education in Sweden and Switzerland. Meng and Gregory (2002) analyze the impact of the
Chinese Cultural Revolution on educational attainment. Because all schools in urban China were
closed during the early years of the Cultural Revolution and the curriculum emphasized manual
labor and university entrance was based on students’ political attitudes in later years, certain
cohorts of Chinese men and women lost up to eight years of education in the traditional sense.
This missed opportunity exerts a toll on the affected cohorts, as each year of missed schooling is
estimated to reduce the probability of obtaining a university degree by 1 percentage point.
Identifying the effects of macroeconomic shocks is often difficult, as macroeconomic
shocks affect the entire economy. With respect to the effects on education, macroeconomic
shocks affect entire cohorts of school-age children. Without a strong identification assumption
that there is no cohort effect, the effect of macroeconomic shocks on education may be
confounded by the unobserved cohort effect. For example, Schady (2004) uses the exposure to a
macroeconomic crisis to identify the effect of Peru’s major recession on education. His exposure
index is measured as the number of years a child was between the ages of 6 and 17 during 1988-
92, and hence is indistinguishable from the unobserved cohort effect.
In this paper, the effect of the Great Depression on educational attainment is identified by
taking advantage of differences in severity of the Great Depression at the state level. As
documented by Wallis (1989), there was considerable variation in the employment index among
states during the 1930’s. Figure 1 presents the employment index at the regional level, taken
from Wallis. This figure clearly demonstrates cross-sectional as well as time-series variation in
the severity of the Great Depression. For example, the Mountain region suffered most severely,
while the South suffered far less than the rest of the country. These differences are largely
attributable to the differences in regional trend in employment growth and industrial composition
(Rosenbloom and Sundstrom 1999). During the contraction phase, regions with a high
concentration of certain industries, such as lumber and cotton suffered larger declines in
employment, while New England and South Atlantic regions suffered less. During the recovery
period (1933-37), on the other hand, New England and South Atlantic experienced slower
growth compared to previously hard-hit regions such as Mountain and East South Central. I
exploit these differences in the employment index, which are plausibly exogenous to educational
attainment, to study the effects of the Great Depression on schooling decisions.
I want to stress at the outset that this paper measures educational attainment by years of
schooling but do not take into account changes in school quality. During the Great Depression,
however, school years were shortened. While the average decline of a school year was 1.5 days
from 1929/30 to 1931/32, there were a handful of states that shortened school days by more than
ten days (United States Office of Education 1938). In addition, instruction in various fields, such
as music and art, was eliminated at many schools. As funds were cut for schools and classrooms
were overcrowded in some regions of the country, many schools promoted and graduated
students who would otherwise have been held back. The quality of instruction thus inevitably
worsened during the Great Depression. Because I use the Census data, however, I am not able to
measure changes in educational attainment due to shortening of school years. Nor does this
paper address the deterioration of quality of instruction during this period. The estimates
provided in this paper could thus be considered as a lower bound for the effect of the Great
Depression on educational attainment of the affected cohorts.
I use the one-percent sample from the U.S. Decennial Censuses of 1960, made available
by the Integrated Public Use Microdata (IPUMS) Series at the University of Minnesota (Ruggles
and Sobek et al. 2003). I limit the sample to U.S. native-born whites and blacks born between
1912 and 1926.3 I exclude from the sample those who were born in the states of Alaska, Hawaii,
the District of Columbia, and Native-American Territories as well as those whose place of birth
is not recorded in IPUMS. I also eliminate observations for which age and birth place were
3 The choice of these birth-year cohorts is dictated by the availability of the employment
index for 1930-1940 and a ‘critical age’ in deciding human capital investment.
Measuring the highest grade achieved in a retrospective data set such as Census is subject
to a host of problems. First, because mortality is correlated with educational attainment,
members of older cohorts who have survived to a Census year are more likely to be better
educated compared to the average of their respective cohorts. Using the sample of men and
women who have survived to a Census, therefore, older cohorts may appear to have higher
education due to selection. My samples are, however, relatively young (the oldest individuals
are 48 at the time of Census) in 1960. Since the association between education and mortality
usually manifests in individuals older than 55, the bias arising from mortality selection bias
would be small for my sample.4
Second, individuals may overstate their educational attainment when societal norms for
education increase. Older individuals in my sample, therefore, could have exaggerated their
highest grade achieved when asked by Census enumerators, resulting in the bias from
“educational creep.” In this respect, the measurement problem in the highest grade achieved in
the 1940 Census is well known (Goldin 1988). Goldin documents that the Census data overstate
the educational attainment of older cohorts by a substantial margin. Comparing the 1940 U.S.
Census and the actual enrollment and graduation data, she concludes that the graduation rate of
the 1892-1902 birth cohort may be overstated by a factor of 1.5 or 2. For my sample, one can
evaluate the magnitude of possible overstatement by looking at the comparison of high school
graduation rates compiled from the 1940 Census and contemporaneous administrative data
4 Another possible problem is that black schools in early twentieth century were ungraded
(Margo 1990). This creates a problem when comparing educational attainment of blacks
and whites of older cohorts. However, as I estimate regressions separately for blacks and
whites, the inter-racial comparison is not an issue. My black samples are relatively
young (born between 1914 and 1926) who went to school in the 1920s and 1930s when
the ungraded school system is less common compared to older cohorts. I also estimate
the relationship between the employment index and educational attainment by region
(South and non-South).
(figure 6 (p. 366) of Goldin). The comparison of the two data sets reveals that there is no
discrepancy in graduation rates between the two distinct data sources for the cohorts born after
1916. For the earlier cohorts, however, the “educational creep” may exist, as the graduation rate
calculated from the 1940 Census is about 20 percent higher than the rate obtained from the
contemporaneous data, while the gap quickly narrows as the cohort ages. If the 1960 data are
comparable to the 1940 data, therefore, the magnitude of educational creep in the 1960 Census
would be negligible for the cohorts born after 1916, while a small “education creep” may exist
for the earlier cohorts.
To assess the magnitude of the overstatement in the 1960 Census, I compare the data
from the 1940, 1950, and 1960 Censuses and plot in figure 2 the average years of schooling
calculated from the three Censuses for the 1900-1925 birth cohorts, separately by race and
gender. Compared with the 1940 Census estimates, the 1960 average years of schooling appear
systematically higher for the older cohorts in question (born between 1912 and 1915), thus
aggravating the educational creep bias even further.5 Compared with the average years of
schooling estimated from the 1950 Census, on the other hand, the 1960 average years are
generally lower, suggesting the magnitude of over-statement in the 1960 Census is not larger
than in the earlier Censuses. Using the 1960 Census, therefore, would not introduce a further
bias for the cohorts born after 1916. For the cohorts born between 1912 and 1915, on the other
hand, their self-reported highest grade achieved may be overstated. In the regression analysis,
however, I control for birth-year as well as the state-of-birth dummy variables. Thus as long as
there is no systematic differences within a birth cohort in a given state, inclusion of the older
birth cohorts would not cause further bias in my results.
5 I cannot deny the possibility of these cohorts having gone back to school after the 1940
The key variable used for identification is the state-level employment index (August
1929=100) for each year from 1930 to 1940 constructed by Wallis (1989). The index was
calculated from the Bureau of Labor Statistics’s establishment surveys of employment. The
establishment sample was a self-reported, nonrandom sample. Furthermore, because the BLS
reported changes in employment over a two-months period for firms that reported in both
months, the reported percentage change is biased due to attrition and entry of new firms. Wallis
constructed the yearly index by benchmarking the estimated employment changes to the known
employment totals, such as Census of Manufacturers, appropriately reweighted by employment
shares (Wallis 1989).
In regression analysis, I use this employment index of the year when individuals turned a
critical age. I define the “critical age” as 16 and 18 for whites and 14 and 16 for blacks. I
choose these ages because these ages are the age when an individual makes a decision to drop
out of school and to attend college or not.6 Age 16 is the minimum age stipulated by the
compulsory schooling laws in many states that one has to remain in school. Age 18 is the age
when one starts attending college. Age 14 is often the minimum age for a work permit in many
states. I justify the choice of age 16 and 18 based on the government document from the period
that asserts that staying in school beyond the compulsory schooling age is the main culprit in
increasing enrollment in the early 1930s. High school enrollment increased considerably during
the early years of the Great Depression. Nationally, the increase in enrollment was 28.9 percent
between 1930 and 1934. A large part of the increase in enrollment was accounted for in the last
two years of high school. In particular, between 1930 and 1934, there was a 37.5 percent
increase in the third-year enrollment, while the fourth-year classes witnessed an increase of 43.4
6 I obtain similar results when the employment indices at ages 19, 17 or 15 are used.
percent. In addition, thousands of high school graduates returned to high school to take
additional courses. From 1932 to 34, there was a 38.4 percent increase in enrollment of post-
graduate students in high school (United States Office of Education 1938). The increases in the
last years of high school indicate that those who would have otherwise left school after reaching
the minimum compulsory schooling age remained in school because of the dismal employment
situation. Furthermore, the increase in post-graduate enrollment suggests that those who had no
means of going to college but could not find employment found schooling an attractive
I assume that individuals received schooling in the state in which they were born. This is
the same assumption used by Card and Krueger (1992) and Lleras-Muney (2002). These authors
assert that the mobility during the first half of the twentieth century was low and errors arising
from mismatches would be small. Contrary to the assertion of the previous authors, however, the
mobility in the 1930s seems quite high. In 1940, about one quarter of the 1912 birth cohort lived
in a state different from the one they were born in. Naturally, the mobility is lower for younger
birth cohorts but still quite substantial; nearly 12 percent of the 1925 birth cohort lived in 1940 in
a state different from the one they were born in. Using the employment index of the state of
birth during the 1930 thus might introduce attenuation bias due to measurement error.
If geographical mobility is correlated with employment index of the state,7 then using the
employment index of the state of birth could introduce additional bias in my estimation.
However, the direction of bias is not clear. I regress the indicator variable whether or not an
individual is living in the same state in which one was born on employment index turning a
critical age of 14, 16, or 18 and dummy variables for year and state of birth, separately for four
7 For relationship between the New Deal programs and internal migration, see Fishback,
Horace, and Kantor (2001).
race-gender categories (results not reported). In any specification, the coefficient estimate on the
employment index at the critical age is numerically small (between –0.0005 and 0.0004) and
statistically not significant from zero. It therefore seems unlikely that the employment index of
the state of birth would introduce additional bias in a systematic way above and beyond the
attenuation bias due to the measurement error.
Wallis’s employment index is not adjusted for population growth. The index, therefore,
may underestimate the severity of the Great Depression and present an optimistic view of the
recovery. Wallis (1989) documents, however, that adjusting the employment index by total
population growth or working population growth between 1930 and 1940 does not change the
relationship across regions. A possible exception to this is the Pacific region. Because there was
a large influx of migrants from depressed agricultural regions to California and the other western
states, adjusting for population growth reveals that the Pacific region did not enjoy fairly high
rates of employment growth, as the unadjusted raw index might suggest. In order to control for
differential population growth in different states, I include in my regressions the population
index of the state in the year sample individuals are turning the critical age. State-level
population figures (16 years and older) between 1930 and 1940 are linearly interpolated using
the 1930 and 1940 Censuses, and the population index was calculated as a ratio of the state
population relative to that of 1930.
II. Analysis of the Relationship between the Great Depression and Educational
(a) Differences in State-Level Employment Index
I regress the years of education on the employment index and the population index in the
year when an individual turns a critical age, the dummy variables for ten birth-years, three
quarters of birth and 48 state of birth for a sub-sample of those who turned the critical age
between 1930 and 1940 (e.g., for whites, those who were born between 1912 and 1922 for the
employment index at 18 regression, and those who were born between 1914 and 1924 for the
employment index at 16 regression, etc.) separately for whites and blacks by gender. Table 1
reports the results of the regressions, separately for race and gender. The estimated relationships
are numerically small, and all but for white females are statistically insignificant. Even for white
females, for which the coefficient is significant nearly at the 1-percent level (p-value of 0.012),
the effect is economically small (6% increase in schooling for a 10-point reduction in
I suspect that various factors were at work to derive these results. For some, schooling
decisions were constrained from below by child labor laws and compulsory schooling laws. As
Lleras-Muney (2002) has documented, there was a steady trend in rising educational attainment
at the bottom of the distribution, who would be normally affected by the compulsory schooling
laws. For some others, dismal job market opportunities lowered the opportunity costs of
schooling and enticed them to stay in school. Yet for some others, a drastic reduction in family
income during the Great Depression propelled them to give up pursuing further studies which
could be otherwise possible. At the end, however, it appears that the substitution effect and
income effect cancelled out each other, and on average the effects of the Great Depression on
educational attainment were negligible.
The stronger effects of the state employment index for white females may be explained
by the differential labor market situations women faced during the 1930s. First, during the 1930s
women faced higher returns on schooling than men. During the early twentieth century, there
was a compression of wage gaps that existed between clerical workers and production workers
due to an increased supply of high school graduates who were trained to do much of the clerical
work. Goldin and Katz (1995) document that for male workers, the estimated clerk
wage/production wage gap narrowed from 1.67 in 1909-14 to 1.15 in 1939 (table 2). For female
workers, however, while the wage gap narrowed substantially between clerical and production
jobs, the premium for clerical work was still substantial throughout the 1930s: in 1909-14, the
clerical workers on average earned double what production workers did, and in 1939, the
premium was still over 50 percent. Provided that many clerical jobs required high school
diplomas, there was an added incentive for women to stay in school longer as returns to high
school education was higher for women than for men. The substitution effect may have worked
stronger for women than for men, and hence I observe a statistically-significant and negative
relationship between the state-level employment index and the years of schooling.
Another possible reason could be that some women stayed in school longer as the
prospect for marriage dimmed substantially during the Great Depression. The age at first
marriage increased during the 1930s. This may be because marriage is often an expensive
proposition and people tend to delay marriage in economic downturns. Instead of marrying early
and dropping out of school, women could have continued education. Using the same data set, I
test the hypothesis if the age at first marriage is related to the state employment index when one
turns 18. While the results are not reported here, I find that the coefficient of the state
employment index is negative and statistically significant after controlling for the state year and
quarter of birth.
I have so far treated the 1930s as a single period. History of the Great Depression is far
from monotonic, however. After the stock market crash of October 1929, the economy did not
go into a free fall right away. However, starting in late 1930, the economy went through a long
excruciating process of contraction; GNP declined by 8.9 percent, 7.7 percent, and 13.2 percent
from 1930 to 1932, respectively and unemployment rate rose from 3.2 percent to 23.6 percent in
the three years. Only after the drastic measure of Bank Holiday on March 5, 1933 and an
aggressive infusion of money supply, the economy’s bottomless fall ended. GNP declined by
modest 2.1 percent in 1933 and started growing again in 1934. Various measures to bolster
employment were undertaken under the New Deal. After a few years of robust growth, however,
the economy again experienced a year-long recession in 1937II-1938II: output declined by 4.0
percent in 1938 while unemployment stayed high. The economy then registered again strong
growth in 1939 and 1940. These colossal fluctuations of the economy would naturally affect
people’s expectations about future income, which in turn would affect education decisions.
I divide the sample period into the two phases, the contraction period (1930 to 1934) and
the recovery period (1935 to 1940) and estimate again the effects of the employment indices at
the critical ages. Table 3 reports the results of this exercise for four categories of race and
gender. None of the estimates are statistically significantly different from zero, or within-group
estimates across two periods are different from each other. A relatively large estimate is
obtained for the black male sample in the contraction phase, indicating a possibly strong
substitution effect for black males in the early 1930s. However, the estimate is only marginally
significant (p-value of 0.055). I thus conclude that the effect of the Great Depression on
education was very small, if it existed at all.8
(b) Regional Variation in Supply-Side Factors
As there was tremendous state-level heterogeneity in the severity of the Great
Depression, there could be heterogeneity of response of teenagers to the local economic
conditions. In this section, I consider three possible sources of state-level heterogeneity that are
related to the provision of educational infrastructure. I posit that the provision of infrastructure is
important because lack of accessibility to schools could act as significant transaction costs which
may hinder smooth adjustment of human capital investment behavior in response to economic
shocks. Because the supply of educational institutions is fixed in the short run, all students may
not be accommodated even when there is strong, growing demand. The supply constraint is
particularly important during the Great Depression when school budgets were severely cut.
Capital outlays for public schools were reduced from $371 million in the 1929/30 academic year
to only about $59 million in 1933/34 (Moreo 1996).
A reduction in school budgets also has implications for transaction costs of attending
schools. Operating budgets were cut for schools across the board, and many rural schools were
8 One could argue that a possible explanation for small and statistically insignificant
coefficient estimates is that the state dummy variables and the employment index are
closely correlated to each other and that by including state dummies I am simply
capturing the collinearity between the two variables. Although this argument may have
some validity, I argue it is not the entire story. First, there is a tremendous variation in
the employment index across years within the states. Thus the employment index
captures changes in educational attainment across birth-year cohort within the state of
birth. Second, if collinearity between the two variables is the main culprit, then I would
expect that coefficient estimates would always be small and statistically insignificant.
However, certain results (e.g., Mid-Atlantic & Northeast state sample and results from
quantile regressions) are numerically large and statistically significant even after
controlling for state-specific fixed effects. The employment index, therefore, must
contain some useful information beyond the state specific dummy variables.
no longer able to provide transportation to students in remote areas. [source?] Even when
education is provided free of charge, an increased transaction cost may hinder students from
attending schools in times of financial distress. Thus I would expect that responses to economic
downturns could be different between the populous states and the states that are sparsely
populated and thus access to school is difficult.
The first feature to consider is variation in population density and resultant accessibility
to schools in different regions of the country. I estimate the same regressions with a sub-sample
of white men and women in the “Mountain” states, which were hard hit by the Great Depression
but very sparsely populated, and in the “Mid-Atlantic” and “East North Central” regions that
were also severely hit by the Great Depression but were more populated. The results of this
exercise are presented in table 4. For both male and female, the estimates from the Mountain
states samples are numerically small and statistically insignificant. On the other hands, the
estimates from the Mid-Atlantic and East North Central regions are numerically large and highly
significant. The estimates indicate that a 10-percentage point decrease in employment index is
associated with a 10- to 11-percent of a year longer stay in school for white men and women.9
My results here, therefore, confirm the assertion of Goldin that “the Great Depression may have
had on positive effect: it enticed the youth to stay in school” (Goldin 1999, p. S80) with a
qualifier; the positive effect of the Great Depression on schooling appears to be limited to certain
states that are populous.
9 The 10-percentage point difference in employment index --- roughly the difference
between Illinois (70.4) and Kentucky (80.2) in 1933 --- was not uncommon during the
early 1930s. Even within the same state, the 10-percentage point change in employment
index over a two-year period was observed. For example, in Michigan, the employment
index changed from 79.8 in 1932 to 68.1 in 1933 and then to 78.9 in 1934.
When discussing supply of post-secondary education in the early 20th century, one cannot
overly stress the importance of the role played by junior colleges. The junior college is truly a
creation of the twentieth century American educational system. Originally intended as a way to
free universities from freshman and sophomore instructions so that universities could be
reconstituted as research and training centers for intellectual elite (Steven Brint and Jerome
Karabel 1989), the first independent junior college was founded in 1901 in Chicago. The idea of
junior college took some time to take root in other parts of the country. By 1920, there were
only 52 junior colleges in nation, of which only ten were funded publicly. The creation of junior
colleges accelerated during the 1920s; by 1928, there were 248 two-year colleges (114 were
public) enrolling nearly 45,000 students (Brint and Karabel 1989).
While four-year colleges faced extremely tough challenges, junior colleges thrived during
the 1930s. The number of junior colleges increased from 277 in 1930 to 456 in 1940.
Concomitantly, the enrollment grew from 55,616 in 1930 to nearly 150,000 in 1940. The
expansion is attributed to junior college’s relatively inexpensive cost structure to operate at a
time when demand for higher education was high (Brint and Karabel 1989). For many high
school graduates, junior colleges presented the only affordable opportunity to continue education
as their tuition charges were lower compared to four-year colleges and universities.
The growth of junior colleges was uneven across regions. As junior colleges were
created with sponsorship of universities to encourage an increasing number of high school
graduates to “give up college work at the end of the sophomore year” (quoted in Brint and
Karabel 1989, p. 25), university sponsorship was critical in establishment of junior colleges in
many states. In 1930, only four states (California, Illinois, Michigan and Missouri), where
university sponsorship was vigorous, plus Texas accounted for 64 percent of total junior college
enrollment. New junior-college systems were introduced in Wisconsin, Georgia, and Utah while
four states opened public systems. Much of the growth in the 1930s came in states that had
already become established centers of junior colleges, however. By 1940, the three largest states
(California, Illinois, and Texas) enrolled 66 percent of all public junior college students, while
less than 2 percent of enrollments nationwide were accounted for by 11 states in the Northeast
(Brint and Karabel 1989). I, therefore, re-estimate the regressions with the sub-sample of
“Junior College” states and present the results in table 4. Again, the estimated substitution effect
of the Great Depression seems strong and significant for white females: a 10-percentage point
decrease in the state employment index is, on average, related to one-seventh of a year longer
stay in school. The estimates for the white males in the “Junior College” states, however, are
small and statistically insignificant from zero. Looking at the results in panel A of table 4,
availability of school infrastructure and ease of access to schools are an important determinant
for the substitution effect to work to increase educational attainment during economic recessions.
Another possible source of state-level heterogeneity is the existence of Jim Crow and de
jure segregation in the South. Schools were not integrated in the Southern states. As many
Federal programs were administered locally, Southern blacks benefited little from programs
implemented in the 1940s and 1950s. Collins (2001) reports that the change in the non-
white/white ratio of defense-related employment was appreciably lower in the South than in the
non-South states under the Fair Employment Practice Committee. Turner and Bound (2003)
demonstrate that the effects of the G.I. Bill on years of college education and the probability of
college completion were considerably lower for Southern blacks. Although decision to invest in
human capital is largely individual, it would not be surprising if blacks in South as they faced
different constraints than those in the non-South states.
The bottom panel of table 4 presents the estimates from the results from the black
samples estimated separately for the South and non-South regions. Although the results are not
statistically significant, the estimates are more negative and numerically larger (in absolute
values) in the non-South states than in the South for black men and women. Particularly for
black men, the estimates is large and reasonably significant (p-value of 0.052), implying a large
substitution effect at work in the non-South regions. Although such differences may reflect the
differences in industrial structure between the South and non-South states, the findings are
consistent with the hypothesis that social institutions in the South were not conducive to blacks
to stay in school longer in response to economic circumstances.
(c) The Effects of the G.I. Bill
Finally, estimating the relationship between the Great Depression and educational
attainment is further complicated by the existence of the G.I. Bill. Many men in my sample
served during WWII. In my sample, 66 percent of white males and 48 percent of black males for
the 1914-24 birth cohorts (those who turned 16 between 1930 and 1940) served during WWII. If
they took advantage of the G.I. Bill after the war, their education level would be higher than in
the absence of the G.I. Bill. If the employment index at the state level and the WWII veteran
status are correlated, estimates obtained from the regressions excluding the WWII veteran status
would be biased.
Controlling for the WWII veteran status creates a host of econometric issues. In the
simplest form, I could include a binary variable for the WWII veteran status in the right-hand
side of the regressions and estimate the effect of the employment index on educational
attainment. However, because literacy, mental and physical fitness were requirement for
enlistment, the veteran status and unobservable ability that might affect educational attainment
would be correlated. This correlation will create a familiar endogeneity bias in estimation.
Ideally, if I could find an instrumental variable that is correlated with the veteran status but not
with the error term, then I could resort to the instrumental variable estimation. Unfortunately,
because the cohorts of interest in my sample (born in 1912-24 for white men, and born in 1914-
26 for black men) volunteered for military service, I am not able to use the quarter of birth
variable as an instrument used by Angrist and Krueger . As the youngest cohort in my sample
are black men and women born in 1926 (who turned 14 in 1940), I cannot use the instrument
(dummy variable for being born after the third quarter of 1927) used by Bound and Turner ,
Alternatively, I can estimate the magnitude of the bias arising from omitting the WWII
veteran status variable from my estimation. Specifically, the size of the omitted variable bias is
determined by the true effect of the omitted variable on the outcome variable times the
relationship between the omitted variable (WWII veteran status) and the variable of interest
(state-level employment index). I thus regress the binary variable of WWII veteran status on the
state employment index and other control variables to assess the magnitude of association
between these two variables. The results are reported in table 5, which indicate the association is
numerically very small, almost indistinguishable from zero (and statistically insignificant),
ranging from 0.00002 to 0.0004 for white men and –0.0002 to 0.0007 for black men. The
estimates of the effect of WWII service on years of college completed are between 0.23 and 0.28
years for white men (Bound and Turner 2002). Multiplying this with the estimate of the
regression coefficient of the state-level employment index at age 16 would yield 0.0001, which is
about 3.3 percent of the estimate obtained in table 1, column (2). The estimate of the effect of
the Great Depression on educational attainment could therefore be biased upward, but its
magnitude would be extremely small.
(d) Income Effect
That there appears little effect of the Great Depression on the average years of schooling
does not mean that there was little effect on the overall population. For example, Lleras-Muney
plots years of education for each decile (p. 421) for the 1901-1925 birth cohorts. In her Figure 2,
there is a noticeable dip in educational attainment for the 90-percentile of the distribution for the
birth cohorts born during the 1910s. The Great Depression was the time when those who were
born in the 1910s were turning college age. In the 1930s, while high-school education and junior
colleges were mostly free, colleges and universities were not. At public universities, the average
in-state tuition and fees were $61 in 1933 ($753 in 1997 dollars). At private institutions, the
average was $265, or $3,272 in 1997 dollars (Goldin and Katz 1999, p. 50). Hence high-school
graduates whose families suffered from lower income during the Great Depression might have
given up on higher education. Indeed, the percentage of high-school graduates attending college
a year after graduation declined from 31.5 percent in 1929 to 22.9 percent in 1933, and the
enrollment at four-year colleges and universities declined by 8.6 percent between 1932 and 1934
(United States Office of Education 1935). If the Great Depression had a profound impact on the
decision to attend college, then that effect would show up in the top tail of the distribution but
might not at the mean.10
10 Anecdotal evidence abounds; over 60 percent of the freshmen at the University of
Chicago who did not return in the fall semester of 1931 did so because of financial
reasons. Only 54 percent of students who entered Stanford in 1930 graduated four years
later (Levine 1986).
I formally test whether or not the Great Depression has affected the educational
attainment at the top tail of the distribution. Specifically, as before, I regress the years of
education on the employment and population indices at age 18, the dummy variables for 48
states, ten birth-years, and three quarters of birth for a sub-sample of white men who were born
between 1912 and 1922 but this time using the quantile regression, estimating the effect at the 90
percentile. Table 6 presents the results of the quantile regressions, which indicate that there was
a strong income effect.11 For white males at the 90-percentile of the educational attainment
distribution, a 10-point decrease in the employment index has led to a quarter of a year shorter
stay in school, even after controlling for the state of birth and the year of birth. Thus for white
males, there is evidence that the severity of the Great Depression might have affected college-
attendance decisions of those who were turning 18 during the 1930s.
The effect of the Great Depression on white males’ educational attainment can be seen in
changes in distribution of years of schooling. For example, only 8.5 percent of the 1913 birth
cohort has more than 16 years of education, in contrast with 9.2 percent of the 1909 birth cohort
and 9.4 percent of the 1911 birth cohort. This decrease in college graduation was compensated
for by an increase in high school graduation. Of the 1913 birth cohort, fully 24.9 percent had
exactly 12 years of schooling, while the comparable numbers are 19.7 percent and 22.5 percent
for the 1909 and 1911 birth cohorts, respectively. Thus it appears that white men who were
graduating from high school in the early 1930s were giving up college and went into the labor
force more so than the cohorts who were merely a few years older than they.
For a sub-sample of black men born between 1914 and 1924, I estimate the effect of the
state-level employment index at age 16 at the 70 percentile. The second column of table 6
11 I find no statistically significant results for females (white and black) at any decile.
reports the results of the quantile regression. As in the case of white males, I find a strong
income effect at work; a 10-point decrease in employment index has led to one-fifth of a year
shorter stay in school. Compared to white males, the income effect for black men seems to have
manifested earlier in life, at 10th grade, when they were no longer bound by the state compulsory
The Great Depression has had long-lasting effects on those who were growing up during
this period. One such effect could be that the Great Depression increased the educational
attainment of the cohort of white men and women who were turning college age during the early
1930’s through the reduced outside opportunities during this deep and prolonged economic
downturn. My estimates suggest that the effects were small and by and large statistically not
different from zero. However, by splitting the sample into different regions, I find numerically
larger and statistically significant results in states where the population density was higher and/or
access to alternative schooling system, such as junior colleges, was readily available. In such
states, the substitution effect of the recession has dominated the income effect. As a result, the
net effect of the Great Depression was to increase the average years of schooling.
On the other hand, for certain segments of the population, the Great Depression had an
effect to reduce education attainment. In particular, at the top end of educational distribution of
white men, the income effect seems to have dominated the substitution effect and the severity of
the Great Depression is associated with a substantial decrease in the years of schooling.
Similarly a strong income effect is observed for black men, at the 70-percentile of the
distribution. From these results, the main impact of the Great Depression on educational
attainment appears to compress the overall distribution of educational attainment.
Anderson, Charles J. Fact Book on Higher Education, 1997 Edition, Phoenix: The Oryx Press
for the American Council on Education, 1998.
Angrist, Joshua D. and Krueger, Alan B. “Does Compulsory Schooling Attendance Affect
Schooling and Earnings?” Quarterly Journal of Economics, 106 (4), November 1991,
Angrist, Joshua D. and Krueger, Alan B. “Why Do World War II Veterans Earn More than
Nonveterans?” Journal of Labor Economics, 12 (1), January 1994, pp. 74-97.
Bernanke, Ben. “Nonmonetary Effects of the Financial Crisis in Propagation of the Great
Depression.” American Economic Review, 73 (3), June 1983, pp. 257-76.
Bound, John and Turner, Sarah. “Going to War and Going to College: Did World War II and the
G.I. Bill Increase Educational Attainment for Returning Veterans?” Journal of Labor
Economics, 20 (4), October 2002, pp. 784-813.
Brint, Steven, and Karabel, Jerome. The Diverted Dream: Community Colleges and the Promise
of Educational Opportunity in America, 1900-1985. New York: Oxford University
Cecchetti, Stephen G. “Prices During the Great Depression: Was the Deflation of 1930-1932
Really Unanticipated?” American Economic Review, 82 (1), March 1992, pp 141-156.
Collins, William J. “Race, Roosevelt, and Wartime Production: Fair Employment in World War
II Labor Markets.” American Economic Review, 91 (1), March 2001, pp. 272-286.
Fishback, Price V., Horrace, William C., and Kantor, Shawn. “Do Federal Programs Affect
Internal Migration? The Impact of New Deal Expenditures on Mobility during the Great
Depression.” Cambridge, MA: NBER Working Paper No. 8283, May 2001.
Fishback, Price V., Haines, Michael R. and Kantor, Shawn. “Birth, Death, and New Deal Relief
During the Great Depression.” Review of Economics and Statistics, 89 (1), February
2007, pp. 1-14.
Friedman, Milton, and Schwartz, Anna J. A Monetary History of the United States: 1867-1960.
Princeton, NJ: Princeton University Press, 1971.
Goldin, Claudia. “America’s Graduation from High School: The Evolution and Spread of
Secondary Schooling in the Twentieth Century.” Journal of Economic History, 58 (2),
June 1998, pp. 345-74.
Goldin, Claudia. “Egalitarianism and the Returns to Education during the Great Transformation
of American Education.” Journal of Political Economy, 107 (6) part 2, December 1999,
Goldin, Claudia, and Katz, Laurence F. “The Decline of ‘Non-Competing Groups’: Changes in
the Premium to Education, 1890 to 1940.” NBER Working Paper, No. 6144, August
Goldin, Claudia, and Katz, Laurence F. “The Shaping of Higher Education: The Formative Years
in the United States, 1890 to 1940.” Journal of Economic Perspectives, 13 (1), Winter
1999, pp. 37-62.
Goldin, Claudia, and Katz, Laurence. “Mass Secondary Schooling and the State: The Role of
State Compulsion in the High School Movement.” NBER Working Paper, No. 10075,
Hamilton, James D. “Monetary Factors in the Great Depression.” Journal of Monetary
Economics, 19 (2), March 1987, pp. 145-69.
Hamilton, James D. “Was the Deflation during the Great Depression Anticipated? Evidence from
the Commodity Futures Market.” American Economic Review, 82 (1), March 1992, pp
Ichino, Andrea, and Winter-Ebmer, Rudolf. “The Long-Run Educational Cost of World War II.”
Journal of Labor Economics, 22 (1), January 2004, pp. 57-86.
Lleras-Muney, Adriana. “Were Compulsory Attendance and Child Labor Laws Effective? An
Analysis from 1915 to 1939.” Journal of Law and Economics, 45 (2), October 2002, pp.
Margo, Robert A. Race and Schooling in the South, 1880-1950: An Economic History. Chicago,
IL: University of Chicago Press for NBER, 1990.
Margo, Robert A. “The Microeconomics of Depression Unemployment.” Journal of Economic
History, 51 (2), June 1991, pp. 333-341.
Margo, Robert A. “Employment and Unemployment in the 1930s.” Journal of Economic
Perspectives, 7 (2), Spring, 1993, pp. 41-59.
Meng, Xin, and Gregory, Robert. “The Impact of Interrupted Education on Subsequent
Educational Attainment: A Cost of the Chinese Cultural Revolution.” Economic
Development and Cultural Change, 50 (4), July 2002, pp. 935-59.
Moreo, Dominic W. Schools in the Great Depression, New York: Garland Publisher, 1996.
Romer, Christina D. “The Great Crash and the Onset of the Great Depression.” Quarterly
Journal of Economics, 105 (3), August 1990, pp. 597-624.
Rosenbloom, Joshua L. and Sundstrom, William. “The Sources of Regional Variation in the
Severity of the Great Depression: Evidence from U.S. Manufacturing, 1919-1937.”
Journal of Economic History, 59 (3), September 1999, pp. 714-47.
Ruggles, Steven and Sobek, Matthew et al. Integrated Public Use Microdata Series: Version
3.0. Minneapolis: Historical Census Projects, Univ. of Minnesota, 2003.
Schady, Norbert R. “Do Macroeconomic Crises Always Slow Human Capital Accumulation?”
World Bank Economic Review, 18 (2), January 2004, pp. 131-154.
Stanley, Marcus “College Education and the Midcentury GI Bills.” Quarterly Journal of
Economics, 118 (2), May 2003, pp. 671-708.
Thomas, Duncan; Beegle, Kathleen; Frankenberg, Elizabeth; Sikoki, Bondan; Strauss, John, and
Teruel, Graciela. “Education in a Crisis.” Journal of Development Economics, 74 (1),
June 2004, pp. 53-85.
Turner, Sarah, and Bound, John. “Closing the Gap or Widening the Divide: The Effects of the
G.I Bill and World War II on the Educational Outcomes of Black Americans.” Journal
of Economic History, 63 (1), March 2003, pp. 145-177.
United States Office of Education. Biennial Survey of Education in the United States: 1933-34.
Washington, DC: Government Printing Office, 1935.
United States Office of Education. Biennial Survey of Education in the United States: 1934-36.
Washington, DC: Government Printing Office, 1938.
Wallis, John J. “Employment in the Great Depression: New Data and Hypotheses.” Explorations
in Economic History, 26 (1), January 1989, pp. 45-72.
Table 1 Relationship between Employment Indices and Educational Attainment, by Race and
(1) (2) (3)
index at age
−0.006* −0.005 Employment
index at age
index at age
Adjusted R2 0.070 0.068 0.055
No. of obs 10,291 105,474 105,924 108,738 10,306 10,599 11,808 12,231
Note: Dependent variable is the highest grade achieved in years. Regressions also control for
the dummy variables for year, quarter and state of birth and the population index of the
year an individual is turning a critical age in the state of birth. Robust standard errors are
reported in parenthesis and account for clustering on year and state of birth.
* significant at the 5% level
0.055 0.117 0.112 0.094 0.094
Table 2 Return on Education by Sex and Occupation, in 1926-29 and 1939
Ratio of earnings of clerical group to those of production workers in manufacturing
All clericals Clerks
Source: Goldin and Katz, 1999
Table 3 Estimates by Birth-Year Group, by Race and Gender
A. Whites, by Gender
44,598 58,314 46,481 58,993 45,735
index at age 18
index at age 16
No. of obs.
0.066 0.050 0.058
B. Blacks, by Gender
index at age 16
index at age 14
No. of obs.
Note: See note in table 1.
Table 4 Regressions by Region, by Race and Gender
A. Whites, by Gender
& East North
& East North
index at age 18
No. of obs.
B. Blacks, by Gender
index at age 16
No. of obs.
Note: See note in table 1. The Mountain states are Arizona, Colorado, Idaho, Montana,
Nevada, New Mexico, Utah, and Wyoming, the Mid-Atlantic states are New Jersey, New
York, and Pennsylvania, and the East North Central includes Illinois, Indiana, Michigan,
Ohio, and Wisconsin. The “Junior College” states are defined as California, Iowa,
Illinois, Kansas, Michigan, Mississippi, Oklahoma and Texas, and the South are defined
as Alabama, Florida, Georgia, Mississippi, North and South Carolinas, Virginia,
Tennessee, Arkansas, Louisiana, Texas, Kentucky, and Missouri.
** significant at the 1% level
Table 5 Regressions of WWII Veteran Status, Males by Race
Employment Index at 18
Employment Index at 16
Employment Index at 14
No. of obs.
White males Black males
Note: Dependent variable is the binary variable indicating the veteran status during WWII.
Regressions also control for the dummy variables for year, quarter and state of birth and
the population index of the year an individual is turning a critical age in the state of birth.
Robust standard errors are reported in parenthesis and account for clustering on year and
state of birth.
Table 6 Quantile Regressions for the Employment Indices on Education, Males, by Gender
Employment index at age 18
Employment index at age 16
No. of observations
(90 percentile) (70 percentile)
Note: See note in table 1.
** significant at the 1% level
Figure 1 Total Employment Index 1929-1940 (August 1929=100) by Census Region
Source: Wallis (1989), Table 2
Figure 2 Comparison of Self-Reported Years of Schooling in the 1940, 1950, and 1960 Censuses
Average Self-Reported Years of Schooling
Year of Birth