## Abstract

Did the World War II (WWII) GI Bill increase the probability of completing high school and further affect the probability of poverty and employment for the cohorts for whom it benefited? This paper studies whether the GI Bill, one of the largest public financial aid policies for education, affected low education levels in addition to its documented effects on college education, and whether it increased economic well-being for its beneficiaries. I use the 1970 Census and the variation in WWII military participation rate across birth cohorts and states of birth for men. I find that the WWII GI Bill significantly increased the probability of completing high school by 13 percentage points and reduced the probability of being below the poverty line by 4 percentage points for black and white men. It also increased the probability of being employed by 3 percentage points and the number of weeks worked by two weeks.

## 1.  Introduction

The World War II GI Bill of 1944 is one of the largest welfare policies implemented in the United States. Its aim was to help returning soldiers reintegrate into civilian life, primarily by providing generous financial support to attend educational institutions. It is well known that the bill had a considerable impact on the lives of veterans and on society, and is often referred to as “The law that changed America.”1

A sizeable academic literature has documented many benefits of the legislation for the cohort of men who fought in World War II (WWII). This body of work has shown that the GI Bill increased college education (Bound and Turner 2002; Stanley 2003), the next generation's education (Page 2007), home-ownership (Fetter 2013), mortality (Bedard and Deschenes 2006), and led to assortative mating (Larsen et al. 2015). This paper extends this literature by investigating the effects of the bill on individuals who otherwise would have been high school dropouts—focusing on the effects of the bill on high school completion. It further studies the effect of the GI Bill on poverty and employment. I build on the identification strategy proposed by Bound and Turner (2002) and Larsen et al. (2015), which exploits variation in WWII military participation rates for men across birth cohorts. I also incorporate into the empirical strategy differences in mobilization rates across states. This additional level of variation draws from Acemoglu, Autor, and Lyle (2004) who use the variation to study the effect of female labor supply on wage structure. Specifically, I use variation in WWII military participation (and thus GI Bill eligibility) at the birth year by birth quarter by state of birth level.2

I find that the WWII GI Bill had significant impacts for veterans at lower levels of education, just as it affected other veterans. I find that the WWII GI Bill substantially increased the probability of completing high school. Men from birth cohorts that fought in WWII are 13 percentage points more likely to have finished high school than those from birth cohorts who did not fight in WWII and were not eligible for the GI Bill. I also look for effects on other measures of well-being because education can be a safeguard against economic hardship.3 I look at the effects on employment and poverty. For the combined sample of black and white men, I find that the bill significantly reduced the probability of living in poverty by 4 percentage points, increased the probability of being employed by 3 percentage points, and increased the number of weeks worked by two. Separating the effects by race, I find that the bill affected black and white men equally for high school completion, poverty, and employment.

The paper is organized as follows. Section 2 reviews the GI Bill literature in more detail and describes the provisions in the bill for veterans without a high school degree. Section 3 describes the empirical strategy, statistical model, and data. Section 4 presents the results and section 5 concludes.

## 2.  The GI Bill and High School Completion

The GI Bill of 1944 provided returning veterans financial assistance to pursue education. A veteran who had served in the war for ninety days or had been discharged before ninety days because of disabilities acquired during service between September 1940 and July 1947 was eligible for the benefits under the GI Bill. The GI Bill education benefits were available for a maximum of four years, and the years available for each veteran depended on his length of service and age. Most veterans were eligible for maximum benefits as lengths of service often exceeded three years (Bound and Turner 2002). The benefits included tuition and expenses related to education, such as books, up to $500, as well as a monthly stipend. The amount of the stipend depended on family size, with married veterans receiving a more generous stipend than single veterans. Veterans were required to start their education by July 1951. Other provisions in the GI Bill included, among other benefits, access to loans to promote home ownership, and unemployment allowance to help veterans readjust. As mentioned earlier, previous literature has documented many effects of the WWII GI Bill on the lives of veterans and the American society. Bound and Turner (2002) use the sharp fall in WWII participation rates for men born after the third quarter of 1927 and compare college education across cohorts with high and low participation rates. They find that the GI Bill education benefits increased years of college for white men by 0.3 years. Stanley (2003) studies the effect of the bill on college education by comparing cohorts (1921–1922) who had high probability of completing their education prior to war, to cohorts (1923–1926) who went to war right after finishing high school. He finds that the GI Bill increased college education for the 1923–1926 birth cohorts by 20 percent. Turner and Bound (2003) find that black veterans benefited less than white veterans because of segregation barriers in the southern region of the country. Angrist and Krueger (1994) use the chronological order of birth dates, which was used as a method of enlistment, as an instrumental variable for veteran status and find no difference between veterans’ and non-veterans’ earnings. Lemieux and Card (2001) use the Canadian GI Bill and find that it increased education and earnings of the affected cohorts. Studies have also found that the GI Bill led to assortative mating (Larsen et al. 2015), lower grade repetition of the next generation (Page 2007), and more home ownership (Fetter 2013). Bedard and Deschenes (2006) find that mortality is higher among veterans of WWII, a consequence of the subsidized provision of tobacco products on military bases. Although many people associate the GI Bill with higher education, the provisions of the GI Bill allowed the use of funds (tuition and stipend benefits) toward finishing secondary school as well. The following extract from the Servicemen's Readjustment Act of 1944 states that both secondary and post-secondary education levels could be pursued by returning soldiers: the term educational or training institutions shall include all public or private elementary, secondary, and other schools furnishing education for adults, business schools and colleges, scientific and technical institutions, colleges, vocational schools, junior colleges, teachers colleges, normal schools, professional schools, universities, and other educational institutions. (Servicemen's Readjustment Act of 19444; Title II, Chapter IV, Part VIII, section 11) The GI Bill reduced the cost of attending secondary school at the same time it reduced the cost of attending post-secondary institutions. Prior to war, a future soldier may have dropped out of high school as the total cost, including the opportunity cost, of attending school may have been too high for him. After the war, on receiving financial support for attending high school from the GI Bill, he would have been able to afford to go back and complete high school. The reduction in the cost of post-secondary education might have also influenced the decision to go back to school. The GI Bill gave the beneficiary up to July 1951 to start his education and covered the costs up to four years of education—a sufficiently long time horizon of financial aid to accomplish a variety of possible education paths. It would have opened many opportunities beyond just high school for a veteran who was a high school dropout. The veteran would now consider not just the returns from completing high school but also the returns from further attending college or acquiring training. The GI Bill would have covered a significant portion, if not all, of the total costs, including the opportunity cost of pursuing such an education path. By reducing the cost of both secondary education and post-secondary education, the GI Bill may have caused high school dropout veterans to reevaluate their optimal decisions regarding high school completion. Studies have found that programs that expose high school students to opportunities of higher education, such as Early College High Schools that give students the opportunity to earn college credit while in high school, significantly improve high school graduation rates (Berger et al. 2013). Similarly, the GI Bill, by opening up opportunities of higher education, could have incentivized beneficiaries to cross the high school completion hurdle in order to access higher education. The 1979 National Survey on Veterans Summary Report (Hammond 1980) states that only 2.4 percent of WWII veterans who used the GI Bill used it for high school completion and the remaining were distributed over college, on-the-job training, apprenticeship, and other forms of training. However, the Final Report on Educational Assistance to Veterans (Bowman 1973) states that 28.6 percent of beneficiaries used the GI Bill for college education and 44.6 percent used it for below college education (including flight training). The discrepancy in statistics arises because in the 1979 National Survey of Veterans Summary Report beneficiaries who used the GI Bill for college and other types of education were put under the “college” category. It is thus unknown to what extent the GI Bill affected high school education because a substantial number of beneficiaries could have used the GI Bill for both high school completion and further training and education. Given that almost half of the returning veterans had not completed high school (46.8 percent)5 and almost half of the GI Bill beneficiaries used the benefits for below college education (44.6 percent),6 there was the potential of a large number of men using the education benefits to cross the high school completion threshold. It is also possible that the GI Bill caused veterans to complete high school or pass the General Education Development test (GED) on their own finances and use the bill to fund further studies. In this case they would not be counted as having used the GI Bill for completing high school, though it was the causal factor. It is also important to note that the GED was initiated in 1942 for veterans, which would have enabled them to access many entry-level jobs and access higher education. It is possible that many veterans who are classified as high school graduates in the data are in fact GED holders, which would imply that the GI Bill was used both for human capital accumulation and for acquiring a signal that would open doors to employment and training. Given that 44.6 percent of GI Bill beneficiaries used the education benefits for below-college level education, it is quite possible that many veterans who were school dropouts obtained the GED and then pursued on-the-job training, apprenticeship, or some other form of training. Because of data limitations, I am unable to separate a regular high school graduate from a GED holder in my analysis or study the impact on training as a separate outcome. ## 3. Empirical Strategy ### Research Design As noted before, this paper applies the regression discontinuity strategy used by Bound and Turner (2002) and Larsen et al. (2015). There is a drastic fall in the WWII military participation rates for men born toward the end of 1927. Birth cohorts 1923–1927 have high WWII military participation rates and birth cohorts from 1928 onwards have much lower WWII participation rates. Thus, one may estimate the impacts of the GI Bill by comparing birth cohorts just before 1928, with high WWII participation rates (and high GI Bill eligibility rates), to birth cohorts 1928–1938, with low WWII participation rates (and low GI Bill eligibility rates). In addition to this time series variation, however, there is also variation in military participation across states. The differences in mobilization rates across states were determined by state characteristics, such as farming, schooling, and ethnicity (Acemoglu, Autor, and Lyle 2004). Incorporating the differences across states allows me to capture variation in WWII GI Bill eligibility at a finer level.7 I use variation in GI Bill eligibility captured by the fraction of men who served in WWII in a birth year, birth quarter, and state of birth cell. I include state of birth fixed effects interacted with a time trend to control for state-level differences in characteristics that may have determined mobilization and may have changed over time. These state-specific trend controls prevent any differential trends in schooling across states from biasing the coefficient on the WWII variable.8,9 The benefit of using a specification that includes variation across birth cohorts and states is that while it still captures the variation used in previous strategies, that is, across birth cohorts, it also captures variation in military participation that existed across states, thus using more variation and at a finer level to estimate the impact of the GI Bill. Analyses that only use variation across birth cohorts essentially provide estimates that aggregate over the differential effects of the GI Bill across states, due to differential military participation rates, and thus would be missing out on variation that could potentially provide useful information on how the GI Bill affects education. Robustness checks in section 4 compare estimates for specifications that include and exclude the variation across states. The results show that for high school completion the specification that does not include variation across states gives estimates that are almost three times as large (increase of 35 percentage points) as the specification that does include variation across states (increase of 13 percentage points), the former being an implausibly large estimate. The impact of the WWII GI Bill on education, when only using variation across birth cohorts, is driven by states that did witness large changes in the WWII military participation rate, and thus the estimates would be imprecise if these states had trends in schooling that deviated from the national trend but are not controlled for in this specification. The states that do not witness a large change in WWII military participation rate contribute to creating noise in the estimation of the effects of the WWII GI Bill if they are included in the treatment group. Essentially, if there were no variation at the state level, using just the variation at the birth cohort level would be the best strategy. Given that there is variation at the state level, however, it is better to exploit this variation to obtain a more precise estimate of the effect of the GI Bill. The drawback of including variation across states is that states with high and low military participation rates could be different in other dimensions. The WWII variable could then be picking up these differences, which would bias the coefficient on the WWII variable. I include exhaustive controls to parse out any such differences, effectively eliminating potential bias. I include state of birth fixed effects that control for any level differences between states and state-specific time trends that control for any changes across birth cohorts in the differences between states. On controlling for level and time-varying differences between states, the variation captured in the WWII variable is arising solely from sudden changes in the WWII military participation rate for men born after 1927. Thus, the thought experiment is that, all else held constant, a state that has a 100 percent increase in WWII military participation had a 13-percentage point increase in the probability of high school graduation compared with a state that had a zero percent increase in WWII military participation. It is important to control for the trends in education to obtain unbiased estimates for the effect of the WWII GI Bill. Bound and Turner (2002) note that the changes in WWII participation across birth cohorts occur concurrently with changes in men's schooling across birth cohorts. Thus, without controlling for underlying trends in schooling, the WWII variable might just capture the changes in schooling attainment across birth cohorts and not really the effect of the GI Bill. In the main specification I use a quadratic polynomial in trend. Appendix table A.2 shows results from different polynomial specifications in trend. The table shows that the estimates stabilize after a quadratic specification in trend is included, thus a quadratic polynomial in trend adequately controls for concurrent changes in education and other characteristics across birth cohorts. A potential confounding factor is the participation of the younger birth cohorts of the control group in the Korean War. Veterans of the Korean War received education benefits as well. In addition, unlike WWII, the Korean War draft allowed education deferments. To control for the confounding effects from participation in the Korean War, the regressions control for the fraction of men in a birth cohort-state of birth cell who fought in the Korean War (but not in WWII). The Korean War variable is interacted with a time trend to allow differential effects of the Korean War across cohorts, with younger cohorts potentially affected more by the Korean War education deferments. I construct the Korean War variable following the strategy from Larsen et al. (2015). I include all men who say they served in the Korean War or at any other time without specifying any particular war. Larsen and his co-authors show that a significant fraction of the youngest cohorts answered questions on veteran status by saying that they were veterans at “any other time” but without specifying any particular war. They note that this was most likely because men from these cohorts started service post January 1951 but were still eligible for the education benefits of the Korean War GI Bill.10 ### Statistical Model The empirical analysis uses the following reduced-form regression equation: 1 Yicb is the outcome of interest for individual i in birth cohort c and state of birth b. The outcomes I examine are high school completion, employment, weeks worked, and poverty. %WWII is the percentage of men who participated in WWII in birth cohort c (birth year by birth quarter) and in state of birth b. %KOREA is the percentage of men who participated in the Korean War or in the military at any other time (but not in WWII) in birth cohort c (birth year by birth quarter) and in state of birth b, and this variable is further interacted with the trend variables. SB is a vector of dummy variables for state of birth, which is subsequently interacted with the trend variables. The trend variable is defined as (birth year − 1929 + [birth quarter/4]), following the specification by Bound and Turner (2002). As my sample consists of black and white men born in the United States, I include race fixed effects by the indicator variable, Black. α2 is the coefficient of interest, which is interpreted as the percentage point increase in the probability of the outcome caused by a 100 percent increase in the probability of serving in WWII. Finally, two elements of the research design are important to note. First, like all prior work, the empirical strategy cannot separate the effect of the education benefits under the GI Bill from the direct effect of military participation. Thus, I report the reduced form effects of the WWII GI Bill on the outcomes of interest, and these estimates are interpreted as the combined effect of education benefits under the GI Bill and military service in WWII.11 Second, the strength of the empirical strategy depends on the similarities in pre-war characteristics between the cohorts. Larsen et al. (2015) use Panel Study of Income Dynamics and Occupational Changes in a Generation datasets and show that pre-war characteristics like father's education, father's occupation, and whether they lived with both parents (among other characteristics) were similar across cohorts (they fail to reject that they were different). Their results show that the 1923–1927 and 1928–1938 cohorts are comparable in pre-war characteristics. ### Data Consistent with previous literature, I use three 1-percent samples from the 1970 Census for my analysis. The data consist of black and white men born in the United States in the years 1923–1938. I drop observations for which age, sex, race, or veteran status is imputed and observations for which state of birth is missing. Bound and Turner (2002) and Larsen et al. (2015) have used the 1970 Census, noting that the 1960 Census has a much smaller sample size than 1970, and that the latter gives all the cohorts sufficient time to complete education. They note that the 1980 Census shows higher schooling levels for the relevant cohorts, which seems to be due to reasons other than the GI Bill, such as differential mortality and over-reporting of education with age. For these reasons I also use the 1970 Census. Analysis with the 1960 and 1980 Census is discussed in section 4. The Census contains information on an individual's veteran status. It has information on the number of weeks worked in intervals last year12 and whether the individual's family fell below the poverty threshold. Summary statistics from the 1970 Census are shown in table 1 for the pooled sample of white and black men. There is a large difference in WWII military participation rate for cohorts born before and after 1928. The fraction of men who participated in WWII from the birth cohort 1923–1927 is high (0.74) whereas for the birth cohort 1928–1932 it is much lower (0.1). The opposite trend holds true for participation in the Korean War, with the latter birth cohorts having much higher participation rates. The fraction of men completing high school is slightly higher for the latter cohorts, whereas the fraction of men living below the poverty line and employed is similar for the birth cohorts 1923–1927 and 1928–1932. The number of weeks worked is also similar across the two cohorts. Figure 1 shows the graphical representation of the trend for WWII military participation and high school completion by birth cohort (year by quarter). As the solid curve depicts, national WWII military participation across different birth cohorts ranges from 0 to 0.8 and drops sharply at the birth cohort 1928. The dashed line shows that there is an upward trend in high school completion over time. Empirical analysis thus requires the inclusion of a time trend, as is specified in equation 1. Table 1. Summary Statistics for Men of Birth Cohorts 1923—1927 and 1928—1932 1923—19271928—1932 In World War II 0.742 0.102 In Korean War 0.053 0.559 College graduate 0.163 0.181 Completed high school 0.571 0.639 Below poverty line 0.058 0.064 Employed 0.964 0.971 Weeks worked 47.04 47.5 No. of observations 148,337 140,473 1923—19271928—1932 In World War II 0.742 0.102 In Korean War 0.053 0.559 College graduate 0.163 0.181 Completed high school 0.571 0.639 Below poverty line 0.058 0.064 Employed 0.964 0.971 Weeks worked 47.04 47.5 No. of observations 148,337 140,473 Notes: The statistics for all the variables, except number of weeks worked, are reported as fractions because this is how the corresponding variables are measured in the regressions. Source: 1970 Integrated Public Use Microdata Series (see Ruggles et al. 2010). Figure 1. Graphical Representation of WWII Participation Rate and High School Completion Rate for the Combined Sample of Black and White Men. Figure 1. Graphical Representation of WWII Participation Rate and High School Completion Rate for the Combined Sample of Black and White Men. ## 4. Results ### High School Completion Table 2 shows the effect of the WWII GI Bill on high school completion for the pooled sample of white and black men. The first column shows the estimates from the preferred time window of five years before and five years after the discontinuity point of 1928. Columns 2 and 3 use narrower time windows, and columns 4, 5, and 6 use wider time windows. The time windows are chosen by consecutively adding two more years to the control group. The dependent variable is a dummy variable capturing whether the individual completed high school. Table 2. Effect of the GI Bill on High School Completion for Men 1923—19321923—19281923—19301923—19341923—19361923—1938 Time Windows(1)(2)(3)(4)(5)(6) Dependent Variable: Dummy Variable = 1 for completing high school WWII 0.134*** 0.179*** 0.141*** 0.126*** 0.109*** 0.0677*** (0.0227) (0.0279) (0.0235) (0.0220) (0.0198) (0.0170) Korea 0.209*** 0.296*** 0.228*** 0.214*** 0.211*** 0.164*** (0.0254) (0.0448) (0.0309) (0.0247) (0.0223) (0.0188) N 288,810 177,149 233,986 341,677 394,946 449,729 R2 0.074 0.068 0.071 0.075 0.077 0.079 1923—19321923—19281923—19301923—19341923—19361923—1938 Time Windows(1)(2)(3)(4)(5)(6) Dependent Variable: Dummy Variable = 1 for completing high school WWII 0.134*** 0.179*** 0.141*** 0.126*** 0.109*** 0.0677*** (0.0227) (0.0279) (0.0235) (0.0220) (0.0198) (0.0170) Korea 0.209*** 0.296*** 0.228*** 0.214*** 0.211*** 0.164*** (0.0254) (0.0448) (0.0309) (0.0247) (0.0223) (0.0188) N 288,810 177,149 233,986 341,677 394,946 449,729 R2 0.074 0.068 0.071 0.075 0.077 0.079 Notes: The table shows the effect of the WWII GI Bill on the probability of high school completion for males. The independent variable of interest is the fraction of men who participated in WWII. The dependent variable is a dummy = 1 if the person completed high school. The first column with a symmetrical ten-year window around the discontinuity point (end of 1927) gives the preferred estimates. The sample consists of black and white men born in the United States between 1923 and 1938. WWII is the fraction of men who participated in WWII in a birth year-quarter by state of birth cell. Similarly, Korea is the fraction of men who participated in the Korean War or at “any other time” but did not serve in WWII, in a birth year-quarter by state of birth cell. Standard errors are corrected for heteroskedasticity and are clustered at the level of birth year-quarter by state of birth. Controls include trend which is defined as (birth year − 1929 + [quarter of birth/4]), trend squared, “Korea” interacted with trend variables, state of birth dummies, and their interactions with trend variables. ***Significant at the 1% level. Source: 1970 Integrated Public Use Microdata Series (see Ruggles et al. 2010). Table 2, column 1, shows that the WWII GI Bill increased the probability of completing high school by 13 percentage points for men. The estimates are similar in other time windows, as shown in columns 2 to 6. Because the estimates are similar across different time windows, the regression model fits the data well in the preferred time window of 1923–1932. The results demonstrate that the WWII GI Bill was effective in increasing the attainment of complete secondary education. Conversely, it was effective in reducing the prevalence of low human capital accumulation for the cohorts that benefited from it. Because this is a new specification, I also replicate the findings of earlier studies that have established the significant effect of the WWII GI Bill on college education and find similar effects. Table A.1 in the appendix reports the effect on college education for the pooled sample of white and black men. The results show that the WWII GI Bill increased years of college education by 0.4 years. This result is similar to that of Larsen et al. (2015), who obtain an estimate of 0.4 years for white men, and Turner and Bound (2003), who obtain an estimate of 0.3 years for both black and white men. The results on college graduation show that the probability of graduating from college increased by 8 percentage points. The results on the acquisition of some college education show that the probability of attending some college increased by 13 percentage points. The previous studies have used variation at the birth year-quarter level while my specification uses variation at the birth year-quarter-state level, thus also exploiting the variation obtained from differences in WWII mobilization rate across states. The GI Bill could have moved people into more education attainment in three ways: from below high school to just high school, from high school to more than high school, and from below high school to more than high school. I attempt to disentangle these flows by breaking the “High School Completion” outcome into six categories: less than high school, only high school, one year, two years, three years, and four or more years of college. If a significantly larger fraction of people were at the high school level of education, then more people were moved into this level than were moved out for college. On the other hand, if a significantly smaller fraction of people were at the high school level of education then more people were moved out for college than were moved in from below high school. However, if there was no significant change in the fraction of people at the high school level then an equal number of people could have been moved in and out of this level, and at the same time people could have been moved from below high school to college. Thus, all three channels above could have been in operation in net, resulting in no change in the fraction who remain at the high school level. Table 3 presents the results for the six categories. As the results show, I find no significant change in the fraction of people at just high school level of education. Most of the decrease in the fraction of people below high school level of education seems to be made up by increases in the fraction of people at two years and four or more years of college. These results imply that the GI Bill did not just move people from high school level to college level, but also moved people from below high school to at least high school and possibly more. While it is not possible to determine exactly how people transitioned across the high school threshold, that is, the extent to which people below high school moved to just high school or beyond high school, the results do imply that the transition from below high school to high school was balanced by the transition from high school to college. In fact, table 2 shows that below high school level of education decreased by 13 percentage points and appendix table A.1 shows that some college level of education increased by 13 percentage points, implying that on net there was no change in the percentage of people with exactly high school level of education (which holds true in the data). Considering that roughly equal percentages of the WWII veterans were high school dropouts and high school graduates without college degree when they left the war, it is reasonable to find similar increases in education at both the high school level and at the college level, plausibly the former often accompanied by some form of training. Table 3. Effects of the WWII GI Bill on Only High School Completion Time Window: 1923—1932 Less ThanOnly High4 or More DependentHighSchool1 Year of2 Years of3 Years ofYears of VariableSchoolCompletionCollegeCollegeCollegeCollege Explanatory variables WW II −0.134*** −0.000381 0.00557 0.0373*** 0.00526 0.0860*** (0.0227) (0.0219) (0.00964) (0.0114) (0.00656) (0.0167) Korea −0.209*** 0.112*** 0.00604 0.0289** 0.00168 0.0609*** (0.0254) (0.0254) (0.0111) (0.0134) (0.00768) (0.0199) R2 0.074 0.021 0.003 0.004 0.002 0.024 Observations 288,810 288,810 288,810 288,810 288,810 288,810 Time Window: 1923—1932 Less ThanOnly High4 or More DependentHighSchool1 Year of2 Years of3 Years ofYears of VariableSchoolCompletionCollegeCollegeCollegeCollege Explanatory variables WW II −0.134*** −0.000381 0.00557 0.0373*** 0.00526 0.0860*** (0.0227) (0.0219) (0.00964) (0.0114) (0.00656) (0.0167) Korea −0.209*** 0.112*** 0.00604 0.0289** 0.00168 0.0609*** (0.0254) (0.0254) (0.0111) (0.0134) (0.00768) (0.0199) R2 0.074 0.021 0.003 0.004 0.002 0.024 Observations 288,810 288,810 288,810 288,810 288,810 288,810 Notes: The table shows the effect of the WWII GI Bill on less than high school completion, only high school completion, 1 year, 2 years, 3 years, and 4 or more years of college for men. The independent variable of interest is the fraction of men who participated in WWII. The dependent variables are the dummy variables = 1 if not completed high school, dummy = 1 if completed only high school, dummy = 1 if completed up to 1 year of college, and so on till up to 4 or more years of college. The sample consists of black and white men born in the United States between 1923 and 1932. WWII is the fraction of men who participated in World War II in a birth year-quarter by state of birth cell. Similarly, Korea is the fraction of men who participated in the Korean War or at “any other time” but did not serve in WWII, in a birth year-quarter by state of birth cell. Standard errors are corrected for heteroskedasticity and are clustered at the level of birth year-quarter by state of birth. Controls include trend which is defined as (birth year − 1929 + [birth quarter/4]), trend squared, “Korea” interacted with trend variables, state of birth dummies, and their interactions with trend variables. **Significant at the 5% level; ***significant at the 1% level. Source: 1970 Integrated Public Use Microdata Series (see Ruggles et al. 2010). The result of a 13-percentage point increase in the probability of high school completion is a fairly large effect of the GI Bill program. Note that 46.8 percent of veterans had not completed high school when they returned from the war (Hammond 1980). Over 7 million men used the GI Bill education benefits and almost half of these beneficiaries used the benefits for below college level education (Bowman 1973). Thus, the GI Bill targeted a vast number of high school dropouts and had the potential to generate a substantial improvement at the high school level of education. A small fraction of the beneficiaries (2.4 percent) was recorded as having used the GI Bill education benefits solely for high school education. Therefore, it is likely that many of the beneficiaries combined high school completion with some sort of training and apprenticeship. Table 2 also shows positive and significant coefficients for the Korean War variable, implying that the Korean War GI Bill also significantly increased high school completion. Because the Korean War allowed education deferments, however, it is not as straightforward to assign the effect to the Korean War GI Bill, as the coefficients also capture the effect of the incentives to pursue education to avoid being enlisted in the war. ### Poverty and Employment Previous studies have shown that education can play an important role in avoiding poverty and obtaining employment.13 By increasing overall education attainment, the WWII GI Bill could have had a positive impact in securing the economic future of the cohorts for whom it benefited. Such an impact is even more conceivable because of the bill's positive effect on low education levels than if it had only affected higher education levels. People at low human capital levels (less than high school completion) are more susceptible to poverty and unemployment than people at higher human capital levels. Whereas college education would enable people to move into higher income brackets, high school completion would play the major role in moving people across the threshold of poverty. Similarly, whereas college education would enable people to move into higher paying jobs and occupations, high school completion would play the major role in moving people into employment. Understanding the impact of the GI Bill on economic outcomes such as poverty and employment is particularly important because the GI Bill was intended to ease veterans’ transition into civilian life by providing them with financial benefits to help them become economically independent and secure. Around 16 million veterans returned from the war and in the process of the economy trying to absorb these many men into the workforce, several veterans would potentially have had a very difficult time starting their careers, especially if they had acquired health problems during service. In a way, the outcomes of poverty and employment are better tests of the success of the GI Bill than the outcomes of education, as the latter was perceived as the channel through which veterans would acquire economically stable lives. Table 4 shows the effect of the WWII GI Bill on poverty and employment for the pooled sample of white and black men. Column 1 gives the estimates in the preferred time window. The estimates are consistent across the various time windows. The results show that the WWII GI Bill reduced the probability of being poor by 4 percentage points, increased the probability of employment by 3 percentage points, and increased the number of weeks worked by two. Thus, the WWII GI Bill was effective in providing the benefited cohort of men with greater economic well-being, by reducing the incidence of poverty and increasing their employment levels. These effects are quite large when compared to the base means of these outcomes, as shown in table 1. For instance, the base mean for poverty is around 6 percent and the WWII GI Bill reduced the probability of poverty to 2 percent. More than 7 million veterans utilized the WWII GI Bill education benefits and almost half of them used the benefits to access below college level education and training. These statistics suggest that many of the large number of GI Bill beneficiaries were those who would have faced a nontrivial chance of poverty. The GI Bill, by reaching the lower half of the education distribution in large numbers, was able to substantially reduce the probability of severe economic outcomes, such as poverty and unemployment. Table 4. Effect of GI Bill on Poverty and Employment for Men 1923—19321923—19281923—19301923—19341923—19361923—1938 Time Windows:(1)(2)(3)(4)(5)(6) Panel A Dependent Variable: Dummy Variable = 1 if family is below poverty line WW II −0.0407*** −0.0555*** −0.0406*** −0.0473*** −0.0458*** −0.0359*** (0.0128) (0.0159) (0.0130) (0.0126) (0.0114) (0.00954) Korea −0.0535*** −0.0752*** −0.0450*** −0.0573*** −0.0568*** −0.0421*** (0.0147) (0.0242) (0.0168) (0.0140) (0.0126) (0.0106) R2 0.035 0.036 0.036 0.034 0.034 0.033 Panel B Dependent Variable: Dummy Variable = 1 if employed last year WW II 0.0284*** 0.0262** 0.0309*** 0.0302*** 0.0274*** 0.0210*** (0.00876) (0.0110) (0.00910) (0.00846) (0.00763) (0.00649) Korea 0.0329*** 0.0295* 0.0377*** 0.0393*** 0.0381*** 0.0265*** (0.00958) (0.0173) (0.0112) (0.00934) (0.00838) (0.00701) R2 0.010 0.010 0.010 0.010 0.010 0.010 Panel C Dependent Variable: Number of weeks worked WW II 2.012*** 1.883*** 2.146*** 2.068*** 1.927*** 1.341*** (0.545) (0.685) (0.567) (0.528) (0.478) (0.411) Korea 2.238*** 1.962* 2.571*** 2.569*** 2.506*** 1.607*** (0.608) (1.097) (0.712) (0.584) (0.525) (0.444) R2 0.018 0.017 0.018 0.018 0.018 0.018 N 288,810 177,149 233,986 341,677 394,946 449,729 1923—19321923—19281923—19301923—19341923—19361923—1938 Time Windows:(1)(2)(3)(4)(5)(6) Panel A Dependent Variable: Dummy Variable = 1 if family is below poverty line WW II −0.0407*** −0.0555*** −0.0406*** −0.0473*** −0.0458*** −0.0359*** (0.0128) (0.0159) (0.0130) (0.0126) (0.0114) (0.00954) Korea −0.0535*** −0.0752*** −0.0450*** −0.0573*** −0.0568*** −0.0421*** (0.0147) (0.0242) (0.0168) (0.0140) (0.0126) (0.0106) R2 0.035 0.036 0.036 0.034 0.034 0.033 Panel B Dependent Variable: Dummy Variable = 1 if employed last year WW II 0.0284*** 0.0262** 0.0309*** 0.0302*** 0.0274*** 0.0210*** (0.00876) (0.0110) (0.00910) (0.00846) (0.00763) (0.00649) Korea 0.0329*** 0.0295* 0.0377*** 0.0393*** 0.0381*** 0.0265*** (0.00958) (0.0173) (0.0112) (0.00934) (0.00838) (0.00701) R2 0.010 0.010 0.010 0.010 0.010 0.010 Panel C Dependent Variable: Number of weeks worked WW II 2.012*** 1.883*** 2.146*** 2.068*** 1.927*** 1.341*** (0.545) (0.685) (0.567) (0.528) (0.478) (0.411) Korea 2.238*** 1.962* 2.571*** 2.569*** 2.506*** 1.607*** (0.608) (1.097) (0.712) (0.584) (0.525) (0.444) R2 0.018 0.017 0.018 0.018 0.018 0.018 N 288,810 177,149 233,986 341,677 394,946 449,729 Notes: The table shows the effect of the WWII GI Bill on poverty and employment for men. The independent variable of interest is the fraction of men who participated in WWII. The dependent variables are the dummy variables = 1 for family living under poverty line, dummy = 1 if worked positive number of weeks last year and number of weeks worked last year. The first column with a symmetrical ten-year window around the discontinuity point (end of 1927) gives the preferred estimates. The sample consists of black and white men born in the United States between 1923 and 1938. WWII is the fraction of men who participated in WWII in a birth year-quarter by state of birth cell. Similarly, Korea is the fraction of men who participated in the Korean War or at “any other time” but did not serve in WWII, in a birth year-quarter by state of birth cell. Standard errors are corrected for heteroskedasticity and are clustered at the level of birth year-quarter by state of birth. Controls include trend which is defined as (birth year − 1929 + [birth quarter/4]), trend squared, “Korea” interacted with trend variables, state of birth dummies, and their interactions with trend variables. *Significant at the 10% level; **significant at the 5% level; ***significant at the 1% level. Source: 1970 Integrated Public Use Microdata Series (see Ruggles et al. 2010). It is important to note that the GI Bill also provided an unemployment allowance of$20/week up to 52 weeks. This could potentially have affected veterans’ long-term employment outcomes. However, Veterans Administration records state that fewer than 20 percent of the unemployment benefits were claimed and only one out of nineteen veterans used the benefits for the full 52 weeks. Because the unemployment benefits were not as widely used as the education benefits (approximately 50 percent of veterans used the education benefits), the positive effects on employment must, to a large extent, be driven by increased human capital accumulation.

As noted earlier in the Empirical Strategy section, the coefficients capture the effect of both military participation and the education benefits of the GI Bill. It is important to keep in mind, however, that the outcomes of employment and poverty might be influenced to a larger extent by military participation than the outcome of high school completion. Bedard and Deschenes (2006) find that WWII military participation worsened health conditions for veterans. Post-traumatic stress disorder is also a common ailment among veterans. Health problems caused by military participation might adversely affect the employment outcomes of veterans. On the other hand, WWII veterans were famously regarded as heroes when they returned home. This positive public sentiment toward veterans might have favorably affected the employment outcomes of veterans. Because it is not possible to isolate all these channels, the coefficients in table 4 show the combined effect of the GI Bill and the possible positive or negative effects of military participation.

As with high school completion, the Korean War variable also shows similar effects on the outcomes of poverty and employment. Keeping in mind the previously described reservations on interpreting the Korean War variable, it is interesting to find that the Korean War shows similar results as WWII.

### Robustness Checks

To further check the validity of the previous results, I perform a number of robustness analyses. These results are presented in tables 5 through 8. First, I try to address the fact that the effect from the GI Bill cannot be separated from the effect of military participation. Although World War I (WWI) was a very different war, it is the closest event to WWII that did not have its own GI Bill. I estimate in table 5 the effects of military participation in WWI for men born between 1891 and 1902. The 1930 Census is a 5 percent sample with information on veteran status for WWI, although it does not have information on completed level of education and employment. I use the 1 percent sample from the 1940 Census for high school completion and employment and match veteran information from the 1930 Census. I do not have information on poverty status from either census. An advantage of the 1940 Census is that the number of weeks worked is coded as a continuous variable as opposed to intervals as in the 1970 Census. The 1930 and 1940 Censuses do not provide quarter of birth information and, thus, I am restricted to using birth-year by state of birth as the level of variation for WWI participation rates. I do not use the 1950 Census because it occurred after WWII. The results show that military participation lowers the probability of employment and has an insignificant effect on high school completion and number of weeks worked. Thus, the GI Bill overrode the negative effect of military participation on employment and caused an increase in the probability of employment.14

Table 5.
Analysis with WWI: Effect of Military Participation
Time Window: 1891—1902
Dependent VariableHigh School CompletionEmployed Last YearWeeks Worked Last Year
Explanatory variables
WWI 0.0337 −0.0412** −0.465
(0.0280) (0.0192) (1.251)
R2 0.034 0.005 0.011
N 87,390 87,390 87,390
Time Window: 1891—1902
Dependent VariableHigh School CompletionEmployed Last YearWeeks Worked Last Year
Explanatory variables
WWI 0.0337 −0.0412** −0.465
(0.0280) (0.0192) (1.251)
R2 0.034 0.005 0.011
N 87,390 87,390 87,390

Notes: The table shows the effect of WWI on high school completion and employment for men. The independent variable of interest is the fraction of men who participated in WWI. The dependent variables are dummy = 1 for having completed high school, dummy = 1 if worked positive number of weeks last year, and number of weeks worked last year. The sample consists of white and black men born in the United States between 1891 and 1902. WWI is the fraction of men who participated in WWI in a birth year by state of birth cell. WWI is calculated from the 1930 Census and matched to the 1940 Census. The regressions for high school completion, employment, and number of weeks worked are run on the 1940 Census. Standard errors are corrected for heteroskedasticity and are clustered at the level of birth year by state of birth. Controls include trend, which is defined as year of birth-1890, trend squared, state of birth dummies and their interactions with trend variables.

**Significant at the 5% level.

Source: 1930 and 1940 Integrated Public Use Microdata Series (see Ruggles et al. 2010).

Next, I test the robustness of my results to a variety of specifications. These estimates are presented in table 6. For reference, column 1 reproduces the main results from the above analysis for the preferred time window, 1923–1932. Columns 2–8 present the analysis with different specifications in this preferred time window.

Table 6.
Robustness Checks with Different Specifications
Larsen et al. Specification:Bound and Turner
Data aggregatedData aggregated to birthSpecification: Data aggregated
Main Specification:to birthyear-quarter with birthto birth year-quarter with
Individual level datayear-quarter-stateKorean War Variableyear-quarter level ofbirth year-quarter level of
with birthwith birthIndividual level datanot includingvariation and Korean Warvariation and Korean WarData from 1960Data from 1980
year-quarter-stateyear-quarter-statewith birth year-quarterparticipation at “anyincluding participationnot including participationCensus: MainCensus: Main
level of variationalevel of variationblevel of variationcother time”dat “any other time”eat “any other time”fspecificationgspecificationh
(1)(2)(3)(4)(5)(6)(7)(8)
Time Window1923—19321923—19321923—19321923—19321923—19321923—19321923—19321923—1932
Panel A
Dependent Variable: Dummy Variable = 1 for completing high school
WWII 0.134*** 0.149*** 0.348*** 0.135*** 0.405*** 0.214* 0.149*** 0.148***
(0.0227) (0.0243) (0.0962) (0.0224) (0.117) (0.116) (0.0274) (0.0197)
R2 0.074 0.912 0.074 0.074 0.959 0.953 0.073 0.077
Panel B
Dependent Variable: Dummy Variable = 1 if family if below poverty line
WWII −0.0407*** −0.0488*** −0.127*** −0.0368*** −0.150** −0.115*** −0.0960*** −0.0343***
(0.0128) (0.0135) (0.0470) (0.0124) (0.0594) (0.0379) (0.0192) (0.00939)
R2 0.035 0.741 0.035 0.035 0.554 0.552 0.078 0.025
Panel C
Dependent Variable: Dummy Variable = 1 if employed last year
WWII 0.0284*** 0.0322*** 0.0757** 0.0242*** 0.0876** 0.0637*** 0.0370*** 0.0454***
(0.00876) (0.00915) (0.0292) (0.00860) (0.0338) (0.0219) (0.00834) (0.0127)
R2 0.010 0.422 0.010 0.010 0.845 0.838 0.012 0.024
Panel D
Dependent Variable: Number of weeks worked
WWII 2.012*** 2.318*** 4.402** 1.697*** 5.375** 3.721*** 2.417*** 3.195***
(0.545) (0.570) (1.754) (0.529) (2.043) (1.325) (0.587) (0.676)
R2 0.018 0.547 0.018 0.018 0.869 0.861 0.031 0.028
N 288,810 2,030 288,810 288,810 40 40 104,416 778,862
Larsen et al. Specification:Bound and Turner
Data aggregatedData aggregated to birthSpecification: Data aggregated
Main Specification:to birthyear-quarter with birthto birth year-quarter with
Individual level datayear-quarter-stateKorean War Variableyear-quarter level ofbirth year-quarter level of
with birthwith birthIndividual level datanot includingvariation and Korean Warvariation and Korean WarData from 1960Data from 1980
year-quarter-stateyear-quarter-statewith birth year-quarterparticipation at “anyincluding participationnot including participationCensus: MainCensus: Main
level of variationalevel of variationblevel of variationcother time”dat “any other time”eat “any other time”fspecificationgspecificationh
(1)(2)(3)(4)(5)(6)(7)(8)
Time Window1923—19321923—19321923—19321923—19321923—19321923—19321923—19321923—1932
Panel A
Dependent Variable: Dummy Variable = 1 for completing high school
WWII 0.134*** 0.149*** 0.348*** 0.135*** 0.405*** 0.214* 0.149*** 0.148***
(0.0227) (0.0243) (0.0962) (0.0224) (0.117) (0.116) (0.0274) (0.0197)
R2 0.074 0.912 0.074 0.074 0.959 0.953 0.073 0.077
Panel B
Dependent Variable: Dummy Variable = 1 if family if below poverty line
WWII −0.0407*** −0.0488*** −0.127*** −0.0368*** −0.150** −0.115*** −0.0960*** −0.0343***
(0.0128) (0.0135) (0.0470) (0.0124) (0.0594) (0.0379) (0.0192) (0.00939)
R2 0.035 0.741 0.035 0.035 0.554 0.552 0.078 0.025
Panel C
Dependent Variable: Dummy Variable = 1 if employed last year
WWII 0.0284*** 0.0322*** 0.0757** 0.0242*** 0.0876** 0.0637*** 0.0370*** 0.0454***
(0.00876) (0.00915) (0.0292) (0.00860) (0.0338) (0.0219) (0.00834) (0.0127)
R2 0.010 0.422 0.010 0.010 0.845 0.838 0.012 0.024
Panel D
Dependent Variable: Number of weeks worked
WWII 2.012*** 2.318*** 4.402** 1.697*** 5.375** 3.721*** 2.417*** 3.195***
(0.545) (0.570) (1.754) (0.529) (2.043) (1.325) (0.587) (0.676)
R2 0.018 0.547 0.018 0.018 0.869 0.861 0.031 0.028
N 288,810 2,030 288,810 288,810 40 40 104,416 778,862

Notes:aResults from the preferred specification in this paper.

bSpecification aggregates data to the level of birth year-birth quarter-state of birth and uses number of observations in each of these cells as weights.

cSpecification uses level of variation at birth year-birth quarter.

dSpecification uses the Korean War variable definition that does not include participation in the military at “any other time.” The preferred specification in column 1 includes this in the Korean War variable definition.

eSpecification is from Larsen et al. (2015), data are aggregated to the level of birth year-quarter, the level of variation is birth year-quarter and Korean War includes participation at “any other time.”

fSpecification is from Bound and Turner (2002), data are aggregated to the level of birth year-quarter, the level of variation is birth year-quarter and Korean War does not include participation at “any other time.”

gRegression replicates the preferred specification in column 1 using the 1960 Census.

hRegression replicates the preferred specification in column 1 using the 1980 Census.

*Significant at the 10% level; **significant at the 5% level; ***significant at the 1% level.

Previous studies on the effect of the WWII GI Bill have used aggregated data. This strategy should provide similar results as the analysis with individual level data if the standard errors have been properly corrected. I check the robustness of the results by aggregating the data to birth year-quarter-state level, as this is the level of variation used in the preferred specification. To make the estimates consistent, these use number of observations in each birth year-quarter-state cell as weights. Column 2 shows that the results are similar to those obtained in the main specification of column 1.

Previous studies have used birth year-quarter as the level of variation for participation in WWII. The preferred specification of this paper adds state of birth as another level of variation for participation in WWII. I check the robustness of the results when this additional level of variation is not included. Column 3 presents the results using birth year-quarter as the level of variation for WWII GI Bill eligibility with individual level data. The results show significant but larger estimates for the effect of WWII GI Bill as compared to estimates in column 1. Thus, the effect of the WWII GI Bill holds with and without the use of state of birth as a source of variation for WWII participation.

When controlling for the confounding effects of the Korean War, Bound and Turner (2002) and Larsen et al. (2015) use different definitions for the Korean War variable. Bound and Turner (2002) define the variable as the fraction who participated in Korean War but not in WWII, whereas Larsen et al. (2015) define the variable as the fraction who participated in Korean War or the military at any other time but not in WWII. The preferred specification of this paper uses the definition by Larsen et al. (2015) for the reasons mentioned in section 3. I check the robustness of the results with the definition used in Bound and Turner (2002). Column 4 presents the preferred specification with the only difference that the Korean War control variable is defined as the fraction who participated in Korean War but not in WWII. The results show estimates similar to those in column 1.

I further check the robustness of the results to the choice of identifying variation. I use the same specifications that were used in previous studies. Column 5 presents the specification used in Larsen et al. (2015). This specification aggregates the data to birth year-quarter cells, uses variation in WWII participation at the birth year-quarter level, and defines the Korean War variable as the fraction who participated in Korean War or the military at any other time but not in WWII. The number of observations in each birth year-quarter cell is used as weights in the regressions. The results give significant but larger estimates for the effect of the WWII GI Bill as compared to the estimates obtained in the main specification of Column 1. Column 6 presents the specification used in Bound and Turner (2002). This specification aggregates the data to birth year-quarter cells, uses variation in WWII participation at the birth year-quarter level, and defines the Korean War variable as the fraction who participated in Korean War but not in WWII. The number of observations in each birth year-quarter cell is used as weights in the regressions. These results also give significant and slightly larger estimates for the effect of the WWII GI Bill as obtained in the main specification of column 1.

The 1970 Census is the preferred dataset to study the effects of the WWII GI Bill for reasons mentioned above. I also check the robustness of the results by using the 1960 and 1980 Census. Columns 7 and 8 estimate the preferred specification using the 1960 and 1980 Censuses, respectively. The results in the two columns give similar estimates as those in Column 1, obtained from the 1970 Census.

To check whether trends other than military participation and GI Bill eligibility might be driving the results, I study the impact on the outcome variables for women. Only 3 percent of women volunteered to participate in WWII and the GI Bill take-up rates were lower for women than for men (Larsen et al. 2015). A significant difference between women of birth cohorts 1923–1927 and women of birth cohorts 1928–1932 for high school completion, poverty, and employment would suggest that something other than military participation and GI Bill eligibility might be driving the results for men. I match male WWII military participation rates to women at the birth year-quarter-state level. Table 7 shows that there was no significant impact on any of the outcome variables for women. Thus, the significant results obtained for men are the causal effects of the GI Bill education benefits and WWII military participation.

Table 7.
Effects of the WWII GI Bill on Women
Time Window: 1923—1932
High SchoolBelow PovertyEmployed LastWeeks Worked
Dependent VariableCompletionLineYearLast Year
Explanatory variables
WWII 0.0218 −0.0174 −0.00501 −0.433
(0.0228) (0.0139) (0.0241) (1.111)
Korea 0.0550** −0.0221 0.00563 −0.164
(0.0253) (0.0160) (0.0266) (1.234)
R2 0.083 0.069 0.008 0.009
N 309,375 309,375 309,375 309,375
Time Window: 1923—1932
High SchoolBelow PovertyEmployed LastWeeks Worked
Dependent VariableCompletionLineYearLast Year
Explanatory variables
WWII 0.0218 −0.0174 −0.00501 −0.433
(0.0228) (0.0139) (0.0241) (1.111)
Korea 0.0550** −0.0221 0.00563 −0.164
(0.0253) (0.0160) (0.0266) (1.234)
R2 0.083 0.069 0.008 0.009
N 309,375 309,375 309,375 309,375

Notes: The table shows the effect of the WWII GI Bill on high school completion, poverty, and employment for women. Male WWII military participation rates from the 1970 Census are matched to women at the birth year-quarter-state level. The independent variable of interest is the fraction of men who participated in WWII. The dependent variables are dummy = 1 for high school completion, dummy = 1 for family living under poverty line, dummy = 1 if worked positive number of weeks last year, and number of weeks worked last year. The sample consists of black and white women born in the United States between 1923 and 1932 (the preferred time window). WWII is the fraction of men who participated in WWII in a birth year-quarter by state of birth cell. Similarly, Korea is the fraction of men who participated in the Korean War or at “any other time” but did not participate in WWII, in a birth year-quarter by state of birth cell. Standard errors are corrected for heteroskedasticity and are clustered at the level of birth year-quarter by state of birth. Controls include trend which is defined as (birth year − 1929 + [birth quarter/4]), trend squared, “Korea” interacted with trend variables, state of birth dummies, and their interactions with trend variables.

Source: 1970 Integrated Public Use Microdata Series (see Ruggles et al. 2010).

Lastly, I check the robustness of the results to different trend specifications and time windows. The main results in tables 2 and 4 present the impacts for various time windows. As the tables show, the results for the four outcomes of interest are consistent across these time windows. The magnitudes become smaller in size in wider time windows. This is to be expected as wider time windows include additional younger cohorts, which witnessed other influences that increased high school completion, such as the Korean War, and are increasingly dissimilar to the treated cohort. Appendix table A.2 shows the results for specifications with no trend variables, and linear, quadratic, and cubic polynomials of the trend variable in the preferred time window of 1923–1932. It is important to include a trend control in order to isolate the WWII variable from concurrent changes in schooling and other characteristics across birth cohorts. The results are robust across the various trend specifications. The linear specification gives smaller magnitudes for the effects. This is likely due to an upward trend in high school completion rate, which is more wholly controlled for by higher-order polynomials.

### The Effect of the WWII GI Bill across Race

There are two reasons why the GI Bill may have differentially impacted white and black men. First, for any given birth cohort, a higher percentage of white men fought in the war than black men. Thus, a larger fraction of the white male population was eligible for the benefits than of the black male population. Second, Turner and Bound (2003) find that black veterans benefited less in gaining college education than white veterans in southern states because of the segregation barriers present in educational institutes. These segregation barriers could have been relevant for high school completion as well, though to a lesser extent than for college education. Through these channels, the benefits of the GI Bill could have been sufficiently biased toward whites to exacerbate existing inequalities between blacks and whites in education, poverty, and employment. It is also possible, however, that the GI Bill benefited blacks more at the secondary education level because there were more blacks dropping out of school to begin with. I analyze my results by race in southern and non-southern states to determine any differences in the impact of the bill on high school completion, poverty, and employment.

Table 8 presents the results from regressions that interact the WWII variable with the indicator variable for being black. The analysis is done separately for southern and non-southern states. The results show that black men have a significantly lower probability of high school completion and employment, higher probability of poverty, and work for fewer weeks as compared to white men. The impacts of the bill on high school completion and poverty are larger in magnitude for the southern states. The interaction term of the WWII variable and the indicator for black race is not statistically significant for any of the outcomes in both southern and non-southern states. Thus, the WWII GI Bill had similar effects across race (black and white) for the outcomes of high school completion, poverty, and employment.

Table 8.
Comparing the Effects of the WWII GI Bill on Black Men and White Men
Time Window: 1923—1932
High SchoolBelow PovertyEmployed LastWeeks Worked
Dependent VariableCompletionLineYearLast Year
Panel A: Southern States
Explanatory variables
Black −0.221*** 0.129*** −0.0448*** −4.028***
(0.00821) (0.00532) (0.00321) (0.193)
WWII 0.176*** −0.0531** 0.0308* 1.909*
(0.0362) (0.0236) (0.0164) (0.977)
Black*WWII 0.00997 0.00139 −0.00137 0.289
(0.0163) (0.0114) (0.00712) (0.427)
R2 0.057 0.033 0.010 0.018
N 99,123 99,123 99,123 99,123
Panel B: Non-Southern States
Explanatory variables
Black −0.184*** 0.0653*** −0.0743*** −5.281***
(0.0109) (0.00690) (0.00626) (0.367)
WWII 0.0959*** −0.0294** 0.0264*** 1.960***
(0.0294) (0.0143) (0.00980) (0.636)
Black*WWII −0.0304 −0.00645 0.0124 0.600
(0.0189) (0.0118) (0.0113) (0.626)
R2 0.022 0.008 0.007 0.010
N 189,687 189,687 189,687 189,687
Time Window: 1923—1932
High SchoolBelow PovertyEmployed LastWeeks Worked
Dependent VariableCompletionLineYearLast Year
Panel A: Southern States
Explanatory variables
Black −0.221*** 0.129*** −0.0448*** −4.028***
(0.00821) (0.00532) (0.00321) (0.193)
WWII 0.176*** −0.0531** 0.0308* 1.909*
(0.0362) (0.0236) (0.0164) (0.977)
Black*WWII 0.00997 0.00139 −0.00137 0.289
(0.0163) (0.0114) (0.00712) (0.427)
R2 0.057 0.033 0.010 0.018
N 99,123 99,123 99,123 99,123
Panel B: Non-Southern States
Explanatory variables
Black −0.184*** 0.0653*** −0.0743*** −5.281***
(0.0109) (0.00690) (0.00626) (0.367)
WWII 0.0959*** −0.0294** 0.0264*** 1.960***
(0.0294) (0.0143) (0.00980) (0.636)
Black*WWII −0.0304 −0.00645 0.0124 0.600
(0.0189) (0.0118) (0.0113) (0.626)
R2 0.022 0.008 0.007 0.010
N 189,687 189,687 189,687 189,687

Notes: The table shows the differential effect of the WWII GI Bill on high school completion, poverty, and employment for black and white men. The analysis is done separately for southern and non-southern states. The regressions interact the WWII variable with the dummy variable for being black. The independent variables of interest are the fraction of men who participated in WWII and the interaction of this variable with the dummy variable for being black. The dependent variables are dummy = 1 for high school completion, dummy = 1 for family living under poverty line, dummy = 1 if worked positive number of weeks last year and number of weeks worked last year. The sample consists of black and white men born in the United States between 1923 and 1932, the preferred time window. WWII is the fraction of men who participated in World War II in a birth year-quarter by state of birth cell. Standard errors are corrected for heteroskedasticity and are clustered at the level of birth year-quarter-state of birth. Controls include “Korea” which is defined as the fraction of men who participated in the Korean War or at “any other time” but did not participate in WWII, in a birth year-quarter by state of birth cell, trend which is defined as (birth year − 1929 + [birth quarter/4]), trend squared, “Korea” interacted with trend variables, state of birth dummies, and their interactions with trend variables.

*Significant at the 10% level; **significant at the 5% level; ***significant at the 1% level.

Source: 1970 Integrated Public Use Microdata Series (see Ruggles et al. 2010).

These results are in contrast to the findings of Turner and Bound (2003), which show that black men in southern states benefitted significantly less in college education because of segregation barriers. The segregation barriers would have been much more significant at the college level than at the high school level, and black veterans could have potentially returned to the school in which they were enrolled before they became high school dropouts. Overall, the segregation barriers might not have been strong enough at the high school level to produce significantly different effects for black and white men. There are also no significant differences between whites and blacks for the outcomes of poverty and employment. Because these outcomes respond more to high school level of education, and the difference in poverty rates and unemployment rates is the largest between high school dropouts and high school graduates rather than between high school graduates and those with some college education, it is reasonable to find similar effects for these outcomes given the similar effects for high school completion.

## 5.  Conclusion

The WWII GI Bill had many significant effects on the cohorts of men for whom it benefited. Apart from increasing higher education, the WWII GI Bill also increased the probability of high school completion, decreased the probability of being in poverty, and increased employment. By shifting people out of low levels of education, it had effects at both the upper end and lower end of the education distribution. I find that the WWII GI Bill increased the probability of completing high school by 13 percentage points, decreased the probability of being poor by 4 percentage points, increased the probability of being employed by 3 percentage points, and increased the number of weeks worked by two for the pooled sample of black and white men.

The findings show that this massive financial aid program for pursuing education allowed people to move themselves out of the bleaker economic future of being a high school dropout and avoid the poverty and unemployment that can result from inadequate human capital accumulation. In this way, the GI Bill not only benefited the educationally more promising veterans by making college education affordable and attractive but also benefited those who in the absence of the GI Bill would not have completed even high school and who would have been economically less secure in life.

## Acknowledgments

I am grateful to Leigh Linden, Sandra E. Black, Stephen J. Trejo, Gerald S. Oettinger, Marika Cabral, Dayanand Manoli, and seminar participants in the Department of Economics at the University of Texas at Austin for valuable comments. The paper was accepted while the author was at the University of Texas at Austin. All views expressed and any remaining errors in the paper are solely the author's responsibility.

## REFERENCES

Acemoglu
,
Daron
,
David H.
Autor
, and
David
Lyle
.
2004
.
Women, war and wages: The effect of female labor supply on the wage structure at mid-century
.
Journal of Political Economy
112
(
3
):
497
551
. doi:10.1086/383100.
Angrist
,
Joshua D.
, and
Alan B.
Krueger
.
1994
.
Why do World War II veterans earn more than nonveterans
?
Journal of Labor Economics
12
(
1
):
74
97
. doi:10.1086/298344.
Ashenfelter
,
Orley
, and
John
Ham
.
1979
.
Education, unemployment and earnings
.
Journal of Political Economy
87
(
5
):
99
116
. doi:10.1086/260824.
Becker
,
Gary S.
1995
.
Human capital and poverty alleviation
.
Human Resources Development and Operations Policy Working Paper 52
.
Washington, DC
:
The World Bank
.
Bedard
,
Kelly
, and
Olivier
Deschenes
.
2006
.
The long-term impact of military service on health: Evidence from World War II and Korean War veterans
.
American Economic Review
96
(
1
):
176
194
. doi:10.1257/000282806776157731.
Berger
,
Andrea
,
Lori
Turk-Bicakci
,
Michael
Garet
,
Mengli
Song
,
Joel
Knudson
,
Clarisse
Haxton
,
Kristina
Zeiser
, et al.
2013
.
Early college, early success: Early college high school initiative impact study
.
Washington, DC
:
American Institutes for Research
.
Bound
,
John
, and
Sarah E.
Turner
.
2002
.
Going to war and going to college: Did World War II and the GI Bill increase educational attainment for returning veterans
?
Journal of Labor Economics
20
(
4
):
784
815
. doi:10.1086/342012.
Bowman
,
James L.
1973
.
Educational assistance to veterans: A comparative study of three G.I. Bills. Final report
.
Washington, DC
:
Superintendent of Documents, U.S. Government Printing Office Report No. SCP-18
.
Fetter
,
Daniel K.
2013
.
How do mortgage subsidies affect home ownership? Evidence from mid-century GI Bills
.
Economic Policy
5
(
2
):
111
147
.
Greenberg
,
Milton
.
1997
.
The law that changed America
.
New York
:
Lickle Publishing Inc
.
Hammond
,
Richard
.
1980
.
1979 National survey of veterans: Summary report
.
Washington, DC
:
Veterans Administration, Reports and Statistics Service, Office of Controller, Superintendent of Documents, U.S. Government Printing Office
.
Larsen
,
Matthew
,
Theodore J.
McCarthy
,
Jeremy
Moulton
,
Marianne
Page
, and
Ankur
Patel
.
2015
.
War and marriage: Assortative mating and the World War II GI Bill
.
Demography
52
(
5
):
1431
1461
. doi:10.1007/s13524-015-0426-x.
Lemieux
,
Thomas
, and
David
Card
.
2001
.
Education, earnings and the Canadian GI Bill
.
34
(
2
):
313
344
. doi:10.1111/0008-4085.00077.
Nickell
,
Stephen
.
1979
.
Education and lifetime patterns of unemployment
.
Journal of Political Economy
87
(
5
):
117
131
. doi:10.1086/260825.
Page
,
Marianne
.
2007
.
Fathers’ education and children's human capital: Evidence from the WWII GI Bill
.
Unpublished paper
,
University of California, Davis
.
Squire
,
Lyn
.
1993
.
Fighting poverty
.
American Economic Review
83
(
2
):
377
382
.
Stanley
,
Marcus
.
2003
.
College education and the midcentury GI Bills
.
Quarterly Journal of Economics
118
(
2
):
671
708
. doi:10.1162/003355303321675482.
Ruggles
,
Steven J.
,
Trent
Alexander
,
Katie
,
Ronald
Goeken
,
Matthew B.
Schroeder
, and
Matthew
Sobek
.
2010
.
Integrated public use microdata series: Version 5.0
.
:
University of Minnesota
.
Turner
,
Sarah E.
, and
John
Bound
.
2003
.
Closing the gap or widening the divide: The effects of the GI Bill and World War II on the educational outcomes of black Americans
.
Journal of Economic History
63
(
1
):
145
177
. doi:10.1017/S0022050703001761.

Table A.1.
Effect of the WWII GI Bill on College Education for Men
1923—19321923—19281923—19301923—19341923—19361923—1938
Time Windows(1)(2)(3)(4)(5)(6)
Panel A
Dependent Variable: Years of College
WWII 0.440*** 0.573*** 0.485*** 0.441*** 0.452*** 0.464***
(0.0680) (0.0833) (0.0703) (0.0666) (0.0617) (0.0554)
Korea 0.313*** 0.572*** 0.471*** 0.309*** 0.374*** 0.407***
(0.0799) (0.138) (0.0937) (0.0760) (0.0699) (0.0611)
R2 0.032 0.029 0.030 0.033 0.034 0.034
Panel B
WWII 0.0860*** 0.134*** 0.0966*** 0.0882*** 0.0949*** 0.106***
(0.0167) (0.0199) (0.0173) (0.0163) (0.0151) (0.0136)
Korea 0.0609*** 0.164*** 0.106*** 0.0573*** 0.0714*** 0.0889***
(0.0199) (0.0335) (0.0233) (0.0188) (0.0172) (0.0150)
R2 0.024 0.021 0.022 0.025 0.025 0.025
Panel C
Dependent Variable: Some College
WWII 0.134*** 0.156*** 0.150*** 0.131*** 0.126*** 0.118***
(0.0209) (0.0256) (0.0215) (0.0204) (0.0189) (0.0166)
Korea 0.0975*** 0.128*** 0.139*** 0.0996*** 0.114*** 0.109***
(0.0242) (0.0411) (0.0280) (0.0231) (0.0212) (0.0185)
R2 0.033 0.030 0.031 0.035 0.036 0.037
N 288,810 177,149 233,986 341,677 394,946 449,729
1923—19321923—19281923—19301923—19341923—19361923—1938
Time Windows(1)(2)(3)(4)(5)(6)
Panel A
Dependent Variable: Years of College
WWII 0.440*** 0.573*** 0.485*** 0.441*** 0.452*** 0.464***
(0.0680) (0.0833) (0.0703) (0.0666) (0.0617) (0.0554)
Korea 0.313*** 0.572*** 0.471*** 0.309*** 0.374*** 0.407***
(0.0799) (0.138) (0.0937) (0.0760) (0.0699) (0.0611)
R2 0.032 0.029 0.030 0.033 0.034 0.034
Panel B
WWII 0.0860*** 0.134*** 0.0966*** 0.0882*** 0.0949*** 0.106***
(0.0167) (0.0199) (0.0173) (0.0163) (0.0151) (0.0136)
Korea 0.0609*** 0.164*** 0.106*** 0.0573*** 0.0714*** 0.0889***
(0.0199) (0.0335) (0.0233) (0.0188) (0.0172) (0.0150)
R2 0.024 0.021 0.022 0.025 0.025 0.025
Panel C
Dependent Variable: Some College
WWII 0.134*** 0.156*** 0.150*** 0.131*** 0.126*** 0.118***
(0.0209) (0.0256) (0.0215) (0.0204) (0.0189) (0.0166)
Korea 0.0975*** 0.128*** 0.139*** 0.0996*** 0.114*** 0.109***
(0.0242) (0.0411) (0.0280) (0.0231) (0.0212) (0.0185)
R2 0.033 0.030 0.031 0.035 0.036 0.037
N 288,810 177,149 233,986 341,677 394,946 449,729

Notes: The table shows the effect of the WWII GI Bill on college attainment for men. The independent variable of interest is the fraction of men who participated in WWII. The dependent variables are years of college, dummy = 1 for some college and dummy = 1 for college graduation. The first column with a symmetrical ten-year window around the discontinuity point (end of 1927) gives the preferred estimates. The sample consists of black and white men born in the United States between 1923 and 1938. WWII is the fraction of men who participated in World War II in a birth year-quarter by state of birth cell. Similarly, Korea is the fraction of men who participated in the Korean War or at “any other time” but did not participate in WWII, in a birth year-quarter by state of birth cell. Standard errors are corrected for heteroskedasticity and are clustered at the level of birth year-quarter by state of birth. Controls include trend which is defined as (birth year − 1929 + [birth quarter/4]), trend squared, “Korea” interacted with trend variables, state of birth dummies and their interactions with trend variables, and current state of residence dummies.

***Significant at the 1% level.

Source: 1970 Integrated Public Use Microdata Series (see Ruggles et al. 2010).

Table A.2.
Effects of the WWII GI Bill with Different Trend Specifications
Time Window: 1923—1932
Panel A
Dependent Variable: Dummy Variable = 1 for completing high school
WWII 0.0839*** 0.0595*** 0.134*** 0.149***
(0.0167) (0.0185) (0.0227) (0.0244)
R2 0.073 0.074 0.074 0.075
Panel B
Dependent Variable: Dummy Variable = 1 if family is below poverty line
WWII −0.0352*** −0.0372*** −0.0407*** −0.0433***
(0.00906) (0.0104) (0.0128) (0.0138)
R2 0.035 0.035 0.035 0.035
Panel C
Dependent Variable: Dummy Variable = 1 if employed last year
WWII 0.00106 0.0225*** 0.0284*** 0.0283***
(0.00633) (0.00689) (0.00876) (0.00926)
R2 0.010 0.010 0.010 0.011
Panel D
Dependent Variable: Number of weeks worked
WWII 0.145 1.611*** 2.012*** 2.067***
(0.409) (0.443) (0.545) (0.582)
R2 0.017 0.018 0.018 0.018
N 288,810 288,810 288,810 288,810
Korean War
State of birth
Trend
Trend square
Trend cube
Trend interactions with Korean War and state of birth
Trend square interactions with Korean War and state of birth
Trend cube interactions with Korean War and state of birth
Time Window: 1923—1932
Panel A
Dependent Variable: Dummy Variable = 1 for completing high school
WWII 0.0839*** 0.0595*** 0.134*** 0.149***
(0.0167) (0.0185) (0.0227) (0.0244)
R2 0.073 0.074 0.074 0.075
Panel B
Dependent Variable: Dummy Variable = 1 if family is below poverty line
WWII −0.0352*** −0.0372*** −0.0407*** −0.0433***
(0.00906) (0.0104) (0.0128) (0.0138)
R2 0.035 0.035 0.035 0.035
Panel C
Dependent Variable: Dummy Variable = 1 if employed last year
WWII 0.00106 0.0225*** 0.0284*** 0.0283***
(0.00633) (0.00689) (0.00876) (0.00926)
R2 0.010 0.010 0.010 0.011
Panel D
Dependent Variable: Number of weeks worked
WWII 0.145 1.611*** 2.012*** 2.067***
(0.409) (0.443) (0.545) (0.582)
R2 0.017 0.018 0.018 0.018
N 288,810 288,810 288,810 288,810
Korean War
State of birth
Trend
Trend square
Trend cube
Trend interactions with Korean War and state of birth
Trend square interactions with Korean War and state of birth
Trend cube interactions with Korean War and state of birth

Notes: The table shows the effect of the WWII GI Bill on high school completion, poverty, and employment for men. The independent variable of interest is the fraction of men who participated in WWII. The dependent variables are the dummy variables = 1 if completed high school, dummy = 1 for family living under poverty line, dummy = 1 if worked positive number of weeks last year, and number of weeks worked last year. The sample consists of black and white men born in the United States between 1923 and 1932. WWII is the fraction of men who participated in World War II in a birth year-quarter by state of birth cell. Standard errors are corrected for heteroskedasticity and are clustered at the level of birth year-quarter by state of birth. Controls include “Korea” which is defined as the fraction of men who participated in the Korean War or at “any other time” but did not participate in WWII, in a birth year-quarter by state of birth cell, different polynomials in trend which is defined as (birth year − 1929 + [birth quarter/4]). “Korea” interacted with the trend polynomials, state of birth dummies and their interactions with the trend polynomials.

***Significant at the 1% level.

Source: 1970 Integrated Public Use Microdata Series (see Ruggles et al. 2010).

### Appendix B: Data

The 1970 U.S. Census data in IPUMS (Integrated Public Use Microdata Series; see Ruggles et al. 2010) is the main data source for this paper. The study uses the three 1-percent samples that are made available from the 15 percent of the population that was given the long-form questionnaire. The 15-percent questionnaire contains information on veteran service, which is used in this study. A different questionnaire was also given to 5 percent of the population in 1970, which does not have information on veteran service. Each of the three 1-percent samples of the 15-percent questionnaire identifies states, county groups, or neighborhoods as the smallest geographic unit. The sample in the study consists only of black and white men born in the United States between the years 1923 and 1938. Observations for which age, sex, race, or veteran status is imputed and for which state of birth is missing are dropped. The key variable of interest, WWII participation rate, is constructed from the variable VETWWII, which asks respondents if they served in World War II. The WWII participation rate in a birth state-year-quarter cell is calculated as the fraction of men who say they served in WWII in that birth state-year-quarter cell. The control variable for the Korean War is similarly constructed from the variables VETKOREA and VETOTHER. The participation rate for the Korean War in a birth state-year-quarter cell is calculated as the fraction of men who said they served in the Korean War or the military at any other time, but did not serve in WWII, in that birth state-year-quarter cell. The High School Completion variable in table 2 is constructed from the EDUC variable, which gives the highest grade or year completed. A person is coded as having finished high school if he has completed at least grade 12. All other education outcome variables are also constructed from the EDUC variable. The poverty outcome variable is constructed from the POVERTY variable and a person is coded as living under the poverty line if his family income is below the poverty threshold, that is, has a value of less than or equal to 100 in the POVERTY variable. The employment and number of weeks worked variables are constructed from the WKSWORK2 variable which gives the number of weeks worked last year in intervals. A person is coded as employed if he worked a positive number of weeks last year. The number of weeks worked outcome is calculated as the mid-point of the interval in which the person's number of weeks worked last year falls. The trend variable is defined as (birth year − 1929 + [birth quarter/4]).

In my robustness checks I use the IPUMS 1-percent sample from the 1960 Census and the IPUMS 5-percent sample from the 1980 Census and follow the same procedure as above to construct my final samples.

In my robustness check looking at the impact on women I use the three 1-percent samples from the 1970 Census as described above, with the only change that it is a sample of black and white women, not men.

For the WWI analysis, I use the IPUMS 5-percent sample from the 1930 Census and the IPUMS 1-percent sample from the 1940 Census. I construct the high school completion and employment outcomes from the EDUC and WKSWORK1 variables in the 1940 Census. I construct the WWI participation rate in a birth state-year cell from the VET1930 variable in the 1930 Census and define it as the fraction of men who served in WWI in that birth state-year cell. The WWI participation rate from the 1930 Census is matched to the 1940 Census by birth state and birth year. The trend variable is defined as (year of birth) − 1890.

## Notes

1.

The phrase was coined by Milton Greenberg (1997).

2.

As with previous work, the limitation of this empirical strategy is that it cannot distinguish between the effects of participating in war from the effects of the education benefits under the GI Bill. The estimates obtained are, thus, interpreted as the combined effect of war and education benefits.

3.

See also Ashenfelter and Ham (1979), Nickell (1979), Becker (1995), and Squire (1993).

4.

Recent revisions of the GI Bill require veterans to have completed high school but the Servicemen's Readjustment Act of 1944 allowed veterans of WWII to pursue secondary schooling as well.

5.

See Hammond 1980.

6.

Bowman (1973).

7.

The additional level of variation makes the implicit assumption that there was not significant migration before adulthood for the relevant birth cohorts.

8.

Previous studies have used birth year by birth quarter as the level of variation for WWII participation rates. I find similar results with this level of variation. The results are presented in table 6.

9.

Earlier papers have aggregated data to the birth year by birth quarter level. I keep data at the individual level and cluster the standard errors at the level of variation. Analysis gives similar results for data aggregated to the level of variation and number of observations in each aggregated cell used as weights. The results are presented in table 6.

10.

A robustness check using a definition for the Korea War variable that does not include participation at “any other time” gives similar results. It is presented in table 6.

11.

A robustness check studying the impact of WWI, which had no GI Bill, shows no impact of military participation on high school completion and weeks worked. It shows a negative effect of military participation on the probability of being employed. The results are presented in table 5.

12.

Because the data are in intervals for this variable, I assign the midpoint of the interval as the number of weeks worked by the individual in the last year.

13.

Ashenfelter and Ham (1979) and Nickell (1979) find that schooling reduces the incidence and the number of unemployment spells. Becker (1995) and Squire (1993) discuss that the provision of education is important for reducing poverty, especially in the context of developing countries.

14.

As another check, I follow Stanley's (2003) within-veteran analysis. Stanley compares cohorts 1921–1922 to cohorts 1923–1926. The latter left for war right after high school and thus had a higher probability of taking up the GI Bill benefits on return, as opposed to the older cohorts of 1921–1922 who had, to a larger extent, already decided their optimal education levels prior to leaving for war. I carry out the same analyses but do not find that the 1921–1922 cohorts differ significantly from the 1923–1926 cohorts in high school completion, poverty, and employment. When I compare cohorts 1921–1922 and 1923–1926 independently to cohort 1928–1938 I obtain similar results. Thus, veterans from both cohorts 1921–1922 and 1923–1926 responded quite similarly to the GI Bill on the outcomes studied in this paper. The effect of the GI Bill should still be understood as a combined effect of war and GI benefits.