Responsible researcher: Eduarda Miller de Figueiredo
Article Name: Why Do World War II Veterans Earn More than Nonveterans?
Authors: Joshua Angrist and Alan B. Krueger
Intervention Location: United States
Sample Size: 335,989 male individuals.
Sector: Job Market
Variable of Main Interest: Salaries
Type of Intervention: World War II
Methodology: 2SLS
Summary
About 10 million men served in the World War II era, in which the recruitment process followed a birth chronology rule. Based on the results of a salary comparison between veterans and non-veterans, the authors seek to explore alternative explanations for the apparent success in the civilian job market of World War II veterans. Based on estimations using methods such as OLS and 2SLS, which control for possible endogeneities in this analysis, the authors concluded that World War II veterans do not earn more than comparable non-veterans.
The objective of this work was to explore alternative explanations for the apparent labor market success of World War II veterans. This study is based on the fact that a comparison showed that, on average, World War II veterans earn more and have lower unemployment rates than non-veterans of the same age. Vietnam War veterans earn less and have higher unemployment rates when compared to non-veterans.
These differences in the consequences of veteran status at different times raise questions about the functioning of the labor market and the military. One hypothesis is that the Vietnam War was an unpopular cause, in which men who served during the war were discriminated against when they returned to the civilian job market. While World War II was widely supported, veterans of that period received different treatment in civilian employment. Furthermore, according to the authors, positive feelings about World War II may also have been transformed into political support for relatively generous subsidies for education and preferential treatment in hiring.
Another factor that may have influenced this differentiation of treatments, as the authors explain in the text, was the fact that young people from wealthy families avoided military service during the Vietnam period, while those with low income did not have the same option (Cooper, 1977). Already in World War II, almost 75% of men in eligible cohorts served in the armed forces. Thus, these selection forces likely induced a positive bias in veteran premium estimates for World War II veterans and a negative bias for Vietnam War veterans.
According to data obtained by the authors, approximately 16,354,000 men served in the armed forces during the World War II era (1940-1947), in which 10 million of these men were drafted. As the recruitment process followed a birth chronology rule, once the number of men necessary to meet the quota required at that time had been recruited, the “rest” were no longer at risk of recruitment. Thus, this process generated a correlation between date of birth and the probability of military service.
Therefore, the authors' analysis focuses on those registered in the second part of the 6th register (born after 1925). 64% of those registered were exempted or rejected because they were classified as physically or mentally incapable, while the other 34% were “deferred” for professional reasons – wartime production –.
The minimum length of service in the armed forces for draftees was 18 months, with an additional requirement to serve as reserves after the end of the service period. This requirement was waived for recruits who agreed to remain in the regular army for an additional 24 months of service.
There were different recall dates, totaling 7 national recalls during this period, in which almost all United States citizens and residents were required to register. The calls took place with the following rules:
This study aimed to re-examine the World War II veterans' premium using econometric techniques to control for non-random selection into the armed forces. In which instrumental variables correlated with veteran status are used, but not correlated with other determinants of earnings.
The authors used a 5% public use sample from the 1980 census, where the data set contains information for a self-weighted sample of 11 million individuals in the US. Therefore, after properly cleaning the database, the study contained information from 335,989 men.
An OLS was estimated with a human capital gains function, where the dependent variable is the log of wages [2] . However, because there is some level of endogeneity due to the fact that the Selective Service tends to reject registrants with low earning capacity, the authors ran a 2SLS, in which the first stage is used to determine the veteran's status (dependent variable) [3 ] . Therefore, an instrumental variables strategy was used that explores the relationship between date of birth and veteran status.
The estimates from the OLS and 2SLS models showed very different results, in which the salary of a veteran according to the OLS estimate is -22% while according to the 2SLS estimate it is -13%. There is also a contrast between the binary variables of race, marital status, residence status and whether the individual works in a metropolitan region.
When looking at the covariates of education and disability status, according to the 2SLS estimate, the veteran effect is still negative. According to the authors, the attenuation in the 2SLS estimates may also reflect the fact that education levels were increasing for successive generations of men at this time. Like the effects of compulsory attendance laws, this would also lead individuals born earlier in the year, who are more likely to be veterans, to have less education and therefore earn less.
Estimates based on the three censuses (1960, 1970, 1980), which follow the same cohort of men over time, show from the OLS estimate that veterans' earnings reduce as the cohort ages. For example, when controlling only for year of birth, veterans earned 31% more than non-veterans in 1960, 26% more in 1970 and 22% more in 1980. For the authors, this occurs because selection for services military action occurred in the 1940s, and the characteristics on which the selection was based may have become diluted over time. For example, some individuals who were excluded from service because of a physical disability at age 18 may have recovered at least partially from the disability over the years.
Other estimates, using the 2SLS model, demonstrate that the veteran's salary is more negative in the 1980 census than in the 1960 and 1970 censuses. The explanation for this result, according to the authors, is the correlation of age between the quarter of birth and income. Between 1960 and 1970, the cohorts examined were between 31-45 years old, that is, they would be in the upward-sloping portion of the cross-sectional earnings-by-age profile. On the other hand, in 1980, this group of men reached 50 years or more, therefore, they would be at the peak or at the bottom of the slope of this cross-sectional income-by-age profile. This is based on Mincer's (1974) estimate that earnings peak at age 52 for those with high school education and at age 56 for graduates. In other words, in 1960 and 1970 men born at the beginning of the year were expected to earn more than men born at the end of the year, while in 1980 the opposite was expected.
Through estimations, using robust endogeneity controls, the authors concluded that World War II veterans do not earn more than comparable non-veterans, and may well earn less. Demonstrating that military service, in fact, may have reduced the earnings of World War II veterans from what they would have otherwise been.
References
Cooper, VL Military Manpower and the All-Volunteer Force . Report R-1450-ARPA. Santa Monica, CA:RAND Corp., 1977.
Mincer, Jacob. Schooling, Experience and Earnings . New York: National Bureau of Economic Research, 1974.
[1] Each element can be numbered using separate pieces of paper and placed in a container. After shuffling, each piece is randomly chosen one by one.
[2], j = (1926, 1927, 1928)
[3] , j = (1926, 1927, 1928)
Check the Institution's registration in the e-MEC System here
© 2025 All Rights Reserved
