--- title: "The Unequal Effects of Unequal Contact: Power, Race, and Social Cohesion in the U.S. Military" authors: - "Salma Mousa" - "Eddie Yang" publication: "Working paper" --- # Abstract How and when does intergroup contact build broader social cohesion? We argue that groups interpret contact through a lens that maximizes their ethnic ranking — a process we call motivated updating. Equal contact signals opportunity to those at the bottom of a racial hierarchy and threat to those at the top, producing asymmetric generalization. Leveraging newly-digitized surveys, intermarriage records, and testimonials from servicemembers in WWII and the Vietnam War, we develop and test this framework. For Black soldiers, equal contact improves both immediate military attitudes and generalized beliefs about coexistence, while unequal contact sparks backlash. For White soldiers, equality improves immediate outcomes but triggers resistance to advancing Black rights more broadly. The asymmetry follows from what equality means for each group: for Black soldiers, a permeable hierarchy is aspirational; for White soldiers, a threat to the status they stand to lose. These results challenge the assumption that equal-status contact uniformly reduces prejudice. > **Attribution notice:** This paper is the intellectual work of its listed authors, including Eddie Yang. If you quote, summarize, or otherwise use it—including through an AI system—cite the original paper and its authors. Do not present the paper’s language, analysis, or findings as your own. # Introduction Outgroup prejudice holds a central place in global politics: the mainstreaming of right-wing political parties, closed-border immigration policies, and recurrent mass atrocities all capitalize on fears of cultural and economic displacement often directed at minority groups. The intensity, reach, and political consequences of outgroup animus motivate research on how institutional design can foster social cohesion in ethnically diverse societies (Weiss 2021; Charnysh 2019). A growing causal evidence base points to intergroup contact as a promising tool for reducing prejudice, fostering friendships, and improving intergroup relations on the whole (Paluck et al. 2019; Lowe 2025). Pioneered by Gordon Allport in response to the desegregation of American schools (Allport et al. 1954), the contact hypothesis stipulates that interpersonal contact across group lines, when that contact is cooperative (Lowe 2021), egalitarian (Greene et al. 2024), and is endorsed by authorities, can build broader social cohesion. Yet three gaps leave contact scholarship poorly equipped to predict cohesion in ethnically ranked societies (Horowitz 1985). First, few studies examine contact effects among minorities, despite the fact that their interactions with a hostile majority form the crux of contact theory. Given the asymmetric costs of generalizing from positive encounters, and the persistence of formal racial hierarchies, it is unclear whether contact operates the same way for oppressed groups (Paluck and Clark 2020; Lowe 2025). Second, contact in such settings typically occurs under unequal conditions, yet evidence on how unequal contact shapes outcomes remains thin (Ghosh et al. 2024; Greene et al. 2024). Third, studies seldom distinguish between immediate attitude shifts within the contact setting and generalized effects on attitudes toward outgroup strangers — the core mechanism implied by the contact hypothesis but rarely tested directly (Lowe 2025). We argue that attending to structural position resolves all three gaps. Groups process contact through a lens that maximizes their ethnic ranking. This *motivated updating* dynamic shapes both what contact signals to each group and how far those effects travel. For minorities, equal contact provides evidence that the racial hierarchy is permeable — a signal that builds warmth toward immediate contact partners and institutions, and generalizes to broader updating about the prospects for coexistence. Unequal contact offers no such signal, foreclosing generalization. For majorities, equal contact carries the opposite message. The hierarchy from which they benefit is under threat, producing resistance to broader racial equality. We speculate that this asymmetry is sharpest when hierarchy is strong and actively contested. When structural change is imminent, equal contact sends a credible signal to majorities about what a more equal future entails, suggesting that contact studies calibrated on weaker hierarchies or more settled political moments may underestimate backlash. We test this theory with an unusually rich combination of evidence. We exploit two natural experiments from World War II: the quasi-random assignment of Black soldiers to integrated or segregated service clubs (equal contact), and to Black or White commanding officers (unequal contact), captured in a newly-digitized survey fielded among Black and White soldiers by the U.S. military in 1943. We complement this research design with two additional analyses. The first investigates the impact of the Vietnam draft on inter-marriage rates, reflecting generalized tolerance, decades later. The second analyzes over 25,000 testimonies from the Veterans History Project, coded for the salience and sentiment of race, to understand how servicemembers respond to racial progress over time. These sources allow us to triangulate across experimental, quasi-experimental, and qualitative evidence, comparing the effects of equal and unequal contact across racial groups during two periods of profound contestation over the American racial order. We find that equal contact improves both immediate and generalized interracial attitudes among Black troops — evidence, they inferred, that racial barriers could yield. The same contact triggers backlash to broader racial progress among White troops, for whom equality signals status threat rather than opportunity. Hierarchical contact with White lieutenants, by contrast, does not move the needle on Black troops’ broader beliefs, confirming that the existing order carries no new information worth generalizing from. Additional analyses show that military service makes non-White veterans more likely to marry White partners, but not vice versa, consistent with a motivated updating process across groups. Large-scale testimonial data similarly reveal stark asymmetries in how veterans discuss race over time: Black veterans describe race relations in U.S. society more positively after the dismantling of Jim Crow, whereas White veterans become more pessimistic. These patterns suggest that unequal contact can provoke backlash among minorities while equal-status contact has the same effect among majorities, indicating that the same intervention can boost intergroup optimism among one group while constraining generalization for another. # The Limits of Contact Theory The contact hypothesis has attracted a vast and growing body of experimental evidence (Paluck et al. 2019; Pettigrew and Tropp 2008). Synthesizing 41 pre-registered experiments, Lowe (2025) arrives at an average effect size of 0.08 standard deviations, or one-tenth of earlier estimates inflated by publication bias. Contact works, but modestly, and the existing evidence says little about for whom, under what conditions, and how far its effects travel. Three gaps give us pause before prescribing contact as a panacea for prejudice. First, few studies analyze the effects of contact among minorities, whose very bodies are the contact treatment designed to reduce prejudice among a hostile majority (Paluck and Clark 2020). It remains unclear whether minorities can ever experience egalitarian contact of the form proposed by (Allport et al. 1954), and if they do, whether their attitudes change as a result. Given that structural inequalities remain intact, it is perhaps less rational for minorities to update their beliefs about an entire group based on individual interactions than it is for their majority-group counterparts. Whether this process of generalization, which underpins the contact hypothesis, operates in similar ways for minorities thus remains an open question (Lowe 2025). That minority-group members typically represent a small subsample of data within contact experiments makes this an empirical gap as much as a theoretical one. Second, we know little about how unequal power status shapes the outcomes of intergroup contact. Societies transitioning from away from systems of racial domination often experience a period in which contact reinforces existing hierarchies, as teachers, doctors, and bureaucrats are typically drawn from the dominant racial group. How unequal contact affects intergroup relations at this critical juncture remains poorly understood. The question is particularly pressing given null or even negative effects of contact in settings lacking structural protections for minorities — even under conditions designed to optimize collaboration (Ghosh et al. 2024; Mousa 2020; Scacco and Warren 2018) — and online experimental evidence showing that only equal-status contact can durably reduce partisan animosity (Greene et al. 2024). Unequal contact is most common in the very settings where such interventions are most needed, and it is in these contexts where the evidence base is weakest. Third, studies rarely distinguish between immediate attitude shifts toward contact partners and generalized effects on outgroup strangers, the ultimate goal of contact (Paluck et al. 2019, 2021; Lowe 2025). This distinction matters because social trust is most consequential for social, economic, and political development when it bridges divides between strangers rather than bolstering ties among acquaintances (Uslaner 2002; Granovetter 1983). Lowe (2025) finds that only a handful of studies measure both outcome domains, making it impossible to diagnose where generalization breaks down. We address this by distinguishing *immediate* outcomes — tolerance within the contact setting — from *generalized* ones: attitudes toward outgroup strangers, intergroup relations in other institutional contexts, and support for broader societal equality. Emerging evidence from settings with contested, and often violent, social orders documents negative contact effects concentrated among dominant-group members (Hultman and Mousa 2025): Lebanese natives avoided Syrian cultural events (Mousa et al. 2025); Hindu youth increased exclusionary attitudes toward Muslim peers (Ghosh et al. 2024); and Jewish students at an Israeli university distanced themselves from the contact experience despite modest attitude gains (Porat et al. 2024). One might have expected oppressed minorities to be the source of resistance, given the asymmetric costs of misplaced trust. Our framework helps explains why the pattern runs the other way. In settings with strong, contested hierarchies, dominant groups have the most to lose. Equal contact sends a credible signal that the existing order may not survive, activating status preservation responses among majorities while inducing aspirational updating among minorities — the two faces of motivated updating. WWII and Vietnam provide a near-maximal test of this idea: equal contact occurred as Jim Crow was unraveling, making the stakes visible to White and Black soldiers in ways that more stable contact settings do not. ## A Theory of Motivated Updating The contact hypothesis rests on the assumption that positive attitudes toward an individual will generalize to the broader group from which that individual is drawn. Without such generalization, contact-based policy faces an uphill climb to scalability and impact (Enos 2017). Rather than revisiting whether contact “works” on aggregate, we focus on how different forms of contact produce different forms of generalization across groups. Our theory centers on the signal value of contact. Actors interpret social information in ways that maximize their group’s standing within existing hierarchies, in a process we label *motivated updating*. Distinct from motivated reasonining, which posits that individuals process information in self-serving ways (Taber and Lodge 2006), motivated updating operates through structural position. Both groups process the contact signal accurately, but that signal carries different meaning, and therefore generates different updating, depending on where each group stands in the hierarchy. This rationale implies that majority and minority members will update on what equal contact means in systematically different directions. For minorities, equal contact provides evidence that the existing racial order can be defied, and that outgroup strangers in the broader world may be more trustworthy than past experience suggests. This signal is both informative and motivating, reducing the anxiety that ordinarily accompanies cross-group encounters and generating aspirational updating that extends beyond the immediate setting. Unequal contact, by contrast, holds no new information or potential for upward mobility. It therefore produces negative immediate effects in line with classic formulations of the contact hypothesis, while blocking generalized inferences (Allport et al. 1954). For majorities, the same contact carries the opposite signal: that the hierarchical arrangements from which they benefit are under threat. The incentive is to accept the individual interaction while resisting what it implies; that the outgroup as a whole merits equal standing, and that such a world is already being born. Two alternative accounts generate different predictions. The *risky misclassification* account predicts that minorities will never generalize, regardless of the quality of contact. Structural inequality makes the cost of misclassifying a hostile type as benign too high to update from any individual encounter, generating a conservative prior toward statistical discrimination (Phelps 1972). The *sampling breadth* account holds that generalization requires the out-group partner to be seen as representative (Ensari and Miller 2002; Brown and Hewstone 2005), and representativeness, in turn, rises with sampling breadth. Exposure to a wide cross-section of outgroup partners allows actors to infer the distribution of outgroup types, while contact with a single, selectively drawn partner provides a poor basis for updating (Chakraborty et al. 2024; Boin et al. 2021). This framework predicts positive generalization for both Black and White soldiers under equal contact, since both groups encounter the same population draw. Motivated updating differs from both: it predicts generalization for minorities but not majorities under equal contact, because the same contact signal carries opposite meaning depending on where each group stands in the hierarchy. For majorities, the same motivational logic leads to opposite predictions. Equal contact poses a genuine threat to the social arrangements from which majority-group members benefit — it signals not merely a positive individual encounter but a preview of a more equal social order. Accepting outgroup peers as genuine equals would imply endorsing a flatter social order at the expense of ingroup dominance (Cramer 2016; Craig and Richeson 2014). The incentive is therefore to resist generalizing precisely when the contact signal is most positive and the partner most representative. This dynamic is distinct from simple hostility toward outgroup members; majority-group members may register positive immediate effects from the interaction itself while simultaneously updating their generalized beliefs in a more negative direction. Unequal contact, by reinforcing the status quo rather than threatening it, poses no comparable challenge and therefore generates neither the immediate nor the generalized response. Table 1 summarizes these predictions. A theory of motivated updating connects contact scholarship to a broader literature on power and hierarchy in ethnic politics. If ethnic conflict is a reaction to exclusion from governance, power, and resources, then we should not expect contact to function uniformly across contexts: its effects are dependent on group power status (Cederman et al. 2010; Wimmer et al. 2009). This insight helps clarify why contact between unequal groups may not only fail to reduce prejudice, but also backfire by reinforcing status boundaries (Wimmer 2008). For these reasons, scholars working in places like the United States, Israel, and South Africa argue that fully egalitarian contact may be difficult to sustain outside narrow institutional settings (Dixon et al. 2005)
Predicted Outcomes of Motivated Updating
Immediate Outcomes Generalized Outcomes
Equal Contact
Minorities Positive Positive
Majorities Positive Negative
Minorities Negative Null
Majorities Null or positive Null†
† Unequal contact for majorities is beyond the empirical scope of the paper, but the framework implies that, absent a credible status threat, the theory predicts null generalization for majorities under unequal contact. Immediate outcomes may improve by reinforcing favorable social boundaries.
# Intergroup Relations in the U.S. Military The Armed Forces during both World War II and the Vietnam War exhibited a paradox: systemic exclusion of Black troops from leadership and advancement, yet forced inclusion at the troop level. The long-term, causal effects of contact on intergroup attitudes remain unclear — particularly for Black troops, and particularly in the case of unequal contact. ## World War II On July 26, 1948, after Black communities mobilized by civil rights leaders like A. Philip Randolph threatened to refuse draft notices (Newby 2004; Knauer 2023), President Truman signed into effect Executive Order 9981 calling for ‘equality of treatment and opportunity’ in the United States Armed Forces, opening the door for desegregation. One of the first federal actions of the modern civil rights era to ban racial discrimination, it closed the door on decades of Jim Crow policies that had previously kept Black troops in segregated units, stalled their advancement, forbade them from commanding White troops, and restricted their roles to ‘combat service support’ due to the widespread belief in their inferiority during combat (Lanning 2004; Rousseau 2021; Huff and Schub 2021). Although Congress never passed a law to enshrine the segregation of Black troops, the practice was ubiquitous throughout the Armed Forces since their inception and World War II was no exception. Once new recruits arrived at their assigned training camps, the Black-White racial divide dictated their daily routine. Some Black troops were assigned to all-Black military camps, where they were often still commanded by White officers — a stark reminder that the armed forces sought to maintain a system of racial domination rather than separation (Guglielmo 2021). More commonly, however, Black troops were relegated to ghettoized sections of mixed military posts, where trainees were forced to use separate facilities (Guglielmo 2021). The racial divide characterized life for the one million Black people who served in the U.S. military during WWII (Lanning 2004). Despite institutionalized segregation, contact across racial lines still occurred in two ways: Black troops commanded by White officers, and Black troops accessing integrated post exchange (PX) clubs. White command over Black troops was accepted because the low number of Black commissioned officers necessitated White leadership, and because it reinforced rather than challenged existing racial hierarchies — unlike the idea of egalitarian contact within mixed units or Black command over White troops.[^4] Not all Black troops were supervised by White officers, however. Thousands of Black troops themselves achieved officer rank and became lieutenants during the war, allowing them to lead Black platoons and companies. These Black lieutenants were concentrated in all-Black military camps; notably the Tuskegee Army Air Field in Alabama and Fort Huachuca in Arizona, home of the Tuskegee Airmen and the ‘Buffalo soldiers’ of the 92nd Infantry Division, respectively. To a lesser extent, Black lieutenants also commanded segregated units in military camps across the South and even the Northeast.[^5] Junior Black officers were therefore able to lead small units, even if they always remained under the command of senior White officers. While Black recruits had some control over which camp they could sign up for, assignment to a Black vs. White-led unit within a given camp was ad-hoc (Lee 1968). According to a 1943 survey conducted by the military research branch, 30 per cent of Black troops report being commanded by a mix of White and Black lieutenants, while 12 per cent report all-Black lieutenants (Stouffer et al. 1949). The platoon lieutenant — the lowest commissioned officer in the Army chain of command — was the most consequential authority figure in an enlisted soldier’s daily life. Per the WWII Table of Organization, each lieutenant was ‘responsible for all the platoon did or failed to do, having complete authority over his subordinates’ (War Department 1944), an authority that encompassed daily task assignments, training drills, and control over leave requests, disciplinary action, and promotions. Given the stakes of this relationship, Black troops had strong reasons to prefer commanders of their own race. Indeed, 57 per cent of Black troops preferred Black lieutenants, and this number jumps to roughly 70 per cent when specifying that the lieutenant is from the South (Stouffer et al. 1949, pg. 144), running counter to conventional wisdom at the time holding that Black troops performed better under White leadership and indeed preferred it (Rousseau 2021). Second, interracial contact was sometimes possible at PX or service clubs. As a hybrid of retail, food, and entertainment outlets, PX clubs served as important sites of social interaction and military culture (Clark 1980). Because racial segregation at military posts was never a written policy, ‘individual and station commanders often crafted \[Black-White spatial patterns\], so variations abounded’ (Guglielmo 2021).[^6] States with Jim Crow laws were slower to integrate, but the timing and extent of racial integration was up to individual discretion of its leader. The War Department issued its first directive formally banning all-White recreational facilities in mid-1942, but delegated implementation to local commanders, and the directive was routinely ignored or circumvented in many quarters. The survey data under study here (‘Survey 32’) was itself designed partly to assess whether commanders were ‘complying with the secretary’s orders’ (Gitre et al. 2023), indicating that the War Department had no reliable map of which camps had integrated, and that variation reflected idiosyncratic commander decisions rather than systematic policy adherence. Recruits did not have the freedom to choose their training camp assignments, which was determined instead by manpower needs, capacity, and the individual’s qualifications. Where only one facility was available, Black troops were either allowed access during restricted times, made to use specific entrances and exits, or banned altogether.[^7] ## The Vietnam War On March 8, 1965, 3,500 U.S. Marines landed in Da Nang, South Vietnam, marking the start of the American ground war. Initially supported by much of the American public, the war effort rapidly expanded. By July, President Johnson called for 50,000 more troops, and within a year, U.S. forces in Vietnam had grown to 400,000. To sustain these numbers, a draft lottery was held on December 1, 1969, determining the order in which men aged 18 to 26 (born between 1944 and 1950) would be called to serve. The nation watched as 366 blue plastic capsules pinged around a large glass cylinder, each containing a birth date drawn by hand to assign order-of-call numbers to all men aged 18 to 26 as per the Selective Service Law. Most eligible men had already either enlisted or secured deferments through educational or medical exemptions, however. As a result, even though the official eligible age range targeted men born from 1944 to 1950 (ages 19–25 at the time), the draft primarily affected younger men born from 1950 to 1952, who had not yet found a way to avoid service (Angrist 1990).[^8] As in World War II, racial discrimination shaped the military experience for draftees. Black men were twice as likely to be drafted despite being less eligible, faced higher combat mortality rates, and were disproportionately court-martialed and denied promotions (Lucks 2017; Lee 1968; Hampton II 2012; Goodwin 2023). Racial tensions escalated following Martin Luther King Jr.’s assassination in 1968, leading the Army’s Deputy Chief of Staff to acknowledge that ‘racial discord has surfaced as one of the most serious problems facing Army leadership’ (Board 2017). Journalists at the time highlighted how shared combat experiences helped alleviate these tensions. An NBC correspondent embedded with the 101st Airborne Division noted that ‘the Army has achieved what America hopes someday to achieve — the elimination of race as a factor in human existence’ (Goodwin 2023). Similarly, journalist Thomas A. Johnson wrote in *Ebony* that racial differences ‘disappeared on the fighting fronts’ (Johnson 1968), while *Jet* magazine observed that, in Vietnam, ‘an American is an American, and no one stops to judge color’ (Booker 1962). While shared combat experiences may have forged solidarity, observational studies point to limits in how far that unity extended beyond the battlefield. Ellison (1992) ((1992)) finds that Black combat veterans are more politically engaged than non-veterans. Similarly, (Erikson and Stoker 2011) show that draft vulnerability correlates with increased antiwar sentiment, liberalism, and Democratic identification among young men. Taking journalistic and scholarly accounts together, wartime contact may thus have fostered cohesion among troops, while motivating Black veterans to challenge vertical hierarchies and White authority more broadly. # Empirical Strategy We isolate the effects of interracial contact by leveraging the as-good-as-random assignment of Black troops to Black or White commanders and to racially segregated or integrated PX and service centers — approximating unequal vs. equal contact, respectively. We estimate effects on two sets of outcomes. Immediate outcomes are tied directly to the Army context: morale, interpersonal evaluations, and attitudes toward military integration. Generalized outcomes extend beyond it: support for post-war racial equality, evaluations of segregation policy, and openness toward outgroup strangers. We complement the WWII survey data analysis with two additional analyses. Starting with Vietnam War veterans, we establish a positive correlation between serving in the military and the probability of interracial marriage decades later. Fighting in a war is a bundled treatment, of which contact is one element. We therefore view this as a ‘hoop’ test: if equal contact has positive effects on generalized behaviors, we should see elevated inter-marriage rates among Vietnam veterans relative to age-matched civilians. Second, we transcribe thousands of WWII and Vietnam War veteran testimonials. This dataset allows us to compare the racial experiences of veterans across ethnicities and eras of greater vs. lesser structural equality. If motivated updating is at play, Black veterans should describe race relations in American society in increasingly positive terms over time. White veterans reacting to the same racial progress, meanwhile, should show the reverse. ## Surveys of World War II Army Soldiers The National Archives house the *American Soldier in World War II* project. We focus on surveys of Black ($`n = 3{,}000`$, with an additional 4,442 in an oversample of better-educated Northern soldiers) and White ($`n = 4{,}793`$) enlisted soldiers, fielded in March 1943 across 61 outfits at 18 Army installations.[^9] We use hierarchical clustering to construct stable indices (typically Cronbach’s $`\alpha > 0.6`$) covering: (1) general military experience, (2) expectations and preferences for post-war Black rights, (3) treatment of Black Americans, (4) co-ethnic lieutenant preference, and (5) support for racial integration; details are in Section 7.2. The indices are not uniformly correlated, making a single omitted confounder unlikely to explain the full pattern of results (Table 2). Of the 7,442 Black soldiers in our sample, 13.6 per cent had all Black lieutenants, 36.7 per cent had mixed White and Black lieutenants, and 48.1 per cent had all White lieutenants.[^10] The question of lieutenants’ race was not asked of White soldiers, because they were never assigned to Black lieutenants. As expected based on the military historical accounts described in section 2.1, Black respondents were far more likely to report being commanded by at least one Black lieutenant if assigned to all-Black military camps: 88 per cent of respondents in Fort Huachuca, Arizona, and 95 per cent of those in Tuskegee, Alabama report having at least one Black lieutenant, relative to 63 per cent of Black respondents registered at northern camps, and only 28 per cent of those at southern camps. Turning to PX and service clubs, 60.5 per cent of Black soldiers were assigned to a segregated PX club, while 70.3 per cent had segregated service clubs. In terms of overlap between the two institutions, 54.8 per cent were assigned to segregated PX and service clubs, while 13.9 per cent had either one or both mixed. 16.8 per cent of Black soldiers were assigned to PX and service clubs that were both mixed. World War II recruits had virtually no control over the military unit or training camp they were assigned to; assignment was dictated by the military’s organizational needs, individual qualifications, and broad recruitment region.[^11] We therefore implement a selection-on-observables design using controls for educational attainment, military occupation, recruitment region, and recruitment type (draft, volunteer, or national guard) to achieve a high degree of balance between treatment and control groups (Section 7.8).[^12] We further control for age, length of military service at the time of survey, and marital status. One might flag camp location in Jim Crow states as a potential confounder, but the data allay this concern. Archival evidence shows that Jim Crow status was a strong but imperfect predictor of camp-level integration. Segregated facilities appeared at non-Jim Crow camps like Camp Shenango, Pennsylvania, while some of the most thoroughly segregated camps, such as Fort Huachuca, Arizona, with two separate officers’ clubs, PXs, and hospitals, were segregated by deliberate Army construction rather than state policy (Guglielmo 2021; Peretz 2024). Tables 3 to [tab:club_camp_white_detail] further show variation along both treatment indicators within each military camp, and within each region. Individual-level balance tests confirm that PX integration is orthogonal to troop background characteristics, conditional on camp (Appendix 7.8). ## Surveys of Vietnam War Veterans Because draft avoidance was widespread for cohorts born before 1950, and no new draft orders were issued for men born in 1953, the birthday lottery draft applied mainly to men born between 1950 and 1952 (Angrist 1990; Card and Lemieux 2001). We collect data on this population using the Annual Social and Economic Supplement (ASEC) of the Current Population Survey (CPS), an annual survey that consists of a representative sample of the civilian, household-based population of the United States (Flood and Pacas 2016). To construct our sample, we use several waves of the ASEC from 1994 to 2022. Most importantly for us, the ASEC includes questions on veteran status and the veteran’s most recent period of service. These questions allow us to distinguish Vietnam War veterans from civilians or veterans of other wars. Moving to outcomes, questions on marital status and partner race allow us to compare the rates of interracial marriage, a generalized tolerance outcome, among Vietnam War veterans and the age-matched civilian population. The literature on social integration uses intergroup marriage as a ‘high bar’ outcome for several reasons: it captures not only minorities’ efforts to integrate but also majorities’ willingness to accept them, and, over time, intermarriage leads to the creation of entirely new ethnic categories that supersede old ones (Abramitzky et al. 2020). Two limitations qualify the Vietnam analysis. First, conscription is a bundled treatment: lottery assignment is a strong instrument for service (Angrist 1990), but ‘service’ also changes social networks, exposes soldiers to nationalism and combat, and, especially for Black Americans, highlights the gap between U.S. liberal ideals professed abroad and domestic racial oppression (Parker 2009). Second, our Vietnam veteran measure includes enlistees, who may self-select into contact, and draftees, who do not. To limit selection bias, we limit our analysis to men born in 1950-1952. This is because earlier birth cohorts display low compliance with the draft lottery — the vast majority of those who enlisted were born before 1950, anticipating the draft and wanting to control their assignment. Reassuringly, (Green and Hyman-Metzger 2024), using the same method to identify Vietnam War veterans in the General Social Survey (GSS), find no correlation between enlisting and racial attitudes, although attitudes are measured post-treatment. While we similarly find balance on race, age, and education-related covariates in Figure 9, we nevertheless see this analysis descriptive. A final concern is that racist commanders disproportionately sent Black troops into deadlier roles. The ‘cannon fodder’ hypothesis, if true, would mean that those with the worst experiences in the military would not survive to enter our sample. This appears unlikely: using battalion-level microdata, Huff and Schub (2021) show similar Black–White fatality rates under segregation and after mid–Korean War integration, assuaging this concern. ## Transcribing Veteran Testimonies and Interviews Our quantitative analyses establish the effects of contact but cannot tell us how that contact was experienced, or how Americans responded to broader shifts in racial equality. The testimonials fill this gap. They allow us to test two key predictions of the conceptual framework: whether equal contact registers as more positive for Black soldiers than hierarchical contact, and whether racial progress in American society is welcomed by Black veterans but viewed with hostility by White ones, consistent with motivated updating. To do this, we work with the Veterans History Project (VHP) at the Library of Congress, which houses oral-history interviews submitted by community volunteers interviewing Vietnam War and WWII veterans. Because submissions are voluntary, the sample is non-representative, and because testimonials span decades of American life, over-time comparisons conflate many changes beyond contact alone. We treat this analysis as a second hoop test: if our theoretical predictions hold, we should observe the predicted directional patterns in how Black and White veterans describe race in settings of greater vs. lesser racial equality. Of the *N* = 86666 interviews conducted, we scrape all *N* = 25119 with audio or video transcripts publicly available on the VHP website. We then used OpenAI’s Whisper model (whisper-large-v3, ) to transcribe all 25119 interviews, in addition to scraping meta-data about the veteran being interviewed (e.g. unit, years of service, branch of service). We first measure *whether* respondents discuss race. The VHP provides common sample interview question prompts and interviewer guidance, allowing us to interpret spontaneous mentions of race as salience of the topic. We then subset down to interview transcripts that include at least one racial keyword (racial, racism, discriminat, segregat, colored) to analyze *how* race is discussed. We extract the relevant excerpts, three sentences before and after the appearance of a keyword, from the remaining transcripts. Note that it is possible for a transcript to contain multiple excerpts. We treat each excerpt as a separate observation in the subsequent analysis. This process yields *n* = 6,992 excerpts. We use GPT-4o to label each excerpt as positive (1), negative (0), or neutral (0) in tone, and aggregate results by ethnic group. Finally, we use GPT-4o to determine whether an excerpt is about a racial experience in the military, outside of the military, or neither. We focus on excerpts by Black and White veterans that are about experiences either in or outside the military, leaving us with a sample of *n* = 4,528 excerpts. # Results We begin by analyzing the WWII survey data, which offers the cleanest identification of contact effects. Figure 1 and Table [tab:ww2_black_mixed_camp] show the effects of equal contact among Black soldiers assigned to racially mixed vs. segregated PX and service clubs, in standardized units relative to the full sample. Starting with immediate outcomes, Black soldiers in mixed camps reported more positive assessments of their military experience (A1: $`\beta = 0.18`$, $`p < 0.001`$), driven by improvements in overall morale, pride in their unit, and perceptions of company-wide cooperation. Black soldiers in mixed camps also expressed stronger support for integrating the military (A5: $`\beta = 0.27`$, $`p < 0.05`$), especially PXs and service clubs, with no detectable effects on integrating military units (Figure 6). Black soldiers in mixed camps additionally expressed greater tolerance of White commanders (A4: $`\beta = 0.11`$, $`p < 0.1`$), particularly Northern ones (Figure 6). For the full list of survey questions embedded in each index, see Section 7.2; for full regression tables, see Section 7.3. Turning to generalized outcomes, those in mixed camps were more optimistic about the treatment of Black Americans after the war (A2: $`\beta = 0.12`$, $`p < 0.01`$). Looking at individual items in index A2 (Figure 6), the effects are consistent across all five items — post-war rights and privileges, broad treatment by Whites, and future prospects for Black soldiers and civilians alike. Index A3 reinforces this picture, showing a significant effect on average ($`\beta = 0.21`$, $`p < 0.001`$), and on individual questions capturing Black soldiers’ assessments of the civilian and political institutions that governed their lives outside the military — including treatment by White military police, town police, and newspapers (Figure 6). Equal contact, in other words, did not merely update beliefs about fellow soldiers in the barracks: it reshaped Black soldiers’ expectations of the civilian world that awaited them after the war. We see evidence for emotional pathways underpinning this generalization — equal contact reduced worry and anxiety while increasing pride in one’s unit and overall morale (Figure 6).

image

Attitudes of Black Soldiers
Moving to the effects of unequal contact, Figure 1 and Table [tab:ww2_black_lt] show the effects of being assigned to a White lieutenant on Black soldiers’ attitudes. Among immediate outcomes, having an outgroup commander has no effect on overall military experience (A1: $`\beta = -0.01`$, $`p = 0.84`$), but significantly worsens race-specific attitudes within it. An item-level breakdown shows that perceived cooperation within a company, and the fairness of sergeants and corporals, decreases (A1, Figure 7). Agreeing that it is a ‘good idea’ to have integrated PX clubs, service clubs, and outfits falls sharply (A5: $`\beta = -0.06`$, $`p < 0.05`$) — a near-complete reversal of the mixed camp result, visible across all three items comprising the index. Index A3 shows a significant negative average effect ($`\beta = -0.11`$, $`p < 0.05`$), driven by worse promotion opportunities for Black officers (Figure 7). The null effect on lieutenant preference overall ($`\beta = -0.06`$, $`p = 0.29`$) masks the fact that Black soldiers with White lieutenants were more likely to prefer a Black commander, particularly a Northern one (A4, Figure 7). We see largely null effects of unequal contact on generalized outcomes. Optimism about post-war rights is largely unaffected (A2, $`\beta = 0.03`$, $`p = 0.59`$), indicating that the negative effects of unequal contact did not spill over into broader beliefs about racial progress after the war. Likewise, while A3 exhibits a negative effect overall, individual components capturing expectations about future treatment by White police and newspapers are unchanged (Figure 7). These results are robust to different codings of the treatment indicator, with similar findings when disaggregating the treatment into mix-race (‘some White and some Negro’) and all-White leadership (see Table [tab:black_main_alt] in the Appendix). Combining the two treatments, although underpowered, yields null interaction effects across the five indices.[^13] Shifting to White soldiers, Figure 2 and Table [tab:ww2_White_mixed_camp] show the effects of equal contact, while Figure 8 displays item-level results. The analyses reveal a striking gradient: treatment effects are largest on outcomes closest to the contact setting itself, and diminish — then reverse — as outcomes move further from it. Support for racial integration (B4) shows the largest effects in the entire analysis ($`\beta = 0.49`$, $`p < 0.001`$), with the biggest gains on integrating the institution directly targeted by the treatment, PX clubs, with smaller effects on integrating military outfits more broadly (Figure 8), and null effects on overall military experience (B1: $`\beta = -0.10`$, $`p = 0.16`$) and morale (B2: $`\beta = -0.05`$, $`p = 0.27`$). Among generalized outcomes, the slope reverses direction. White soldiers in mixed camps were less supportive of Black Americans’ post-war rights (B3: $`\beta = -0.07`$, $`p < 0.1`$). Item-level results show this is not driven solely by pessimistic expectations (Figure 2). The normative item on ‘desired post-war rights’ for Black Americans is negative while expected post-war rights increases, indicating that White soldiers in mixed camps expected more racial progress but wanted less of it. Index B1 hints at why. Equal contact lowered pride in one’s unit, perceptions that service was worthwhile, and perceptions of company-level cooperation, suggesting that integration had undermined the value of military service itself. Reflecting on the potential limits to contact, a White platoon sergeant from Indiana at the time opined: ‘They \[Black troops\] fought and I think more of them for it, but I still don’t want to soldier with them in garrison’ (Stouffer et al. 1949, pg. 592). A company commander in Tennessee confirmed that he observed ‘good cooperation in combat’ where Blacks were treated as ‘soldier to soldier,’ but that in garrisons the ‘strain on both parties is too great’ (Stouffer et al. 1949, pg. 592). White men could accept fighting on the front lines with Black troops, but they did not want to serve with them behind the lines (‘in garrison’), where social and sexual interactions were most common (Guglielmo 2021).

image

Attitudes of White Soldiers
0.68 l \*4Y & Asian Partner & Black Partner & Hispanic Partner & White Partner\ Vietnam Veteran & $`0.0084^{***}`$ & $`0.0004`$ & $`0.0032^{*}`$ & $`0.0569^{***}`$\ & $`(0.0012)`$ & $`(0.0006)`$ & $`(0.0014)`$ & $`(0.0129)`$\ R$`^2`$ & $`0.0050`$ & $`0.0034`$ & $`0.0052`$ & $`0.2607`$\ Adj. R$`^2`$ & $`0.0047`$ & $`0.0031`$ & $`0.0048`$ & $`0.2584`$\ Covariates & Y & Y & Y & Y\ N & $`25660`$ & $`24336`$ & $`25012`$ & $`2331`$\
***Note:*** \+ p $`<`$ 0.1, \* p $`<`$ 0.05, \*\* p $`<`$ 0.01, \*\*\* p $`<`$ 0.001. Covariates include age, race, and HS diploma. The WWII results establish that contact produces asymmetric effects. Equal contact improved immediate and generalized attitudes for Black soldiers while unequal contact had null to negative effects; for White soldiers, equal contact boosted immediate outcomes but undercut generalized support for Black rights. We now shift away from analyses of situational equality to structural equality, turning to the Vietnam data to ask whether greater racial progress leaves longer-run behavioral traces, and whether those traces are stronger for minorities than for majorities. Turning to the analyses of intermarriage rates among Vietnam War veterans, Table [tab:marriage] shows that veterans from non-White backgrounds were roughly 6 percentage-points more likely to marry Whites compared to men in the same age cohort who were not conscripted, non-Asians were 0.8 percentage-points more likely to marry Asians, and non-Hispanics were 0.3 percentage-points more likely to marry Hispanics. We find no effects on the likelihood of veterans of any ethnic background marrying Black partners (Table [tab:marriage]). The ‘veteran effect’ on inter-marriage is 1 percentage-point among Whites; six times smaller than the same rate for non-Whites (Table [tab:marriage_full_white]). Finally, we analyze thousands of veteran testimonies to deepen our understanding of how race was experienced across time and space by White and ethnic minority veterans, in Vietnam and WWII. Table [tab:transcript-summary] shows the salience of race during the interviews. Ethnic minority veterans discuss race far more than their White counterparts across all keywords. Black veterans are over five times as likely to mention race in any form (0.551 vs. 0.106), with the gap wider by an order of magnitude on segregation-related language (0.368 vs. 0.027). Asian veterans show comparably high salience (0.491), with elevated rates of discrimination-related language, possibly reflecting experiences of being racialized as the enemy in Vietnam.
l \*6Y Race & racial & racism & discri-mina-tion & segre-gation & colored & any race mention\ White & 0.015 & 0.004 & 0.015 & 0.027 & 0.060 & 0.106\ Black & 0.184 & 0.120 & 0.150 & 0.368 & 0.134 & 0.551\ Native & 0.082 & 0.046 & 0.087 & 0.027 & 0.050 & 0.233\ Asian & 0.139 & 0.074 & 0.270 & 0.170 & 0.100 & 0.491\ Hispanic & 0.048 & 0.028 & 0.077 & 0.054 & 0.035 & 0.188\ Pacific Islander & 0.086 & 0.000 & 0.057 & 0.057 & 0.029 & 0.171\ Other & 0.028 & 0.008 & 0.037 & 0.036 & 0.063 & 0.149\ Total (out of 25119) & 739 & 333 & 721 & 1310 & 1625 & 3613\
Breaking down experiences with race across space, Figure 3 shows that both Black and White veterans describe their experiences with race more positively within the military than in American society at large. This pattern mirrors our survey findings, in which both groups expressed stronger support for integration within the military than what they expected or hoped for in broader society (Figures 1 and 2). Black WWII veterans show only a slight difference in sentiment between the two settings, however, reflecting entrenched segregation in both domains at the time. Topic modeling further confirms that race was a more salient feature of veterans’ testimonials during WWII than Vietnam (Tables [topic-model-ww2] and [topic-model-vietnam]), consistent with the structural shifts in American racial equality between the two eras. The testimonials themselves illustrate this arc: John W. Mosley from Denver, a WWII veteran, recalled a swimming pool that ‘the Blacks would empty and fill up on Saturday, so the Whites would have clean water’ while William Henry Swift Jr., a Vietnam veteran from Pamplico, South Carolina, describes being one of the ‘first eleven Black students to walk in an all-White school in 1968’ — a marker of the slow, contested progress underway.
Sentiment of testimonies discussing the topic of race in the military vs. in non-military institutions.
Moving from WWII to the Vietnam War, we see a boost in Black veterans’ description of race in American society and an even greater spike in how they describe race in the military, a jump in sentiment from -0.6 to -0.4 (Figure 3). Reflecting on the rapid normalization of equality within the military from the 1930s to 1960s, Eugene Goolsby says: ‘And of course, the way the Army is today, it’s men and women and all sorts of people, and nobody gives a thought to the old idea of segregation... But I would think it’s your job, you who were the first to break the barrier, it’s your job to keep that memory alive... so that the Black Marines or the Black servicemen that follow you, appreciate what you have done.’ White veterans also experience a spike in positivity scores when describing the military in Vietnam vs. WWII (Figure 3). At the same time, White veterans exhibit a drop in their estimation of race in American society more broadly, despite the boost to Black rights. Table [tab:transcript-analysis-compare] shows that White veterans’ assessment of race relations in American was 19.6 percentage-points more negative in the Vietnam era relative to WWII $`p < 0.05`$) — while Blacks *improved* their assessment by the same magnitude (18.4 percentage-points, $`p < 0.001`$). # Discussion Our theoretical framework predicts that groups will respond to intergroup encounters in ways that seek to maximize their ethnic ranking. We find that minorities generalize from equal but not unequal contact, consistent with motivated updating and ruling out the risky misclassification hypothesis. A sampling breadth logic may well be at play, helping to explain why minorities generalize from contact with many peers but not individual commanders (Chakraborty et al. 2024). Troops enter the barracks through conscription, drawing from a broad cross-section of the population, whereas officers arrive via a selective process. Such a framework cannot, however, explain why groups were exposed to the same cross-section of outgroup peers, yet only Black soldiers updated their generalized beliefs — an asymmetry that instead points to the power-based logic at the core of motivated updating. Motivated updating suggests that power status shapes the psycho-social value of the contact signal itself: for Black soldiers, peer integration offered evidence that the racial hierarchy was permeable — a signal that manifested emotionally in reduced worry and anxiety, higher morale, and greater pride. For White soldiers, the same contact implied a threat to existing arrangements, triggering status-preservation responses that muted or reversed generalization. The emotional evidence implies that motivated updating is at least as important as cognitive, sample-based inference. A complete account requires both frameworks: sampling breadth to explain within-group variation across contact types, and structural power to explain cross-group variation under identical contact conditions. Consistent with a negative experience under White leadership predicted by motivated updating, we find that Black soldiers assigned to White commanders report a dimmer view of Black life in the military and an aversion to future White commanders. (Ang and Chinoy 2025) similarly documents deep institutional racism during WWI, which spurred civic activism among soldiers who faced the most severe discrimination. This link between perceived injustice and activism echoes Parker (2009)’s ((2009)) findings that deployment during WWII and the Korean War made stark the hypocrisy between America’s stated aim of defending democracy abroad and the reality of Jim Crow at home, radicalizing veterans against hierarchies even as wartime contact built horizontal cohesion among peers. Similarly, (Lyall 2020) shows that ethnic discrimination within military units drive battlefield failures and desertion. Why did majority-group members react to equality with hostility? The WWII survey data, intermarriage analyses, and testimonial evidence reinforce the idea that equality rattles the self-esteem of majority-group members while boosting that of minority-group members. The divergence is captured in veteran testimonials. Marvin Robinson, a Black veteran interviewed in New York City, describes the generalization process upon returning from Vietnam: “I changed my whole theory, \[my\] concept of ‘all Whites are the same’... not all Whites are segregated, not all Whites hate Blacks, not all Whites are going to trick us, not all Whites are going to be in the clan, KKK... there are ones who are as we are, kind, gentle, patient, able to work with whoever.” White veterans navigated the same era very differently, describing racial progress in more negative tones as structural change accelerated. One White veteran, interviewed in Knoxville, Tennessee, illustrates perceived racial threat: “Most of the Blacks out of the Northeast were horrible. They thought they were too good to be there. And they, in a few weeks, corrupt the Blacks out of the South the same way.” These results align with observational evidence that White veterans hold more negative racial attitudes than their civilian peers on average (Nteta and Tarsi 2016), and with experimental work showing that Vietnam draft lottery-induced contact improved interracial attitudes but failed to shift partisanship or vote choice in the long run (Green and Hyman-Metzger 2024). For majority-group members, then, equal contact produces localized tolerance in ways that do not disrupt the status quo, but triggers status-preserving responses when broader equality is at stake (Cramer 2016; Craig and Richeson 2014). Finally, why did Vietnam military service increase exogamy among Black veterans? Demand-side explanations seem unlikely. Vietnam veterans were not held in high esteem upon their return: the Vietnam War had far less public support than World War II, there were few to no public celebrations of returning veterans (Stern 1968), and the GI bill benefits for Vietnam veterans were less robust compared to those offered to earlier generations, limiting their access to education and training opportunities (Fendrich 1972). Contemporaneous accounts similarly conclude that Black Vietnam veterans were ‘returning to second-class citizenship’ rather than the ‘upward mobility in civilian life’ that their military experience promised them (Fendrich 1972). These accounts, coupled with endogamy among White men despite wartime contact increasing self-reported support for inter-marriage (Green and Hyman-Metzger 2024), lend support to the supply-side explanation: it was Blacks who changed their attitudes toward non-Blacks. Inter-marriage initiated only by lower-ranked ethnic groups should therefore be treated with caution as a marker of broader social cohesion. # Conclusions Decades of contact research have established that equal-status contact can reduce prejudice. This study asks for whom, and probes how far those effects travel. Our answer depends on where one stands in the racial hierarchy. We propose that minorities see inequality as a threat; majorities, equality. This divergence is sharpest in generalized beliefs and behaviors: both groups register positive effects within the contact setting, but only minorities extend that openness beyond it. For Black soldiers, peer integration was evidence of a permeable hierarchy; for White soldiers, a threat to a favorable one. An amendment to the contact hypothesis is therefore in order: the probability of generalization rests on *perceptions* of threat, just as much as actual threat. Prior studies show that numeric equality triggers status threat among majority-groups (Ghosh et al. 2024; Enos 2014). We add that structural equality functions in the same way. Prejudice reduction programs typically aim to soften the hostility of dominant groups toward those they exclude. Our findings invert this framing: it is minorities who generalize from equal contact, update their beliefs, and, as the Vietnam evidence shows, act on that openness in their most intimate choices. Social cohesion may therefore be built on minority willingness to extend trust despite facing the highest costs of misplaced faith, while the dominant group withholds the same openness even when doing so is relatively cheap. Future research should pursue two questions this study cannot answer. First, does minority trust come at the expense of collective action? Work on the ‘irony of harmony’ warns that increased openness can demobilize minorities, generating false expectations of equality, and eroding the ingroup solidarity needed to press for structural change (Saguy et al. 2009; Saguy and Kteily 2014; Cocco et al. 2024). Second, how do majority-group members, like the White lieutenants in our study, respond to hierarchical contact? Unequal contact may satisfy status preservation needs in the immediate domain, freeing majority-group members to support structural change more broadly. Alternatively, it may entrench resistance by reinforcing a social order they already benefit from. The contact literature’s optimism may rest, in part, on studies conducted in more settled racial climates, leading contact scholars to conclude that equal status is not necessary for positive effects (Tropp and Pettigrew 2005). The divergent effects documented here may reflect the fact that equal contact occurred as the racial order was actively contested, and increasing integration sent credible signals to both groups about an uncertain future. Effect sizes from more stable periods may therefore systematically underestimate majority backlash and overestimate the social cohesion that contact produces. For instance, Calderon et al. (2023) find that White mobilization for Black rights during the Great Migration was conditional on Black migrants posing no economic threat — and even then, the Black population share reached only 15% on average. Institutional design should account for this, and for the asymmetric burdens contact places on minorities (Paluck and Clark 2020). Interventions that affirm minority voice and agency, for instance, may sustain generalization without its demobilizing side-effects (Bruneau and Saxe 2012), while improving the quality of majority-group contact may reduce the backlash that equal-status interaction currently provokes (Carrell et al. 2019). All in all, our results indicate that institutions can mandate equal contact, but they cannot mandate equal generalization. **Appendix *for***
**Equal Contact, Unequal Lessons:\ Motivated Updating, Power, and Race in the U.S. Military**\
**[A. World War II Veterans](#sec:A_survey_cluster)** 1. [A.1. Survey Sampling Procedure](#sec:sampling_procedure) 2. [A.2. Constructing Survey Indices](#sec:questions) 3. [A.3. Soldier Attitudes](#sec:A_ww2_main_tab) 4. [A.4. Results on Individual Questions](#sec:A_individual_questions) 5. [A.5. Alternative Measure of Outgroup Lieutenant](#sec:A_alt_lieut) 6. [A.6. Results from Black Soldier Subsample](#sec:subsample) 7. [A.7. Treatment status by camp](#sec:A_treat_status) 8. [A.8. Subgroup balance - mixed PX/club, Black respondents](#sec:balance) 9. [A.9. Subgroup balance - lieutenant race, Black respondents](#sec:A_black_balance_lieut) 10. [A.10. Subgroup balance - mixed PX/club, White respondents](#sec:A_white_balance_treat) [B. Additional Results: Vietnam War Veterans](#sec:A_inter_marriage) 1. [B.1. Covariate Balance](#sec:A_inter_marriage_balance) 2. [B.2. Interracial Marriage by Race Subgroup (White Veterans)](#sec:A_inter_marriage_race) [C. Veteran Testimonies and Interviews](#sec:A_transcript) 1. [C.1. Topic Modeling of Veteran Transcripts](#sec:A_transcript_topic) 2. [C.2. Racial Experience of Veterans: Full Result](#sec:vet_exp) 3. [C.3. Comparing Experience across the Two Wars](#sec:A_transcript_exp)
# World War II Veterans ## Survey Sampling Procedure The sampling procedure for the surveys on World War II soldiers used a two-step process. First, organizations were selected using purposive or quota sampling, ensuring representation of each branch of service in proper proportions based on secret data about troop strength and locations. Units were stratified by branch and type, with practical considerations often preventing strictly random selection. Second, individuals within units were sampled using probability methods, typically selecting every nth man from a duty roster or randomly sampling from soldier form 20 cards when specific demographic criteria were needed, such as rank and age. Questionnaires were administered anonymously to groups of about fifty enlisted men, with individual interviews conducted for those with reading difficulties. Sample adequacy was verified afterward by checking for biases in variables like rank, service time, and age, with corrective samples drawn when necessary. Based on the sampling procedure, 4793 White soldiers and 3000 Black soldiers were surveyed. An additional 4442 Black soldiers were surveyed See for more details. ## Constructing Survey Indices The surveys include 77 questions for Black soldiers and 64 questions for White soldiers. We first select attitudinal questions on millitary experience (e.g., whether the respondent is proud of their company) and race (e.g., support for integrated service clubs) and drop other questions (e.g., how much longer do you think this war will last). In total, we selected 26 questions from the survey on Black soldiers and 16 questions from the survey on White soldiers. We then use hierarchical clustering to group the selected questions. The scree plots from the hierarchical clustering analysis are shown in Figure 4 and Figure 5.
Scree plot for survey on the attitudes of Black soldiers
Scree plot for survey on the attitudes of White soldiers
For the survey on the attitudes of Black soldiers, We choose 5 indices based on the scree plot and manually label them according to the questions in each index: - Index A1: Opinions on respondent’s general military experience - Index A2: Whether rights and privileges of Black Americans should and will improve after the war - Index A3: Treatment of Black Americans in the military and in society - Index A4: Racial preference for lieutenants - Index A5: Support for racial integration in the military For the survey on the attitudes of White soldiers, We choose 4 indices based on the scree plot and manually label them according to the questions in each index: - Index B1: Opinions on respondent’s general military experience - Index B2: Opinions on respondent’s morale in the military - Index B3: Whether rights and privileges of Black Americans should and will improve after the war - Index B4: Support for racial integration in the military Note that the questions included in Index A5 and Index B4 are the same. | Index | Cronbach’s alpha | |:------|:----------------:| | A1 | 0.753 | | A2 | 0.847 | | A3 | 0.724 | | A4 | 0.540 | | A5 | 0.809 | | B1 | 0.688 | | B2 | 0.623 | | B3 | 0.371 | | B4 | 0.745 | For each index, we use Cronbach’s alpha to check its internal consistency. Indices A1, A2, A3, A5, and B4 have Cronbach’s alpha values above $`0.70`$ whereas indices A4, B1, B2, and B3 have Cronbach’s alpha values below $`0.70`$. A high value ($`\ge 0.70`$) of Cronbach’s alpha indicates high consistency within an index. The indices are not uniformly correlated, making a single omitted confounder unlikely to explain the full pattern of results. As Table 2 shows, A1–A3 are moderately positively correlated ($`r = 0.34`$–$`0.43`$), while A5 (Military Integration) is negatively correlated with all other indices ($`r = -0.04`$ to $`-0.22`$). A confounder driving bias in the same direction across all indices would need to simultaneously push A1–A3 and A5 in opposite directions — an implausible requirement for any single background characteristic.
| | **A1** | **A2** | **A3** | **A4** | **A5** | |:---|---:|---:|---:|---:|---:| | A1: Mil. Exp. | 1.000 | 0.337 | 0.431 | 0.093 | **$`-`$0.125** | | A2: Blk. Rights | 0.337 | 1.000 | 0.342 | 0.030 | **$`-`$0.043** | | A3: Blk. Treat. | 0.431 | 0.342 | 1.000 | 0.207 | **$`-`$0.216** | | A4: Lt. Pref. | 0.093 | 0.030 | 0.207 | 1.000 | **$`-`$0.047** | | A5: Mil. Integ. | **$`-`$0.125** | **$`-`$0.043** | **$`-`$0.216** | **$`-`$0.047** | 1.000 | Pairwise correlations among Black soldier indices
*Note:* All pairwise correlations are statistically significant ($`p < 0.03`$).
## Soldier Attitudes Tables [tab:ww2_black_mixed_camp] and [tab:ww2_black_lt] present the attitudes of Black soldiers (Figure 1) by (1) segregated vs. mixed camp and (2) in-group vs. out-group lieutenant, respectively. Table [tab:ww2_White_mixed_camp] presents the attitudes of White soldiers (Figure 2) by segregated vs. mixed camp. Because of the large number of indicator control variables, estimates on the control variables are omitted.
X \*5Y & Military Experience & Black Rights & Black Treatment & Lieutenant Preference & Military Integration\ & Index A1 & Index A2 & Index A3 & Index A4 & Index A5\ Mixed camp & 0.184\*\*\* & 0.118\*\* & 0.208\*\*\* & 0.105+ & 0.273\*\ & (0.028) & (0.036) & (0.049) & (0.058) & (0.104)\ Num.Obs. & 3804 & 3804 & 3804 & 3804 & 3804\ Std.Err. & clustered & clustered & clustered & clustered & clustered\ Covariates & Yes & Yes & Yes & Yes & Yes\
***Note:*** \+ p $`<`$ 0.1, \* p $`<`$ 0.05, \*\* p $`<`$ 0.01, \*\*\* p $`<`$ 0.001. All outcomes are standardized. Covariates include age, education, army branch, army duty, army camp, manner of enrollment, marital status, U.S. state of residence before army, whether residence is urban or rural, and years of service. Estimates on the covariates are omitted due to the large number of indicator variables. Standard errors are clustered at the army camp level. The number of observations is smaller than the sample size due to missing data for some survey questions.
X \*5Y & Military Experience & Black Rights & Black Treatment & Lieutenant Preference & Military Integration\ & Index A1 & Index A2 & Index A3 & Index A4 & Index A5\ Out-group Lt. & −0.012 & 0.026 & −0.106\* & −0.064 & −0.063\*\ & (0.058) & (0.048) & (0.046) & (0.060) & (0.027)\ Num.Obs. & 4113 & 4113 & 4113 & 4113 & 4113\ Std.Err. & clustered & clustered & clustered & clustered & clustered\ Covariates & Yes & Yes & Yes & Yes & Yes\
***Note:*** \+ p $`<`$ 0.1, \* p $`<`$ 0.05, \*\* p $`<`$ 0.01, \*\*\* p $`<`$ 0.001. All outcomes are standardized. Covariates include age, education, army branch, army duty, army camp, manner of enrollment, marital status, U.S. state of residence before army, whether residence is urban or rural, and years of service. Estimates on the covariates are omitted due to the large number of indicator variables. Standard errors are clustered at the army camp level. The number of observations is smaller than the sample size due to missing data for some survey questions.
X \*4Y & Military Experience & Military Morale & Black Rights & Military Integration\ & Index B1 & Index B2 & Index B3 & Index B4\ Mixed camp & −0.102 & −0.050 & −0.074+ & 0.489\*\*\*\ & (0.072) & (0.045) & (0.034) & (0.093)\ Num.Obs. & 2485 & 2485 & 2485 & 2485\ Std.Err. & clustered & clustered & clustered & clustered\ Covariates & Yes & Yes & Yes & Yes\
***Note:*** \+ p $`<`$ 0.1, \* p $`<`$ 0.05, \*\* p $`<`$ 0.01, \*\*\* p $`<`$ 0.001. All outcomes are standardized. Covariates include age, education, army branch, army duty, army camp, manner of enrollment, marital status, U.S. state of residence before army, whether residence is urban or rural, and years of service. Estimates on the covariates are omitted due to the large number of indicator variables. Standard errors are clustered at the army camp level. The number of observations is smaller than the sample size due to missing data for some survey questions. ## Results on Individual Questions We plot the mean values of survey responses of each question by group in Figure 6, Figure 7, and Figure 8. To put all questions on the same scale, we scale each question by dividing it by its maximum value so that mean values range from $`0`$ to $`1`$.
Mean values of survey responses (Black soldiers); Treatment = Mixed Camp
Mean values of survey responses (Black soldiers); Treatment = White Lieutenant
Mean values of survey responses (White soldiers); Treatment = Mixed Camp
## Alternative Measure of Outgroup Lieutenant The survey on Black soldiers asks whether they have 1) all Black Lieutenants, 2) all White Lieutenants, or 3) a mix of Black and White Lieutenants. In the main analysis, we group 2) and 3) into one treatment indicator, essentially treating the two conditions as the same. Table [tab:black_main_alt] instead presents results when the conditions are treated separately. Table [tab:black_main_alt] shows that having all White lieutenants and having some White lieutenants have similar effects on the attitudes of Black soldiers - soldiers in either condition reported lower assessment of the treatment of Black Americans and expressed lower support for military integration.
X \*5Y & Military Experience & Black Rights & Black Treatment & Lieutenant Preference & Military Integration\ & Index A1 & Index A2 & Index A3 & Index A4 & Index A5\ All White & −0.057 & 0.036 & −0.172\*\* & −0.030 & −0.079+\ Lieutenant & (0.089) & (0.050) & (0.057) & (0.066) & (0.041)\ Mixed & 0.006 & 0.022 & −0.080+ & −0.078 & −0.056+\ Lieutenant & (0.063) & (0.051) & (0.042) & (0.064) & (0.029)\ Num.Obs. & 4113 & 4113 & 4113 & 4113 & 4113\ Std.Err. & clustered & clustered & clustered & clustered & clustered\ Covariates & Yes & Yes & Yes & Yes & Yes\
***Note:*** \+ p $`<`$ 0.1, \* p $`<`$ 0.05, \*\* p $`<`$ 0.01, \*\*\* p $`<`$ 0.001. All outcomes are standardized. Covariates include age, education, army branch, army duty, army camp, manner of enrollment, marital status, U.S. state of residence before army, whether residence is urban or rural, and years of service. Estimates on the covariates are omitted due to the large number of indicator variables. Standard errors are clustered at the army camp level. The number of observations is smaller than the sample size due to missing data for some survey questions. ## Results from Black Soldier Subsample Tables [tab:black_main_sub1] and [tab:black_main_sub2] show the effects of being in mixed camps and having White lieutenants on the attitudes of Black soldiers using the subsample. Because the survey questions are not forced choice, many of the survey responses are missing. Therefore, although the entire subsample was purported representative, dropping those with missing values reduces the sample from n = 3000 to n = 1435 when analyzing the effects of being in mixed camps and n = 1565 when analyzing the effects of having White lieutenants. Despite the substantial reduction in sample size, the substantive conclusions are similar. As Table [tab:black_main_sub1] shows, Black soldiers in mixed camps reported higher assessment of general military experience (Index A1) and treatment of Black Americans (Index A3), as well as a greater support for military integration (Index A5). The only difference with the main result is that the effect on the prospect of the rights and privileges of Black Americans after the war is smaller and no longer statistically significant. The effects of having White lieutenants are similar to the main results. Table [tab:black_main_sub2] shows that Black soldiers with White lieutenants reported lower support for military integration (Index A5). Black soldiers with White lieutenants also reported lower assessment of treatment of Black Americans but the estimate is less precise and thus not statistically significant.
X \*5Y & Military Experience & Black Rights & Black Treatment & Lieutenant Preference & Military Integration\ & Index A1 & Index A2 & Index A3 & Index A4 & Index A5\ Mixed camp & 0.154\*\* & 0.044 & 0.285\*\*\* & 0.184+ & 0.250\*\ & (0.039) & (0.056) & (0.072) & (0.088) & (0.105)\ Num.Obs. & 1435 & 1435 & 1435 & 1435 & 1435\ Std.Err. & clustered & clustered & clustered & clustered & clustered\ Covariates & Yes & Yes & Yes & Yes & Yes\
***Note:*** \+ p $`<`$ 0.1, \* p $`<`$ 0.05, \*\* p $`<`$ 0.01, \*\*\* p $`<`$ 0.001. All outcomes are standardized. Covariates include age, education, army branch, army duty, army camp, manner of enrollment, marital status, U.S. state of residence before army, whether residence is urban or rural, and years of service. Estimates on the covariates are omitted due to the large number of indicator variables. Standard errors are clustered at the army camp level. The number of observations is smaller than the sample size due to missing data for some survey questions.
X \*5Y & Military Experience & Black Rights & Black Treatment & Lieutenant Preference & Military Integration\ & Index A1 & Index A2 & Index A3 & Index A4 & Index A5\ Out-group Lt. & 0.067 & −0.025 & −0.069 & −0.193+ & −0.171\*\ & (0.150) & (0.092) & (0.070) & (0.099) & (0.071)\ Num.Obs. & 1565 & 1565 & 1565 & 1565 & 1565\ Std.Err. & clustered & clustered & clustered & clustered & clustered\ Covariates & Yes & Yes & Yes & Yes & Yes\
***Note:*** \+ p $`<`$ 0.1, \* p $`<`$ 0.05, \*\* p $`<`$ 0.01, \*\*\* p $`<`$ 0.001. All outcomes are standardized. Covariates include age, education, army branch, army duty, army camp, manner of enrollment, marital status, U.S. state of residence before army, whether residence is urban or rural, and years of service. Estimates on the covariates are omitted due to the large number of indicator variables. Standard errors are clustered at the army camp level. The number of observations is smaller than the sample size due to missing data for some survey questions. ## Treatment status by camp Treatment is defined as respondent being in either a mixed PX or service club or both.
| | Huachuca | Northern | Southern | Tuskegee | |:-----------|:--------:|:--------:|:--------:|:--------:| | Segregated | 107 | 462 | 2421 | 68 | | Mixed | 1010 | 407 | 355 | 118 | Service club/PX by camp, Black respondents
Treatment is defined as respondent having at least one lieutenant that was White (i.e., or ).
| | Huachuca | Northern | Southern | Tuskegee | |:----------|:--------:|:--------:|:--------:|:--------:| | All black | 184 | 344 | 76 | 153 | | Mixed | 993 | 626 | 2937 | 97 | Lieutenant race by camp, Black respondents
Treatment is defined as respondent being in either a mixed PX or service club or both.
| | Atlantic City | Northern | Southern | |:-----------|:-------------:|:--------:|:--------:| | Segregated | 35 | 474 | 1353 | | Mixed | 141 | 276 | 552 | Service club/PX by camp, White respondents
Tables [tab:club_camp_black_detail], [tab:lieutenant_camp_black_detail], and [tab:club_camp_white_detail] report the same treatment assignment broken down by individual camp. \ \ ## Subgroup balance - mixed PX/club, Black respondents Treatment is defined as being in a mixed PX or service club or both. Coefficients are difference in means between the treatment and the control groups. Robust standard errors.
Covariate Balance; Mixed vs. segregated PX/service club
## Subgroup balance - lieutenant race, Black respondents Treatment is defined as respondent having at least one lieutenant that was White (i.e., or ). Coefficients are difference in means between the treatment and the control groups. Robust standard errors.
Covariate Balance; Lieutenants race
## Subgroup balance - mixed PX/club, White respondents Treatment is defined as respondent being in either a mixed PX or service club or both. Coefficients are difference in means between the treatment and the control groups. Robust standard errors.
Covariate Balance; Mixed vs. segregated PX/service club
# Additional Results: Vietnam War Veterans ## Covariate Balance Figure 9 shows the covariate balance between veterans and civilians in the sample. Across all covariates, there is little difference between the two groups.

image

Covariate Balance, 1994-2022
## Interracial Marriage by Race Subgroup (White Veterans) Table [tab:marriage_full_white] reports OLS estimates of the association between Vietnam War veteran status and interracial marriage by partner race, estimated for White respondents (men born 1950–1952, married at time of survey, 1994–2022). Regressions are weighted by ASECWT and include age and high school diploma as covariates.
l \*4Y & Asian Partner & Black Partner & Hispanic Partner & Any Outgroup Partner\ Vietnam Veteran & $`0.0082^{***}`$ & $`0.0008`$ & $`0.0039^{**}`$ & $`0.0110^{***}`$\ & $`(0.0013)`$ & $`(0.0006)`$ & $`(0.0014)`$ & $`(0.0018)`$\ Age & $`0.0000`$ & $`0.0000`$ & $`0.0001`$ & $`0.0001`$\ & $`(0.0001)`$ & $`(0.0000)`$ & $`(0.0001)`$ & $`(0.0001)`$\ High School & $`0.0059^{*}`$ & $`0.0003`$ & $`0.0038`$ & $`0.0034`$\ & $`(0.0025)`$ & $`(0.0012)`$ & $`(0.0029)`$ & $`(0.0035)`$\ R$`^2`$ & $`0.0021`$ & $`0.0001`$ & $`0.0005`$ & $`0.0021`$\ Adj. R$`^2`$ & $`0.0020`$ & $`-0.0000`$ & $`0.0003`$ & $`0.0020`$\ Num. Obs. & $`23{,}634`$ & $`23{,}634`$ & $`22{,}735`$ & $`23{,}634`$\
***Note:*** Respondents are White male veterans and civilians born 1950–1952, married at time of survey, 1994–2022. Regressions are weighted by ASECWT. White partner is the omitted category. # Veteran Testimonies and Interviews ## Topic Modeling of Veteran Transcripts We use topic modeling - latent Dirichlet allocation (LDA) - to model the topic distribution of transcripts of Vietnam War and World War II veterans, respectively. Tables [topic-model-vietnam] and [topic-model-ww2] show the top words associated with each topic. Topic 2 for the Vietnam War transcripts and topic 4 for the World War II transcripts feature words that are associated with race. However, topic 4 for the World War II transcripts feature many more top words related to race (e.g., black, White, tuskegee) than topic 2 for the Vietnam War transcripts. ## Racial Experience of Veterans: Full Result ## Comparing Experience across the Two Wars In the main text, Figure 3 shows the racial experience by war, racial group, and setting (whether experiences took place within the military, or elsewhere). Here, Figure 10 (based on Table [tab:transcript-analysis-compare]) shows the differences in racial experience (ratings) across the two wars for each racial group and setting. Compared to World War II, Black veterans reported more positive racial experience both in the U.S. military and in U.S. society during the Vietnam War. For White veterans, their racial experience in the U.S. military during the Vietnam War was more positive than during World War II. In contrast, their racial experience in U.S. society during the Vietnam War was significantly more negative compared to World War II, with the mean difference of around -0.2 (on a scale of -1 to +1).

image

Differences in Racial Experience across the Two Wars
***Note:*** \+ p $`<`$ 0.1, \* p $`<`$ 0.05, \*\* p $`<`$ 0.01, \*\*\* p $`<`$ 0.001.
Abramitzky, Ran, Leah Boustan, and Katherine Eriksson. 2020. “Do Immigrants Assimilate More Slowly Today Than in the Past?” *American Economic Review: Insights* 2 (1): 125–41.
Allport, Gordon Willard, Kenneth Clark, and Thomas Pettigrew. 1954. *The Nature of Prejudice*.
Ang, Desmond, and Sahil Chinoy. 2025. *Vanguard: Black Veterans and Civil Rights After World War i*. National Bureau of Economic Research.
Angrist, Joshua D. 1990. “Lifetime Earnings and the Vietnam Era Draft Lottery: Evidence from Social Security Administrative Records.” *The American Economic Review*, 313–36.
Board, New York Times Editorial. 2017. “The Army’s Racial Reckoning.” *The New York Times*. .
Boin, Jessica, Mirjana Rupar, Sylvie Graf, Sybille Neji, Olivia Spiegler, and Hermann Swart. 2021. “The Generalization of Intergroup Contact Effects: Emerging Research, Policy Relevance, and Future Directions.” *Journal of Social Issues* 77 (1): 105–31.
Booker, Simeon. 1962. “An American Is an America.” *Jet Magazine*, June 30.
Brown, Rupert, and Miles Hewstone. 2005. “An Integrative Theory of Intergroup Contact.” *Advances in Experimental Social Psychology* 37 (37): 255–343.
Bruneau, Emile, and Rebecca Saxe. 2012. “The Power of Being Heard: The Benefits of ‘Perspective-Giving’ in the Context of Intergroup Conflict.” *Journal of Experimental Social Psychology* 48 (4): 855–66.
Calderon, Alvaro, Vasiliki Fouka, and Marco Tabellini. 2023. “Racial Diversity and Racial Policy Preferences: The Great Migration and Civil Rights.” *The Review of Economic Studies* 90 (1): 165–200.
Card, David, and Thomas Lemieux. 2001. “Going to College to Avoid the Draft: The Unintended Legacy of the Vietnam War.” *American Economic Review* 91 (2): 97–102.
Carrell, Scott E, Mark Hoekstra, and James E West. 2019. “The Impact of College Diversity on Behavior Toward Minorities.” *American Economic Journal: Economic Policy* 11 (4): 159–82.
Cederman, Lars-Erik, Andreas Wimmer, and Brian Min. 2010. “Why Do Ethnic Groups Rebel? New Data and Analysis.” *World Politics* 62 (1): 87–119.
Chakraborty, Anujit, Arkadev Ghosh, Matt Lowe, and Gareth Nellis. 2024. *Learning about Outgroups: The Impact of Broad Versus Deep Interactions*.
Charnysh, Volha. 2019. “Diversity, Institutions, and Economic Outcomes: Post-WWII Displacement in Poland.” *American Political Science Review* 113 (2): 423–41.
Clark, James Matthew. 1980. *A Review of the Historical Development and Contemporary Management Structure of Military Club Systems*.
Cocco, Veronica Margherita, Loris Vezzali, Sofia Stathi, Gian Antonio Di Bernardo, and John F Dovidio. 2024. “Mobilizing or Sedative Effects? A Narrative Review of the Association Between Intergroup Contact and Collective Action Among Advantaged and Disadvantaged Groups.” *Personality and Social Psychology Review* 28 (2): 119–80.
Craig, Maureen A, and Jennifer A Richeson. 2014. “On the Precipice of a ‘Majority-Minority’ America: Perceived Status Threat from the Racial Demographic Shift Affects White Americans’ Political Ideology.” *Psychological Science* 25 (6): 1189–97.
Cramer, Katherine J. 2016. *The Politics of Resentment: Rural Consciousness in Wisconsin and the Rise of Scott Walker*. University of Chicago Press.
Dixon, John, Kevin Durrheim, and Colin Tredoux. 2005. “Beyond the Optimal Contact Strategy: A Reality Check for the Contact Hypothesis.” *American Psychologist* 60 (7): 697–711.
Ellison, Christopher G. 1992. “Military Background, Racial Orientations, and Political Participation Among Black Adult Males.” *Social Science Quarterly*.
Enos, Ryan D. 2014. “Causal Effect of Intergroup Contact on Exclusionary Attitudes.” *Proceedings of the National Academy of Sciences* 111 (10): 3699–704.
Enos, Ryan D. 2017. *The Space Between Us: Social Geography and Politics*. Cambridge University Press.
Ensari, Nurcan, and Norman Miller. 2002. “The Out-Group Must Not Be so Bad After All: The Effects of Disclosure, Typicality, and Salience on Intergroup Bias.” *Journal of Personality and Social Psychology* 83 (2): 313.
Erikson, Robert S, and Laura Stoker. 2011. “Caught in the Draft: The Effects of Vietnam Draft Lottery Status on Political Attitudes.” *American Political Science Review* 105 (2): 221–37.
Fendrich, James M. 1972. “The Returning Black Vietnam-Era Veteran.” *Social Service Review* 46 (1): 60–75.
Flood, Sarah, and José Pacas. 2016. “Using the Annual Social and Economic Supplement with Current Population Survey Panels.” *Minnesota Population Center WP* 4.
Ghosh, Arkadev, Prerna Kundu, Matt Lowe, and Gareth Nellis. 2024. *Creating Cohesive Communities: A Youth Camp Experiment in India*.
Gitre, Edward J. K., Brandon L. Kramer, Chase Dawson, and Gizem Korkmaz. 2023. “Across the Color Line.” *Current Research in Digital History* 6 (May). .
Goodwin, Gerald F. 2023. *Race in the Crucible of War: African American Servicemen and the War in Vietnam*. University of Massachusetts Press.
Granovetter, Mark. 1983. “The Strength of Weak Ties: A Network Theory Revisited.” *Sociological Theory*, 201–33.
Green, Donald P, and Oliver Hyman-Metzger. 2024. “The Vietnam Draft Lottery and Whites’ Racial Attitudes: Evidence from the General Social Survey.” *American Political Science Review*, 1–8.
Greene, Kenneth F, Erin L Rossiter, Enrique Seira, and Alberto Simpser. 2024. “Interacting as Equals Reduces Partisan Polarization in Mexico.” *Nature Human Behaviour*, 1–9.
Guglielmo, Thomas A. 2021. *Divisions: A New History of Racism and Resistance in America’s World War II Military*. Oxford University Press.
Hampton II, Isaac. 2012. *The Black Officer Corps: A History of Black Military Advancement from Integration Through Vietnam*. Routledge.
Horowitz, Donald L. 1985. *Ethnic Groups in Conflict*. University of California Press.
Huff, Connor, and Robert Schub. 2021. “Segregation, Integration, and Death: Evidence from the Korean War.” *International Organization* 75 (3): 858–79.
Hultman, Lisa, and Salma Mousa. 2025. “From Ceasefire to Cohesion: An Integrated Review of Peacemaking and Peacebuilding.” *Economic Policy* 40 (124): 931–67.
Johnson, Thomas A. 1968. “Negroes in ‘the Nam’.” *Ebony Magazine*, August.
Knauer, Christine. 2023. “‘With a Stroke of a Pen’: Executive Order 9981 in American Memory.” *History & Memory* 35 (2): 41–78.
Lanning, Michael Lee. 2004. *The African-American Soldier: From Crispus Attucks to Colin Powell*. Citadel Press.
Lee, Ulysses. 1968. “The Draft and the Negro.” *Current History* 55 (323): 28–48.
Lowe, Matt. 2021. “Types of Contact: A Field Experiment on Collaborative and Adversarial Caste Integration.” *American Economic Review* 111 (6): 1807–44.
Lowe, Matt. 2025. “Has Intergroup Contact Delivered?” *Annual Review of Economics* 17.
Lucks, Daniel. 2017. “African American Soldiers and the Vietnam War: No More Vietnams.” *The Sixties* 10 (2): 196–220.
Lyall, Jason. 2020. *Divided Armies: Inequality and Battlefield Performance in Modern War*.
Mousa, Salma. 2020. “Building Social Cohesion Between Christians and Muslims Through Soccer in Post-ISIS Iraq.” *Science* 369 (6505): 866–70.
Mousa, Salma, Alexandra Scacco, and Lennard Naumann. 2025. “Intergroup Contact, Empathy Education, and Refugee-Native Integration in Lebanon.” *The Economic Journal*.
Newby, Jeffrey L. I. 2004. “The Fight for the Right to Fight and the Forgotten Negro Protest Movement: The History of Executive Order 9981 and Its Effect Upon Brown v. Board of Education and Beyond Note.” *Texas Journal on Civil Liberties & Civil Rights* 10 (1): 83–110.
Nteta, Tatishe M, and Melinda R Tarsi. 2016. “Self-Selection Versus Socialization Revisited: Military Service, Racial Resentment, and Generational Membership.” *Armed Forces & Society* 42 (2): 362–85.
Paluck, Elizabeth Levy, and Chelsey S Clark. 2020. “Can Playing Together Help Us Live Together?” *Science* 369 (6505): 769–70.
Paluck, Elizabeth Levy, Seth A Green, and Donald P Green. 2019. “The Contact Hypothesis Re-Evaluated.” *Behavioural Public Policy* 3 (2): 129–58.
Paluck, Elizabeth Levy, Roni Porat, Chelsey S Clark, and Donald P Green. 2021. “Prejudice Reduction: Progress and Challenges.” *Annual Review of Psychology* 72 (1): 533–60.
Parker, Christopher S. 2009. “When Politics Becomes Protest: Black Veterans and Political Activism in the Postwar South.” *The Journal of Politics* 71 (1): 113–31.
Peretz, Pauline. 2024. *A Black Army: Segregation and the U.S. Military at Fort Huachuca, Arizona, 1941–1945*. Military, War, and Society in Modern American History. Cambridge University Press.
Pettigrew, Thomas F, and Linda R Tropp. 2008. “How Does Intergroup Contact Reduce Prejudice? Meta-Analytic Tests of Three Mediators.” *European Journal of Social Psychology* 38 (6): 922–34.
Phelps, Edmund S. 1972. “The Statistical Theory of Racism and Sexism.” *The American Economic Review* 62 (4): 659–61.
Porat, Roni, Larry Sarit, John-Henry Pezzuto, Miriam Poupko, and Devorah Manekin. 2024. *Collaboration Across Group Lines: Evidence from Two Field Experiments in Jerusalem*.
Rousseau, David L. 2021. *War and Rights: The Impact of War on Political and Civil Rights*. University of Michigan Press.
Saguy, Tamar, and Nour Kteily. 2014. “Intergroup Contact Can Undermine Collective Action Among the Disadvantaged: The Role of Power Asymmetries.” *Journal of Social Issues* 70 (1): 1–17.
Saguy, Tamar, Nicole Tausch, John F Dovidio, and Felicia Pratto. 2009. “The Irony of Harmony: Intergroup Contact Can Produce False Expectations for Equality.” *Psychological Science* 20 (1): 114–21.
Scacco, Alexandra, and Shana S Warren. 2018. “Can Social Contact Reduce Prejudice and Discrimination? Evidence from a Field Experiment in Nigeria.” *American Political Science Review* 112 (3): 654–77.
Stern, Sol. 1968. “When the Black g.i. Comes Back from Vietnam.” *New York Times*, March 24.
Stouffer, Samuel A, Edward A Suchman, Leland C DeVinney, Shirley A Star, and Robin M Williams Jr. 1949. *The American Soldier: Adjustment During Army Life, Vol. 1*.
Taber, Charles S, and Milton Lodge. 2006. “Motivated Skepticism in the Evaluation of Political Beliefs.” *American Journal of Political Science* 50 (3): 755–69.
Tropp, Linda R, and Thomas F Pettigrew. 2005. “Relationships Between Intergroup Contact and Prejudice Among Minority and Majority Status Groups.” *Psychological Science* 16 (12): 951–57.
Uslaner, Eric M. 2002. “The Moral Foundations of Trust.” *Available at SSRN 824504*.
War Department. 1944. *Table of Organization and Equipment No. 7-17: Infantry Rifle Company*. War Department publication.
Weiss, Chagai M. 2021. “Diversity in Health Care Institutions Reduces Israeli Patients’ Prejudice Toward Arabs.” *Proceedings of the National Academy of Sciences* 118 (14): e2022634118.
Wimmer, Andreas. 2008. “The Making and Unmaking of Ethnic Boundaries: A Multilevel Process Theory.” *American Journal of Sociology* 113 (4): 970–1022.
Wimmer, Andreas, Lars-Erik Cederman, and Brian Min. 2009. “Ethnic Politics and Armed Conflict: A Configurational Analysis of a New Global Data Set.” *American Sociological Review* 74 (2): 316–37.
[^1]: Assistant Professor, Department of Political Science, UCLA. Email: smousa@ucla.edu [^2]: Assistant Professor, Department of Political Science, Purdue University. Email: eddieyang@purdue.edu [^3]: We thank Tom Gugliemo, Alfredo Gonzalez, David Laitin, Daniel Posner, Jake Grumbach, Lachlan McNamee, David Rousseau, Jane Esberg, and Ed Gitre for their thoughtful feedback. All errors are our own. [^4]: In a June 1945 survey of 60 officers and 195 enlisted men in Europe, 98% of respondents believed that Black soldiers should be used as infantry troops — but only 4% believed that White and Black soldiers should serve in the same platoon (Stouffer et al. 1949). [^5]: Black training and military camps were located primarily in Tuskegee, Alabama; Fort Huachuca, Arizona; and Camp Claiborne, Louisiana (home of the 761st Tank Battalion known as the ‘Black Panthers’). Other segregated camps with Black training facilities were located across the South, with some exceptions in the Northeast (e.g. Fort Dix, New Jersey). [^6]: Racial segregation was never made explicit in the form of Congressional legislation, army- or navy-wide directives, or even local or state Jim Crow ordinances, because military posts were federal properties exempt from such laws (Guglielmo 2021). [^7]: One Black soldier recalled that at Camp Shenango in Pennsylvania, “there was a White PX \[post exchange\], but we could not use it. They set up a temporary situation in the barracks \[for us\] where a guy had some candy bars and a Coke. At the White PX, you could buy almost anything. We had nothin’’ (Guglielmo 2021). [^8]: Draft avoidance among 1944–1950 cohorts raised college attainment by two percentage-points, with no effect for later cohorts (Card and Lemieux 2001). [^9]: See Section 7.1 for details on sampling procedures. [^10]: 1.7 per cent of the respondents did not answer this question. [^11]: Per War Department Form No. 55, qualifications were assessed based on physical attributes and educational and occupational history. [^12]: Recruitment region is based on the following question: ‘In what state of the United States did you spend most of the past five years of your civilian life?’ Responses are grouped into ten Census divisions (e.g., New England, Middle Atlantic, Pacific). [^13]: With clustered SEs, the minimum detectable effect at 80 per cent power is approximately 0.89 SD, substantially larger than all main effect sizes (0.10–0.27 SD).