The Effect of Service-Length Obligations on Occupational Selection: Evidence from West Point Graduates

ABSTRACT This paper estimates the effect of the U.S. Army’s June 2020 decision to increase the service obligation for Army aviators from six to ten years on West Point cadets’ preferences for aviation. We use a difference-in-differences identification strategy, exploiting how the policy took effect after the Class of 2021 submitted four sets of interim preferences but before cadets submitted two final sets of preferences, using preferences from cadets in the Classes of 2015 to 2020 to form counterfactual outcomes. The increased service obligation made aviation less popular, reducing the percentage of cadets who ranked aviation as their first choice by 4.1 percentage points (baseline of 18.5%) and increasing the percentage of cadets who ranked aviation as among their three lowest choices by 6.9 percentage points (baseline of 8.9%). The reduction in aviation’s popularity is most pronounced among cadets with above-median grade-point-averages, above-median aviation talent scores, and above-median SAT scores.


Introduction
A topic of substantial interest to policymakers and economists is how nonwage characteristics, or job amenities, influence occupational choice.Recent research suggests workers are willing to accept jobs that pay lower wages in exchange for alternative work arrangements (Mas and  Pallais 2017), the opportunity to direct their own work (Stern 2004), and the opportunity to work in safer environments (Kniesner et al. 2012; Greenberg et al. 2021).For the U.S. military, one of the most important non-wage amenities is the length of each servicemember's contractual obligation.Longer service obligations increase the opportunity cost of service, but shorter service obligations force the military to bear higher training costs and make it difficult to fill mid-level enlisted and officer ranks.
Balancing this trade-off is especially important when establishing contract lengths for new pilots in the U.S. Army.Army aviators spend a year or longer completing initial flight training that is very costly: training an additional 100 Army pilots per year costs an additional $76 M, or a marginal cost of $760,000 per pilot (Department of Defense 2019).Moreover, flight training is highly valued in the commercial airline industry, where recent shortages of civilian pilots has increased commercial demand for trained military pilots (Dillingham 2014; Military Pilot Shortage, 2017; Department of  Defense 2019).In response to these challenges, the Secretary of the Army announced in June 2020 that the Active Duty Service Obligation (ADSO) for aviation officers would increase from six years to ten years for any officers who commissioned in October 2020 or later (Brading 2020).
One-third of Army aviators who commission as Second Lieutenants graduate from the United States Military Academy at West Point.Two important features of West Point's process for assigning cadets to Army occupations, or branches, permit us to estimate the causal effect of the increased service obligation on cadets' preferences for becoming aviators.First, West Point cadets submit preferences over branches six different times during their four years at the academy.The Army announced the increased service obligation after cadets graduating in the Class of 2021 had submitted their first four sets of branch preferences but before they had submitted their fifth and sixth sets of branch preferences.We use preferences of cadets from the West Point Classes of 2015 through 2020, all of whom graduated and commissioned into the Army before June 2020, as a comparison group in a difference-in-difference identification strategy to account for the possibility that cadets in the Class of 2021 might have changed their preferences for aviation after submitting their fourth set of preferences even in the absence of the June 2020 policy change.As a second important feature, the Army announced the increased service obligation after cadets in the Class of 2021 were legally obligated to serve in the Army regardless of their eventual graduation status, thus permitting us to avoid concerns that the policy impacted attrition among cadets in the Class of 2021. 1 We find that the increased service obligation made aviation less popular among West Point cadets.Across three broad measures of preferences, popularity for aviation among cadets in the Class of 2021 fell in preference sets five and six, after the announcement of the increased service obligation, compared to preferences among cadets in the Classes of 2015 through 2020, despite all cohorts having similar preferences for aviation through their first four preference sets.Using cadets' numerical rank-order preferences for aviation, where a cadet's most preferred branch is ranked number 1 and a cadet's least preferred branch is ranked 17, we find that the increased service obligation increased the average cadet ranking for aviation by 1.138 positions, implying that the policy change made aviation less popular.
The change in average rankings for aviation is reflected in two other measures: the propensity for cadets to rank aviation as their most preferred branch and the propensity for cadets to rank aviation as their 15 th , 16 th , or 17 th choice, where the latter outcome reflects Army rules limiting how and when women could rank order infantry and armor among their branch preferences during our study. 2The longer service obligation appears to have decreased the percentage of cadets who ranked aviation as their most preferred branch by 4.1 percentage points, a 22% decrease on the mean.The longer obligation also increased the percentage of cadets who ranked aviation as their 15th choice or worse by 6.9 percentage points, a 78% increase on the mean.Although the reduction in cadets who ranked aviation as their first choice suggests the impact of the increased service obligation was not limited to cadets with only an inframarginal interest in aviation, we directly test this hypothesis by reporting estimates for the subset of cadets who ranked aviation as their most preferred branch during their fourth preference set.Even among these cadets with a strong interest in aviation, we find the policy change increased the average numerical ranking for aviation by 1.35 positions and reduced the propensity for these cadets to continue ranking aviation as their top choice by 19.1 percentage points.
Exploring heterogeneity, we find that aviation's reduction in popularity was most pronounced among cadets with relatively high levels of aptitude and achievement.Our estimates suggest the increased service obligation reduced the percentage of cadets with above-median GPAs who ranked aviation as their first choice by 8.8 percentage points, which is significantly different from the precisely estimated zero effect we find among cadets with below-median GPAs.We find similar results when we compare cadets with high aviation talent scores to cadets with low aviation talent scores, and when we compare cadets with high SAT scores to those with relatively low SAT scores. 3stablishing policies that balance short-and long-term requirements for trained aviators are critical for overcoming the U.S. Army's pilot shortage.On one hand, since roughly 40% of officers who branch aviation leave the Army within four years of the conclusion of their initial service obligation, a four-year increase in the service obligation for trained pilots will undoubtedly improve retention among officers who decide to become pilots. 4 On the other hand, many in the Army aviation community expressed concern that the longer service obligation would negatively impact the quality of future aviation officers and hurt the Army's ability to recruit future pilots (Randel 2020).Our finding that the reductions in aviation's popularity following the increased service obligation were predominately concentrated among cadets with high levels of baseline ability, coupled with evidence that average GPAs and talent scores among cadets in the Class of 2021 who ultimately branched aviation were lower than average ability scores among cadets in earlier classes who branched aviation, lends some support to concerns that the longer commitment could reduce the quality of future pilots.However, even after the announcement of the increased service obligation, the number of cadets in the Class of 2021 who ranked aviation as their most preferred branch exceeded the number of aviation allocations available to West Point cadets, suggesting that the most negative predictions surrounding the increased service commitment might not materialize.
Our study is also relevant to other military forces, as pilot shortages are not unique to the U.S. Army.Since 2016, the Royal Canadian Air Force has had more aircraft than pilots available to fly them, causing the Canadian government to explore monetary incentives to improve aviator retention (Berthiaume 2019; Canadian Press Staff 2019).As recently as 2019, the German Air Force was only able to fill two-thirds of its combat pilot positions (Witting 2019).The Chinese military has also suffered from readiness and training issues due to a lack of naval pilots, which it has attempted to fix through increased recruitment efforts (Pickrell 2018; Huang 2020).Beyond pilot shortfalls, our analysis offers insights on how potential military officers might respond to changes in their contractual military service obligations, a policy relevant topic.For example, West Point recently tested the feasibility of extending U.S. Army officer service commitments from five to six years. 5Indeed, our finding that the longer aviation service obligation caused a 6.9 percentage point increase in cadets who ranked aviation among their lowest branch choices suggests that at a nontrivial portion of West Point cadets strongly oppose longer service obligations.More broadly, officer retention has become an enormously important topic for the U.S. military in recent years (Colarusso, Wardynski, and Lyle  2010; Colarusso et al. 2016).The Department of Defense is the largest employer in the United States and remains an important segment of the U.S. labor market: 7.5% of men and 1.7% of women between 25 and 44 years old are former or current U.S. servicemembers according to the 2019 American Community Survey (U.S. Census Bureau, Population Division 2019).The U.S. military commissions roughly 17,000 new officers and enlists another 160,0000 enlisted servicemembers each year (Department of Defense 2020).Policies that influence the size, quality, and retention of future officer cohorts could have a profound impact on the servicemembers they will lead, especially considering recent evidence that military leaders influence their subordinates' performance, retention, and career decisions (Lyle and Smith 2014; Carter et al. 2019; Kofoed and mcGovney 2019).
Our paper also contributes to a large literature that estimates how workers respond to nonwage job amenities, as first described in the theory of hedonic pricing (Rosen 1974, 1986). 6To our knowledge, our study is the first to explore the relationship between contract length and occupational choice.Furthermore, our identification strategy allows us to avoid concerns that contractual service obligations might be correlated with unobserved characteristics of Army jobs, a problem that often generates omitted variable bias in cross-sectional estimates of compensating wage differentials (Brown 1980; Roback 1982; Hwang, Mortensen, and Reed 1998; Lang and Majumdar 2004;  Bonhomme and Jolivet, 2009).Our results also broadly relate to recent research by Bingley,  Lundborg, and Lyk-Jensen (2020), who find that the opportunity cost of mandatory military service is higher among Danish men with higher ability levels, and Borgschulte and Martorell (2018), who find that U.S. servicemembers are more likely to reenlist when the opportunity cost of leaving the service, as measured by home-state unemployment rates, is higher, and a growing literature related to cadet branching (Greenberg, Pathak, and Sonmez 2021; Jones and Kofoed 2020; Kofoed and mcGovney  2017).

Institutional Details: Cadet Branching and Service Obligations
Each year West Point matriculates roughly 1,200 cadets who complete a four-year post-secondary education that culminates with cadets graduating with a Bachelor of Science degree and commissioning as officers in the U.S. Army.Although West Point's military mission results in a unique student body that is predominately male, its undergraduate coursework, student-to-faculty ratios, and SAT and ACT test scores are comparable to top liberal arts colleges (Carter, Greenberg, and Walker 2017).Cadets commission into one of 17 Army occupational specialities, or branches. 7Cadets submit preferences for branches six different times: in the spring of their freshman year (P1), in the spring of their sophomore year (P2), in the fall of their junior year (P3), in the spring of their junior year (P4), in August just after starting their senior year (P5), and their sixth and final round of preferences in October of their senior year (P6).Preference sets 1 through 5 encourage cadets to educate themselves on different branches and allow West Point to evaluate how its branching program influences cadets' decisions.However, these interim preferences do not influence the mechanism that determines a cadet's final branch assignment.West Point uses the sixth and final set of preferences in a cumulative offer mechanism to determine each cadet's branch assignment. 8Before cadets submit their fifth set of branch preferences, the Army determines the number of allocations by branch available to West Point cadets.Interim preferences submitted before preference set 5 do not influence the allocation of branches to each source of commission.
West Point graduates who do not branch aviation typically incur a five-year Active Duty Service Obligation (ADSO), which stipulates the minimum time they must serve in the Active Duty Army.Prior to 2021, the standard ADSO for cadets who branched aviation was six years after completing flight school and all necessary training pre-requisites, which usually takes at least a full year (Army Regulation 350-100).However, in June 2020, the Secretary of the Army unexpectedly announced plans to increase the ADSO for aviation officers to ten years: "Beginning fiscal year 2021, Regular Army officers who attend initial entry flight training will incur a 10-year active duty service obligation on completion of the course and attaining an aeronautical rating of Army Aviator." 9 The timing of this announcement forms the basis of our natural experiment.This change in policy did not impact West Point cadets who graduated in the Class of 2020 or earlier, but cadets in the Class of 2021 who branched aviation were subject to the longer service obligation.The Class of 2021 learned of the increased service obligation after submitting their fourth set of preferences in the spring of their junior year but before submitting their fifth and sixth round of preferences in August and October of 2021.Due to the unexpected nature of the change in policy, cadets were unlikely to anticipate the increased service obligation prior to submitting their fifth round of preferences, a point we have confirmed with the Army's Talent Based Branching Program Manager and through informal conversations with cadets.
Officers are rarely eligible to leave the Army before completing their ADSO.Service obligations are rarely waived, and only for 'the convenience of the Government or personal hardship reasons' when approved by the Assistant Secretary of the Army for Manpower and Reserve Affairs (Army Regulation 350-100).Graduates of West Point who do not complete their ADSO and who fail to receive a waiver must reimburse the cost of their education to the government (Army Regulation 150-1).For West Point Graduates in the Class of 2018 this cost was $236,052 per student. 10Due to the rarity of service obligation waivers and the high financial costs associated with repaying the government, roughly 95% of West Point graduates remain on active duty through their initial service obligation. 11

Empirical Strategy
The timing of the Army's announcement of the ADSO change for aviators, coupled with our ability to observe cadet's preferences at six different times, enables us to identify the causal effect of the increased service obligation on cadet's preferences for aviation.We begin our empirical analysis with standard difference-in-difference plots that compare average preferences for aviation among cadets in the Class of 2021 (the treated group) to cadets in the Classes of 2015 through 2020 (the untreated or control group) at each preference set.Any differential changes in preferences for aviation among cadets in the Class of 2021 between preference sets 4 and 5 of provide suggestive evidence that the increased service obligation altered cadets' preferences for aviation.It is important to clarify that cadets in the treated group submitted their fifth and sixth sets of preferences after the June 2020 policy change while cadets in the control group submitted all sets of their preferences before the policy change.However, cadets in both groups submitted preferences at the same time relative to the start of their four-year education at West Point.For example, cadets in the treated group (the Class of 2021) submitted their fifth round of preferences in August of 2020, the start of their senior year at West Point, while cadets in the control group (the Classes of 2015-2020) submitted their preferences in the August of their respective senior years at West Point.
To test formally for changes in preferences for aviation among the Class of 2021 relative to cadets in earlier cohorts, we estimate the following event study equation: Equation ( 1) includes individual cadet fixed effects, which absorb cadet class (i.e.cohort) fixed effects that would be in a standard event study or difference-in-difference specification.CL21 i is an indicator variable that equals 1 if cadet i belongs to the Class of 2021, which we interact with a fixed effect for each preference set.Y it is one of our three primary outcomes: 1) cadet i's numerical ranking for aviation (e.g. 1 through 17) in preference set t, where a value of 1 implies aviation is cadet i's most desired branch and a value of 17 implies aviation is cadet i's least desired branch, 2) an indicator variable that equals 1 if aviation is cadet i's most preferred branch in preference set t and equals 0 otherwise, or 3) an indicator variable that equals 1 if aviation is among cadet i's 15 th , 16 th , or 17 th choice in preference set t and equals 0 otherwise.We define this last outcome as aviation among a cadet's three least preferred branches, rather than as a cadet's least preferred branch, because of Army rules regarding when and how women could rank order infantry and armor.Prior to 2016, female cadets could not rank infantry or armor.Between 2016 and 2020, female cadets were permitted to branch infantry and armor, but were not required to list them among their 17 branch preferences.In all cases where we observe a female cadet who has not ranked infantry or armor, we assume the unranked branch(es) is her least preferred branch(es).
Estimates of β 1 , β 2 , β 3 , β 5 , and β 6 from equation ( 1) reflect the average difference in outcomes between the Class of 2021 and cadets in the Classes of 2015 through 2020 at each preference set.By excluding the interaction of Class of 2021 with preference set 4, we can interpret each event study estimate as the average difference in outcomes between cadets in the Class of 2021 and cadets in the Classes of 2015 through 2020 relative to the average difference between these two groups at the fourth preference set.
We complement event study estimates with difference-in-difference estimates that compare changes in preferences for aviation among cadets in the Class of 2021 following the announcement of the increased service obligation between the fourth and fifth set of preferences to changes in preference for aviation among cadets in earlier classes that did not experience an increased service obligation: Equation ( 2) includes fixed effects for each preference set and individual cadet fixed effects, analogous to Equation (1).We interact CL21 i , the indicator variable for being in the Class of 2021, with 1 t > 4 ð Þ, an indicator variable that equals one in preference sets 5 and 6.Finally, we cluster standard errors on individual cadet when estimating Equations ( 1) and (2).
Estimates of β from Equation ( 2) reflect how average differences in preferences for aviation between cadets in the Class of 2021 and cadets in the Classes of 2015-2020 changed following the Army's announcement of an increased service obligation for aviation.To interpret β from our difference-in-difference Equation ( 2) as the causal effect of the increased service obligation on preferences for aviation, two key assumptions must hold.The first is the standard parallel trends assumption: in the absence of the increased service obligation, we assume preferences for aviation among cadets in the Class of 2021 would have followed the same (i.e.parallel) trend as preferences for aviation among cadets in Classes of 2015-2020 who were not affected by the increased service obligation.The second assumption is that other than the increased service obligation for Army aviators, nothing else happened around June 2020 that would have changed the Class of 2021ʹs preferences for aviation between preference set 4 and preference set 5 in a way that differed from changes in preferences for aviation between preference sets 4 and 5 among cadets in earlier classes.Our difference-in-difference plots and estimates of β 1 , β 2 , and β 3 from our event study Equation ( 1) help us test the validity of our parallel trends assumption.Regarding the second assumption, we are not aware of any other policy changes or external events that may have differentially affected preferences for aviation among cadets in the Class of 2021 relative to cadets in earlier classes.Our conversations with the Army's Talent Based Branching Program Manager and additional robustness checks we report on in the Online Appendix lend further evidence that this assumption is valid.

Data and Summary Statistics
We use administrative cadet branching data from a sample of 7,412 West Point cadets from the Classes of 2015 through 2021.This is the universe of West Point cadets from the Classes of 2015-2021 after excluding roughly 1,250 cadets who either attrited from West Point (about 1,000 of the 1,250) or who were identified as not requiring an initial Army occupation (about 250 of the 1,250) before submitting their fifth preference set, the preference set where the increased service obligation first took effect for the Class of 2021.The data are a panel of cadets' rank-order preferences for all branches, where each observation corresponds to a specific cadet's preferences at one of six preference sets.Our data include cadet grade point averages (GPAs) at the end of each academic year, standardized test scores, and demographic information.Our data also include aviation branch scores, or aviation talent scores, derived from assessments administered as part of the Army's Talent Based Branching program.Aviation talent scores are intended to predict a cadet's potential for success as an Army aviator.Cadets only receive one talent score for each branch while at the Academy -this variable does not vary across preference set time periods.
As described above, prior to 2016 cadets did not rank cyber and thus only ranked 16 branches.To facilitate comparison of cadets' preferences before and after this change, when our outcome of interest is a cadet's rank for aviation, we multiply the cadet's preference for aviation by 17 16 if the ranking corresponds to a preference submitted before cyber was an option.We also adjust preference rankings for cadets in the Class of 2021, who were permitted to separately rank explosive ordnance disposal (EOD), by assigning a cadet's preference for ordnance as the more preferred of EOD and ordnance, dropping the least preferred of EOD and ordnance, then re-ordering remaining branches so that cadets have 17 ranked branches.
Table A.1 (Online Appendix) compares summary statistics for cadets in the class of 2021 to cadets in the classes of 2015-2020.Roughly 80% of cadets in our sample are male, about 70% are white, and average SAT scores are near 1280 on a 1600 scale.Aviation is relatively popular: the average cadet ranks it as their 7 th most preferred branch, 16% of cadets rank aviation as their most preferred branch, and only 9% of cadets rank aviation as their 15 th , 16 th , or 17 th choice.Cadets in the Class of 2021 have more women, more minorities, and slightly stronger preferences for aviation (as of preference set 4) than cadets in earlier classes, and could differ from cadets in earlier classes on unobservable dimensions, further necessitating a difference-in-difference research design.Although not reported in Table A.1, 78.5% of cadets in our sample were ultimately assigned to the branch they rated as their most preferred choice during preference set 6, 14.9% received their second or third branch choice, and only 1.1% received their 15 th , 16 th , or 17 th choice.

Results
We begin our empirical analysis with a visual comparison of trends in preferences for aviation among cadets in the Class of 2021 (the treated group) to preferences for aviation among cadets in the Classes of 2015-2020 (the untreated or control group) before proceeding to estimates of Equations ( 1) and ( 2) for the full sample and important subsamples.

Main Results
Panel A of Figure 1 plots the average preference ranking for aviation among cadets in the Class of 2021 (the solid black line) and the average preference ranking for aviation among cadets in the Classes of 2015-2020 (the dashed black line) at each of the six different preference sets.The vertical red line between preference sets 4 and 5 reflects the timing of the increased service obligation for aviation, which the Army announced after cadets in the Class of 2021 submitted their 4 th set of preferences but before they submitted their 5 th set of preferences.As a reminder, a higher preference-rank for aviation implies cadets prefer aviation less, so the figure reveals that the desire to branch aviation decreases as cadets draw closer to graduation among all classes.Importantly, however, the figure shows how preferences for aviation among cadets in the Class of 2021 changed after the announcement of the increased service obligation in ways that did not follow the same pattern of preferences for aviation among cadets in earlier classes.Specifically, among preference sets 1 through 4, cadets in the Class of 2021 had slightly stronger preferences for aviation (i.e.aviation has lower average rankings) than cadets in earlier classes.However, this reversed in preference sets 5 and 6, where cadets in the Class of 2021 had weaker preferences for aviation than their counterparts in earlier classes during the same preference periods.
Panel B of Figure 1 plots event study estimates of Equation ( 1) where the outcome is a cadet's numerical preference ranking for aviation.During preference sets 1-3, the average difference in preferences for aviation between cadets in the Class of 2021 and cadets in earlier classes fluctuated between 0.3 positions closer and 0.4 positions larger than the difference in average preferences observed in preference set 4. These pre-treatment fluctuations do not exhibit a clear pattern, which lends support to the parallel trends assumption.However, in both preference set 5 and 6, the estimated difference in preference rankings for aviation among cadets in the Class of 2021 relative to cadets in earlier classes grew to roughly 1.25 positions larger than the average difference in aviation rankings observed in preference set 4.
Analysis of our two other outcomes suggests that the service obligation reduced the likelihood that cadets in the Class of 2021 ranked aviation as their top choice and increased the likelihood that they ranked aviation among their lowest three choices.Panel A of Figure 2 displays the difference-in-difference plot and panel B of the same figure shows the event study plot when the outcome is an indicator variable that equals 1 if aviation is a cadet's most preferred branch in the given preference period.Prior to the announcement of the increased service obligation, the share of cadets in the Class of 2021 who had aviation as their most preferred branch exceeded the share of cadets in earlier classes who had aviation as their most preferred branch during every preference set.This reversed after the announcement of the increased service obligation: relative to earlier classes, fewer cadets in the Class of 2021 listed aviation as their most preferred branch in preference sets 5 and 6.     2 by reporting on the difference-in-difference plot and event study estimates when the outcome is a variable that equals one if a cadet ranks aviation among her bottom three choices and is zero otherwise.Both plots reveal that the percentage of cadets with low preferences for aviation had similar levels and trends in both groups prior to the increased service obligation, but diverged substantially following the policy change.Even though roughly 10% of cadets in both groups ranked aviation as their 15 th , 16 th , or 17 th choice during preference set 4, in preference sets 5 and 6, cadets in the Class of 2021 were 7 percentage points more likely to rank aviation among their three least preferred branches.
To formalize our results thus far, panel A of Table 1 reports estimates of β from Equation ( 2).Column 1 reports estimates where the outcome is a cadet's numerical preference rank for aviation during a particular preference set (corresponding to Figure 1), column 2 reports estimates where the outcome is an indicator variable that equals 1 if a cadet ranks aviation as her top choice and is 0 otherwise (corresponding to Figure 2), and column 3 reports estimates where the outcome is an indicator variable that equals 1 if a cadet ranks aviation as one of his bottom three choices and is 0 otherwise (corresponding to Figure 3).The point estimate of 1.138 implies that the increased service obligation increased cadets' numerical preference ranking of aviation by an average of just over one position among the 17 possible branches.Meanwhile, the point estimate reported in column 2 suggests that the increased service obligation reduced the propensity for cadets to rank aviation as their first choice by 4.1 percentage points, roughly 22% of the mean percentage of cadets in the Class of 2021 who ranked aviation as their first choice in preference set 4. Finally, the point estimate in column 3 indicates that the increased service obligation caused a 6.9 percentage point increase in the propensity for cadets to rank aviation as one of their bottom three branch choices, a 78% increase from the percentage of cadets in the Class of 2021 who ranked aviation as their 15 th , 16 th , or 17 th choice in preference set 4.
Before proceeding, we cautiously note that Figure 2 reveals a large share of cadets in the Class of 2021 ranked aviation as their top choice during preference set 3. This appears to generate an aberration from the otherwise parallel trends seen in preference sets 1, 2, and 4.This anomaly could cause our estimates of the effect of the increased service obligation on cadets' propensity to rank aviation as their top choice to be biased downwards (i.e. more negative than the true effect).We revisit this point and present results to several other robustness checks in our Online Appendix.The results of these robustness checks all support our general conclusion: the increased service obligation appears to have decreased cadet preferences for aviation, and some of this is due to a reduction in the propensity for cadets to rank aviation as their top choice.

Impacts on Cadets with Strong Prior Preferences for Aviation
While our results thus far suggest that the increased service obligation reduced cadets' desire to branch aviation, the policy change may have had relatively little impact on the group of cadets who ultimately branched aviation if it primarily impacted cadets who would normally rank aviation near the middle or bottom of their rank-ordered list of 17 branches.Since only about 7% of cadets receive branches they rank below their third choice, such a scenario would have little impact on cadets with a reasonable chance of branching aviation.On the other hand, the increased service obligation could alter who ultimately branches aviation if the policy influences preferences of cadets who would otherwise have a strong desire to branch aviation.To test whether the increased service obligation impacted preferences among cadets with a strong desire to branch aviation, we conduct a separate analysis that restricts attention to cadets who ranked aviation as their top choice in preference set 4. As a matter of level setting, 70% of cadets in the Classes of 2015 through 2020 who ranked aviation as their most preferred branch in preference set 4 also ranked aviation as their most preferred branch in preference set 6. Panel A of Figure 4 plots cadets' average preference rank for aviation among cadets who ranked aviation as their most preferred branch in preference set 4, separately for cadets in the Class of 2021 and for cadets in earlier classes.Among this group, cadets in the Class of 2021 gravitated towards aviation at nearly the same rate as cadets in earlier classes during preference sets 1-4.However, these trends diverged dramatically after that.Cadets in the Class of 2021 who ranked aviation as their most preferred branch during preference set 4 gave aviation an average ranking of roughly 3.6 in preference sets 5 and 6, more than a full position higher than the average ranking among cadets  2) for the subsamples identified in the panel and column headings.In columns ( 1) and ( 2) the outcome is a cadet's preference rank for aviation.In columns ( 3) and ( 4) the outcome is an indicator that equals 1 if a cadet ranks aviation as her most preferred branch.In columns ( 5) and ( 6) the outcome is an indicator that equals in earlier classes who had also indicated that aviation was their most preferred branch in preference set 4. Panel B of Figure 4 shows a similar pattern.Whereas 70% of cadets in the Classes of 2015-2020 who ranked aviation as their top choice in preference set 4 still ranked aviation as their most preferred branch in preference set 6, this was only true for 57% of cadets in the Class of 2021 who ranked aviation as their most preferred branch in preference set 4. Moreover, panel C of Figure 4 suggests that the policy change caused a non-trivial share of cadets to move aviation from their top choice in preference set 4 to one of their bottom three branch choices in preference sets 5 and 6.Panel B, Table 1, reports estimates of Equation ( 2) when we restrict the sample to cadets who ranked aviation as their most preferred branch during preference set 4. The estimate reported in panel B, column (1) indicates that, among cadets who ranked aviation as their most preferred branch in preference set 4, the average preference rank for aviation among cadets in the Class of 2021 was nearly 1.5 places higher than the average preference rank for aviation among cadets in earlier classes.Panel B, column (2) further suggests that the increased service obligation reduced the propensity for cadets to continue ranking aviation as their most preferred branch by 19 percentage points.Finally, although the point estimates in column (3) of panel B is not statistically significant, it is directionally consistent with the notion that the increased service obligation caused some cadets to move aviation from their most preferred branch to among their least preferred branches.
The results reported in Figure 4 and panel B of Table 1 suggest that much of the estimated effect of the increased service obligation on changes in cadet preferences for aviation was driven by changes in preferences among cadets with a strong interest in aviation prior to the policy change.Still, there is also evidence that the increased service obligation caused cadets who normally would not have ranked aviation as their top choice to further reduce their interest in the branch.This is most apparent in column (3) of Table 1, which shows that the propensity for cadets to rank aviation among their three lowest branches because of the policy change was larger in the full sample (panel A) than in the subsample of cadets who ranked aviation as their most preferred branch in preference set 4 (panel B).Decreased interest in aviation even among cadets with relatively low baseline preferences for the branch is still important from a policy perspective because it is evidence that some cadets have a strong aversion to longer contractual service obligations in general.

Heterogeneous Effects by Subgroups
Of particular interest to the Army is understanding which types of cadets are most responsive to the increased service obligation.We begin our subgroup analysis by examining whether cadets with high baseline levels of achievement and ability, as measured by cadet grade point averages (GPAs) calculated at the end of their junior year, after cadets submit preference set 4 but prior to submitting preference set 5. Panel A of Table 2 reports estimates of Equation ( 2) separately for cadets with low (below-median) and high (above-median) GPAs for all three outcomes.As reported in columns ( 1) and ( 2), the increased service obligation increased preference rankings for aviation by 0.712 positions, on average, for cadets with low baseline GPAs, but this effect is more than double in size (1.559 positions) for cadets with high baseline GPAs.We can easily reject the null hypothesis that effect sizes are equal for low-GPA and high-GPA cadets (P-value = .002).Interestingly, columns (3) and ( 4) suggest that the increased service obligation had no impact on the propensity for cadets with low GPAs to rank aviation as their most preferred branch, but decreased the propensity for cadets with high GPAs to rank aviation as their most preferred branch by 8.8 percentage points, a 32% reduction from the mean.Although columns ( 5) and ( 6) reveal similar point estimates for low-GPA and high-GPA cadets when the outcome is an indicator for aviation being among a cadet's 15 th , 16 th , or 17 th branch choice, this is still notable considering that cadets with low GPAs were nearly twice as likely to rank aviation among their least preferred branches in preference set 4 relative to cadets with high GPAs.In fact, the estimates in column ( 6), panel A, of Table 2 suggests that the policy change increased the likelihood of high-GPA cadets ranking aviation as one of their least preferred branches by 7.1 percentage points, more than double the baseline mean (6.0%) of high-GPA cadets with very low preferences for aviation.
Panels B and C of Table 2 reveal that splitting our sample by two other measures of ability, cadet aviation branch scores and composite SAT scores, yield similar estimates to splits by cadet GPA.Cadets with higher aviation branch scores appear to have a stronger response to the increased service obligation than cadets with relatively low aviation branch scores.In particular, the policy change reduced the likelihood that cadets with high aviation branch scores ranked aviation as their most preferred branch by 8.3 percentage points, but had no effect on the same outcome among cadets with relatively low aviation branch scores.We observe similar patterns by SAT score.
Columns ( 1) and ( 2) of panel D, Table 2 suggest that the increased service obligation may have elicited a stronger response among women than men.The policy change caused increased women's average preference ranking for aviation by 1.6 positions, an effect that is at least marginally significantly larger than the estimated effect of 1.0 positions for men.Subsequent columns of panel D, Table 2 suggest that the larger response among women is potentially the result of the increased service obligation causing more women to rank aviation among their least preferred branches relative to men, as point estimates when the outcome is aviation as a cadet's most preferred branch are indistinguishable between men and women.Panel E does not reveal significant differences in effects between nonwhite and white cadets.
To examine for heterogeneity further among more finely divided subgroups, Figure 5 reports estimates of Equation ( 2) where the outcome is a cadet's preference ranking for aviation when we split the sample based on GPA quartiles, aviation branch score quartiles, SAT quartiles, and four  2) for different subgroups.The outcome variable is a cadet's preference rank for aviation.A cadet's most preferred branch is ranked 1, so a positive point estimate implies that the Army's increased service obligation for aviation makes aviation less popular.The whiskers display 95% confidence intervals derived from standard errors clustered on individual.
different race groups (Black, Hispanic, Other, and White).Estimates for women, men, and the full sample are included to facilitate comparison.Importantly, this figure reveals that cadets in the highest quartiles of GPA, aviation branch scores, and SAT scores exhibit the largest responses to the increased service obligation.

Implications of Results
An important question for the U.S. Army is whether the increased service obligation could negatively impact the quality of future Army aviators (Randel 2020).To explore this further, Figure A.2 (Online Appendix) reports the average GPA, the average aviation branch score, and the average SAT score among cadets who were assigned to an aviation allocation at the conclusion of the branching process by each class year in our study.For this figure we standardize GPAs, aviation branch scores, and SAT scores to have a mean of zero and a standard deviation of one among all cadets in the same class year.Panel A shows that the average GPA of cadets who branched aviation in the Classes of 2015-2020 was typically around 95% of a standard deviation higher than the average GPA of cadets in their corresponding classes.However, the average GPA of cadets who branched aviation in the Class of 2021 was only 75% of a standard deviation above the average GPA of the entire class.If we attribute this change to the impact of the increased service obligation, then it suggests that the June 2020 policy may have reduced the average GPA of Army aviators by 20% of a standard deviation.Panel B also shows a substantial reduction in average aviation branch scores among cadets who branched aviation in 2021, from an average branch score of 76% of a standard deviation above class means for cadets who branched aviation between 2015 and 2020 to an average branch score only 47% of a standard deviation above the class mean for cadets who branched aviation in 2021.However, panel C does not show a similar reduction in SAT scores.
Our comparison of achievement measures across cohorts requires some important qualifications.First, lower average GPAs and lower average branch scores might not translate to weaker performance as Army aviators.Second, cohort changes in the average GPA and the average aviation branch score only offer suggestive evidence on the extent to which the increased aviation service obligation might impact the average quality of cadet who branches aviation.Other factors, such as the number of aviation allocations available to West Point cadets, can change the composition of cadets who branch aviation each year.Still, the trends reported in Figure A.2 are consistent with the possibility that the Army's decision to increase the aviation service obligation could lower the average quality of cadets who branch aviation when all other factors remain equal.

Conclusion
In response to a severe shortage of Army pilots, the Secretary of the Army increased the required service obligation for Army aviators from six years to ten years in June 2020.This paper estimates the impact of the increased service obligation on cadet preferences using a difference-indifference identification strategy that compares changes in preferences for aviation among cadets in the Class of 2021, who were impacted by the policy change after submitting four sets of preferences for branches but before submitting their final two sets of preferences, to changes in preferences for aviation among cadets in the Classes of 2015 through 2020, who were not impacted by the policy change.We find the increased service obligation made aviation less popular, and that cadets with higher GPAs, higher aviation branch scores, and higher SAT scores were more responsive to the increased service obligation than cadets with lower GPAs, branch scores, and SAT scores.
Further research on how this policy change impacts the eventual performance and retention of Army aviators is warranted, as is research on how the increased service obligation might impact recruitment of future aviators.Our findings, particularly the large increase in cadets who ranked aviation among their lowest branch choices, also suggests an underlying aversion to longer contractual obligations, regardless of branch, among a nontrivial portion of West Point cadets.The U.S. military could benefit from additional research on how variation in initial service obligations for officers and enlisted personnel impact the Army's ability to recruit servicemembers.At the same time, even though the policy change made the aviation branch less popular, there were still more cadets in the West Point Class of 2021 who ranked aviation as their first choice than the total number of cadets who received an aviation allocation.This suggests the Army might be able to improve retention, or balance the quality of officers across branches, through additional policies that increase service obligations for relatively popular branches while decreasing service obligations for relatively unpopular branches.Finally, the impact of this policy may differ at other U.S. Army commissioning sources (e.g. the Reserve Officer Training Corps and Officer Candidate School).Considering our heterogeneity analysis and other evidence that West Point cadets are positively selected relative to cadets from other commissioning sources (Colarusso, Wardynski, and Lyle 2010), our results may be an upper bound on the policy's impact across all commissioning sources.

Notes
Figure A.1 in the Online Appendix describes preference submission timelines by graduating class.

Figure 1 .
Figure 1.Aviation preferences over time: Class of 2021 v. Classes of 2015-2020.(A) average aviation preference rank over time.(B) average aviation preference rank event study.Panel (A) reports average cadet preferences for aviation at the six different preference set submission periods among cadets receiving branches in the Class of 2021 and among cadets receiving branches in the Classes of 2015-2020.A cadet's most preferred branch is ranked 1 and a cadet's least preferred branch is ranked 17.The vertical line corresponds to June 2020 aviation ADSO increase, which took place after cadets in the Class of 2021 submitted their fourth round of preferences but before they submitted their fifth round of preferences.Panel (B) reports event study estimates of Equation (1) where the outcome is a cadet's ranking for aviation.The whiskers display 95% confidence intervals derived from standard errors clustered on individual.

Figure 2 .
Figure 2. Aviation as most preferred branch: class of 2021 v. classes of 2015-2020.(A) aviation as top branch choice over time.(B) aviation as top branch choice event study.Panel (A) reports the percentage of cadets who ranked aviation as their most preferred branch at different preference submission periods among cadets in the Class of 2021 and among cadets in the Classes of 2015-2020.Panel (B) reports event study estimates of Equation (1) where the outcome is a binary variable that equals one if the cadet's most preferred branch is aviation.The whiskers display 95% confidence intervals derived from standard errors clustered on individual.See the Figure1notes for more details.

Figure 3 .
Figure 3. Aviation among least preferred choices: class of 2021 v. classes of 2015-2020.(A) aviation as 15 th , 16 th , or 17 th branch choice over time.(B) aviation as 15 th , 16 th , or 17 th branch choice event study.Panel (A) reports the percentage of cadets who ranked aviation as their 15 th , 16 th , or 17 th branch choice (among 17 possible branches) at different preference submission periods among cadets in the Class of 2021 and among cadets in the Classes of 2015-2020.Panel (B) reports event study estimates of Equation (1) where the outcome is a binary variable that equals one if aviation is among the cadet's 15 th , 16 th , or 17 th branch choice.The whiskers display 95% confidence intervals derived from standard errors clustered on individual.See the Figure1notes for more details.

Figure 3
Figure3complements Figure2by reporting on the difference-in-difference plot and event study estimates when the outcome is a variable that equals one if a cadet ranks aviation among her bottom three choices and is zero otherwise.Both plots reveal that the percentage of cadets with low preferences for aviation had similar levels and trends in both groups prior to the increased service obligation, but diverged substantially following the policy change.Even though roughly 10% of cadets in both groups ranked aviation as their 15 th , 16 th , or 17 th choice during preference set 4, in preference sets 5 and 6, cadets in the Class of 2021 were 7 percentage points more likely to rank aviation among their three least preferred branches.To formalize our results thus far, panel A of Table1reports estimates of β from Equation (2).Column 1 reports estimates where the outcome is a cadet's numerical preference rank for aviation during a particular preference set (corresponding to Figure1), column 2 reports estimates where the outcome is an indicator variable that equals 1 if a cadet ranks aviation as her top choice and is 0 otherwise (corresponding to Figure2), and column 3 reports estimates where the outcome is an indicator variable that equals 1 if a cadet ranks aviation as one of his bottom three choices and is 0 otherwise (corresponding to Figure3).The point estimate of 1.138 implies that the increased service obligation increased cadets' numerical preference ranking of aviation by an average of just over one position among the 17 possible branches.Meanwhile, the point estimate reported in column 2 suggests that the increased service obligation reduced the propensity for cadets to rank aviation as their first choice by 4.1 percentage points, roughly 22% of the mean percentage of cadets in the Class of 2021 who ranked aviation as their first choice in preference set 4. Finally, the point estimate in column 3 indicates that the increased service obligation caused a 6.9 percentage point increase in the propensity for cadets to rank aviation as one of their bottom three branch choices, a 78% increase from the percentage of cadets in the Class of 2021 who ranked aviation as their 15 th , 16 th , or 17 th choice in preference set 4.Before proceeding, we cautiously note that Figure2reveals a large share of cadets in the Class of 2021 ranked aviation as their top choice during preference set 3. This appears to generate an aberration from the otherwise parallel trends seen in preference sets 1, 2, and 4.This anomaly could cause our estimates of the effect of the increased service obligation on cadets' propensity to rank aviation as their top choice to be biased downwards (i.e. more negative than the true effect).We

Figure 4 .
Figure 4. Preferences for aviation among cadets who ranked aviation as their most preferred branch in preference set 4. (A) average aviation preference rank.(B) aviation as top branch choice (C) aviation as 15 th , 16 th , or 17 th branch choice.Panel (A) reports average cadet preferences for aviation at different preference submission periods among cadets who ranked aviation as their most preferred branch in the 4th preference set.Panels (B) and (C) report the percentage of the same cadets who ranked aviation as their top choice and who ranked aviation as their 15th, 16th, or 17th choices, respectively.The solid black lines correspond to cadets in the Class of 2021 and the dashed black lines correspond to cadets in the Classes of 2015-2020.See the notes for Figures 1, 2, and 3 for additional details.

Figure 5 .
Figure 5.Estimated effects of the increased service obligation by subgroup.This figure reports difference-in-difference estimates of β 1 from Equation (2) for different subgroups.The outcome variable is a cadet's preference rank for aviation.A cadet's most preferred branch is ranked 1, so a positive point estimate implies that the Army's increased service obligation for aviation makes aviation less popular.The whiskers display 95% confidence intervals derived from standard errors clustered on individual.

Table 1 .
Impact of increased service obligation on preferences for aviation.)wherethe outcome variable is a cadet's preference rank for aviation (column 1), an indicator variable that equals 1 if a cadet ranks aviation as his most preferred branch (column 2), or an indicator variable that equals 1 if a cadet ranks aviation among as her 15 th , 16 th , or 17 th branch choice (column 3).Panel B reports analogous estimates when we restrict the sample to cadets who ranked aviation as their most preferred branch in preference set 4. See the notes for Figures1 and 4for additional details.Robust standard errors, clustered on individual cadet, are reported in parentheses.Significance levels: *: 10%, **: 5%, ***: 1%.

Table 2 .
Impact of increased service obligation by subsamples.
1 if a cadet ranks aviation as her 15 th , 16 th , or 17 th choice.We divide cadets into High and Low categories of GPA, branch score, and SAT according to median values within each cadet class.Robust standard errors, clustered on individual cadet, are shown in parentheses.Significance levels: *: 10%, **: 5%, ***: 1%.