The effect of deployment on first‐ and second‐term re‐enlistment in the US active duty force

Abstract

Why should deployment affect re‐enlistment? In our model, members enter the military with naïve beliefs about deployment and use actual deployment experience to update their beliefs and revise their expected utility of re‐enlisting. Empirically, re‐enlistment is related to the type and number of deployments, consistent with the learning model. Non‐hostile deployment increases first‐term re‐enlistment but hostile deployment has little effect except for the Army, where the effect is positive. Both types increase second‐term re‐enlistment. Interestingly, first‐term members with dependants tend to respond to deployment like second‐term members. In addition, deployment acts directly to affect re‐enlistment, not indirectly through time to promotion.

Keywords:

• Deployment
• Re‐enlistment
• Bayesian learning
• Expected utility
• JEL codes: J22
• J24
• J28

Notes

The authors thank Beth Asch, Susan Everingham, Ron Fricker, Curt Gilroy, Glenn Gotz, Carole Roan Gresenz, Susan Hosek, Michael Mattock, Stan Panis, John Warner, the Defense Manpower Data Center for providing the data, and the Office of the Secretary of Defense (Personnel & Readiness) for research support.

It is possible that regime shifts occur in which the stochastic distributions of the nature, prevalence, and duration of deployment change abruptly. In a Bayesian framework, a natural way of allowing for large shifts in the distribution parameters is to establish a wide support for the prior distribution. Abrupt shifts lead to large revisions in estimates of the parameter values and decreased precision in the estimates. See‐sawing from one regime to another would decrease the predictive value of deployment experience.

Some deployments may be voluntary. For example, active duty officers may apply for assignments to open positions, including those with unaccompanied tours (e.g. the Army in the Republic of Korea or the Marines in Okinawa), and this can be thought of as a voluntary deployment that supplements the top‐down assignment process. In addition, reserve call‐ups may permit voluntary service.

Because the data contain the entire population of enlisted personnel making re‐enlistment decisions in FY 96–99, they are representative of the population at that time. Variable means and standard deviations can be referenced at www.rand.org/publications/MR/MR1594/. The re‐enlistment indicator is based on stay/leave behavior drawn from the services’ personnel records, and the measures of deployment are derived from individual pay records. Deployment is indicated from the receipt of Family Separation Allowance (FSA) and/or Hostile Fire Pay (HFP). We defined re‐enlistment as a new service obligation of at least 24 months. We used FSA to define non‐hostile deployments of at least 30 days and HFP to define deployments involving hostile duty at some time during the deployment. For single members, ineligible to receive FSA, non‐hostile deployments were imputed on the basis of FSA receipt by unit members with dependants. The measures appear to be a reliable basis for analyzing the relationship between re‐enlistment and long or hostile deployment. However, the measured number of deployments may slightly undercount the number of deployments, and measured months of deployment probably undercounts months by one to two weeks per deployment, which average three to four months in length.

Deployments and months of deployment involve counts over a three‐year window ending three months before the month of the re‐enlist/leave decision. The three‐month buffer was intended to control for possible reverse causality. If a member near the end of the term, learned that the unit would be deployed, and wanted to go on the deployment, then that specific deployment could influence the member’s decision to re‐enlist. But that deployment, falling in the three‐month window, was not included in our count. To implement a count over a three‐year window, we focused on members with enlistment terms of three and a half years or longer. Members with two‐ and three‐year terms were excluded from our analysis sample unless they had extended the length of their term.

We also estimated a two‐equation model of time to promotion and re‐enlistment, where deployment affects re‐enlistment directly and through its influence on the speed of promotion.

In addition, we estimated models that interacted the hostile and non‐hostile deployment indicators. Likelihood tests indicated that these models fit better for the Navy and the Air Force but not for the Army and the Marine Corps, and they provided no further insights and are not reported here.

Because we do not have bonus data, we cannot estimate the effect of bonuses on re‐enlistment. In general, bonus effects are difficult to estimate because bonuses tend to be increased when re‐enlistment is low, hence there can be a simultaneity bias. In our case, the argument would be that deployment caused a decrease in re‐enlistment, which then led to an increase in the re‐enlistment bonus. Assuming a higher bonus had a positive effect on re‐enlistment, the omission of a bonus variable (or more specifically, the omission of an instrument for bonus) would then cause an upward bias in the deployment coefficients. The bias would be mitigated to some extent because although a bonus applies to everyone up for re‐enlistment in a specialty at a given time, only a portion of personnel will have had any deployment (see Tables I and II). In addition, we are dealing with deployment in a three‐year period prior to re‐enlistment, and other factors occurring after deployment might affect bonus usage. For instance, private sector job opportunities improved steadily in the late 1990s, and military pay fell relative to civilian pay. These changes led to the several‐year increase in bonuses starting in FY 1999, and this increase may have had little connection to deployment in prior years.

Although the Army is the largest service, the number of observations in the Army regression is similar to that for the Navy and less than for the Marine Corps and the Air Force. The reason has to do with sample exclusions. The first‐term regressions include members between the ages of 16 and 26, the follow‐up period to determine whether a re‐enlistment (and not an extension) occurred begins after 35 months in the current term and must fall within the data window, and the first obligation is 42 to 72 months long. Members in any service with first obligations of three years or less are excluded.

To the extent that deployments are endogenous, the estimated effect of deployment on re‐enlistment may be biased upward. An upward bias would occur if members preferring more deployment were selected for deployment, or selected frequently deploying occupations or units, and had a higher propensity to re‐enlist (apart from deployment).

The predictions were made for a member with a given set of characteristics: high school diploma graduate, AFQT Category IIIA (score of 50–64), electrical or mechanical equipment repairer, white male, no dependants (first‐term) or dependants (second term), 6.6% unemployment rate at accession, 4.9% unemployment rate at current re‐enlistment, and fiscal year 1999. A different set of characteristics would change the predictions but have little effect on the shape of the relationship.

Dependency status is measured at the time of re‐enlistment decision. For the vast majority of members, a change in dependency status from no dependants to dependants indicates marriage.

The estimated first‐term error correlation was ‘large’ and negative for the Army, Navy, and Marine Corps, around −0.3, but that of the Air Force was small, −0.06. The second‐term estimates were about −0.5 for the Army, Navy, and Air Force and −0.2 for the Marine Corps.