CAREER EARNINGS AND RETENTION OF U.S. MILITARY PHYSICIANS

Abstract

Military physicians consider potential career earnings when making their stay/leave decisions. Moreover, they consider the effects of differences between military and civilian compensation at three distinct decision periods in the military career: the first year of unobligated service, the years after that point but before military retirement eligibility, and the year of retirement eligibility. We find that military retention is highly sensitive to compensation differences at the first decision and substantially less sensitive after that. We also account for endogeneity of military physician pay and retention with an instrumental variable technique, without which, estimates substantially underestimate the wage effect.

Keywords:

• Military physicians
• Physician retention
• Military retention
• Physician labor markets
• Physician compensation
• Instrumental variable estimation

JEL Codes:

• J30
• J33
• J40
• J44
• J45
• I11

ACKNOWLEDGMENTS

We would like to thank our data programmer, Mr. Kletus Lawler at CNA, for his good work in collecting the military personnel data and putting it into usable format.

Notes

¹ This figure assumes a 10% personal discount rate, a twenty‐year military career and that the first unobligated year of service is in year 4.

² We define ‘retention’ in our model as continuation in the military service from one fiscal year to the next. In the data, we know if the physician is retained because at the end of each fiscal year a physician in the data is coded as having been either retained or lost.

³ Nearly half of medical students are women, and around 30% of DOD medical scholarships are awarded to women. It is only for convenience that we refer to our notional medical student with the masculine ‘he’ rather than the more technically correct ‘he or she’.

⁴ Since the medical specialty he will ultimately choose is typically not known until later, there is some uncertainty as to the length of the residency and for the length of the service obligation. Thus, we express total wages for the period in expected‐value terms.

⁵ Note that this model can be generalized for any labor market in which an employer is offering a scholarship combined with a contractual obligation.

⁶ Service members become eligible for retirement after completing 19.5 years of service if they are not obligated by a reenlistment contract.

⁷ It is possible that differences in average skill level between veteran and civilian physicians could result in different expected wages, and would explain some of the stay/leave decisions. But there is not really evidence that such differences exist. Nor is it clear that the market would have more or less stringent quality requirements than the military. In any case, our theoretical model captures these differences via the differences between the respective wages and the hedonic values γ. That is, we would see the extent that average skills differed in the difference between W^M + γ^M and W^C + γ^C .

⁸ We exclude from our retention estimates the years some physicians spend as a general medical officer (GMO) prior to going into a residency program.

⁹ We use hospital‐based group practice physicians who are employees—and exclude self‐employed physicians—because they are more comparable to the military, and because using self‐employed physicians would co‐mingle the returns from being a physician with the entrepreneurial returns of self‐employment.

¹⁰ RMC consists of basic pay (BP) plus housing and subsistence allowances (BAH and BAS) plus the tax advantage of not having to pay income taxes on these allowances. The BP is a function only of paygrade and years of service. The BAH is based on paygrade, dependent status (i.e., with or without dependents), and housing location. They are set by the Secretary of Defense to average housing costs in an area, and thus average real BAH is the same for all service members of the same rank and family status. Thus, we assume all service members earn the average BAH.

¹¹ It is conceivable that physicians who are on the margin of leaving the military do not expect that their promotion patterns will match those reflected by physicians who stay. More specifically, they may be on the verge of leaving because they believe that their promotion pattern will be slower than what we observe in the data. For these physicians, our career compensation figures will underestimate their real expected career compensation.

Conversely, physicians who stay in the military could expect their earnings in the civilian sector would be less than what is observed in our data. To the degree that this is true, the military versus civilian difference we observe will be greater than their actual expected difference. If we had data on the actual compensation received by military physicians who move to the civilian sector, we could model these affects more directly.

Since we do not have these data, how might that bias the results? This question gets back to the issue of selection. Raising career compensation of physicians conditional on rank, affects the type of physician who stays, which will, in turn, affect their promotion pattern and their career compensation. This could lead to either higher compensation costs to the DOD and higher retention rates or the reverse depending on the nature of the selection.

¹² We assume that physicians take advantage of all of the medical special pays available to them, including the maximum possible Multiyear Special Pay (MSP). Our logic is that we presume physicians consider all potential military compensation when deciding whether to stay or go, even if they ultimately decide to obligate themselves for less than the maximum length.

¹³ We supplemented these physician income data with data from the American Medical Association. (American Medical Association (1991–2005)).

¹⁴ We deflate pay and retirement benefit values to present value using a personal discount rate of 10%. This is consistent with Black (1984) who asked military personnel in surveys about their preferences for alternative hypothetical military retirement plans. The author found that discount rates declined with income, education, and age, but estimated an average discount rate of 10.3% for officers and 12.5% for enlisted personnel. No study that we know of has explicitly estimated discount rates for physicians. We believe that physician discount rates are likely to be on the low side of estimates because, by the length of their education, they have demonstrated an ability to forgo compensation for future rewards. That said, we examined the sensitivity of the results we present to a discount rate of 20%. Doing so did not materially impact our results (see sensitivity to discount rate in the Results section).

¹⁵ Our estimates of the number of years service members expect to collect retirement is based on the average age at which service members reach retirement eligibility and the actuarial tables gleaned from the Social Security benefits website: http://www.ssa.gov/OACT/STATS/table4c6.html

¹⁶ The ‘accession source’ refers to one of four types: the AFHSPS programs include a civilian medical school scholarship and either military hospital residency (AFHSPS Direct 70%) or civilian residency (AFHSPS Indirect 10%), a scholarship to the armed forces own Uniformed Services University of Health Sciences (USUHS 16%) followed by military residency, and the Financial Aid Program (FAP 4%), which provides an accession bonus of about $30,000 for a fully trained physician.

¹⁷ YOS 0 is the initial year of service.

¹⁸ Note that special pays and allowances do not impact the value of the military retirement benefit, which is function of only military basic pay.

¹⁹ In our data, there are a limited number of cases (less than 10%) where the doctor left the military before the estimated obligation period. For these physicians we estimate what their compensation would have been at the point where we estimate their obligation is up. We do this by hot‐decking compensation estimates from other physicians of similar obligation, training, tenure, type of recruit and branch of the military. This method eliminates bias associated with low military pay due to short periods of tenure at the time of exit.

²⁰ Variations in pay differences within a specialty come mostly from differences in paygrade, YOS, and changes in military and civilian wages over time.

²¹ Because we only include a limited number of specialty indicators, this variable may capture the impact of specialty training on retention that is correlated with time in service, but uncorrelated to specialty pay.

²² USUHS is the omitted category in the estimating equations.

²³ We adjusted all standard errors to account for correlated errors across specialties using the robust cluster adjustment in STATA (White, 1984).

²⁴ Due to correlations between the PV of the annual compensation measures and the PV of the retirement variable we were not able to separately estimate the differential impact of these two characteristics of compensation.

²⁵ Although a given percent change in compensation has a much larger effect on retention for physicians in Period 1 than in Period 3, the estimates of the coefficients of retention for physicians in Period 3 are greater than for those in Period 1. This is because of the larger absolute size of the career compensation for Period 1.

²⁶ For an excellent summary of studies of military pay elasticities, see Warner and Asch (1995).

²⁷ Recall that AFHSPS Direct is the medical school scholarship (to civilian school) and the military residency. The military residency adds a few years to the obligation.

²⁸ We are unable to account for the impact of compensation changes in the number or composition of newly recruited physicians. To the extent that changes in physician compensation also change recruitment, our simulations are likely to under‐ or overestimate the impact of compensation on total person‐years.

²⁹ Although applicable sample sizes are small (22 for FP and 13 for orthopedic surgeons), simulated base case probabilities are inline with actual increases in retention probabilities (about 10 percentage points higher for both FP and orthopedic surgeons). Across the whole sample, simulated retention is nearly identical to actual retention.

³⁰ This method assumes that the difference in the coefficients from the Probit to the IV Probit models is the bias in the model.

³¹ The statistical significance of the underlying coefficients was not affected by the change in discount rate.