Public and Private Options in Practice: The Military Health System

Michael D Frakes; Jonathan Gruber; Timothy Justicz

doi:10.1257/pol.20210625

. Author manuscript; available in PMC: 2023 Nov 29.

Published in final edited form as: Am Econ J Econ Policy. 2023 Nov;15(4):37–74. doi: 10.1257/pol.20210625

Public and Private Options in Practice: The Military Health System

Michael D Frakes ¹, Jonathan Gruber ², Timothy Justicz ^3,^*

PMCID: PMC10686321 NIHMSID: NIHMS1892662 PMID: 38031535

Abstract

Recent debates over health care reform, including in the context of the Military Health System (MHS) and Veterans Administration, highlight the dispute between public and private provision of health care services. Using novel data on childbirth claims from the MHS and drawing on the combination of plausibly exogenous patient moves and heterogeneity across bases in the availability of base hospitals, we identify the impact of receiving obstetrical care on versus off military bases. We find evidence that off-base care is associated with slightly greater resource intensity, but also notably better outcomes, suggesting marginal efficiency gains from care privatization.

Health systems throughout the world typically consist of some mix of public and private health care delivery. In the U.S., most health care is delivered privately, albeit sometimes under contract with public payers, but health care for the U.S. military and veterans has historically been delivered largely through public systems. In many European nations, health care is typically delivered by public systems, while individuals can often “top up” their care using private arrangements with (often the same) providers. But one thing that appears common to all systems is a constant sense of dissatisfaction with the mix. For example, in the UK, the Prime Minister faces pressure to undo the increased tendering of private contracts that began in 2012 (Hussain 2020). On the other hand, in the United States, former President Trump signed the Veterans Affairs’ Mission Act, allowed some veterans to receive care from community providers outside of public VA Medical Centers – and was opposed by many stakeholders, including the former head of the Veteran Affairs Department David Shulkin, who has argued that the VA has special expertise in treating certain issues (Shulkin 2018).

Therefore, understanding the implications of shifting care from public to private health systems, and vice versa, is a key issue worldwide. But this has been a difficult issue to study because the populations using these different systems are quite heterogeneous – and when individuals have a choice between systems, they may do so in a way which is correlated with unobserved determinants of health.

Perhaps for this reason, there is little work of which we are aware comparing delivery between public and private systems of care (particularly in the U.S).² There is a large literature that compares public versus private financing of care, such as through the Medicare Advantage program versus traditional Medicare (e.g. Duggan, Gruber and Vabson, 2018; Curto et al., 2019), or through private Medicaid HMOs versus government direct contracting with physicians (e.g. Geruso et al., 2019). But the actual care delivered through these different financing mechanisms are often at similar sets of providers, so that these differences primarily reflect contracting and care management differences across the public and private sectors.

Closer in spirit are a number of papers that compare delivery of care at publicly owned versus privately owned hospitals; for example, Villa and Kane (2012) find that privatizing public hospitals in the United States results in increased occupancy, shorter stays, and the loss of unprofitable services. Various related studies in the international context use survey instruments to compare outcomes at public and private hospitals, finding that private facilities have both higher prices and higher quality.³ But these international investigations are often subject to selection bias, with papers using propensity matching (Bjorvatn, 2018) and instrumental variables strategies (Pérotin et al., 2013) finding substantial patient differences across the types of providers and little causal difference in quality.⁴ Methodological challenges aside, inquiries into public-versus-private ownership of a particular hospital are unable to speak to the impact of an entirely public-versus-private system of care, which includes the entirety of outpatient and inpatient care, and which incorporates different payment mechanisms and care management strategies.

Surprisingly, the best context for studying a mixed public-private system may be in the U.S.: the Military Health System (MHS). The MHS is a $50 billion/year program that provides care to active-duty military, their dependents, and military retirees, covering over 9 million eligible individuals. Crucially, MHS beneficiaries have access to government-owned and run facilities on military bases, as well as private providers that are contracted to the military through an insurance company. Care is split roughly equally between the two sources; 49 percent of outpatient encounters and more than 67 percent of hospitalizations for MHS beneficiaries take place with private providers (TRICARE Management Activity, 2020).

The proper division of care on and off military bases is a source of controversy and policy action. In particular, the MHS announced plans in 2019 to repurpose roughly 17,000 medical providers (Philpott, 2019)—a development that is likely to push MHS beneficiaries to receive care from civilian facilities—and the Pentagon’s proposed 2021 budget called for transitioning roughly 200,000 non-active-duty beneficiaries to receiving care from civilian providers (Kime, 2020). Yet despite these policy actions, there is no convincing evidence as to whether moving care to civilian providers raises or lowers costs, or improves or deteriorates outcomes.

In this paper, we provide such evidence. This evidence is immediately relevant to the $50 billion MHS, one of our nation’s largest health care programs. But it also has important implications for other systems in the U.S., such as the VA; more broadly, it speaks to the rigorous and ongoing debate between expanding or contracting the role of the private sector in delivering care in the U.S.⁵

A key feature of the military allows us to create a quasi-experimental analysis of the impact of systems of care on patient treatment and outcomes: exogenous moves. Moves across military bases are dictated by the needs of the military and are not solely determined by the preferences of individuals which might be endogenous to their tastes for medical care. Preferences may only be expressed within the set of available bases, which are dictated by the combination of rank, occupation, and time (Lleras-Muney, 2010). Combining beneficiary moves with another key feature of the military context—i.e., variation across bases in the availability of base hospitals—provides us with exogenous variation in access to a public system of care. A further advantage of studying the MHS is that there is a large sample of individuals who themselves receive care both on and off base for a major medical intervention: childbirth. This allows us to hold patient characteristics constant in comparing care delivered from different sources, as well as to take advantage of a detailed set of quality measures facilitated by the childbirth setting. During our sample period there are 776,074 births to 590,353 dependents of active-duty military, and 43.63% of children are delivered at MHS bases.

To carry out this study, we turn to an innovative data set that has been little used by economists, the Military Health System Data Repository data. These data provide complete medical claims for all military personnel and their dependents over the 2003–2013 period. Critically, they provide detail on care delivered on and off military bases.

Our analysis follows the approach of Finkelstein et al. (2016) in a discrete event framework. We consider every mother in our sample who gives birth at least twice. We then restrict our sample to all mothers who move to a new base between births, and we compare those mothers who see a change in the availability of military hospitals. That is, we compare mothers who move but both before and after the move do (or do not) have a military hospital nearby, to mothers who move and who see a change in their nearby access to military hospitals. In this way, we control for both underlying mother characteristics, as well as any effect of moves per se. Following Finkelstein et al. (2016), we provide evidence that our estimates may be causal by examining the previous births for all such mothers. We begin with a transparent presentation of mean birth outcomes among movers. We then move to an instrumental variables (IV) framework which allows us to carry out overidentification tests to confirm that the results are similar for mothers moving either closer or further from base hospitals when they give birth.

We find that mothers delivering off-base use more resources than those delivering on-base; total resource utilization appears to be about 1% higher for those using the private- rather than public-care system, primarily driven by higher Cesarean section rates off-base. At the same time, we find that the quality of care appears to be higher for mothers delivering and receiving prenatal care off-base. We find that mothers and babies receiving off-base care have fewer severe complications. There is also a substantially lower rate of unplanned readmissions within 30 days of discharge. Our results suggest that, at current levels, shifting childbirth from on- to off-base is likely to be cost effective.

Our paper proceeds as follows. Section I provides background on the MHS and a comparison to the Veterans Health Administration (VHA). Section II discusses our data, while Section III describes our empirical strategy. Section IV shows results, and Section V concludes.

Part I: Institutional Background

The Military Health System (MHS) is the primary insurer for all active-duty military, their dependents, and many military retirees through the TRICARE program. Of the nearly 9 million patients covered by TRICARE, only 20% are actively serving in the US military. TRICARE is not involved in health care delivery in combat zones and operates separately from the Department of Veterans Affairs’ Veterans Health Administration (Schoenfeld et al. 2017). For TRICARE enrollees, care can be delivered in one of two ways: either directly at Military Treatment Facilities (MTFs) on military bases (direct care), or purchased from private providers (purchased care).

The 42 MTF hospitals in the U.S. provide both outpatient and inpatient care to TRICARE enrollees.⁶ MTF care represents the classic public delivery mechanism. All resources are owned by one of the branches of the U.S. military. Individual MTFs are managed by active-duty military officers who draw a salary that is not tied to the performance of their MTF. Their incentives to perform well come from a sense of duty as service-members and physicians, and from a desire to advance their careers and attain positions of greater responsibility. Medical departments or services, such as OBGYN, are led by physicians or other providers as appointed by the MTF commander (HQDA, 2004).

For those treated at MTFs, the care is delivered by a mix of providers including active-duty military providers (49 percent), federal civilian employee providers (40 percent), and providers hired using contract mechanisms who work full time at the MTF (11 percent) (Defense Health Program, 2018). Providers are primarily salaried with payment not explicitly tied to either quantity or quality of care delivered.⁷ In almost every aspect, care at MTFs approximates the type of publicly provided care delivered in the UK and other systems with public health care delivery.

Alternatively, TRICARE enrollees can receive “purchased care” outside the MTF. This care is delivered by a network run by a contracting insurer—United HealthCare, in the case of our sample period—where patients go to private providers within the insurer’s contracted network. This purchased care system very much parallels the public purchase of private care, such as through the Medicare Advantage program. However, contractors for TRICARE are not paid a capitated rate per beneficiary per month. They are instead paid for their services and manage payment to providers on a per-claim basis. Therefore, by comparing direct to purchased care, we are comparing care provided directly by the government to care provided by private contractors paid by the government.

Providers under purchased care are paid Medicare rates, and all inpatient stays are subject to bundling into a single Diagnosis Related Group (DRG) based on age, sex, diagnoses, procedures, and discharge status using the Medicare Grouper. Individual discharges may also be further reimbursed for high cost and length-of-stay outliers. The TRICARE Reimbursement Manual notes that, while this system is functionally identical to Medicare’s prospective payment system, weights and payments may vary because of the characteristics and needs of the beneficiaries (OASD, 2018). While Medicare is associated with people over the age of 65, it also provides care for Americans with permanent disabilities or end-stage renal disease. According to Medicare.org, “there are over 1 million female Medicare recipients under the age of 65.” Therefore, Medicare rates are applicable to childbirth services.

The network of MTFs is maintained primarily to ensure military readiness. The MHS is organized with three principal goals: to maintain the health of uniformed personnel, to provide medical professionals that can deploy worldwide, and to provide a medical benefit to active-duty personnel, retirees, and their families. In practice, the families of uniformed personnel require more medical procedures, and this is reflected in the composition of the most common directly provided and purchased procedures. In 2013, six of the seven most common procedures provided at MTFs were related to delivery. From most to least common, these were vaccination, circumcision, repair of obstetric laceration, Cesarean section, manually assisted delivery, and two different procedures for the induction of labor. Similarly, four of the top seven privately provided procedures purchased by the MHS were related to childbirth. In order these were: manually assisted delivery, total knee replacement, Cesarean section, circumcision, vaccination, repair of obstetric laceration, and insertion of a stent. In terms of volume, 12,400 C-sections were performed on-base in 2013, while 20,400 were purchased from private providers by the MHS.

In principle, enrollees who live within the “catchment area” of an MTF are supposed to go to the MTF for care. This area was defined as 40 miles originally, though the military has shifted to time-based boundaries. Our data show clearly that a mileage boundary rule was not rigorously enforced during our sample period. Those who live closer to an MTF are much more likely to go there, but with a more gradual fall off rather than a strong distance discontinuity.

Importantly, moves are not under the control of active-duty enrollees. Moves within the military are driven by Department of Defense personnel management strategies, motivated by staffing needs across units. As personnel leave or are promoted, units must be reorganized so that each formation has the appropriate composition. Official policy is that these decisions are made according to “current qualifications and ability to fill a valid requirement,” and this takes priority over all other considerations, including individual preferences or time at current locations. Furthermore, assignments “will not be influenced by the employment, school enrollment, volunteer activities, or health of a Service member’s family member” (DoD 2015). Several papers have demonstrated that the timing and frequency of relocations is not subject to soldier preferences, including Lyle (2006), Lleras-Muney (2010), Carter and Skimmyhorn (2017), and Carter and Wozniak (2018). While the choice of location may be influenced by Soldier preferences, these choices are constrained by the timing of the move, and the sponsor’s rank and occupation. Accordingly, we follow Lleras-Muney (2010) in controlling for time, rank, and occupation throughout our estimates. Importantly, we argue only that moves across bases are orthogonal to family health-care system preferences. Moves within a catchment area may be endogenous since they are unlikely to be driven by personnel management needs; we therefore exclude such moves from our analysis. A related concern is the relationship between distance to the nearest MTF and preferences over health care delivery, since all mothers in our sample exercise a degree of control over their residential locations. To allay concerns, we control for distance in Table A7, finding that it has little effect on our estimates.

Part II: Data

Our data come from the Military Health System Data Repository, a comprehensive set of health records collected at MTFs and from civilian providers reimbursed under TRICARE. The data are comprised of several separate repositories, including administrative eligibility files and separate sets of claims data for direct and purchased care, both inpatient and outpatient. While TRICARE does not pay claims for direct care, the data are structured as if they were claims and include detailed information such as physician identifiers, diagnosis and procedure codes, length of hospitalization, and measures of treatment intensity. The sample period covers fiscal years 2003 through 2013.

In our comparison of the care provided at MTF hospitals versus civilian facilities, we focus on inpatient care for childbirth, for a number of reasons. First, we have a very large sample of births, as this is the most common type of hospital admission for the MHS population. Second, many women in the sample period deliver more than once, allowing us to control for unobservable heterogeneity by focusing on within-mother differences in outcomes across births. Third, the delivery context affords us the ability to explore both birth-specific measures of treatment intensity, like the prevalence of C-sections, and detailed health outcomes, such as preventable complications. Finally, admission for birth is nearly universal among pregnant women, minimizing the potential for selection into the sample on the dimension of whether to receive care at all.

We construct a sample consisting of the wives of active-duty service-members to facilitate the comparability of on and off base care. This sample limitation is important for two reasons. First, we exclude active-duty service members (who account for 21 percent of births) because, as Frakes and Gruber (2019, 2020) document, they face different medical malpractice liability environments which would limit our ability to extrapolate our findings to other settings. We use the active-duty partner’s rank, occupation, and race since these characteristics are not in the data for dependents. Second, we remove retirees and their dependents (who account for only 3.5 percent of births) since under TRICARE rules they have different cost sharing on- and off-base. We also remove all transfers from on to off-base or vice versa.

After restricting to mothers who give birth at least twice in our sample window, and who move across military installations between births (for reasons discussed below), we have a sample of 87,005 women who are admitted for labor and delivery 182,779 times. Out of these admissions, 55 percent are off-base, while 45 percent are at MTF hospitals.

We use several variables to measure treatment intensity. The most important is Relative Weighted Product (RWP), an encounter-specific metric created by the MHS to compare the intensities of inpatient resource use across the direct- and purchased-care systems. This measure is a function of both the DRG weight associated with the admission and the length of the inpatient stay. A RAND study found an 80 percent correlation between RWP and cost for surgical cases (Farley et al. 1999). The use of this RWP measure is necessary because, while the MHS records cost allocations to each inpatient stay both on and off the base, they only record charges for stays off the base.⁸ In addition to RWP, we consider separately the rate of C-sections, the number of additional (non-C-section) procedures, the length of stay for mothers (bed days) and newborns (bassinet days), and the number of diagnostic procedures.⁹

In addition to treatment intensity, we consider several measures of care quality and patient outcomes. First, we consider measures indicating severe and preventable complications. Complications are very common during childbirth, occurring in more than two-thirds of admissions in our sample. For example, the single most common is trauma or lacerations of the perineum, occurring in 40 percent of admissions. However, this complication is rarely present in cases of delivery by C-section. Therefore we construct indicators for severe and preventable complications. Severe complications are a set of dangerous complications identified by the World Health Organization, including for example severe preeclampsia, a potentially life-threatening condition. Other complications classified as severe are third stage hemorrhage, rupture of uterus, and sepsis. Note that these complications are not prevented by C-sections. Preventable complications include bleeding problems, such as post-partum hemorrhage and post-partum coagulopathy, and several conditions of labor such as unusually fast delivery, called precipitate delivery, or unusually long labor. Some of these latter preventable complications can be mitigated using C-sections. Severe and preventable complications are not mutually exclusive—bleeding problems are considered both severe and preventable. We also consider a measure called severe acute maternal morbidity, or SAMM. SAMM consists of life-threatening conditions for the mother such as “embolism, acute renal failure, stroke, acute myocardial infarction,” and others. Callaghan et al. (2012), who developed an updated version of the measure, estimate that SAMM is roughly 100 times more common than death. Maternal mortality outcomes were, fortunately, too rare to offer precise estimates.

We are, however, able to measure neonatal mortality. Typically, infant mortality is defined as death during the first 28 days after birth. To consider a longer-term outcome, we also measure infant mortality within one year of birth. Third, we look at unplanned readmissions of the mother within 30 days of discharge. Unplanned readmissions are inpatient admissions to short-term acute care hospitals within 30 days of discharge from an “index admission” (CMS 2016).¹⁰ We also consider unplanned readmissions of the child within 30 days and within one year of discharge. Given the importance of child health to understanding quality of care, we also consider any hospital admission for the child within the same time frames.

One disadvantage of limiting our analysis to birth admissions is that pre-hospitalization care—prenatal care—affects birth outcomes. This is potentially problematic to the extent women may receive on-base prenatal care and deliver off-base, or vice versa. Indeed, there is a sizeable share of such “mixed care”. Of women who live within 40 miles of an MTF and who deliver on-base and for whom we can clearly identify pre-natal care, 22% get at least some of their pre-natal care off-base; conversely, of women who live more than 40 miles from an MTF and deliver off-base, 20% get at least some of their pre-natal care on-base.¹¹ We do not have a separate instrument for receipt of pre-natal versus delivery care, so we are treating these as a package test of public-versus-private delivery. Below, we show separately results for the three-quarters of our sample for whom all pre-natal and childbirth care is delivered within the same system. We also attempt to assess the relative role of these two factors by looking at outcomes mostly determined by pre-hospital care (low birthweight and premature births) and by looking at mothers who move during pregnancy.

Sample Generalizability

To understand the nature of selection into our sample, we compare the women in our sample to those in the National Health Interview Survey (NHIS) using similar sample selection criteria—married, recent mothers from the same time periods. The results are presented in Table A1. In comparison to the NHIS sample, we find that women in our sample are on average three years younger, and less than half as likely to have a baby with low birth weight. We also find that the husbands of women in our sample are less likely to be White, more likely to be Black, and less likely to have a four-year college degree, as proxied by status as a military officer.

Part III: Empirical Strategy

The objective of this paper is to determine whether treatment intensity and quality differ between the “direct” (public) and “purchased-care” (private) systems—i.e., between MTF-based care / management and private-based care / management. We frame this empirical comparison by considering off-base private care the treatment, making those who give birth at MTFs the untreated group. The key inferential challenge for this exercise is that selection to purchased care is non-random. Simply comparing outcomes of direct and purchased-care encounters will be confounded by observable and non-observable differences between patients.

Finkelstein et al. (2016) develop an approach to address this concern in the context of Medicare spending. They face a similar challenge: they are interested in determining the extent to which geographic area differences in spending are due to true differences in area characteristics as opposed to the selection of patients who choose to live in those areas. Their approach is to create a sample of Medicare movers and to compare the change in movers spending based on the differences in average area spending in their ex-ante and ex-post locations. In their context, the concern is that such moves are endogenous – e.g. that unobservably sicker people will move to high spending areas. To address this, Finkelstein et al. (2016) study the outcomes of movers before they move – showing that those who move to more and less expensive areas are quite similar before the move but differ thereafter in ways correlated with area expense. We can apply this same approach to the context of discrete medical events - with the further advantages that the timing of military moves is exogenous, and the set of locations is constrained by military requirements.

Fundamentally, both approaches rely on the logic of differences-in-differences; as Finkelstein et al. state, the “key identifying assumption is that such differential trends do not vary systematically with the migrant’s origin and destination.” However, our context has several significant advantages when compared to Finkelstein et al. (2016). First, and perhaps most important, we do not need to assume that moves to one location type have the same effect on individuals from more than one other location type. We consider only one treatment (private care) relative to one untreated control (public care), while Finkelstein et al. (2016) estimate 305 separate healthcare market treatments. Second, we do not rely on non-movers to identify effects on movers. Finkelstein et al. must assume that the place effects that are relevant for movers are the same as those that are relevant for non-movers. Finally, given our sample restrictions, we know a priori that moves in our data are induced by military relocations, which gives us a plausibly exogenous (non-health-related) source of moves across location types. By contrast, Finkelstein et al. “cannot allow for shocks to utilization that coincide exactly with the timing of the move and that are correlated with utilization in the origin and destination,” for example, newly diagnosed patients could respond to illness by moving to areas with higher utilization. We will show that, conditional solely on initial location type, we appear to have conditional random assignment to subsequent location types.

In particular, let Y_it (d) be potential outcome Y for individual i in period t, given treatment status d, where d = 0 if a woman lives within 40 miles of an MTF hospital, and d = 1 for women who live more than 40 miles away from an MTF hospital.¹² For any outcome Y, we can measure the impact of losing access to an MTF hospital as Y_i1 (1) – Y_i0 (0). Likewise, we can define the impact of gaining access to an MTF hospital as Y_i1 (0) – Y_i0 (1).

There are two fundamental threats to identification in this context. The first is that movers and non-movers are not comparable. As we argued above, moves are assigned by the military and should not be endogenous to individual characteristics once we have controlled for time, occupation, and rank. But it is conceivable that pregnancy could impact the timing of moves. Moreover, it is possible that moving itself impacts birth outcomes. We can readily address both of these concerns by restricting our analysis only to mothers who move between births.

The second is the potential endogeneity of move location, despite the institutional factors that suggest exogenous move timing. To assess this possibility, we pursue the discrete analogy to the event-study approach of Finkelstein et al. (2016) and examine the previous birth of movers. That is, if those movers who are high risk are systematically moved to a base with an available base hospital, then this should show up in those same mothers having higher-risk first births.

On a final note regarding potential endogeneity of moves, we restrict our inquiry to mothers with multiple births who move between births to a new base location. That is, the variation we capture in whether patients live within 40 miles of a base hospital derives from moves across bases that differ in their availability of a base hospital, rather than from internal moves within bases that cross this 40-mile threshold. Such within-base moves are more likely to be endogenous with respect to treatment decisions since they are not tied to military reassignments.

Given these sample restrictions, we have four groups:

mothers who move from a base close to an MTF hospital to a different base close to an MTF hospital
mothers who move from a base close to an MTF hospital to a base far from an MTF hospital
mothers who move from a base far from an MTF hospital to a different base far from an MTF hospital
mothers who move from a base far from an MTF hospital to a base close to an MTF hospital

To transparently illustrate the power of this approach, we begin by showing simple comparisons of our key measures for these four sets of possible move combinations. We start with the sample of mothers whose first birth occurs when they live within 40 miles of an MTF hospital; within this sample, we then compare those whose second birth takes place after they move within 40 miles of a different MTF hospital, group (a), to those whose second birth takes place after they move more than 40 miles away from any MTF hospital, group (b). Likewise, we consider a sample of mothers whose first birth occurs when they live more than 40 miles from an MTF hospital; within this sample, we then compare those whose second birth takes place after they move more than 40 miles from a different MTF hospital, group (c), to those whose second birth takes place after they move to within 40 miles of an MTF hospital, group (d).

Of course, our approach has one other key difference from Finkelstein et al. (2016), which is that we are only measuring an intent-to-treat effect, since some of our movers will be close to military hospitals but still use non-military providers. To address this, we can implement an instrumental variables strategy. In particular, the reduced form is the following formalization of our simple intent-to-treat:

Δ Y_{i t} = α_{i} + γ_{t} + β Δ D_{i t} + δ D_{i t - 1} + μ X_{i t} + ε_{i t}

(1)

where ΔY_it is the change in outcome Y for individual i at period t relative to period t − 1. ΔD=0 for groups (a) and (c), ΔD = 1 for group (b), and ΔD = −1 for group (d). We control for certain observable birth characteristics (as of the post-move birth) in X_it: dummies for calendar year, age group, sponsor’s pay grade and occupation, and sponsor’s race (because we do not observe maternal race); in a specification check, as discussed below, X_it also includes dummies for various maternal risk-factors. This reduced form is causally identified under the usual parallel trends assumption between groups (a) and (b) and between groups (c) and (d).¹³

Given this framework, we can estimate second stage estimates of the form:

Δ Y_{i t} = α_{i} + γ_{t} + β Δ O_{i t} + δ O_{i t - 1} + μ X_{i t} + ε_{i t}

(2)

where ΔO_it is equal to 1 for women who give birth off-base in period t = 1 and on-base in period t = 0. ΔO_it is equal to −1 for women who give birth at an MTF in period t = 1, but who give birth off-base in period t = 0. To address the fact that choice of delivery location is endogenous, we instrument the endogenous regressor ΔO in the equation above. To identify the coefficient of interest, β, we use two instruments:

1(ΔD = 1) is an indicator for moving from within 40 miles of an MTF to a new location more than 40 miles from an MTF.

1(ΔD = −1) is an indicator for moving from a location that is more than 40 miles from an MTF to a new location within 40 miles of an MTF.

These instruments use the plausibly exogenous variation in MTF access from move location as an instrument for the actual location of the birth. Moreover, the restriction of our model to incorporate symmetric responses to moves both away from and towards military facilities allows us to implement an overidentification test of our estimates.

For this instrumental variable estimate to be causally identified, we must assume that the exclusion restriction holds. In our context this means that the only way that a military-reassignment-induced change in distance from an MTF hospital affects a change in birth outcomes is by causing a change in birth location on- or off-base – and not other area factors that might drive outcomes when individuals move. To assess this issue, Table A2 compares key features of the catchment areas with and without MTFs available, using both information on births from vital statistics, and population from the American Community Survey (ACS). We find that birth characteristics are very similar across the two types of areas, with very similar age at birth and rates of prematurity, and slight differences in low birth weight. The areas with MTFs do have significantly lower rates of tobacco use, which would suggest if anything that outcomes should be better in those areas (and not worse as we find). As far as population characteristics, the areas with MTFs are slightly higher income, higher share black, and have higher education and lower poverty rates. The major differences, not surprisingly, is a higher share of the population with TRICARE insurance and a higher share veterans, and that the areas with MTFs are significantly larger with more births.

While area characteristics may differ, TRICARE beneficiaries are unique in that their access to comprehensive health insurance is unaffected by moves, and the active-duty sponsor’s compensation is also unaffected by location.¹⁴ Additionally, for the majority of couples, the active-duty husband is the main income provider. During this period, roughly 60% of military wives did not work, and “80% of military wives [earned] less than $30,000” (Rapino and Beckhusen, 2013). We are also reassured by the robustness of our results to controlling for the volume of area surgical inpatient cases.¹⁵ Thus, not only are area differences modest, military families are insulated from some of the effects of moves relative to their civilian counterparts.

We show the first stage for this IV model in Table A3. Both instruments are highly significant with nearly equal and opposite signs. The F-statistic for the set of instruments is 557, well above the standard suggested by Lee et al. (2020). The first stage itself shows us that nearly two-thirds of the sample are compliers with the instrument. We can also use the first stage to estimate the proportion of the treated that are compliers—this is a relatively high 56%. By assumption, the remainder of the sample is split between always-takers (who would always give birth off-base), and never-takers (who would always find a way to give birth on-base, even if they must travel to a MTF hospital). We assume no defiers. Such individuals would only give birth off-base if the nearest base had an MTF-hospital. Such behavior would not only be implausible, but it would also violate the monotonicity of our instrument.

We further characterize compliers in Table 1. We can see that compliers have slightly fewer bed days and lower utilization. In large part, this is because compliers have a C-section rate that is almost three percentage points below average. Importantly, compliers have higher rates of low birth weight, and higher rates of infant mortality measured at both 28 days and at one year. We can also see that compliers are slightly older, are almost nine percentage points less likely to have White sponsors and are three percentage points more likely to have Black sponsors.

Table 1:

Descriptive Statistics

	Mean	SD	25^th Pctl	75^th Pctl	2^nd Birth Mean	2^nd Birth Complier Mean
Bed Days	2.315	1.890	2	3	2.159	2.097
Bassinet Days	2.467	4.535	1	2	2.361	2.142
RWP	0.499	0.154	0.392	0.695	0.503	0.490
Procedures	2.256	1.263	1	3	2.134	2.191
Diagnostics	0.085	0.748	0	0	0.100	0.106
C-Section	0.262	0.440	0	1	0.282	0.252
Severe Complication	0.041	0.199	0	0	0.036	0.031
Preventable Complication	0.223	0.416	0	0	0.189	0.210
Severe Acute Maternal Morbidity	0.011	0.103	0	0	0.012	0.013
28-Day Infant Mortality (per 1,000)	1.118	33.251	0	0	0.804	0.935
1-Year Infant Mortality (per 1,000)	2.126	45.777	0	0	1.576	2.266
Unplanned Readmission (per 1,000)	12.372	110.540	0	0	12.089	10.499
Unplanned Child Readmission (per 1,000)	37.855	190.846	0	0	36.080	33.891
Any Child Hospitalization within 30 Days (per 1,000)	55.886	229.702	0	0	53.917	55.197
1-Year Unplanned Child Hospitalization (per 1,000)	85.562	279.717	0	0	84.343	97.857
Any Child Hospitalization within 30 Days (per 1,000)	103.472	304.576	0	0	102.091	119.412
Low Birth Weight	0.033	0.178	0	0	0.031	0.034
Premature	0.057	0.232	0	0	0.057	0.050
Age	26.933	4.844	23	30	28.454	28.654
White	0.777	0.416	1	1	0.779	0.693
Black	0.104	0.305	0	0	0.102	0.132
N Admissions	182,777				95,774
N Mothers	87,004

	Starting with MTF Access				Starting Without MTF Access				(3)+(7)=0
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	MTF/ MTF	MTF/ No MTF	Difference	t-stat	No MTF/ No MTF	No MTF/ MTF	Difference	t-stat	t-stat
Panel 1: Utilization
Bed Days	2.535	2.530	−0.005	0.262	2.475	2.519	0.045	2.465*	1.767
Bassinet Days	2.596	2.613	0.018	0.388	2.620	2.622	0.002	0.046	0.372
RWP	0.496	0.496	−0.001	0.478	0.497	0.498	0.001	0.316	0.130
Procedures	2.492	2.486	−0.005	0.389	2.316	2.289	−0.027	2.020	2.007
Diagnostics	0.062	0.059	−0.003	0.513	0.068	0.088	0.020	2.879*	2.285*
C-Section	0.236	0.232	−0.004	1.088	0.251	0.254	0.003	0.719	0.287
Panel 2: Outcomes
Severe Complication	0.053	0.056	0.004	1.757	0.041	0.040	−0.001	0.312	1.348
Preventable Complication	0.292	0.287	−0.005	1.222	0.233	0.225	−0.008	1.687	2.506*
Severe Acute Maternal Morbidity	0.010	0.010	0.000	0.366	0.008	0.010	0.002	2.095	2.128
28-Day Infant Mortality (per 1,000)	1.420	1.462	0.042	0.116	1.286	1.223	−0.063	0.161	0.035
1-Year Infant Mortality (per 1,000)	2.543	2.775	0.232	0.472	3.025	2.194	−0.831	1.480	0.888
Unplanned Readmission (per 1,000)	13.876	15.372	1.496	1.332	10.304	11.125	0.820	0.731	1.803
Unplanned Child Readmission (per 1,000)	40.705	42.311	1.606	0.837	36.819	38.574	1.755	0.820	1.439
30-Day Child Hosp. (per 1,000)	60.493	62.194	1.701	0.738	53.166	55.085	1.919	0.757	1.295
1-Year Unplanned Child Hosp. (per 1,000)	88.807	90.795	1.988	0.722	78.160	85.546	7.386	2.394*	2.700*
Any 1-Year Child Hosp. (per 1,000)	108.266	111.018	2.752	0.916	94.112	101.807	7.696	2.306*	2.787*
Sup-T Critical Value (0.05)				2.153				2.168	2.154
Share Giving Birth Off-Base	0.266	0.267			0.971	0.971
N	31,671	17,581			14,967	19,820

		OLS	IV
	(1) Mean	(2) β	(3) β	(4) p-value
Bed Days	2.159	0.233*** (0.034)	0.147*** (0.046)	0.580
Bassinet Days	2.361	0.430*** (0.069)	0.120 (0.092)	0.905
RWP	0.503	0.014*** (0.002)	0.011*** (0.003)	0.477
Procedures	2.134	0.029 (0.076)	0.105 (0.108)	0.822
Diagnostics	0.100	0.104*** (0.009)	0.062*** (0.015)	0.774
C-Section	0.282	0.036*** (0.004)	0.029*** (0.005)	0.479
Severe Complication	0.036	−0.019*** (0.003)	−0.020*** (0.004)	0.053
Preventable Complication	0.189	−0.081*** (0.016)	−0.087*** (0.019)	0.726
Severe Acute Maternal Morbidity	0.012	0.000 (0.001)	−0.002 (0.002)	0.172
28-Day Infant Mortality (per 1,000)	0.804	−0.149 (0.282)	−0.352 (0.633)	0.997
1-Year Infant Mortality (per 1,000)	1.576	−0.730* (0.380)	−0.487 (0.817)	0.557
Unplanned Readmissions (per 1,000)	12.089	−6.057*** (1.373)	−5.478*** (1.947)	0.801
Unplanned Child Readmission (per 1,000)	36.080	−4.494 (3.081)	−6.515 (4.877)	0.338
30-Day Child Hospitalization (per 1,000)	53.917	−6.657* (3.762)	−16.272*** (5.757)	0.856
1-Year Unplanned Child Hospitalization (per 1,000)	84.343	−4.895 (3.816)	−7.742 (5.860)	0.070
Any 1-Year Child Hospitalization (per 1,000)	102.091	−7.385* (4.091)	−17.894*** (6.668)	0.174
N		95,773	95,773

	1^st Quartile (P < 7.21%)			4^th Quartile (P ≥ 43.90%)
	(1)	(2)	(3)	(4)	(5)	(6)
	Mean	β	p-value	Mean	β	p-value
Bed Days	1.904	0.061 (0.054)	0.372	2.640	0.465*** (0.075)	0.982
Bassinet Days	1.945	0.086 (0.101)	0.478	3.138	0.200 (0.243)	0.487
RWP	0.429	0.007** (0.003)	0.425	0.693	0.020*** (0.005)	0.819
Procedures	2.168	0.239* (0.123)	0.910	1.913	−0.170* (0.101)	0.095
Diagnostics	0.061	0.039** (0.017)	0.863	0.174	0.102*** (0.038)	0.698
C-Section	0.060	0.013* (0.007)	0.327	0.883	0.063*** (0.011)	0.795
Severe Complication	0.034	−0.014* (0.007)	0.562	0.038	−0.025*** (0.008)	0.237
Preventable Complication	0.213	−0.120*** (0.027)	0.594	0.096	−0.045*** (0.015)	0.829
Severe Acute Maternal Morbidity	0.006	−0.001 (0.002)	0.748	0.023	−0.008 (0.006)	0.285
28-Day Infant Mortality (per 1,000)	0.423	−0.080 (1.044)	0.702	1.196	−1.625 (1.291)	0.811
1-Year Infant Mortality (per 1,000)	1.129	1.128 (1.572)	0.463	1.870	−1.582 (2.006)	0.284
Unplanned Readmissions (per 1,000)	7.401	−4.366 (2.617)	0.942	18.739	−8.406* (4.456)	0.407
Unplanned Child Readmission (per 1,000)	35.091	−3.224 (8.744)	0.479	33.231	−8.727 (6.100)	0.107
30-Day Child Hospitalization (per 1,000)	48.108	−6.338 (9.223)	0.681	60.817	−25.122** (10.241)	0.758
1-Year Unplanned Child Hospitalization (per 1,000)	79.945	−5.475 (10.035)	0.659	84.787	−3.014 (9.761)	0.036
Any 1-Year Child Hospitalization (per 1,000)	92.779	−7.974 (10.679)	0.814	111.063	−19.237 (12.093)	0.401
N		22,138			23,957

	1^st Quartile (Rate Diff. < −0.052)			4^th Quartile (Rate Diff. ≥ −0.009)
	(1) Mean	(2) β	(3) p-value	(4) Mean	(5) β	(6) p-value
Bed Days	2.117	0.151 (0.120)	0.270	2.227	0.149* (0.085)	0.485
Bassinet Days	2.390	0.137 (0.253)	0.048	2.422	0.010 (0.184)	0.461
RWP	0.502	0.021*** (0.004)	0.761	0.505	0.002 (0.005)	0.684
Procedures	1.954	0.334*** (0.144)	0.227	2.212	0.090 (0.175)	0.861
Diagnostics	0.104	0.055 (0.038)	0.055	0.097	0.055** (0.027)	0.699
C-Section	0.285	0.050*** (0.008)	0.303	0.285	−0.001 (0.011)	0.939
Severe Complication	0.034	−0.019*** (0.008)	0.978	0.032	−0.021*** (0.007)	0.073
Preventable Complication	0.175	−0.072*** (0.026)	0.550	0.173	−0.090*** (0.023)	0.090
Severe Acute Maternal Morbidity	0.011	−0.002 (0.005)	0.790	0.012	0.002 (0.003)	0.021
28-Day Infant Mort. (per 1,000)	0.569	−1.060 (1.122)	0.563	1.022	−0.046 (1.594)	0.402
1-Year Infant Mort. (per 1,000)	1.184	−1.660** (0.787)	0.820	1.877	0.152 (2.174)	0.991
Unplanned Readmits (per 1,000)	12.347	−3.274 (3.081)	0.437	10.499	−1.875 (5.436)	0.781
Unplanned Child Readmission (per 1,000)	37.786	−20.630*** (7.156)	0.453	35.506	12.080 (9.455)	0.467
30-Day Child Hospitalization (per 1,000)	57.259	−32.835*** (4.906)	0.162	52.460	−3.813 (11.621)	0.836
1-Year Unplanned Child Hospitalization (per 1,000)	90.485	−33.878*** (10.768)	0.111	82.192	17.182 (11.136)	0.361
Any 1-Year Child Hospitalization (per 1,000)	110.600	−49.421*** (10.398)	0.057	98.535	3.583 (12.559)	0.665
N		22,873			24,974

	MHS	NHIS	ACS
	(1)	(2)	(3)
Age	26.93 [4.84]	30.11 [5.61]	30.85 [5.90]
Age at First Birth	24.54 [4.30]	25.05 [5.72]	--
Number of Children/ Number of Children in Household (ACS)	2.27 [1.20]	2.13 [1.12]	1.99 [1.20]
Low Birth Weight	0.033 [0.178]	0.072 [0.250]	--
Husband Information
White	0.777 [0.416]	0.845 [0.362]	0.790 [0.407]
Black	0.104 [0.304]	0.074 [0.262]	0.058 [0.234]
Officer (MHS) or Four-Year College Graduate (NHIS)	0.257 [0.437]	0.317 [0.466]	0.391 [0.488]
In the Labor Force	1 [0]	0.966 [0.182]	0.950 [0.217]
Unemployed	0 [0]	0.041 [0.199]	0.039 [0.194]
N	182,779	1,130	254,361

	No MTF	MTF
	(1)	(2)
Panel 1: Vital Statistics Natality Birth Data (2004)
Age at Birth	27.59	27.86
Number of Prenatal Visits	11.46	11.50
Premature	12.22%	12.27%
Low Birth Weight	8.42%	8.05%
Mother is White	68.80%	63.13%
Mother is Black	9.37%	12.31%
Mother Used Tobacco	8.45%	5.88%
Mother Drank Alcohol	0.49%	0.52%
Births in Catchment Area Counties	17,802.16	24,218.31
N Counties	76	141
Panel 2: ACS Data (2013)
Household Income	$81,542.34	$86,597.00
Proportion Black Population	15.05%	16.28%
Proportion White Population	75.44%	73.70%
Population with Health Insurance	85.56%	86.18%
Medicaid	16.85%	16.42%
Medicare	15.71%	15.04 %
Population with TRICARE	3.65%	7.06%
Labor Force Participation Rate	60.91%	62.10%
Unemployment Rate	8.13%	7.94%
College Graduates	19.94%	21.01%
Poverty Rate	15.15%	14.35%
Veterans	6.50%	7.49%
N Counties	61	119

	ΔOff-Base
	(1) β
MTF Hospital to No MTF Hospital	0.673*** (0.019)
No MTF Hospital to MTF Hospital	−0.636*** (0.013)
F-Statistic	556.989
p-value	0.171
Outcome Mean	0.013
R²	0.441
N	95,773

	(1) Mean	Single Instrument	Two Move Types
	(1) Mean	(2) β₁	(3) β_b	(4) β_d
Bed Days	2.159	0.155*** (0.044)	0.126** (0.056)	−0.143*** (0.046)
Bassinet Days	2.361	0.080 (0.089)	0.122 (0.105)	−0.096 (0.107)
RWP	0.503	0.009*** (0.002)	0.009*** (0.003)	−0.010*** (0.003)
Procedures	2.134	0.101 (0.105)	0.095 (0.107)	−0.096 (0.096)
Diagnostics	0.100	0.064*** (0.014)	0.056*** (0.016)	−0.056*** (0.019)
C-Section	0.282	0.027*** (0.005)	0.025*** (0.006)	−0.027*** (0.005)
Severe Complication	0.036	−0.021*** (0.004)	−0.025*** (0.005)	0.011** (0.005)
Preventable Complication	0.189	−0.086*** (0.018)	−0.081*** (0.021)	0.078*** (0.016)
Severe Acute Maternal Morbidity	0.012	−0.001 (0.002)	−0.004 (0.003)	−0.001 (0.002)
28-Day Infant Mortality (per 1,000)	0.804	−0.263 (0.545)	−0.334 (0.855)	0.306 (0.746)
1-Year Infant Mortality (per 1,000)	1.576	−0.965 (0.704)	−0.035 (1.140)	0.904 (1.017)
Unplanned Readmissions (per 1,000)	12.089	−6.066*** (1.760)	−4.799* (2.485)	5.212** (2.646)
Unplanned Child Readmit (per 1,000)	36.080	−5.354 (4.492)	−9.397 (5.487)	2.033 (5.895)
30-Day Child Hosp. (per 1,000)	53.917	−13.893** (5.453)	−16.323** (6.520)	13.127 (7.731)
1-Year Unplanned Child Hosp. (per 1,000)	84.343	−4.527 (5.517)	−16.302** (7.280)	−3.377 (7.458)
Any 1-Year Child Hosp. (per 1,000)	102.091	−13.470** (6.253)	−25.017*** (8.481)	6.468 (8.837)
N		95,773	95,773

	Move from Base with MTF		Move from Base without MTF
	(1) Mean	(2) β	(3) Mean	(4) β
Bed Days	2.160	0.149*** (0.052)	2.158	0.139** (0.055)
Bassinet Days	2.346	0.125 (0.100)	2.385	0.090 (0.121)
RWP	0.503	0.010*** (0.003)	0.504	0.008** (0.003)
Procedures	2.141	0.102 (0.108)	2.124	0.082 (0.106)
Diagnostics	0.102	0.065*** (0.015)	0.097	0.065*** (0.022)
C-Section	0.279	0.028*** (0.006)	0.286	0.025*** (0.006)
Severe Complication	0.036	−0.026*** (0.004)	0.036	−0.013** (0.005)
Preventable Complication	0.193	−0.086*** (0.020)	0.184	−0.090*** (0.018)
Severe Acute Maternal Morbidity	0.012	−0.003 (0.002)	0.011	0.000 (0.002)
28-Day Infant Mortality (per 1,000)	0.825	−0.539 (0.776)	0.772	−0.450 (0.749)
1-Year Infant Mortality (per 1,000)	1.597	−0.549 (0.977)	1.544	−1.720 (0.982)
Unplanned Readmissions (per 1,000)	12.498	−4.692** (2.089)	11.460	−7.335*** (2.698)
Unplanned Child Readmit (per 1,000)	36.414	−8.434* (4.878)	35.566	−1.290 (6.408)
30-Day Child Hosp. (per 1,000)	55.122	−15.457** (5.880)	52.064	−12.719 (8.895)
1-Year Unplanned Child Hosp. (per 1,000)	85.784	−11.814* (6.416)	82.128	8.801 (7.676)
Any 1-Year Child Hosp. (per 1,000)	104.159	−20.274*** (7.303)	98.913	−1.978 (9.560)
N		58,144		37,629

	ΔOff-Base Enlisted	ΔOff-Base Officer
	(1)	(2)
	β	β
MTF Hospital to No MTF Hospital	0.685*** (0.018)	0.643*** (0.025)
No MTF Hospital to MTF Hospital	−0.645*** (0.013)	−0.610*** (0.023)
F-Statistic	553.191	323.512
p-value	0.124	0.456
Outcome Mean	0.001	0.045
R²	0.449	0.421
N	69,663	25,469

	Below Median			Above Median
	(1)	(2)	(3)	(4)	(5)	(6)
	Mean	β	p-value	Mean	β	p-value
Bed Days	2.119	0.174** (0.066)	0.052	2.189	0.136** (0.058)	0.403
Bassinet Days	2.348	0.069 (0.142)	0.924	2.359	0.224** (0.101)	0.772
RWP	0.502	0.010*** (0.004)	0.263	0.503	0.013*** (0.003)	0.743
Procedures	2.078	0.181 (0.130)	0.114	2.182	0.055 (0.118)	0.039
Diagnostics	0.093	0.052*** (0.021)	0.032	0.103	0.074*** (0.019)	0.246
C-Section	0.281	0.026*** (0.006)	0.320	0.281	0.036*** (0.006)	0.558
Severe Complication	0.036	−0.021*** (0.006)	0.328	0.036	−0.019*** (0.005)	0.130
Preventable Complication	0.193	−0.078** (0.031)	0.625	0.190	−0.096*** (0.015)	0.884
Severe Acute Maternal Morbidity	0.011	−0.003 (0.003)	0.946	0.012	−0.001 (0.003)	0.143
28-Day Infant Mortality (per 1,000)	0.881	−0.707 (0.982)	0.575	0.724	−0.107 (0.780)	0.443
1-Year Infant Mortality (per 1,000)	1.509	−0.896 (1.227)	0.791	1.603	0.244 (1.040)	0.993
Unplanned Readmits (per 1,000)	11.098	−3.344 (3.329)	0.756	12.956	−8.307*** (2.406)	0.394
Unplanned Child Readmit (per 1,000)	36.998	−7.448 (7.680)	0.599	35.932	−5.494 (4.321)	0.718
30-Day Child Hosp. (per 1,000)	54.553	−15.295 (9.606)	0.852	54.393	−13.501** (5.298)	0.858
1-Year Unplanned Child Hosp. (per 1,000)	86.615	−7.494 (8.133)	0.238	82.892	−8.683 (6.610)	0.188
Any 1-Year Child Hosp. (per 1,000)	103.743	−16.475 (9.932)	0.384	101.438	−16.972** (7.350)	0.302
N		41,348			50,281

	(1) Mean	OLS	IV
	(1) Mean	(2) β	(3) β	(4) p-value
Bed Days	2.342	−0.044 (0.263)	−3.632*** (1.319)	0.536
Bassinet Days	3.034	0.994*** (0.336)	−11.077*** (3.871)	0.392
RWP	0.478	0.026** (0.010)	−0.115* (0.070)	0.010
Procedures	2.025	−0.171 (0.134)	−0.099 (0.795)	0.372
Diagnostics	0.212	0.044 (0.062)	−1.247** (0.550)	0.512
C-Section	0.179	0.063*** (0.019)	−0.257** (0.131)	0.017
Severe Complication	0.042	−0.002 (0.016)	−0.223*** (0.080)	0.316
Preventable Complication	0.139	−0.076 (0.046)	−0.276* (0.159)	0.429
Severe Acute Maternal Morbidity	0.017	−0.005 (0.011)	−0.131* (0.072)	0.281
28-Day Infant Mortality (per 1,000)	1.287	−4.544 (2.998)	5.225 (15.963)	0.721
1-Year Infant Mortality (per 1,000)	2.466	−6.010** (2.966)	0.281 (19.168)	0.938
Unplanned Readmissions (per 1,000)	10.507	−10.880 (11.439)	−25.066 (37.872)	0.182
Unplanned Child Readmit (per 1,000)	38.457	−4.012 (15.691)	−29.523 (79.352)	0.415
30-Day Child Hosp. (per 1,000)	59.991	−18.862 (21.520)	−294.187*** (105.027)	0.430
1-Year Unplanned Child Hosp. (per 1,000)	93.073	8.664 (26.213)	−34.913 (105.770)	0.595
Any 1-Year Child Hosp. (per 1,000)	113.973	−5.486 (26.317)	−252.438** (120.058)	0.686
N		9,992	9,992

	Enlisted Sponsor			Officer Sponsor
	(1) Mean	(2) β	(3) p-value	(4) Mean	(5) β	(6) p-value
Bed Days	2.164	0.134*** (0.050)	0.428	2.146	0.184*** (0.057)	0.932
Bassinet Days	2.400	0.122 (0.113)	0.738	2.257	0.103 (0.138)	0.830
RWP	0.506	0.012*** (0.003)	0.591	0.495	0.008** (0.004)	0.036
Procedures	2.130	0.108 (0.114)	0.632	2.144	0.098 (0.110)	0.755
Diagnostics	0.105	0.060*** (0.017)	0.551	0.088	0.063*** (0.024)	0.130
C-Section	0.287	0.031*** (0.005)	0.676	0.268	0.023*** (0.007)	0.028
Severe Complication	0.038	−0.022*** (0.005)	0.217	0.030	−0.017** (0.007)	0.053
Preventable Complication	0.193	−0.084*** (0.021)	0.276	0.180	−0.099*** (0.021)	0.146
Severe Acute Maternal Morbidity	0.012	−0.002 (0.002)	0.540	0.010	−0.004 (0.003)	0.088
28-Day Infant Mortality (per 1,000)	0.809	−0.229 (0.733)	0.725	0.793	−0.855 (0.941)	0.573
1-Year Infant Mortality (per 1,000)	1.677	0.280 (0.945)	0.833	1.308	−2.907** (1.350)	0.342
Unplanned Readmissions (per 1,000)	13.083	−6.094*** (2.279)	0.660	9.428	−3.928 (2.693)	0.832
Unplanned Child Readmit (per 1,000)	38.184	−6.737 (4.939)	0.359	30.516	−5.640 (7.284)	0.646
30-Day Child Hosp. (per 1,000)	57.083	−16.304*** (6.094)	0.962	45.541	−15.663* (8.024)	0.714
1-Year Unplanned Child Hosp. (per 1,000)	90.149	−4.976 (6.086)	0.064	68.990	−15.266 (9.638)	0.487
Any 1-Year Child Hosp. (per 1,000)	108.757	−14.491** (7.184)	0.179	84.463	−27.055** (10.637)	0.528
N		69,766			26,007

	No Previous C-section			Previous C-section
	(1) Mean	(2) β	(3) p-value	(4) Mean	(5) β	(6) p-value
Bed Days	2.038	0.045 (0.049)	0.289	2.550	0.459*** (0.064)	0.616
Bassinet Days	2.166	0.122 (0.083)	0.357	2.985	0.142 (0.256)	0.382
RWP	0.444	0.007*** (0.003)	0.145	0.695	0.021*** (0.005)	0.522
Procedures	2.207	0.188 (0.119)	0.824	1.900	−0.161 (0.099)	0.088
Diagnostics	0.088	0.062*** (0.014)	0.690	0.139	0.063** (0.032)	0.963
C-Section	0.060	0.017*** (0.005)	0.312	0.895	0.066*** (0.011)	0.778
Severe Complication	0.036	−0.019*** (0.004)	0.076	0.036	−0.023*** (0.008)	0.326
Preventable Complication	0.221	−0.101*** (0.022)	0.992	0.089	−0.041*** (0.014)	0.701
Severe Acute Maternal Morbidity	0.009	0.000 (0.002)	0.142	0.021	−0.009 (0.005)	0.552
28-Day Infant Mortality (per 1,000)	0.727	−0.044 (0.585)	0.858	1.051	−1.253 (1.326)	0.836
1-Year Infant Mortality (per 1,000)	1.583	−0.159 (0.753)	0.993	1.553	−1.264 (1.993)	0.381
Unplanned Readmissions (per 1,000)	10.036	−4.398** (1.721)	0.382	18.696	−7.534 (4.647)	0.359
Unplanned Child Readmission (per 1,000)	37.200	−6.319 (5.400)	0.881	32.490	−6.665 (6.213)	0.052
s30-Day Child Hospitalization (per 1,000)	53.542	−15.261** (5.804)	0.837	55.116	−19.344* (10.749)	0.931
1-Year Unplanned Child Hospitalization (per 1,000)	84.628	−9.905 (6.313)	0.334	83.429	−0.908 (9.725)	0.017
Any 1-Year Child Hospitalization (per 1,000)	101.190	−19.434*** (6.862)	0.305	104.977	−13.397 (12.255)	0.183
N		73,059			22,714

PERMALINK

Public and Private Options in Practice: The Military Health System

Michael D Frakes

Jonathan Gruber

Timothy Justicz

Abstract

Part I: Institutional Background

Part II: Data

Sample Generalizability

Part III: Empirical Strategy

Table 1:

Part IV: Results

Mean Outcomes

Table 2:

Table 3:

Regression Analysis

Table 4:

Table 6:

Specification Checks

Table 5:

Mechanisms

Testing the C-section explanation

Figure 1:

Figure 2:

Table 7:

Pre-Natal Care

Table 8:

Differences in Treating Physicians

Table 9:

Differences in Hospital Volume

Heterogeneity

Heterogeneity by Move Type

Heterogeneity for Spouses of Enlisted Personnel and Officers

Comparison to Card, Chan, and Taylor (2020)

Part V: Conclusions

Figure 3:

Acknowledgments

Table A1: Sample Generalizability

Table A2: Characterizing Areas with and without Hospital-Equivalent Military Treatment Facilities

Table A3: First Stage

Table A4: Alternate Specifications

Table A5: Outcomes following Admission via Emergency Department

Table A6: Heterogeneity by Move Type

Table A7: Controlling for Distance

Table A8: Controlling for Area Volume and Hospital Volume

Table A9: First Stage for Enlisted and Officers

Table A10: Heterogeneity by Enlisted or Officer

Table A11: Controlling for Previous C-Section

Table A12: Heterogeneity by Previous C-section

Footnotes

Contributor Information

Works Cited

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases