One of the largest integrated health care delivery systems in the United States, the Veterans Health Administration (VA) provided care to 4.2 million veterans in fiscal year (FY) 2003. Treatment is virtually free for enrollees with high priority status due to service-connected disability or low

income. Others face copayments but the amount is considerably lower than for non-VA care. Many VA enrollees have private insurance or other public coverage such as Medicare. In 1999, 27.4 percent of VA enrollees had no public or private insurance (Shen et al. 2003). Earlier studies have found that alternative insurance coverage was related to VA enrollees' use of VA and non-VA services in addition to sociodemographic information, VA priority status and travel distance to VA (Kosloski, Austin, and Borgatta 1987; Fleming et al. 1992; Burgess and DeFiore 1994; Kashner et al. 1998; Romeis, Gillespie, and Coe 1998; Borowsky and Cowper 1999; Mooney et al. 2000). For example, Kashner et al. (1998) found that veterans with private insurance were less likely to use VA surgical care and more likely to have VA outpatient visits. They concluded that the latter may be due to adverse selection, i.e., sicker people (indicated by more VA outpatient visits) tended to purchase insurance. However, few studies have tried to account for a potential selection effect in insurance coverage for VA enrollees.

Selection effect (sometimes called "adverse selection") has been well studied in the general population (Wolfe and Goddeeris 1991; Davidson, Sofaer, and Gertler 1992; Gertler, Sturm, and Davidson 1994; Ettner 1997; Hurd and McGarry 1997). The positive correlation of private health insurance and the use of health care services may arise from both an insurance effect (i.e., low out-of-pocket cost encourages people to seek more health care) and a selection effect (people with perceived needs for more care tend to be insured).

As in the general population, VA enrollees with alternative insurance coverage may use less VA care because the effective cost of non-VA care has decreased. Unlike the general population, however, selection effects for VA enrollees arise from two aspects: (1) veterans who are sicker may tend to purchase private insurance for security and therefore may use both VA and non-VA care intensively (a security selection effect); and (2) veterans who prefer non-VA care may purchase private insurance and use less VA care (a preference selection effect). Neither aspect can be completely observed, but both affect decisions to purchase insurance and the amount of VA/non-VA care demanded. Ignoring the selection effect by treating insurance as exogenous could either over- or underestimate the insurance effect for this special population, depending on which type of selection effect predominates for VA enrollees.

In this study, using survey data from VA, we examined non-Medicare VA enrollees' private insurance coverage and its relationship to the use of VA care. This study broadens our knowledge of the reliance of VA services for VA enrollees lacking Medicare coverage and is the first to address the complex issue of VA enrollees' use of VA care by incorporating a wider array of explanatory factors, including sociodemographic information, patients' health status measured by Veterans SF-36 (Kazis et al. 1999), self-reported chronic diseases, access to VA and non-VA alternatives including both travel distance and characteristics of VA/non-VA systems, and insurance coverage. To our knowledge, this is the first study to evaluate the impact of various measures of access to non-VA care on veterans' use of care.

Above all, our analysis acknowledges the existence of selection effects by treating insurance coverage as endogenous. Using appropriate models and instrumental variables, we were able to separate an insurance effect from a selection effect. If future premium increases force enrollees to forgo private insurance, the coefficient for the insurance effect allows us to predict the impact on the demand for VA care. Hence it is important to correctly estimate the insurance effect.

METHODS

Data Sources

The study used the 1999 National Health Survey of Veteran Enrollees, the largest and most detailed survey of veterans using VA services ever conducted. A total of 1,400,000 enrollees were surveyed by mail using a stratified random sample design. The survey sample accounted for 41 percent of the VA enrollment file as of March 1999. Patients were sampled randomly in equal numbers between five survey modules. These modules share 31 core questions about enrollees' health status measured by the Veterans SF-36, prevalence of medical and mental conditions, and sociodemographic factors. The overall response rate was 63 percent, but was higher for older enrollees (over 70 percent for those age 60 or higher) and lower for those who were younger (below 50 percent for those under 40) (Department of Veterans Affairs 2000).

We used the Insurance module for this study (N= 152,258). The module asked enrollees whether they had Medicare, Medicaid, or private insurance. The survey data also provided enrollees' VA priority status and the zip codes of their residence. We restricted our study sample to those who had reported having no Medicare coverage (47 percent of the respondents, Shen et al. 2003).

We merged the survey data with the FY 1999 VA cost data from VA's Allocation Resource Center (ARC), which calculates the VA total budget expenditures as well as inpatient, outpatient, and pharmacy expenditure for each patient in each year for use by the Department leadership in the federal budget process. The ARC calculations are based on the national utilization and pharmacy databases merged with financial data. We used the Area Resource File for information on percentage of no insurance at the metropolitan statistical area (MSA) level. Using latitude/longitude data, the residence zip codes and zip codes of hospitals from American Hospital Association data, we identified the closest VA and non-VA hospitals and calculated the shortest distance in miles. We defined the non-VA hospital market as all the hospitals within an area with a radius equal to the distance to the closest non-VA hospital plus ten miles. In addition to travel distance, VA/non-VA hospitals' characteristics were used in the analyses to capture the accessibility of VA/ non-VA care. We focused on VA enrollees residing only in the 48 contiguous states. After excluding survey respondents with any missing values of the independent variables, our final study sample size is 48,448.

Measures of Care

We used several measures of care to examine the impact of private insurance on the use of VA and non-VA care. First, we examined general VA care and VA inpatient care measured by their VA total cost and inpatient cost. The VA pharmacy benefit was free to qualified VA enrollees and the copay was only $2 per prescription per month for others in 1999 (increasing to $7 in 2001). Given the high premium of outpatient prescription coverage and its significant out-of-pocket cost in the private sector at that time, policymakers believed that many veterans used their VA benefits primarily for prescription medicines. We examined VA pharmacy cost to test the hypothesis.

Empirical Model

The empirical model of VA enrollees' choice of purchasing private insurance and use of care was derived from a consumers' choice model. First, VA enrollees made an insurance purchase decision. Let [D.sub.i] be binary discrete choice: [D.sub.i] = 1 if enrollee i had private insurance, otherwise [D.sub.i] = 0. VA enrollees purchased private insurance if its utility was greater than the utility of no insurance. Let [Z.sub.i] be a set of independent variables that affected the utility level of having private insurance and normalize the utility without insurance as 0. Let [U.sub.i] be the latent unobserved utility with: [U.sub.i] = [[alpha].sub.0] + [Z.sub.i][[alpha].sub.1] + [[epsilon].sub.i] x [alpha] was the set of unknown parameters and [[epsilon].sub.i] was a random unobservable. Hence VA enrollee i purchased insurance iff

[[alpha].sub.0] + [Z.sub.i][[alpha].sub.1] + [[epsilon].sub.i] [greater than or equal to] 0 (1)

Next, we modeled how having private insurance affected VA enrollees' use of care. Because almost 20 percent of VA enrollees had no VA care, 90 percent had no VA inpatient care, and 30 percent had no VA pharmacy cost, we used a two-part model. For example, total cost comprises the probability of having any care in VA (positive total cost) and the level of total cost conditional on positive cost. We assumed the utility of having VA care to be [[beta].sub.0] + [X.sub.i][[beta].sub.i] + '" [D.sub.i][[beta].sub.2] + [[delta].sub.i], and normalized the utility without VA care as 0. Then

[TC.sub.i] > 0 iff [[beta].sub.0] + [X.sub.i][[beta].sub.1] + [D.sub.i][[beta].sub.2] + [[delta].sub.i] [greater than or equal to] 0 (2)

where TC was the total cost, X was the set of independent variables that affect the utility function. The insurance effect was captured by [[beta].sub.2].

The level of total cost was modeled as

[IC.sub.I]|[IC.sub.i] > 0 = [[lambda].sub.0] + [X.sub.i][[lambda].sub.1] + [D.sub.i][[lambda].sub.2] + [v.sub.i] (3)

The endogeneity of private insurance existed when [rho]([[epsilon.sub.i], [[delta].sub.i])[not equal to]0 and [rho]([[epsilon.sub.i], [[v.sub.i])[not equal to]0 i.e., when the unobservable error term [[epsilon].sub.i] in the insurance purchase model was correlated with error terms [[delta].sub.i] and [v.sub.i] in the total cost model. Ignoring this correlation would lead to biased estimation of the impact of D and X on IC.

Two binary choice models were interacted with each other when we assumed that [rho]([[epsilon.sub.i], [[delta].sub.i])[not equal to]0. By assuming that ([[epsilon].sub.i], [[delta].sub.i]) ~ bivariate normal distribution with [0, 0, 1, [sigma], [rho]], we used a maximum likelihood (ML) method to obtain consistent estimation of [alpha], [beta], and [rho]. Similarly, we assumed that ([epsilon].sub.i], [v.sub.i]) ~ bivariate normal distribution with [0, 0, 1, [sigma]', [rho]']. As the level of total cost was a continuous variable, we used a Heckman two-stage model (Heckman 1979) to deal with endogeneity.

Model Identification and Instrumental Variables

The key to this study is to identify valid instruments, variables that affect the VA enrollees' insurance status but do not directly affect their use of care. One possible instrument was the percentage of people with no insurance in the MSA. An individual living in an area with a high insurance rate would be more likely to have insurance, but it seemed plausible that the aggregate insurance rate in the MSA was unlikely to affect a specific individual's need for health care. We empirically tested the validity of the instrumental variable.

In the utilization model, several factors were used to capture the non-VA care market and VA care available to VA patients. They included the travel distance to the closest VA hospitals and non-VA hospitals, whether the closest VA hospital was a general hospital or was affiliated with a medical school, and the number of major services provided in the closest VA hospital (Table 1 lists the 18 major services considered here). The non-VA care market was characterized by the number of non-VA hospitals, number of non-VA hospitals affiliated with medical schools, and number of general hospitals within the area with radius of 10 miles plus the distance to the identified closest non-VA hospitals. (The results were similar when we used a radius with + 15 miles.)

Key Explanatory Variables

Enrollees' health status was posited to be a major factor in the decisions to purchase insurance coverage or use care. We used the Veterans SF-36 two component summary scales, a physical component summary (PCS) and a mental component summary (MCS), to measure health status. Higher summary scores indicate better health (Kazis et al. 1999). In addition, we also used the number of physical and mental chronic diseases to measure health status. These two indicators were based on self-reported information on 12 chronic physical diseases and three chronic mental disorders. The full list of explanatory variables is given in Table 1.

RESULTS

Table 2 presents the descriptive statistics of the relevant measures of VA/non-VA utilization by insurance status. In FY 1999, 80 percent of our study sample sought care in VA with an average budget expenditure of $4,400 per patient. For enrollees without private insurance coverage, 88 percent had VA care (mean = $5,300). About 66 percent of those with private insurance received VA care (average = $2,300).

About 9.2 percent of the study sample had VA inpatient care with average cost of $12,100, but enrollees without private insurance coverage had a higher rate of VA inpatient care than those with private insurance (12.9 versus 3.2 percent) and cost $3,600 more per patient. More than two-thirds of people used VA pharmacy with the average cost of $730, which accounted for 19.9 percent of total VA cost. More enrollees without private insurance used VA pharmacy (80 percent) than those with private insurance (51 percent) and their cost was higher ($785 versus $590). The ratio of VA pharmacy cost to total cost was similar for enrollees with and without private insurance.

Validity of the Instrumental Variable

IV estimates are consistent and reliable only if the instrumental variable (percentage of people without insurance at the MSA level) is (1) strongly correlated with the endogenous variable (VA enrollees' insurance status), instrumental variable estimations could be inconsistent if there is weak correlation (Bound, Jaeger, and Baker 1995; Staiger and Stock 1997); and (2) does not have a direct effect on use of VA care. To test the first condition, Staiger and Stock (1997) recommend that the first stage F-statistics, which test the hypothesis that instruments do not enter the first stage regression (in our case, VA enrollees' insurance status), should have a value of at least 10. Given that the F-statistics are developed for linear models, we ran the linear first-stage insurance model (Hadley and Waidmann 2006). The instrumental variable had an F-statistic equal to 242, which indicated that it was unlikely to have bias due to a weak instrument. To test the second assumption, we regressed total patient-level VA cost on the MSA-level noninsurance rate and other covariates. The results showed that the parameter estimate of the noninsurance rate was not statistically significant (p = .418). Similarly, noninsurance was not correlated with VA patient-level inpatient cost (p = .173) or pharmacy cost (p = .382). The tests increased our confidence in the validity of our instrumental variable.

Literature has demonstrated that some MSAs may have more aggressive practice patterns than other MSAs. The aggressive MSAs may have higher premiums for private coverage and lower coverage rates (Wennberg, Fisher, and Skinner 2002). If the aggressive practice patterns spill over into VA practice, areas with high rates of un-insurance will exhibit greater spending in the VA sector even if there is no causal link at the individual level, which has been demonstrated above. The possible spillover effect from the private sector to the VA sector may affect the strength of the instrumental variable.

Unlike the Medicare fee-for-service program, VA medical center budgets provide a financial disincentive for chiefs of medical services to allow more aggressive practice, however. While Medicare providers are paid more if they provide more services, VA medical centers are not. Indeed, internal (unpublished) VA studies suggest that VA does not mirror private sector practice patterns in this way. Nevertheless, we explored this possible confounder (spillover effect) by testing for correlation of the area un-insurance rate with both per capita Medicare expenditures (aggregated to the MSA level from county-level Part A and B reimbursement figures published on the CMS website) and per capita VA expenditures calculated across all VA patients for each medical center. The former were significantly and positively correlated as suggested by the published literature (t = 4.69, [R.sup.2] 0.07). VA per capita expenditures had a negative but statistically insignificant relationship with the no insurance rate, however (t = 1.41, [R.sup.2] 0.02). Consequently, we do not believe that geographic areas with high rates of un-insurance exhibit greater spending in the VA sector, which further supports the validity of the instrumental variable.

Who Purchased Private Insurance

Table 3 presents the estimation of the insurance model, based on the joint estimation of insurance coverage and use of VA care or not. (The results from the simple probit model of insurance coverage without considering its correlation with use of care and others joint estimation were similar and are not presented here.)

Sociodemographic factors were highly associated with insurance coverage. For example, VA enrollees more likely to have private insurance were female, white, highly educated, married, and employed. The likelihood of having private insurance was also associated with VA priority status. Enrollees with extensive disabilities (priority 4) were least likely to have insurance. Those without a service connected disability and whose incomes were above established thresholds (priority 7) were most likely to have insurance.

The correlation between observable health status and private insurance was very weak. The numbers of chronic and mental conditions were not significantly correlated with the likelihood of having insurance coverage. Even though both veteran SF-36 physical and mental component scales (PCS and MCS) had a positive and significant impact on the likelihood of having insurance coverage, their magnitudes were negligible.

The parameter estimate of the instrumental variable (noninsurance rate at the MSA level) was -0.006 (p [less than or equal to] .001). The sign is consistent with the hypothesis that people who live in areas with a high noninsurance rate were less likely to have private insurance.

Endogeneity and Impact of Private Insurance

To provide a baseline estimation and comparison, we ran naive models that did not account for the unobservable correlation between insurance and use of care. These naive models included (1) probit models dealing with dependent variables such as the likelihood of having VA cost, inpatient cost, and pharmacy cost; and (2) least square regressions with dependent variables such as the level of VA cost, inpatient cost, pharmacy cost, and the ratio of pharmacy cost to total cost. Table 4 summarizes and compares the impact of private insurance coverage on the use of care with/without controlling for insurance endogeneity. Overall, having insurance significantly decreased the likelihood of using VA care overall, inpatient care and pharmacy care in particular. However, its effects on the level of VA total cost, inpatient cost, and pharmacy cost were not significant. Models with no control for endogeneity always underestimated the insurance effect on the use of VA care.

For example, in the estimation of the likelihood of having VA care in FY 1999, the magnitude of the insurance effect estimated in the IV model was -1.03 (p [less than or equal to] .001) compared with -0.60 (p [less than or equal to] .001) in the naive model. The correlation of two error terms in the functions for insurance coverage and having VA care in FY 1999 was positive and significant (0.26, p [less than or equal to] .001). This suggests that the security selection effect dominates the preference selection effect. Sicker VA enrollees appear to purchase insurance coverage because of their perceived need for care whether in the VA or not. Because of the positive selection effect, a naive probit model that treats insurance as exogenous would result in an underestimation of the insurance effect on the use of VA care. For those who sought care in VA in FY 1999, private insurance reduced their VA cost, but the impact was not statistically significant.

The results for inpatient care were similar, i.e., private insurance significantly decreased the likelihood of VA hospitalization (-0.96, p [less than or equal to] .001). Once VA enrollees got in VA hospitals, however, the effect of insurance status on VA inpatient cost was not statistically significant. The naive model, which did not take into account insurance endogeneity, underestimated the insurance effect on the use of VA inpatient care.

Private insurance significantly decreased the likelihood of using VA pharmacy (-1.02, p [less than or equal to] .001) and the level of pharmacy cost (-0.24, p [less than or equal to] .05). However, there was no impact of insurance on the ratio of pharmacy cost to total cost. Given that medications are a crucial part of care, the probability of having positive pharmacy cost and the level of total pharmacy cost can be treated as another indicator of intensive use of VA care. With the ratio of pharmacy cost to total cost as a proxy of whether VA enrollees came to VA mainly for pharmacy, there is no evidence to suggest that veterans with private insurance would use VA primarily for prescription.

To illustrate the impact of private insurance coverage on VA cost, we quantified the cost to VA of a 4 percentage point decline in private coverage among VA enrollees without Medicare coverage. Basically we calculated the predicted probabilities of having VA care, VA inpatient care, VA pharmacy use, and the predicted level of the cost items at the population average. Then we generated the predicted values by allowing the average insurance rate of 38 percent to dip to 34 percent. We estimated that a 4 percentage point decline in private coverage would result in 84.4 percent of VA enrollees without Medicare turning to VA for some care, an increase of 1 percentage point; 7.6 percent of VA enrollees would have used VA inpatient care, an increase of 0.5 percentage points; and 73.5 percent of VA enrollees would have used VA pharmacy, an increase of 1.4 percentage points (Table 4).

Regarding the difference between the IV models and naive models, the IV method of modeling predicted probability of having some VA care doubled the impact of a 4 percentage point decline in private coverage (1 versus 0.5 percent). This difference (between the IV and naive models) could have an important impact on forecasts or simulations of aggregate levels of VA services under various scenarios for veterans' insurance coverage. The impact of the IV approach is not as great for other measures of VA utilization, however. For example, the difference in total VA cost following a change in private coverage is only $3 between the IV and naive models.

Finally, Table 5 presents the flail IV regression results for all VA cost items. The results show that other factors are associated with VA enrollees' use of VA. Some results are consistent with previous findings. For example, travel distance to VA was associated with VA use. VA enrollees living farther from VA hospitals were less likely to use VA care (Burgess and DeFiore 1994; Mooney et al. 2000).

CONCLUSIONS

In this study, using VA data for VA enrollees without Medicare coverage, we modeled VA enrollees' decision to purchase insurance coverage and its impact on use of VA/non-VA care. By treating insurance as endogenous, we estimated the true insurance effect by teasing out a potential selection effect from the impact of insurance on the use of VA and non-VA care. We argue that the existence of selection bias in the estimation of an insurance effect can be from two sources: security or preference. VA enrollees with private insurance coverage were less likely to use VA care. Both the models that dealt with selection effect and the naive models without controlling for selection effect led to these conclusions. The naive models consistently underestimated the impact of insurance. The results imply that security selection dominates the preference selection effect.

The latest data showed that private health insurance coverage among people under 65 years of age has dropped from 73 percent in 1999 to 69 percent in 2003 (U.S. Department of Health and Human Services 2005). The continuous drop in the private insurance coverage may result in more pressure in the VA system. We estimated that a 4 percentage point decline in the private coverage among VA enrollees without Medicare coverage may lead to 1 percent more enrollees using VA care. With 2.94 million VA enrollees without Medicare coverage in 2002 (from our analysis of data published by Department of Veterans Affairs 2002), VA would have treated 29,400 new patients due to this decline in private coverage. Given that VA spent $5,386 per patient in 2002, the shift would translate into $158 million in additional VA cost in 2002.

Our empirical results indicated that security effect predominates, i.e., veterans who are sicker may tend to purchase private insurance for security and therefore may use VA care intensively. However, the correlation between measurable health stares and the propensity to purchase private insurance is weak. Other studies have found similar results (Wolfe and Goddeeris 1991; Gertler, Storm, and Davidson 1994; Ettner 1997; Hurd and McGarry 1997). The quality of measurable health status directly affects the detection of observable selection (Mello et al. 2003).

All of our analyses are based on the sample of the 1999 Large Health Survey of VA enrollees, which allows us to study only the self-selected population, i.e., veterans who have made the effort to show up at a VA facility and enroll to be eligible for VA care. The enrollment process is, by itself, a selection process that we do not address. The vast majority of veterans (about 16 million) who have not enrolled for VA care are not represented in this study. In addition, our analysis may have limited generalizability for younger VA enrollees, who were less likely to respond to the survey invitation.

ACKNOWLEDGMENTS

The work was supported by grant #ECI 20-032-2, the Health Services Research and Development Service and Office of Quality and Performance, the Veterans Health Administration. This article was written solely from the perspective of the authors and does not necessarily represent the official policy or position of the Veterans Health Administration.

REFERENCES

Borowsky, S.J., and D. C. Cowper. 1999. "Dual Use of VA and Non-VA Primary Care." Journal of General Internal Medicine 14 (5): 274-80.

Bound, J., D. A. Jaeger, and R. M. Baker. 1995. "Problems with Instrumental Variables Estimation When the Correlation between the Instruments and the Endogenous Explanatory Variable Is Weak." Journal of the American Statistical Association 90 (430): 443-50.

Burgess, J. F., and D. A. DeFiore. 1994. "The Effect of Distance to VA Facilities on the Choice and Level of Utilization of VA Outpatient Services." Social Science and Medicine 39 (1): 95-104.

Davidson, B. N., S. Sofaer, and P. Gertler. 1992. "Consumer Information and Biased Selection in the Demand for Coverage Supplementing Medicare." Social Science and Medicine 34 (9): 1023-34.

Department of Veterans Affairs. 2000. Health Status and Outcomes of Veterans: Physical and Mental Component Summary Scores, Veterans SF-36, 1999 National VIL4 Health Survey of Veteran Enrollees. Executive report. Washington, DC: Office of Quality and Performance, Veterans Health Administration.

--. 2002. Veterans Health Care Enrollment and Expenditure Projections FY 2002-2012. Washington, DC: Office of Policy and Planning, Veterans Health Administration.

Ettner, S. L. 1997. "Adverse Selection and the Purchase of Medigap Insurance by the Elderly." Journal of Health Economics 16 (5): 543-62.

Fleming, C., E. S. Fisher, C. H. Chang, T. A. Bubolz, and D.J. Malenka. 1992. "Studying Outcomes and Hospital Utilization in the Elderly: The Advantages of a Merged Database for Medicare and Veterans Affairs Hospitals." Medical Care 30 (5): 377-91.

Gertler, P., R. Sturm, and B. Davidson. 1994. "Information and the Demand for Supplemental Medicare Insurance." NBER. Working Paper #4700

Hadley, J, and T. Waidmann. 2006. "Health Insurance and Health at Age 65: Implications for Medical Care Spending on New Medicare Beneficiaries." Health Service Research 41 (2): 429-51.

Heckman, J. 1979. "Sample Selection Bias as a Specification Error." Econometrica 47 (1): 153-61.

Hurd, M. D., and K. McGarry. 1997. "Medical Insurance and the Use of Health Care Services by the Elderly." Journal of Health Economics 16 (2): 129-53.

Kashner, T. M., A. Muller, E. Richter, A. Hendricks, C. V. Lucas, and D. R. Stubblefield. 1998. "Private Health Insurance and Veterans' Use of Veterans Affairs Care." Medical Care 36 (7): 1085-97.

Kazis, L. E., X. S. Ren, A. Lee, K. Skinner, W. Rogers, J. Clark, and D. R. Miller. 1999. "Health Status in VA Patients: Results from the Veterans Health Study." American Journal of Medical Quality 14 (1): 28-38.

Kosloski, K., C. Austin, and E. Borgatta. 1987. "Determinants of VA Utilization: The 1983 Survey of Aging Veterans." Medical Care 25 (9): 830-46.

Mello, M. M., S. C. Stearns, E. C. Norton, and T. C. Ricketts. 2003. "Understanding Biased Selection in Medicare HMOs." Health Service Research 38 (3): 961-92.

Mooney, C, J. Zwanziger, C. S. Phibbs, and S. Schmitt. 2000. "Is Travel Distance a Barrier to Veterans' Use of VA Hospitals for Medical Surgical Care?" Social Science and Medicine 50 (12): 1743-55.

Romeis, J. C., K. N. Gillespie, and R. M. Coe. 1998. "Older Veterans' Future Use of VA Health Care Services." Medical Care 26 (9): 854-66.

Shen, Y., A. Hendricks, S. Zhang, and L. E. Kazis. 2003. "VHA Enrollees' Health Care Coverage and Use of Care." Medical Care Research and Review 60 (2): 253-67.

Staiger, D., and J. H. Stock. 1997. "Instrumental Variables Regression with Weak Instruments." Econometrica 65 (3): 557-86.

U.S. Department of Health and Human Services. 2005. "Health, United States, 2005." DHHS Pub No. 2005-1232. Maryland.

Wennberg, J. E., E. S. Fisher, and J. S. Skinner. 2002. "Geography and the Debate over Medicare Reform." Health Affairs Web Exclusive, February 13, 2002.

Wolfe, J. R., and J. H. Goddeeris. 1991. "Adverse Selection, Moral Hazard, and Wealth Effects in the Medigap Insurance Market." Journal of Health Economics 10 (4): 433-59.

Address correspondence to Yujing Shen, Ph.D., VA New Jersey Healthcare System Center for Healthcare Knowledge Management, East Orange VA Medical Center, 385 Tremont Avenue, Mailstop 129, East Orange, NJ 07018-1095. Ann Hendricks, Ph.D., and John Gardner, Ph.D., are with the Health Care Financing & Economics (152H), VA Boston Healthcare System, Boston, MA. Fenghua Wang, Ph.D., is with the Research Center, Shanghai Stock Exchange, Shanghai, China. Lewis E. Kazis, Sc.D., is with the Center for Health Quality, Outcomes & Economic Research, CHOOER (152), Bedford, MA.

Table 1: Descriptive Statistics of Independent Variables
(1999 National Health Survey of Veteran Enrollees--Insurance
Module)

                                                         Private
                         Total Sample,   No Insurance,   Insurance,
                         N = 48,448      N = 29,944      N = 18,504

Personal
    characteristics
  Female                   7.1%            6.2%            8.5%
  Nonwhite                29.2%           31.6%           25.4%
  Age
    Under 45              21.8%           19.1%           26.3%
    45-54                 38.6            37.7            40.2
    55-64                 30.1            30.4            29.7
    65 and older           9.5            12.9             3.9
  Education status
    [less than or          8.6%           11.3%            4.2%
      equal to] 8
      years,
      elementary
    9-11 years, some       7.1             9.5             3.1
      high school
    12 years or GED       31.3            33.7            27.4
      high school
      graduate
    College 1-3 years     37.3            34.5            41.9
    College graduate      15.7            11.0            23.4
      or graduate
      school
  Economic hardship:      18.6%           23.9%            9.9%
    yes = 1
  Live alone: yes = 1     23.8%           29.4%           14.7%
  Marriage status
    Currently married     54.3%           42.9%           72.8%
    Formerly married      33.1            41.5            19.4
    Never married         12.6            15.6             7.8
  Employment status
    Employed              55.2%           41.7%           77.0%
    Retired               16.8            19.1            13.1
    Others                28.0            39.2             9.9
  VA priority
    Priority 1:           12.9%           13.4%           11.9%
      service
      connected
      disability 50%
      or more
    Priority 2:           10.6            8.0             14.9
      service
      connected
      disability
      30-40%
    Priority 3:           17.7            12.9            25.5
      Service
      connected
      disability
      10-20%, plus
      former POWs
    Priority 4: Aid        1.2             1.9             0.2
      and attendance,
      housebound,
      catastrophically
      disabled
    Priority 5: income    43.7            55.9            24.1
      and net worth
      below
      established
      thresholds
    Priority 6: World      2.1             1.3             3.5
      War-1, Agent
      Orange, "Gulf
      War syndrome,"
      compensable 0%
    Priority 7:           11.8             6.7            20.0
      copayments for
      VHA services
      (category C)
  Medicaid enrolled        2.7%            3.3%            1.6%
  Health status
      indicators, mean
      (standard error)
    PCS: VSF-36           35.0 (11.90)    34.0 (11.63)    37.1 (12.13)
      physical
      component scale
    MCS: STD mental       45.2 (13.74)    43.8 (13.91)    47.7 (13.04)
      component scale
    Total # of             3.3 (2.11)      3.3 (2.12)      3.1 (2.08)
      physical
      chronicle
      conditions
      (0-12)
    Total # of mental      0.5 (0.77)      0.6 (0.81)      0.4 (0.67)
      conditions (0-3)
No insurance              19.7 (6.76)     20.1 (6.80)     19.0 (6.63)
  percentage at MSA
  (metropolitan
  statistical area)
  level *
Distance **
  Distance difference    -29.9 (36.61)   -29.5 (36.06)   -30.5 (37.48)
      between the
      closest
    Non-VA hospital
      and the closest
    VA hospital,
      mean (standard
      error)
  Travel distance to
      the closest VA
      hospital
    Less than 10 miles    29.0%           30.8%           26.2%
    10-20 miles           16.5            15.3            18.5
    20-30 miles           10.3             9.9            11.0
    30-40 miles            8.0             7.8             8.4
    40-50 miles            7.6             7.5             7.8
    Over 50 miles         28.5            28.7            28.2
Hospital information
  Percent for whom: **
    The closest VA         8.5             7.6            10.1
      hospital is not
      general hospital
    The closest VA        11.7            11.7            11.8
      hospital is not
      affiliated with
      medical school
  Mean # of major          9.2 (5.92)      9.2 (5.94)      9.2 (5.89)
      services
      provided in the
      closest
    VA hospital
      ([dagger])
  Mean # of non-VA         3.6 (7.09)      3.8 (7.39)      3.3 (6.55)
    hospitals
    affiliated with
    medical school
  Mean # of non-VA         7.1 (9.16)      7.5 (9.65)      6.5 (8.25)
    hospitals being
    general hospital
  Mean # of non-VA         9.5 (12.0)      9.9 (12.57)     8.7 (10.89)
    hospitals

([dagger]) We examined the following 18 major services categories:
adult day care, HIV/AIDS services, alc/ drug abuse, alc/drug
outpatient, arthritis treatment, card cath lab, cardiac ICU,
emergency dept., geriatric services, hospice, med/surg ICU,
oncology services, psychiatric care, psychiatric out-patient,
rehabilitation care, physical rehabilitation, skilled nursing,
urgent care center. Variables with *were used only in model of
choosing insurance type, variables with ** were used only in the
model of using health services. The rest were common variables.

Table 2: Use of VA Care by VA Patient's Insurance Status

                                                             Private
                           Total Sample,   No Insurance,    Insurance,
                            N = 48,448      N = 29,944      N = 18,504

Percentage with VA care        79.9%           88.1%          66.4%
  Total VA cost ($1,000)       4.40         5.3 (11.49)     2.3 (5.75)
Percentage with VA             9.2%            12.9%           3.2%
    inpatient care
  VA inpatient                 12.10       12.6 (17.05)    9.0 (11.91)
    cost ($1,000)
  Percentage with VA            69%             80%            51%
    pharmacy cost > 0
  VA pharmacy cost ($)       730 (135)       785 (138)      590 (129)
  Ratio of pharmacy            19.9%           19.8%      20.1% (0.24)
    cost/total cost

Note: VA cost was FY 1999 cost using VA ARC annual data. Standard
errors are presented in parentheses.

Table 3: Results on Insurance Model (Probability of Having
Private Insurance)

Independent Variables                    Coefficient       Standard
                                                             Error

Female                                     0.24***           0.025
Nonwhite                                  -0.06***           0.015
Age                                      (Under 45:
                                       reference group)
  45-54                                   -0.35***           0.041
  55-64                                   -0.56***           0.042
  65 and older                            -0.33***           0.059
Education                            (College 1-3 years:
                                       reference group)
  [less than or equal to] 8 years,
    elementary                            -0.39***           0.041
  9-11 years, some high school            -0.37***           0.031
  High school graduate                    -0.07***           0.016
  At least college graduate                0.20***           0.019
Medicaid enrolled                          0.26***           0.045
Economic hardship: yes = 1                -0.28***           0.019
Live alone                                 0.04*             0.020
Marriage status                      (Currently married:
                                       reference group)
  Formerly married                        -0.62***           0.018
  Never married                           -0.66***           0.024
Employment status                        (Employed:
                                       reference group)
  Retired                                 -0.35***           0.020
  Other                                   -0.87***           0.023
VA priority status                      (Priority-5:
                                       reference group)
Priority-1                                 0.44***           0.022
Priority-2                                 0.67***           0.022
Priority-3                                 0.68***           0.018
Priority 4                                -0.17***           0.088
Priority 6                                 0.58*             0.043
Priority-7                                 0.83***           0.021
PCs                                        0.01***           0.000
MCS                                        0.00***           0.000
# of physical chronicle conditions         0.01              0.005
# of mental conditions                     0.01              0.011
No insurance rate at MSA level            -0.06***           0.000

* Significance [less than or equal to] .05.
*** Significance [less than or equal to] .001.

Table 4: Comparisons of Insurance Effect for IV Models
versus Naive Models

                                           IV Models

                                 Coefficient of       Predicted
                                Private Insurance      Value 1
Dependent Variables                ([dagger])        ([dagger])

Probability of any VA cost          -1.03 ***           83.4%
  Level of total VA cost            -1.42              $3,906
Probability of any VA               -0.96 ***            7.1%
  inpatient cost
    Level of VA inpatient cost      -3.47              $9,769
Probability of any VA               -1.03 ***           72.1%
  pharmacy cost
    Level of VA pharmacy cost       -0.24 *              $638
    Pharmacy cost/total cost        -0.002              19.5%

                                           IV Models

                                   Predicted
                                    Value 2          Difference
Dependent Variables               ([section])       ([paragraph])

Probability of any VA cost            84.4%               1%
  Level of total VA cost             $3,959              $53
Probability of any VA                  7.6%             0.5%
  inpatient cost
    Level of VA inpatient cost       $9,872             $103
Probability of any VA                 73.5%             1.4%
  pharmacy cost
    Level of VA pharmacy cost          $648              $10
    Pharmacy cost/total cost          19.5%                0

                                          Naive Models

                                  Coefficient         Predicted
                                   of Private      Value 1 ([double
Dependent Variables                Insurance          dagger])

Probability of any VA cost          -0.60 ***           83.7%
  Level of total VA cost            -1.25              $3,906
Probability of any VA               -0.41 ***            6.6%
  inpatient cost
    Level of VA inpatient cost      -2.19              $9,673
Probability of any VA               -0.61 ***           72.4%
  pharmacy cost
    Level of VA pharmacy cost       -0.05 *              $649
    Pharmacy cost/total cost        -0.7%               19.5%

                                          Naive Models

                                   Predicted
                                    Value 2          Difference
Dependent Variables               ([section])       ([paragraph])

Probability of any VA cost           84.3%              0.5%
  Level of total VA cost            $3,956               $50
Probability of any VA                 6.7%              0.1%
  inpatient cost
    Level of VA inpatient cost      $9,761               $88
Probability of any VA                73.2%              0.8%
  pharmacy cost
    Level of VA pharmacy cost         $651                $4
    Pharmacy cost/total cost         19.5%                 0

* Significance [less than or equal to] .05

*** Significance [less than or equal to] .001

([dagger]) From Table 5 first row.

([double dagger]) The predicted value at the population average with
insurance rate of 38%.

([section]) The predicated value at the population average with
insurance rate of 34%.

([paragraph)] The difference between predicted value 2 and predicted
value 1 is the impact of dropping 4% insurance rate.

Table 5, The Full IV Regression Results on VA Total Cost,
Inpatient Cost, and Pharmacy Cost (FY 1999)

                           Prob (Total
                           Cost > 0)

Independent                              Standard
Variables                  Coefficient   Error

Private insurance          -1.03 ***     0.710
Female                      0.21 ***     0.028
Nonwhite                   -0.01         0.016

Age                        (Under 45: reference group)
  45-54                     0.21 ***     0.048
  55-64                     0.31 ***     0.055
  65 and older              0.04         0.072

Education                  (College 1-3 years: reference group)
 [less than or equal to]    0.05         0.045
 8 years, elementary
 9-11 years,               -0.02         0.032
  some high
  school
 High school                0.02         0.017
  graduate
 At least college           0.03         0.021
  graduate
Medicaid                   -0.28 ***     0.044
Economic                   -0.18 ***     0.021
 hardship
Live alone                  0.05 *       0.021

Marriage status            (Currently married: reference group)
 Formerly                  -0.63 **      0.024
  married
 Never married             -0.11 ***     0.029

Employment                 (Employed: reference group)
 status
  Retired                   0.10 ***     0.026
  Others                   -0.04         0.026

VA priority                (Priority_5: reference group)
 Priority_1                 0.38 ***     0.029
 Priority_2                 0.16 ***     0.029
 Priority_3                 0.06 *       0.026
 Priority_4                 0.00         0.109
 Priority_6                -0.16 ***     0.049
 Priority_7                -0.27 ***     0.031
PCs                        -0.01 ***     0.001
MCs                        -0.00         0.001
# of physical               0.1l ***     0.005
  conditions
# of mental                 0.09 ***     0.011
  conditions
Difference in              -0.00         0.000
 travel distance
Travel distance            (<10 miles: reference group)
 to VA
 10-19 miles               -0.18 ***     0.022
 20-29 miles               -0.22 ***     0.027
 30-39 miles               -0.27 ***     0.030
 40-50 miles               -0.24 ***     0.033
 Over 50 miles             -0.22 ***     0.033
VA was not
 general hospital           0.02         0.025
VA was not                  0.08 ***     0.023
 affiliated with
 medical school
# of services               0.00         0.001
 provided in VA
# of non-VA hospitals      -0.01 **      0.003
 affiliated with
 medical school
# of non-VA                -0.01         0.004
 hospitals being
 general hospitals
# of non-VA                 0.00         0.004
 hospitals

                           Level of
                           Total
                           Cost

Independent                              Standard
Variables                  Coefficient   Error

Private insurance          -1.42         0.745
Female                     -0.11         0.199
Nonwhite                   -0.25 *       0.114

Age
  45-54                     0.44         0.271
  55-64                     1.50 ***     0.260
  65 and older              1.84 ***     0.387

Education                  (College 1-3 years: reference group)
 [less than or equal to]   -1.10 ***     0.248
 8 years, elementary
 9-11 years,               -0.42 *       0.207
  some high
  school
 High school               -0.15         0.119
  graduate
 At least college           0.01         0.162
  graduate
Medicaid                    0.26         0.313
Economic                   -0.73 ***     0.144
 hardship
Live alone                 -0.41 **      0.140

Marriage status            (Currently married: reference group)
 Formerly                   0.57 **      0.185
  married
 Never married              0.86 ***     0.222

Employment                 (Employed: reference group)
 status
  Retired                   1.35 ***     0.179
  Others                    1.80 ***     0.219

VA priority                (Priority_5: reference group)
 Priority_1                 1.01 ***     0.188
 Priority_2                -0.61 **      0.238
 Priority_3                -0.50 *       0.220
 Priority_4                 8.93 ***     0.412
 Priority_6                -0.77         0.428
 Priority_7                -0.48         0.262
PCs                        -0.07 ***     0.005
MCs                        -0.03 ***     0.005
# of physical               0.36 ***     0.031
  conditions
# of mental                 0.58 ***     0.075
  conditions
Difference in               0.00         0.002
 travel distance
Travel distance            (<10 miles: reference group)
 to VA
 10-19 miles               -0.53 ***     0.159
 20-29 miles               -0.68 ***     0.195
 30-39 miles               -0.72 ***     0.219
 40-50 miles               -1.22 ***     0.231
 Over 50 miles             -1.44 ***     0.236
VA was not
 general hospital           0.39 *       0.186
VA was not                 -0.36 *       0.159
 affiliated with
 medical school
# of services               0.03 **      0.009
 provided in VA
# of non-VA hospitals      -0.06 **      0.019
 affiliated with
 medical school
# of non-VA                 0.12 ***     0.030
 hospitals being
 general hospitals
# of non-VA                -0.04         0.025

                           Prob (Inpatient
                           Cost > 0)

Independent                Standard
Variables                  Coefficient   Error

Private insurance          -0.96 ***     0.101
Female                     -0.02         0.037
Nonwhite                   -0.05 *       0.019

Age
  45-54                     0.02         0.044
  55-64                     0.06         0.041
  65 and older              0.15 **      0.057

Education                  (College 1-3 years: reference group)
 [less than or equal to]   -0.05         0.039
 8 years, elementary
 9-11 years,               -0.02         0.033
  some high
  school
 High school                0.01         0.020
  graduate
 At least college           0.03         0.029
  graduate
Medicaid                   -0.07         0.048
Economic                   -0.08 ***     0.023
 hardship
Live alone                 -0.03         0.023

Marriage status            (Currently married: reference group)
 Formerly                   0.01         0.030
  married
 Never married             -0.02         0.036

Employment
 status
  Retired                   0.09 **      0.031
  Others                    0.08 *       0.038

VA priority                (Priority_5: reference group)
 Priority_1                 0.02         0.030
 Priority_2                -0.01         0.039
 Priority_3                -0.08 *       0.037
 Priority_4                 0.48 ***     0.055
 Priority_6                -0.09         0.079
 Priority_7                -0.11 *       0.049
PCs                        -0.01 ***     0.001
MCs                        -0.01 ***     0.001
# of physical               0.06 ***     0.005
  conditions
# of mental                 0.05 ***     0.012
  conditions
Difference in               0.00 **      0.000
 travel distance
Travel distance            (<10 miles: reference group)
 to VA
 10-19 miles               -0.02         0.026
 20-29 miles               -0.12 ***     0.033
 30-39 miles               -0.12 **      0.037
 40-50 miles               -0.18 ***     0.041
 Over 50 miles             -0.16 ***     0.042
VA was not
 general hospital          -0.04         0.033
VA was not                 -0.02         0.028
 affiliated with
 medical school
# of services               0.00 *       0.001
 provided in VA
# of non-VA hospitals      -0.00         0.003
 affiliated with
 medical school
# of non-VA                 0.00         0.005
 hospitals being
 general hospitals
# of non-VA                -0.00         0.004

                           Level of
                           Inpatient
                           Cost

Independent                              Standard
Variables                  Coefficient   Error

Private insurance          -3.47         4.158
Female                     -1.84         1.104
Nonwhite                   -0.22         0.50

Age
  45-54                    -0.22         1.151
  55-64                    -0.85         0.979
  65 and older             -2.79 *       1.391

Education                  (College 1-3 years: reference group)
 [less than or equal to]   -1.72         0.989
 8 years, elementary
 9-11 years,               -0.17         0.857
  some high
  school
 High school               -0.17         0.585
  graduate
 At least college          -0.05         0.944
  graduate
Medicaid                    0.55         1.335
Economic                   -2.27 ***     0.608
 hardship
Live alone                 -0.51         0.641

Marriage status            (Currently married: reference group)
 Formerly                   0.90         0.763
  married
 Never married              1.58         1.019

Employment
 status
  Retired                   4.17 ***     0.921
  Others                    2.75 **      0.984

VA priority                (Priority_5: reference group)
 Priority_1                 2.00 **      0.770
 Priority_2                -1.33         1.110
 Priority_3                -0.58         0.933
 Priority_4                 5.59 ***     1.171
 Priority_6                -0.41         2.581
 Priority_7                 0.37         1.400
PCs                        -0.13 ***     0.026
MCs                         0.00         0.022
# of physical               0.18         0.132
  conditions
# of mental                -0.24         0.332
  conditions
Difference in              -0.02         0.015
 travel distance
Travel distance            (<10 miles: reference group)
 to VA
 10-19 miles               -0.48         0.756
 20-29 miles                0.87         0.965
 30-39 miles                1.30         1.072
 40-50 miles               -0.46         1.191
 Over 50 miles             -2.01         1.268
VA was not
 general hospital           1.15         0.975
VA was not                 -2.08 **      0.797
 affiliated with
 medical school
# of services               0.02         0.042
 provided in VA
# of non-VA hospitals      -0.07         0.087
 affiliated with
 medical school
# of non-VA                 0.12         0.137
 hospitals being
 general hospitals
# of non-VA                 0.03         0.113

                           Prob
                           (Pharmacy
                           Cost > 0)

Independent                              Standard
Variables                  Coefficient    Error

Private insurance          -1.03 ***     0.072
Female                      0.25 ***     0.025
Nonwhite                    0.00         0.015

Age
  45-54                     0.25 ***     0.043
  55-64                     0.28 ***     0.045
  65 and older              0.02         0.062

Education                  (College 1-3 years: reference group)
 [less than or equal to]    0.05         0.039
 8 years, elementary
 9-11 years,                0.03         0.029
  some high
  school
 High school                0.05 ***     0.016
  graduate
 At least college          -0.01         0.019
  graduate
Medicaid                   -0.31 ***     0.040
Economic                   -0.19 ***     0.019
 hardship
Live alone                  0.07         0.019

Marriage status            (Currently married: reference group)
 Formerly                  -0.05 *       0.023
  married
 Never married             -0.16 ***     0.027

Employment
 status
  Retired                   0.07 **      0.023
  Others                   -0.05         0.025

VA priority                (Priority_5: reference group)
 Priority_1                 0.28 ***     0.025
 Priority_2                 0.10 ***     0.028
 Priority_3                -0.02         0.025
 Priority_4                 0.29 ***     0.079
 Priority_6                -0.21 ***     0.048
 Priority_7                -0.31 ***     0.031
PCs                        -0.01 ***     0.001
MCs                        -0.00 ***     0.001
# of physical               0.15 ***     0.005
  conditions
# of mental                 0.09 ***     0.011
  conditions
Difference in              -0.00         0.000
 travel distance
Travel distance            (< 10 miles: reference group)
 to VA
 10-19 miles               -0.13 ***     0.020
 20-29 miles               -0.17 ***     0.025
 30-39 miles               -0.20 ***     0.028
 40-50 miles               -0.16 ***     0.030
 Over 50 miles             -0.14 ***     0.031
VA was not
 general hospital          -0.07 **      0.023
VA was not                  0.07 ***     0.021
 affiliated with
 medical school
# of services              -0.00         0.001
 provided in VA
# of non-VA hospitals      -0.01 ***     0.002
 affiliated with
 medical school
# of non-VA                -0.00         0.004
 hospitals being
 general hospitals
# of non-VA                 0.00         0.003

                           Level of
                           Pharmacy
                           Cost

Independent                              Standard
Variables                  Coefficient    Error

Private insurance          -0.24 *       0.110
Female                      0.04         0.029
Nonwhite                   -0.10 ***     0.017

Age
  45-54                    -0.02         0.038
  55-64                    -0.06         0.036
  65 and older             -0.23 ***     0.055

Education                  (College 1-3 years: reference group)
 [less than or equal to]   -0.10 **      0.035
 8 years, elementary
 9-11 years,               -0.07 *       0.029
  some high
  school
 High school               -0.04 *       0.017
  graduate
 At least college           0.07 **      0.024
  graduate
Medicaid                    0.11 *       0.046
Economic                   -0.11 ***     0.021
 hardship
Live alone                  0.02         0.020

Marriage status            (Currently married: reference group)
 Formerly                  -0.07 **      0.026
  married
 Never married              0.03         0.032

Employment
 status
  Retired                   0.16 ***     0.026
  Others                    0.15 ***     0.032

VA priority                (Priority_5: reference group)
 Priority_1                 0.28 ***     0.027
 Priority_2                 0.00         0.034
 Priority_3                -0.06         0.032
 Priority_4                 0.43 ***     0.058
 Priority_6                 0.02         0.065
 Priority_7                 0.00         0.038
PCs                        -0.01 ***     0.001
MCs                        -0.00 ***     0.001
# of physical               0.06 ***     0.005
  conditions
# of mental                 0.13 ***     0.011
  conditions
Difference in               0.00         0.000
 travel distance
Travel distance            (< 10 miles: reference group)
 to VA
 10-19 miles               -0.05 *       0.024
 20-29 miles               -0.03         0.029
 30-39 miles               -0.09 **      0.032
 40-50 miles               -0.09 **      0.034
 Over 50 miles             -0.09 *       0.034
VA was not
 general hospital           0.02         0.028
VA was not                 -0.01         0.023
 affiliated with
 medical school
# of services               0.00         0.001
 provided in VA
# of non-VA hospitals      -0.01 **      0.003
 affiliated with
 medical school
# of non-VA                 0.01 *       0.004
 hospitals being
 general hospitals
# of non-VA                -0.00         0.004

                           Pharmacy
                           Cost/Total
                           Cost

Independent                              Standard
Variables                  Coefficient    Error

Private insurance          -0.002        0.016
Female                      0.00         0.004
Nonwhite                   -0.02 ***     0.002

Age
  45-54                     0.01 *       0.006
  55-64                    -0.01 *       0.006
  65 and older             -0.05 ***     0.008

Education                  (College 1-3 years: reference group)
 [less than or equal to]    0.01 *       0.005
 8 years, elementary
 9-11 years,                0.01 **      0.005
  some high
  school
 High school                0.00         0.003
  graduate
 At least college           0.00         0.004
  graduate
Medicaid                    0.02 **      0.007
Economic                   -0.02 ***     0.003
 hardship
Live alone                  0.01 *       0.003

Marriage status            (Currently married: reference group)
 Formerly                  -0.01 **      0.004
  married
 Never married             -0.01 **      0.005

Employment
 status
  Retired                   0.00         0.004
  Others                   -0.00         0.005

VA priority                (Priority_5: reference group)
 Priority_1                 0.01 **      0.004
 Priority_2                 0.00         0.005
 Priority_3                -0.02 ***     0.005
 Priority_4                -0.02 *       0.009
 Priority_6                 0.00         0.009
 Priority_7                 0.01         0.006
PCs                        -0.00         0.000
MCs                        -0.00 **      0.000
# of physical               0.01 ***     0.001
  conditions
# of mental                 0.10 ***     0.002
  conditions
Difference in              -0.00 *       0.000
 travel distance
Travel distance            (< 10 miles: reference group)
 to VA
 10-19 miles                0.01 *       0.003
 20-29 miles                0.01 *       0.004
 30-39 miles                0.01 **      0.005
 40-50 miles                0.02 ***     0.005
 Over 50 miles              0.03 ***     0.005
VA was not
 general hospital          -0.01         0.004
VA was not                  0.01 *       0.003
 affiliated with
 medical school
# of services              -0.00 ***     0.000
 provided in VA
# of non-VA hospitals      -0.00         0.000
 affiliated with
 medical school
# of non-VA                -0.00 **      0.001
 hospitals being
 general hospitals
# of non-VA                 0.00 *       0.001

* Significance [less than or equal to] .05.

** Significance [less than or equal to] .01.

*** Significance [less than or equal to] .001.