The social role of not-for-profit organizations: hospital provision of charity care.

Thorpe, Kenneth E. ; Phelps, Charles E.

I. INTRODUCTION

Private, as well as public, institutions may be granted tax-exempt status to induce socially desirable behaviors on their part. Specifically, the tax treatment of not-for-profit hospitals exists in part to encourage the provision of charity care to patients without health insurance. These patients often become charity care patients or bad debts for the hospital, collectively referred to as "uncompensated care" patients. Many people regard the financing and provision of uncompensated care as one of the nation's leading health issues, as summarized by Sloan et al. [1986].

A number of states have adopted innovative strategies to reduce uncompensated care. These strategies fall into four general categories: extend Medicaid coverage; establish public hospital systems; develop new private or public sector insurance plans; or subsidize hospitals through uncompensated care pools. Despite the widespread attention directed toward the uncompensated care issue, few studies have examined the actual performance of programs established to solve the problem. This study analyzes the effects of an uncompensated care financing program developed as part of a hospital cost-control system in New York State.

Our analysis depends upon a natural experiment conducted by the New York state government. In 1983, as a part of a comprehensive hospital payment reform, it installed a fund that included both straight block grants and a fund to subsidize the provision of charity care, where the extent of subsidy varied across time. In years preceding 1983, neither the block grant nor the subsidy program existed. We use these policy changes to estimate effects of "income" and "price" effects on the provision of charity care.

Many studies of not-for-profit hospitals model their behavior as analogous to a single utility-maximizing decision maker, typically with the hospital's output and quality as representative "goods" in the utility function. The hospital's response to external price and income shifts should follow a pattern analogous to the predictions of a standard utility-maximizing model, as developed by Newhouse [1970], and expanded by Morrisey et al. [1984] and Phelps and Sened, [1989]. In these models, a "decision rule" emerges that guides the behavior of the hospital, much like a utility function guides an individual's behavior. Peltzman's [1976] model of a regulatory process has similar characteristics.

Using this underlying framework, we address three questions frequently raised about the problem of inducing hospitals to provide charity care: (1) Does the supply of charity care (here, uncompensated hospital care) increase with a hospital's income, and if so, by how much? (2) Does the supply of hospital charity care depend upon the size of the subsidy provided by the state for such care, and if so, by how much? (3) Does the response of hospitals to various grant programs differ systematically by institutional characteristic and market conditions? Estimates of the magnitudes of these effects can help policymakers choose more equitable strategies to provide hospital care for persons without insurance. Our analysis shows that, while price matters, changes in income apparently have little or no effect on the choices made by not-for-profit hospitals in the provision of "charity care."

II. THE NEW YORK HOSPITAL PROSPECTIVE REIMBURSEMENT METHOD

In 1983, New York adopted an all-payer hospital reimbursement program; the New York Prospective Hospital Reimbursement Methodology (NYPHRM). (1)

The major goals of the new reimbursement program included:

* limit cost growth to a trend factor specified by a panel of economists;

* achieve a stable and predictable revenue flow to hospitals from payers;

* reduce the differential between hospital charges and costs;

* assist financially distressed hospitals; and

* assist those hospitals providing charity (uncompensated) care.

We have analyzed the effects of this program on the growth of hospital costs in Thorpe and Phelps [1990]. This paper studies the effects of the reimbursement reforms on the provision of charity care, the primary "distributive" goal of the program. The main tools available to the state through the program included both matching and block grants used to finance uncompensated care and to aid fiscally distressed facilities. We describe these grant programs next.

The "Subsidy" Program: The Bad Debt and Charity Care Pool

The New York program created eight regional pools of funds to pay for charity care. Reimbursement was then paid to hospitals in proportion to their two-year-lagged provision of charity care. This system of pools provided the basis for the charity care subsidy program that we analyze below. The following describes its structure.

Collection and Dispersal of Funds. The State of New York placed a surtax on all third-party payers for use in its grant programs, distributed through the regional pools. The pools provided substantial funds, growing from $150 million in 1983, the program's first year, to over $300 million by 1985 (see Table I); the surtax on each hospital's per diem payment rate increased steadily from zero in 1982 to 2 percent in 1983 and 4 percent in 1985. However, rather than paying this surtax to the hospitals, each payer sent the funds directly to the regional subsidy pools, which were administered by the State Health Department. They dispersed the funds according to the following rules, from which we established the "price" paid for charity care:

TABLE I
Revenues from NYPHRM Pools, 1983-1985
($ millions)

 Pools as
 Bad Debt Percent of
 and Charity All Other Reimbursable
Year Care Pool Pools * Costs

1983 $150.5 $119.3 3.3
1984 242.2 209.1 5.3
1985 318.9 218.9 5.8


 Uncompensated Care
Year Payment Coverage Rate

1981 0
1982 0
1983 43 percent
1984 75 percent

1. They calculated statewide bad debt and charity care "need" (both inpatient and outpatient) for voluntary and proprietary hospitals, based on uncompensated care costs incurred two years earlier. For example, for rate year 1983, they defined each hospital's bad debt and charity care "need" as inpatient uncompensated care costs (reduced from charges) plus an estimate of its outpatient deficit in 1981. Other years' calculations proceeded similarly.

2. Total bad debt and charity care need, divided by total resources available from the bad debt and charity care fund, yielded the percentage of need that the pool could cover. Every hospital, except major, publicly owned hospitals (dominated by New York's Health and Hospitals Corporation) would receive this percentage of need. Major public hospitals received bad-debt pool payments only in proportion to their share of statewide reimbursable costs.

The "Price" for Uncompensated Care Through Time. Our analysis below relies substantially on changes in the rate of compensated care payments to hospitals, so we describe these changes in detail. As the magnitude of the surtax changed through time, the approximate pattern of payment as a percent of each hospital's overall average costs was:

The payment of "only" 43 to 75 percent of average costs and the delay of two years both blunt the financial incentives to provide free care. However, if this payment exceeds the short-run variable costs of providing the care, the hospital can at least partly recover fixed cost, making the choice desirable. As Friedman and Pauly [1981] show, fixed costs vary considerably from setting to setting, but in the short run, they can constitute up to 70 percent of total costs. Thus, payments of 43 and 75 percent of average costs could quite likely make important contributions to the hospital's operating margin above variable costs.

As described above, the 1983 reimbursement methodology provided payments for uncompensated care with a two-year lag, always basing the current year's subsidy payments on the actual provision of charity care from two years previous. This could affect the way hospitals viewed this subsidy. While we can model the effects of the time delay through discounting (see section III), the issue of regulatory instability still remains. If hospital managers feared that essential features of the reimbursement system would change adversely in the intervening years, their willingness to respond to a matching program such as the uncompensated care pool might evaporate. The New York methodology underwent considerable change during this period, so such fears might not be baseless, but we cannot measure or model these concerns. As a separate matter, Rose-Ackerman [1987] suggests that offsetting changes in donations might blur the effects of changes in the government program, but available data suggest that philanthropic contributions to hospitals provided only a very small proportion of their budgets and did not vary meaningfully during the period we study.

Finally, the State Health Department recognized the potential risk that hospitals would decrease their efforts to collect bad debts and specifically included provisions to limit any degradation of bill-collecting efforts. It stipulated that hospitals must maintain and implement reasonable collection procedures before receiving revenues from the uncompensated care pools. Moreover, to ensure that increased public hospital provision of uncompensated care did not result in less effort by private hospitals, the Health Department developed a detailed inter-sector analysis of uncompensated care. If significant changes in charity care provided by major public and private hospitals appeared, the state would make adjustments to the pool allocations to reflect those changes.

The "Income" Program: Other Revenue Pools

The reimbursement and subsidy program initiated in 1983 contained other pools of revenue in addition to the charity-care subsidy pool just described. The state dispersed revenue from these pools independently of charity care provided by the hospitals, through which we can estimate income effects for the hospitals. We describe these pools and their purpose below.

The New York program contained three additional revenue pools, one for distressed hospitals, one for "transitional", and one discretionary pool. Of these, the discretionary pool provided by far the most extensive funds. In rate year 1985, revenue from the discretionary pool totaled $191 million, compared with $27 million in the distressed hospital and transition pools. The discretionary pool based payments on an allowance of approximately 1 to 2 percent of each hospital's reimbursable costs in each rate year. Intended uses of the discretionary pool included the retirement of short-term, noncapital debt, financing uncompensated care not met by the regional uncompensated care pools or to improve a hospital's current financial ratios.

Finally, the state made available to private hospitals revenues from the transition fund equal to .25 percent of each hospital's reimbursable costs. Regulators intended these additional payments to offset potential fiscal hardships to hospitals created when Medicare became part of the state's rate-setting system. (Pre-1983 cost control programs had excluded Medicare.)

The state dispersed funds from each of these "other" pools independently of the actual provision of charity care by the hospitals in New York; thus these funds constitute a true block grant, from which we can estimate income effects.

Summary

The rate setting system instituted in 1983 provides a natural experiment from which we can measure the effects of both subsidies price and block income grants on the extent of charity care provided by not-for-profit hospitals. In our analysis, we make use of data both from the post-reform period (1983-84) and from the previous period (1981-82), during which no price or income subsidies existed. The next section describes the model and the data we use for this analysis.

III. EMPIRICAL SPECIFICATION AND DATA SOURCES

Empirical Specification

We model the volume of uncompensated care provided by hospitals as a function of the underlying demand for free care, institutional characteristics (possibly correlated with the manager's utility function), market characteristics that constrain the hospital's "standard" price, the price subsidy from the uncompensated care pool, and block income grants from other pools.

We measure the underlying excess demand for free care by two county level variables: the county ratio of Medicaid eligibles to poverty population and real per capita income. Factors influencing the supply of free care (e.g., the shape of the hospital's decision rule) include the hospital's teaching status, measured by the ratio of interns and residents to beds, the (lagged) county share of total and free-care admissions provided by public hospitals, and the (lagged) financial status of the hospital. Since we study only not-for-profit hospitals (constituting over 95 percent of all New York hospitals), we did not enter ownership as a separate control variable. A Herfindahl index measures potential monopoly power, using counties to define the market. The price for uncompensated care created for each hospital by the uncompensated care pool (the matching grant) and the income provided by all other pools (block grants) provided the remaining independent variables.

In each year, the uncompensated care variable reflects the product of three variables: (a) each hospital's audited cost of uncompensated care (lagged by two years), (b) each year's system-wide "coverage rate" for uncompensated care, (see section II) and (c) the relevant two-year discount factor caused by the delay in payments. We used yields on municipal bonds for New York State during the years 1984 to 1986 as a proxy for the relevant two-year discount factor. In addition, we adjusted for inflation using the New York state specific consumer price index. Thus, for hospital i in year t, with costs of producing uncompensated care COS[T.sub.it], we defined the uncompensated care "price" as equal to

(2) Uncompensated Care Price = COS[T.sub.it] (1-Coverage [Rate.sub.t]) / [(1+[r.sub.t+1])(1+[r.sub.t+2])].

Most price variation arises due to the inter-temporal changes in the coverage rate described in section II. We estimated a model of the following form:

(3) [(Uncompensated care costs).sub.it] = [[beta].sub.0](Constant term) + [[beta].sub.1](Uncompensated Care Price) + [[beta].sub.2](Revenues from other pools) + [[beta].sub.3] (Per capita income in county of hospital) + [[beta].sub.4](Ratio Medicaid eligibles to county poverty population) + [[beta].sub.5](Dummy Variable if lagged net income was positive) + [[beta].sub.6](Ratio of interns and residents per bed) + [[beta].sub.7] (County Discharge-Based Herfindahl Index) + [[beta].sub.8] (Lagged public hospital share of total county discharges) + [[beta].sub.9] (Dummy Variable designating an upstate hospital) + [[beta].sub.10] (Dummy Variable designating a church-related hospital) + [[beta].sub.11] (Numbers of hospital beds) + [e.sub.it]

where [e.sub.it] is a random variable, assumed normally distributed after appropriate corrections that we discuss next.

The structure of our 1981-1984 data raises a number of econometric issues. First, the residuals from the model discussed above probably do not exhibit mutual independence. The model takes on some characteristics of an autoregressive process since the uncompensated care price depends upon the dependent variable lagged two years. We correct for autocorrelation in the model by alternatively assuming a first- and second-order autoregressive process using a maximum likelihood estimator. (2) Since the results did not differ by our selection of an autoregressive process, we display those from the first-order estimates. The transformed residuals displayed no further autocorrelation.

Second, the use of hospital-specific data results in heteroskedasticity. To account for unequal error variances in our model, we "deflated" the observations on uncompensated care, bad debt and charity care, and other pool contributions by an exogenous measure of size--the number of hospital beds. Thus, all of our regression results refer to the amount of uncompensated care per bed. Even after dividing by bed size, the error terms remained heteroskedastic. In addition, the distribution of the dependent variable (uncompensated care costs divided by numbers of hospital beds) remained somewhat skewed. To correct for these problems, the we used the logarithm of these three variables in the analysis. In addition, per capita income and beds also appear as logarithms. All other variables appear in their natural units. These changes resulted in the nearly best Box-Cox [1964] transformation of the dependent variable and virtually eliminated heteroskedasticity in the residuals of our regressions.

Third, the other pool (block income) revenues may be endogenous because hospital payments from this pool are proportional to reimbursable inpatient costs. To allow for this possibility, we also estimated the model using two-stage least squares, using hospital-specific case mix and wage indices as instruments. (3) The wage index was based on a survey completed by the New York State Department of Health which collected detailed data on prevailing wages. However, the Hausman [1978] test fails to reject the null hypothesis that these other pool costs were exogenous, so we only report the results corrected for autocorrelation and heteroskedasticity.

Finally, to test the robustness of our results, we estimated a fixed effects model. We can estimate fixed effects models by including hospital-specific means, or (if using only two years of data) through first-differenced models.

To allow the price effect to vary over time, we estimated three first-differenced models, one differencing 1982 from 1981 (both years from the pre-reform period, with no income or subsidy grants available), another differencing 1983 from 1982 and finally one differencing 1984 from 1983. We estimated these models with and without the estimated value of the autocorrelation coefficient estimated above. Due to the similarity of the results to the primary models we estimated, we show the first-differenced models in the appendix.

APPENDIX
Regression Results: Uncompensated Care
(First-Differenced Models)

 Year

Variable 1982-1981 1983-1982 1984-1983

Constant -2.809** -0.442 -0.200
 (1.609) (1.440) (1.438)

Log (Discounted Bad *** 0.196 * 0.290 *
Debt Pool Payments) (.055) (.064)

Log (Other Pool Payments) *** -0.016 0.092
 (.070) (.099)

Log (Per Capita Income) 1.255* 0.707 * 0.691
 (.345) (.109) (.282)

Medicaid Enrollees/ -0.504 -.781 * 0.719 *
Poverty Population (.500) (.355) (.330)

Lagged Profit -0.049 -0.0002 -0.076
 (.079) (.065) (.064)

Residents/Bed 3.111 * 2.013 * 2.007 *
 (.377) (.819) (.743)

Herfindahl Index 0.733 -0.104 -0.076
 (.381) (.370) (.311)

Public Hospital Discharges -1.512 * -0.421 -0.44
as Share of Total (.580) (.646) (.518)

Note: Asymptotic standard errors estimated according to
White [1980].

[R.sup.2] .098 .11 .083
F Statistic 2.63 3.12 2.00
(degrees of freedom) (8, 144) (10, 142) (10, 142)
Probability of F .01 .01 .05

* Significant at p < .05

** Significant at p < .1

*** Variable not included in model.

Data

The data used in the study uniquely represent audited and verified reports of the level of bad debt and charity care provided by each hospital. The New York State Department of Health attempted to allow only those costs actually related to providing care to the uninsured in 1981 through 1984, rather than costs associated with different accounting conventions of writing off bad debts. We also used State Health Department data showing actual pool payments distributed under the reimbursement program. Hospital characteristics came from the state's yearly Institutional Cost Reports. Medicaid and poverty data came from the state's Department of Social Services. Summary statistics for the variables used in the analysis appear in Table II. (4)

TABLE II
Summary Statistics for Data
Means and Standard Deviations (in Parentheses)

 Year

Variable 1981 1982

Real Bad Debt and $1,545,541 1,608,726
Charity Care (3,817,038) (3,894,748)

Real Bad Debt and 3,378.7 3,509.1
Charity Care per bed (3,916.4) (4,554.1)

Real Bad Debt Pool 0 0
Payments

Real Bad Debt Pool 0 0

Real Other Pool 0 0
Payments

Real Other Pool 0 0
Payments per bed

Per Capita Income $9,885 11,670
in County (2,683) (3,134)

Medicaid/Poverty .767 .781
Population ratio (.151) (.143)

Lagged Profit .46 .50
(1 = Yes, 0 = No) (.500) (.500)

Public Hospital .096 .095
Discharges as (.126) (.126)
Share of Total

Herfindahl Index .298 .296
(County) (.255) (.252)

Beds per Hospital 285 285
 (253) (252)

Upstate Hospital 494 (same in all years)
(1 = Yes, 0 = No) (.50)

Residents/Bed .057 .061
 (.104) (.119)

 Year

Variable 1983 1984

Real Bad Debt and 1,585,704 1,591,634
Charity Care (3,376,502) (3,319,136)

Real Bad Debt and 3,024.2 3,723.3
Charity Care per bed (2,944.8) (3,723.3)

Real Bad Debt Pool 380,582 621,468
Payments (535,144) (721,030)

Real Bad Debt Pool 1,202 1,762
 (1,108.7) (1,481.2)

Real Other Pool 590,953 928,541
Payments (1,118,716) (1,866,124)

Real Other Pool 1,140.2 1960.3
Payments per bed (912.2) (1386.8)

Per Capita Income 12,462 13,657
in County (3,474) (3,864)

Medicaid/Poverty .851 .865
Population ratio (.163) (.178)

Lagged Profit .455 .609
(1 = Yes, 0 = No) (.500) (.490)

Public Hospital .097 .097
Discharges as (.126) (.126)
Share of Total

Herfindahl Index .297 .299
(County) (.258) (.258)

Beds per Hospital 288 287
 (253) (252)

Upstate Hospital (same in all years)
(1 = Yes, 0 = No)

Residents/Bed .055 .060
 (.096) (.112)

IV. EMPIRICAL RESULTS

Regression Analysis

Table III shows the primary results of our regression analyses, where the dependent variable shows the annual amount of bad-debt and charity care provided by each hospital. (Recall that we scaled our regressions by each hospital's bed size-see our discussion above on heteroskedasticity--so these equations measure charity care on a "per bed" basis.) We estimate the price-elasticity of supply of charity care as .17. Put differently, a 10 percent rise in bad debt (subsidy) pool payments leads to a 1.7 percent rise in the amount of uncompensated care. Hence, the matching nature of the free care pools in New York does increase the volume of spending to support uncompensated care, but by less than dollar-for-dollar. Hospitals either spend remaining revenues from the free care pools to support other hospital activities or retain them to augment working capital funds. However, the "price effect" is both statistically and behaviorally significant. These results support the hypothesis that many hospitals face excess demand for care by the indigent, and therefore restrict its provision.

TABLE III
Regression Results: Uncompensated Hospital Care

 Model
Variable 1 2

Constant 4.793* 4.958 *
 (1.411) (1.399)

Log (Bad Debt Pool Payments/Bed) 0.171* 0.216 *
 (0.038) (0.042)

Log (Other Pool Payments/Bed) -.029 .014
 (0.064) (0.065)

Log (Real Income/Capita) 0.302 * 0.279 **
 (0.155) (0.154)

Medicaid Eligibles/ -.350 ** -.374 **
Poverty Population (0.198) (0.197)

Lagged Net Profit -.036 -.037
(1 = Yes, 0 = No) (0.041) (0.042)

House Staff/Bed 3.024 * 3.442 *
 (0.302) (0.334)

Herfindahl Index .341 * .327
 (0.150) (0.148)

Lagged Public Hospital -.509 ** -.522 *
Discharges as Share of Total (0.238) (0.236)

Upstate Hospital -.637 * -.631
(1 = Yes, 0 = No) (0.073) (0.073)

Church Affiliated -.071 -.072
(1 = Yes, 0 = No) (0.073) (0.073)

Year = 1982 -.023 -.024
 (0.045) (0.045)

Year = 1983 -.968 -1.503 *
 (0.453) (0.489)

Year = 1984 -.840 * -1.408 *
 (0.487) (0.524)

House Staff/Bed x Debt -- -.148 *
Pool Payments/Bed (0.359)

[R.sup.2] .603 .606

Number of Hospitals 624 624

F Statistic 64.30 62.45

(degrees of freedom) (14, 609) (15, 608)

* Significant at p < .05

** Significant at p < .10

The regression results also show that revenues from the other block grant revenue pools did not increase the amount of uncompensated care--we found no meaningful "income effect." Given the competing demands for these unrestricted revenues, it should not prove surprising that hospitals used the bulk of these revenues for purposes other than directly supporting care for the indigent.

These results corroborate other observations that the tax benefits given to not-for-profit hospitals do not proportionately increase care to the indigent. Estimates from California indicate not-for-profit hospitals receive approximately $300 million in tax subsidies, yet provide only $85 million of free care. Some policymakers view this as an inadequate return on the not-for-profit status granted hospitals. (5)

Our results shed light on other puzzles in the problem of "uncompensated care." The existence of public hospitals in any community may reduce the level of free care provided by private hospitals and may reduce the price response of the matching (subsidy) grant. The results in Table III show that private hospitals provide a significantly lower level of uncompensated care in markets with a substantial public hospital presence. Each 10 percent increase in the share of discharges admitted to public facilities creates a 5.1 percent decline in the uncompensated care to bed ratio for private hospitals. These results correspond to similar nationwide studies by Thorpe and Brecher [1987].

Policymakers in New York had a special concern about the stability of the division of uncompensated care between public and private hospitals. Specifically, they did not want to create a two-class system with public hospitals providing all of the charity care. To maintain a balance, private hospitals received over 80 percent of the pool revenue, although they accounted for only 46 percent of total free care (unpublished data, New York Department of Health). Hence, the private hospitals' response to the subsidy pool in markets with a large public hospital presence has considerable special interest. To test whether private hospital response differed in markets with public hospitals, we also estimated the equations with an additional regressor interacting the public hospital share of discharges with the revenues received from the free care pool. It's coefficient did not differ significantly from zero. Thus, private hospitals provide a lower level of charity care in markets with a large public hospital presence, they respond to the uncompensated care subsidy similarly in markets with and without substantial public hospital systems.

Market characteristics affect the level of uncompensated care. Hospitals in more concentrated markets provide more uncompensated care, as formal models of hospital behavior predict, as developed in Phelps and Sened [1989].

Institutional characteristics, in particular the hospital's commitment to graduate medical education, also influence the level of uncompensated care. Each .1 increase in the hospital resident-to-bed ratio leads to a 30 percent increment in uncompensated care. These results augment previous research findings by Feder, Hadley, and Mullner [1984] that major academic medical centers provide a large volume of uncompensated care.

Despite their large commitment to the medically indigent, large teaching hospitals had a significantly lower price response than other hospitals (see Model 2, Table III). In fact, the interaction term indicates that the ratio of free care per bed in major teaching facilities (those with resident to bed ratios greater than .2) remained virtually unchanged after the introduction of the subsidy pools. This result suggests either that large teaching hospitals did not face significant excess demand for care by the indigent, or that they used the pool revenues for other purposes. We found no other statistically significant interactions with the price variable.

Finally, the hospitals' provision of uncompensated care also depended on their counties' level of Medicaid coverage. As expected, increased Medicaid coverage reduces the demand for uncompensated care. Each .1 increase in the ratio of Medicaid eligibles to poverty population results in a 3.5 percent decrease in uncompensated care per bed. Apparently, both the hospital subsidy programs and expanded Medicaid eligibility increase access for uninsured patients, and hence, at least in part, serve as policy substitutes.

Naturally, these estimates require the caveat that increases in government funding did not alter the response of private donors (and efforts to collect bad debts). If anything, one would suspect that larger government grants might push some of the remaining private giving out of the market, thus making our estimates an overstatement of actual forthcoming additional charity care. However, the stability of charitable contributions during the span of the study suggests that this would not create large problems, at least within the range of observed subsidy rates.

V. CONCLUSIONS

Many models suggest that not-for-profit hospitals should respond to government grants in much the same way as any utility-maximizing person or organization responds to changes in prices or income. The analysis presented above provides clear answers to the three questions posed about this issue at the end of section I.

First, the price-subsidy method for financing uncompensated care under the New York program did generate a net increase in care provided to indigent patients. This result requires that there was excess (unmet) demand for care by the indigent before the 1983 subsidy system assumed operation. On a per-bed basis, each 10 percent increase in payment from the uncompensated care pool resulted in approximately a 1.7 percent increase in care to those responsible for bad debts and receiving charity care. The 3.4 percent rise in uninsured inpatient admissions between 1982 and 1984 indicates these additional resources were devoted to treating the indigent rather than merely being reporting artifacts. (6) Moreover, the 3.4 percent rise in care to the indigent occurred despite a very slight state-wide decrease in inpatient admissions.

Second, matching grants from the uncompensated care pool had a greater effect in increasing indigent care than a simple revenue (block grant) transfer. The results indicate that hospitals retained the unrestricted subsidies, used them to purchase goods other than uncompensated care, or both, but they did not expand their charity care upon receipt of increased block grants.

Third, the price response to the uncompensated care pool varied widely. In general, teaching hospitals, predominantly located in New York City, exhibited significantly lower price responses than other hospitals. We cannot determine whether this different response reflects relatively little excess demand, offsetting expansion by the public sector, or institutional decisions to use the funds for other purposes.

The analysis also highlights an important result: greater public hospital provision of indigent care reduces the level of uncompensated care provided by private hospitals. Hence, an increase in tax-supported public hospital treatment of the indigent would be accompanied by a reduction of care for the uninsured by private hospitals. Yet, since we do not find a dollar for dollar reduction, public hospitals do make a difference, adding additional support for previous findings in this area by Thorpe and Brecher [1987]. (See also Roberts [1984] for analysis of a similar issue.) Perhaps in anticipation of some substitution, the New York subsidy program's design deliberately excluded public hospitals from most of the pool proceeds. Hence, crowding-out could differ significantly under alternative subsidy programs.

Finally, the analysis raises more basic issues regarding the most efficient and equitable method of assuring access to care by the indigent. Subsidy programs generally spread the costs of financing the uninsured across multiple payers and may appear more equitable than alternative schemes. However, hospital subsidies appear relatively inefficient in targeting revenues toward the uninsured patient. Hospitals use some of the revenues received from the pool to support uncompensated care, but a substantial fraction went to support other hospital activities or were simply retained. In light of these caveats, innovative methods of extending insurance benefits to a very diverse uninsured population, or more effective means of matching subsidy payments to care for the indigent, appear as useful alternative means of enticing hospitals to increase care to the uninsured patient.

REFERENCES

Beach, Charles M. and James G. MacKinnon. "A Maximum Likelihood Procedure for Regression With Autocorrelated Errors." Econometrica, January 1978, 51-58.

Box G. E. P. and D. R. Cox. "An Analysis of Transformations." Journal of Royal Statistical Society, B-26, 1964, 211-52.

Feder, Judith, Jack Hadley and Ross Mullner. "Falling Through the Cracks: Poverty, Insurance Coverage, and Hospital Care for the Poor." Milbank Memorial Fund Quarterly, Fall 1984, 544-66.

Friedman, Bernard and Mark V. Pauly. "Cost Functions for a Service Firm with Variable Quality and Stochastic Demand." Review of Economics and Statistics, November 1981, 620-24.

Harris, Jeffrey. "The Internal Organization of Hospitals: Some Economic Implications." Bell Journal of Economics, Autumn 1977, 46742.

Hausman, J. A. "Specification Tests in Econometrics." Econometrica, November 1978, 1251-72.

Morrisey, Michael A., Douglas A. Conrad, Stephen M. Shortell and Karen S. Cook, "Hospital Rate Review: A Theory and an Empirical Review." Journal of Health Economics, 1984, 25-47.

Newhouse, Joseph E "Towards a Theory of Non-Profit Institutions: An Economic Model of a Hospital." American Economic Review, March 1970, 64-74.

Peltzman, Sam. "Toward a More General Theory of Regulation." Journal of Law and Economics 19(2), 1976, 211-40.

Phelps, Charles E. and Itai Sened. "Market Equilibria with Not-for-Profit Firms." Rochester Center for Economic Research, 1989.

Roberts, Russell. "A Positive Model of Private Charity and Public Transfers." Journal of Political Economy, February 1984, 136-48.

Rose-Ackerman, Susan. "Ideals Versus Dollars: Donors, Charity Managers, and Government Grants." Journal of Political Economy, August 1987, 810-23.

Sloan, Frank, James Blumstein and James Perrin, eds. Uncompensated Hospital Care Rights and Responsibilities. Baltimore: Johns Hopkins University Press, 1986.

Thorpe, Kenneth E. "The Distributional Implications of Using Relative Prices in DRG Payment Systems." Inquiry, Spring 1987, 85-95.

Thorpe, Kenneth E. and Charles Brecher. "Improved Access to Care for the Uninsured Poor in Large Cities: Do Public Hospitals Make a Difference?" Journal of Health Politics, Policy and Law, Summer 1987, 313-24.

Thorpe, Kenneth E. and Charles E. Phelps. "Regulatory Intensity and Hospital Cost Growth." Journal of Health Economics, September 1990, 143-66.

White, Halbert. "A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity." Econometrica, May 1980, 817-38.

(1.) New Yorkers pronounce this acronym as "nyefrum", the second syllable sounding much like the clearing of a throat.

(2.) Beach and MacKinnon [1978]. We used both the Durbin-Watson and the Durbin's h statistic to detect autocorrelation. Both statistics produced the same results. We used hospital beds to account for hospital size for two reasons. First, bed size is exogenous in the short run, and there was little change in hospital beds over this period. Second, the "flow" variable, occupancy rate, was remarkably consistent across hospitals. This results from New York's strict rate-setting regulations which mandate that hospitals maintain at least an 80 percent occupancy rate.

(3.) The specific methodology used to develop the New York case mix measures is described in Thorpe [1987].

(4.) None of this data has been previously published. It was obtained on a floppy disk from the state government in Albany. For details, contact the authors.

(5.) See comment by Representative Fortney (Pete) Stark, Medicine and Health, June 1987. This is known as "Fortney's Complaint." Of course, the not-for-profit status provides other types of benefits, such as the increased trust patients have that physicians are faithfully executing their agency responsibilities.

* A grant from the Robert Wood Johnson Foundation supported this research. Kenneth E. Thorpe is Associate Professor of Health Policy and Administration at the University of North Carolina School of Public Health, Chapel Hill, N.C., 27599-7400. Charles E. Phelps is Chair, Department of Community and Preventive Medicine, and Professor of Political Science and Economics at the University of Rochester, Rochester, NY. We gratefully acknowledge beneficial comments from Paul Gertler and Tom McGuire and from anonymous referees for this journal.