ANTITRUST TREATMENT OF NONPROFITS: SHOULD HOSPITALS RECEIVE SPECIAL CARE?

Abstract

Nonprofit hospitals receive favorable tax treatment in exchange for providing socially beneficial activities. Extending this rationale suggests that nonprofit hospital mergers should be evaluated differently than mergers of for-profit hospitals because suppression of competition may also allow nonprofits to cross-subsidize care for the poor. Using detailed California data, we find no evidence that nonprofit hospitals are more likely than for-profit hospitals to provide more charity care or offer unprofitable services in response to an increase in market power. Therefore, we find no empirical justification for applying, as some courts have suggested, different antitrust standards for nonprofit hospitals. (JEL I11, L1, L44)

ABBREVIATIONS

CMS

Centers for Medicare and Medicaid Services

DRG

Diagnostic-Related Group

FTC

Federal Trade Commission

Hosp-HHI

Hospital-Specific HHIs

HRR

Hospital Referral Region

MDC

Major Diagnostic Category

1 INTRODUCTION

Private nonprofit hospitals account for 51% of all short-term, nonfederal, general hospitals, with the balance split about evenly between for-profit hospitals and government hospitals (AHA 2018). To maintain nonprofit status, nonprofit hospitals must provide sufficient public benefits, such as charity care. In return, they receive exemptions from income and state property taxes. In contrast, under the federal antitrust laws nonprofit hospitals typically do not receive special recognition for their provision of public benefits. That is, the antitrust agencies (and over the last decade, the courts) will block actions by nonprofit hospitals—whether mergers, joint ventures, collective price setting, or other conduct such as exclusive dealing or tie-in sales—that would create or increase market power, because market power is presumed to harm consumers by elevating price and reducing output. This presents a potential inconsistency. On one hand, tax laws presume that hospitals use their profits to provide public benefits. On the other hand, antitrust laws presume that higher hospital profits will come at the expense of consumers and output. Should the antitrust laws, or courts in antitrust cases, also provide nonprofit hospitals favorable treatment?

We first review theoretical arguments for and against favorable antitrust treatment for nonprofit hospitals in an environment in which the first-best solution of direct funding of charity care through lump sum taxes is not an option.11 Under the Affordable Care Act, not all states expanded Medicaid and, even in those did, uninsurance rates remained at 4% or higher. As of 2017, the national rate of uninsurance was between 9% and 10% (Kaiser Family Foundation 2018; Kaiser Family Foundation State Health Facts 2018).
On the against side, Philipson and Posner (2009) model nonprofits as maximizing a combination of profits and output and show that, as with a for-profit, an increase in a nonprofit hospital's market power can reduce consumer welfare. Accordingly, they argue against special antitrust consideration for nonprofits. However, their model omits solvency constraints that can limit charity care provision. We show that competition-reducing mergers of nonprofits can in theory relax financing constraints and thereby increase the provision of charity care. If this holds in practice, antitrust policy should not necessarily condemn competition-reducing mergers of nonprofit hospitals.

In other words, economic theory indicates that a balancing of social benefits against harm from market power may be appropriate if nonprofits that have greater market power will also provide greater social benefits. In this article, we go beyond theory and evaluate whether there is systematic empirical evidence that nonprofit hospitals do in fact increase their provision of uncompensated care to those without insurance—the main public benefit often claimed by nonprofits—in response to increased market power. Few papers in the empirical literature address this question and, to our knowledge, no paper studies the effect of changes in market power on a direct measure of nonprofit hospitals' volume of uncompensated care. The most closely related study is Garmon (2009), which examines the relationship between competition and the estimated dollar value of charity care provided.

We examine 11 years of detailed data from the State of California on hospital competition and the provision of charity care. A long line of literature finds that hospitals with greater market power—for-profit and nonprofit hospitals alike—charge higher prices, indicating that increases in market power should relax the financing constraint for a nonprofit hospital (Gaynor, Ho, and Town 2015). Yet, we find no significant difference in the propensities of nonprofit and for-profit hospitals to provide more charity care as they face less competition. We confirm our baseline econometric results with a second, independent analysis of market power based on how the travel times of patients change in response to changes in market power. We also examine whether nonprofit hospitals with more market power are more likely to offer important but typically unprofitable services such as psychiatric care and trauma services. We find no evidence to support this, either.

Hospitals in litigated merger cases have appealed to their nonprofit status as part of their defense, and with success in some instances. Our results, however, show no significant difference in the propensities of nonprofit and for-profit hospitals to provide more charity care as their market power increases. We also find no difference in the two types of hospitals' propensity to offer unprofitable services as market power increases. This is qualitatively consistent with the body of research finding that for profit and nonprofit hospitals are similar when it comes to the relationship between pricing and market power (Capps and Dranove 2014; Town 2012).

Overall, therefore, we find no empirical basis for courts in antitrust cases to treat nonprofit hospitals more favorably than for-profit hospitals on the grounds that they may use their market power to provide greater charity care.

We also find that, controlling for size, nonprofit hospitals provide only marginally more uncompensated care than do for-profit hospitals. Questioning the grounds for tax exemption is beyond the scope of this article, but our evidence suggests a potential role for enhanced administrative oversight of nonprofit hospitals. Indeed, multiple government entities have examined whether the public benefits of specific nonprofit hospitals justify the tax benefits they receive and some have rescinded hospitals' nonprofit status.22 For example, in 2010 the Illinois Supreme Court ruled that Provena Covenant Medical Center did not meet the requirements for tax exemption. The court noted that the hospital waived just 0.7% of its revenue, far less than the tax benefits it stood to receive. (Advisory Board 2013; Doyle 2014; Schencker 2015).
In contrast, government-run hospitals provide a disproportionately high amount of charity care.33 Government hospitals may be less efficient than private hospitals (Almond, Currie, and Simeonova 2010).
Tax exemptions for nonprofit hospitals may be a less efficient method of funding charity care than providing an equivalent amount of funding to government run hospitals.

2 BACKGROUND

The rationale for nonprofit status is that such firms provide public benefits that for-profit firms would have little incentive to provide and that governments are unwilling or unable to provide or fund directly. The provision of those benefits could depend on a suppression of competition. For example, rather than compete, food organizations might agree to serve meals in different areas of a city in order to economize on resources. The antitrust laws would generally condemn such market allocation among for-profit firms.

In several antitrust cases, courts have allowed nonprofits to defend behavior that would otherwise violate the antitrust laws by arguing that the public benefits of the challenged behavior outweigh the adverse effects of lessened competition. For example, in the 1990s, several universities were sued by the Department of Justice over an agreement to restrict competition for students through financial aid in order, per the defendants, to leave more funding for need-based financial aid.44 Carlton, Bamberger, and Epstein (1995) provide an economic analysis of this case, U.S. v. Brown University (1993). Carlton served as an expert on behalf of the Massachusetts Institute of Technology.
The district court ignored the claimed public benefit of directing more financial aid to lower income students and found that the agreement violated the antitrust laws. The appellate court overturned and ordered the district court to take account of any public benefit as an offset to any harm from the collective agreement.

Merging nonprofit hospitals have raised their status as a defense in a number of hospital merger cases (Richman 2008). Courts have varied in their receptiveness. One district court held that the Federal Trade Commission (FTC) had met its burden to show that the merger would lessen competition, yet allowed the merger to proceed under the reasoning that, given the hospitals' nonprofit status, the merged entity would not increase price.55 FTC v. Butterworth Health (1996). Courts also credited merging hospitals' nonprofit status in FTC v. Freeman (1995) and United States v. Carilion (1989).
This differs from the rationale described in this article: rather than balancing harm from market power against community benefits funded by that market power, the court concluded that the merged system, because of its nonprofit status, would not raise prices in the first place—despite the presumed ability to do so.

Thus, there are two possible rationales for more lenient antitrust treatment. The first is that nonprofit hospitals would not exercise any market power they might have. The second is that they will do so, but in ways that are socially valuable. The first rationale has been studied and, with the exception of Lynk (1995), rejected in multiple studies of samples of hospitals (Dranove and Ludwick 1999; Keeler, Melnick, and Zwanziger 1999; Simpson and Shin 1998) and in case studies of hospital mergers (Haas-Wilson and Garmon 2011; Krishnan 2001; Tenn 2011; Thompson 2011; Vita and Sacher 2001).

The economic literature has not made comparable progress in studying the second rationale.66 Tax exemptions for nonprofits are (or used to be) justified by a “bargain” that was “struck between the hospital and the community: a hospital would treat patients who were unable to pay, and the government would grant a tax exemption to the hospital” (Pellegrini 1989). At present, nonprofit hospitals are expected to provide “community benefits.” While the definition of “community benefits” is broader (Nicholson et al. 2000), charity care in the form of free or reduced price services remains at the heart of the justification for nonprofit hospital tax exemptions.
After all, public benefits must ultimately be funded, and higher prices to some patients are a potential mechanism. While competition and nonprofit hospitals' provision of charity care has not been closely studied, some research examines whether nonprofit and for-profit hospitals differ generally with regard to the provision of “community benefits” (Duggan 2000, 2002). This literature generally shows that nonprofits and for-profits respond similarly to financial incentives.77 Sloan et al. (2001) find that nonprofit and for-profit hospitals provide similar quality but that for-profit hospitals cost Medicare more, all else equal. Because Medicare prices are regulated, this suggests that for-profit hospitals render more expensive services or higher volumes of services than nonprofit hospitals. Miller and Wilson (2018) document differences in exit propensities between nonprofit and for-profit hospitals, with much of the difference due to the two types of hospitals having different underlying characteristics (scale, system membership, local income, etc.).
There is also evidence that hospitals rely on cross-subsidies to fund unprofitable services (David et al. 2014). However, studies in this literature typically have not focused on the competitive environment or whether any differential behavior between nonprofit and for-profit hospitals depends on market power.88 Stensland, Gaumer, and Miller (2010) find that hospitals that face less competition and have greater private payer revenue tend to have higher costs, which can create a misleading impression that private payer profits subsidize unprofitable care provided to government-insured patients.
One exception is Dranove, Garthwaite, and Ody (2017), which studies the responses of nonprofit hospitals to the financial crisis of 2008. They find, consistent with profit-maximization, that most hospitals did not change their pricing during the crisis. They also find that a small subset of hospitals that likely possess market power did increase their commercial prices following the adverse financial shock, evidence of a form of “dynamic cost-shifting” among the commercially insured over time—that is, sharing the gain when times are good and sharing the pain when times are bad.

Garmon (2009) is closer to our study. He uses data from Texas and Florida for 1999–2002 to analyze the effects of changes in competition on the dollar value of charity, measured as a hospital's cost-to-charge ratio times the sum of the dollar value of charity care and bad debt. He finds that competition and charity care are “if anything, positively related.” However, hospitals vary significantly in how they record charity care and bad debt and differences in accounting practices can create spurious differences in the extent of charity care provided (IRS 2009). Therefore, in addition to financial measures, we study a more direct, volume-based measure of inpatient charity care that is unlikely to be affected by differences in accounting practices. We also extend Garmon's work by examining nonprofit hospitals' offerings of important but likely unprofitable service lines. Finally, we examine 11 years of data, allowing us to capture greater variation in hospitals' market power and ownership status over time.

3 MODELS OF NONPROFIT OBJECTIVES AND CONSTRAINTS

Philipson and Posner (2009) model altruistic nonprofit producers that exhibit some degree of “output preferences,” meaning that nonprofits derive utility directly from both profits and output. They recognize the difference between a nonprofit's utility and utility in the absence of altruism, but show that, nevertheless, competition still maximizes society's surplus. Based on that model, P-P recommend that antitrust doctrine not distinguish between for-profit and nonprofit actors. We show that their result vanishes when nonprofits have a slightly more general objective function than the one postulated in their paper.

The key insight that P-P identify is that if an altruist has an output preference, then competition among altruists, just like competition among profit-maximizing firms, will generate the “correct” marginal pricing conditions, but only if the social welfare function values consumption in the same way as the altruist. For example, if α represents the additional value the altruist attaches to consumers' incremental health consumption, then p = c − α is the socially optimal pricing condition, where p is price and c is marginal cost. However, for this pricing condition to represent optimality, such marginal pricing must also result in financial viability for the firm. Even in the simplest constant returns to scale model, this cannot be true unless the altruist has funding sufficient to subsidize the consumption of the poor (a “rich altruist”), as Philipson and Posner assume. Conversely, if funds must come from the nonprofit altruist's operating profits, the financing constraint will matter. In this circumstance, market power is the key ingredient for the provision of charity.

In addition, the altruist in the P-P model values the incremental health care consumption of all consumers at α. More generally, the altruist may value the consumption of different individuals differently. Suppose that the altruist thinks that rich people can afford a minimal level of health care consumption, but poor people cannot. Hence, the altruist values additional consumption (above the minimum) of rich people at zero and that of poor people (below the minimum) at α.99 By and large, the need to access charity care is discrete (patients either have insurance or they do not). Our assumption can easily be relaxed and our results continue to apply as long as the altruist places a higher marginal value on healthcare consumption by the poor than the rich.
These two changes to the P-P model—a financing constraint and allowing the altruist to differentially value health care consumption by the rich and the poor—can completely reverse their conclusion.

Our point is not that our assumptions are necessarily superior, though the financing constraint is real, but rather that P-P's strong theoretical conclusion is critically sensitive to their assumptions. Given the sensitivity of the theoretical conclusion to assumptions, only an empirical analysis can resolve the issue of the proper role of nonprofit status in antitrust analysis.

Two implications follow from our modifications. First, if the financing constraint matters (as would be the case if the altruist is not the sole source of funds), then the creation of market power through merger, collective action such as coordinated price setting or market allocation, or conduct such as tie-in sales could benefit society because the elimination of competition relaxes the financing constraint. Second, in order for market power to be exercised so as to generate funds to subsidize uncompensated care, the nonprofit firm must be able to charge differential prices to the rich and the poor.1010 While many of the uninsured are not poor, for simplicity, we use the terms “rich” and “poor” to denote patients who pay above competitive rates and patients who benefit from subsidization.
Without this ability, the transfer from the rich to the poor could not occur. But competition makes such price discrimination difficult: with differential pricing, hospitals (even nonprofits) will want to poach their rivals' profitable customers and this erodes the ability to exercise market power. Hence, competition-reducing mergers can allow the merged firm to use its extra profits to subsidize the provision of public benefits, such as uncompensated care. We now briefly describe the model and refer the reader to the appendix for a fuller discussion.

Suppose the altruist values additional health consumption of the poor, q₂, at α, and that the cost function of a hospital that provides care volume q₁ to the rich and q₂ to the poor is given by c(q₁, q₂) = F + c · (q₁ + q₂), where F and c are the fixed and marginal costs of health care production. Assuming for simplicity that there are two hospitals owned by altruists, Bertrand competition will produce the equilibrium p₁ = c – α and p₂ = c. But this equilibrium produces losses for each firm and is not sustainable unless each altruist has a pool of wealth that it can use to subsidize health care. If we abandon the Philipson-Posner assumption that the altruist can finance losses with lump sum transfers, then profits must be non-negative, and we see that Bertrand competition with two firms is not possible. Competition prevents the firms from earning enough money to subsidize care for the poor. (With less vigorous forms of competition, such as Cournot, charity care may be provided but not in the optimal amount. The key point is that regardless of the assumed form of competition, the competition for the rich limits the amount of financing that is available for financing the health care of the poor. See appendix.)

Suppose the two firms merge (or form a cartel) and thereby eliminate competition. Since the combined firm is still altruistic, it can use its market power to raise price above c for the rich and thereby afford to lower the price below c for the poor. In fact, if the altruist seeks to maximize the consumer surplus it provides, inclusive of its extra marginal valuation of health consumption by the poor, it will choose p₁ and p₂ to maximize

(1)

The solution is a modified Ramsey pricing condition:

where λ is the Lagrange multiplier related to zero profits, μ =  − λ/(1 + λ), and ϵ_i is the elasticity of demand for group i. The social planner pays attention not only to the “net mark-up” to the poor (i.e., the amount by which p₂ exceeds c – α), but also to the markup on rich individuals (i.e., the amount by which p₁ exceeds c). This shows that the solution to the social planner's problem cannot be achieved under Bertrand competition even when the altruist's preferences match the social planner's. One could use other models of competition but the point that competition reduces charity care provision would be the same: when market power is needed to provide funds to finance care for the poor, competition can be incompatible with the social optimum.

4 EMPIRICAL ANALYSIS

As explained, some degree of market power is necessary to the provision of charity care when nonprofit hospitals face a financing constraint. But it is not sufficient: nonprofit hospitals may direct profits from insured patients towards care for the uninsured, but other possibilities—opportunistic behavior by nonprofit administrators, dissipation of rents through possibly inefficient nonprice competition, and various forms of regulatory evasion—are also plausible. Accordingly, how nonprofit hospitals with market power use their profits is an empirical question.

4.1 Measuring Charity and Uncompensated Care

Both for-profit and nonprofit hospitals provide substantial amounts of uncompensated care (CBO 2006). At 56% of reported community benefits, uncompensated care is the most prominent category; other categories include medical education and training, research, and community programs (IRS 2009).1111 Young et al. (2013) report that about 85% of community benefit expenditures are related to direct patient care (25% for charity care, 15% for subsidized health services, and 45% for unreimbursed costs under non-Medicare government programs).
These dollar-denominated measures, however, are subject to manipulation; for example, the Internal Revenue Service (IRS) study finds a great deal of variation in how hospitals measure and report uncompensated care.

In California, for-profit hospitals accounted for more than 20% of all uncompensated care. Like nonprofits, they are legally required to treat patients who require immediate medical attention, regardless of ability to pay, and also may treat patients who subsequently turn out to lack insurance or do not pay the out-of-pocket portion of their medical bill. When the hospital approves free or discounted care in advance, that will likely be recorded as “charity care.” When a hospital learns after the fact that it provided partially or entirely uncompensated care, that will likely be recorded as “bad debt.” In practice, the majority of uncompensated care in California is reported as bad debt rather than as charity care, and hospitals vary in how they report charity care and bad debt. As a result, uncompensated care, defined as the sum of charity care and bad debt, is likely the more reliable dollar-denominated measure (CBO 2006; David and Helmchen 2006; Garmon 2009). We study both measures, but view the sum (charity care + bad debt) as more reliable.

The value of uncompensated care reported in hospitals' financial statements may overstate both its market value and its cost. Some hospitals compute charity care and bad debt using list prices, which are rarely actually paid and somewhat arbitrary (IRS 2009; Reinhardt 2006).1212 Private insurers commonly negotiate discounts of 40%–60% off of list prices; Medicare payments are typically lower and Medicaid payments are usually lower still. The uninsured may be billed for undiscounted prices, but they rarely pay.

When inflation of and discounting from list charges are not constant across hospitals, and when the underlying list charges vary widely and unsystematically, financial measures of charity care provision are likely to be difficult to compare reliably across hospitals. For example, hospitals that provide a given volume of free care but have higher list charges could appear to provide more uncompensated care than hospitals with lower list charges.1313 Garmon (2009) uses each hospital's cost-to-charge ratio to adjust for this.

To focus on actual services rendered to the poor, we construct a third measure of charity care: the volume of inpatient services provided to uninsured patients. Each year, the Centers for Medicare and Medicaid Services (CMS) computes and publishes diagnostic-related group (DRG) “weights” that reflect the relative cost of treating patients in a particular DRG.1414 CMS (2019) explains the construction and usage of DRG weights as follows: “[T]he Secretary shall establish a classification system (referred to as DRGs) for inpatient discharges and adjust payments under the [Inpatient Prospective Payment System, or IPPS] based on appropriate weighting factors assigned to each DRG. Therefore, under the IPPS, we pay for inpatient hospital services on a rate per discharge basis that varies according to the DRG to which a beneficiary's stay is assigned. The formula used to calculate payment for a specific case multiplies an individual hospital's payment rate per case by the weight of the DRG to which the case is assigned. Each DRG weight represents the average resources required to care for cases in that particular DRG, relative to the average resources used to treat cases in all DRGs.”
For example, a patient in a DRG with a weight of 4 is four times as costly on average to treat as a patient in a DRG with a weight of 1. The measure, charity volume, is the sum of the DRG weights for inpatient care provided to all patients without insurance.1515 From 2001 to 2008, the set of inpatient services hospitals offer were divided into roughly 550 DRGs. In 2008, CMS changed its DRG system to better account for severity (introducing Medicare Severity-DRGs or MS-DRGs). This resulted in a complete restructuring of the DRG taxonomy, including an expansion to more than 700 MS-DRGs. This change has little effect on our analysis because we focus on the case weights associated with each DRG, and the meaning of those weights did not change (i.e., both before and after the change, a case weight of 2.0 represents a patient with a condition that costs, on average, twice as much to treat as the national average patient).

In summary, we examine three charity care measures: (1) “charity care”—the reported dollars of charity care; (2) “uncompensated care”—charity care plus bad debt; and (3) “charity volume”—total inpatient services volume provided to patients without insurance. The first two span all services offered by hospitals, including outpatient and inpatient, but are subject to the limitations described above. The third captures only inpatient care but is more accurately measured and less subject to manipulation.

Figure 1 presents statewide yearly trends for the charity care measures. From 2001 to 2011, the financial measures, charity care and uncompensated care, more than tripled yet charity volume grew by just half. This suggests that the growth in charity care and uncompensated care is driven by increases in both charges and patient volume, but more so the former.1616 Although nonprofits cannot distribute profits, if profits from excess pricing are dissipated through higher compensation and staff perquisites, a hospital's costs will increase and so will its cost-to-charge ratio. This could misleadingly increase the dollar measures.

Figure 1

Charity Care, Bad Debt, and Charity Volume: 2001–2011.

Note: left scale applies to charity care and bad debt; right scale applies to charity volume

4.2 Measuring Competition

We use a method similar to Kessler and McClellan (2000) to measure competition without defining geographic markets. First, we calculate the standard HHI based on observed shares of commercially insured patients within each unique zip code and Major Diagnostic Category (MDC) combination (“micromarkets”), taking joint ownership into account.1717 We exclude MDCs 19 (psychiatric care) and 20 (alcohol and drug admissions), which standalone psychiatric hospitals and treatment centers also provide. Additionally, these services are used disproportionately by the uninsured population and are considered unprofitable; therefore, they are not likely to contribute to market power. We do include patients with these MDCs in our measure of charity volume. To avoid double-counting labor and delivery admissions, we also exclude normal newborns.
Second, we compute hospital-specific HHIs (Hosp-HHI) as the weighted sum of micromarket HHIs, where the weights are the shares of each hospital's patients that originate from each zip code-MDC combination.1818 Our data for the three measures end in 2011, the last year in which California included five-digit zip codes in its discharge data. We present results that use hospital-specific HHIs calculated without Kaiser hospitals, which compete only indirectly with the hospitals we study (Ho and Lee (2017)). Our main results are robust to the inclusion of Kaiser.

In general, hospitals that draw patients from more concentrated zip codes and service lines will have higher hospital-specific HHIs. The higher a hospital's HHI, the weaker is the competitive pressure that it faces. A number of studies have demonstrated that this modified HHI is a good predictor of hospital prices, supporting its use in our study (Capps and Dranove 2004; Dranove and Ludwick 1999; Gruber 1994; Keeler, Melnick, and Zwanziger 1999).

This measure addresses the problem of prespecifying a geographic market within which to measure competition, but may still raise endogeneity concerns. Kessler and McClellan address this by substituting for the observed shares within each microsegment the predicted shares from a choice model that uses only exogenous factors as predictors. This is less practical in the current setting because, while Kessler and McClellan compute their concentration measures for heart attack admissions only, we study all acute care inpatient admissions.1919 Compared to Kessler and McClellan, our data encompass roughly 20 times as many patients per year and span 11 years rather than four.

We view the endogeneity concern as minimal. Most insured hospital patients face very modest or no variation in prices across in-network hospitals. As a result, market shares and HHIs will be affected by prices only to the extent that prices determine whether hospitals are included in or excluded from insurers' networks. Most hospitals have at least some excess capacity and would find it profitable at the margin to contract with more insurers and insurers also have a strategic incentive to contract broadly with hospitals (Capps, Dranove, and Satterthwaite 2003). Therefore, most managed care networks have historically included most hospitals.2020 Narrower networks have grown somewhat in recent years, but that was nascent as of 2011, the final year of our data.
Prices serve primarily to divide the gains from trade between hospitals and insurers (and insurers' customers). Because the direct effect of price on patients' choices among in-network hospitals is minimal, we do not think endogeneity poses a problem.

4.3 Data

We use an 11-year panel of data on California hospitals that combines information from hospitals' annual financial reports with inpatient discharge data. Both are from the California Office of Statewide Health Planning and Development. The financial reports provide data on hospitals' revenue, profit, and the dollar-denominated measures of charity care. The discharge data allow us to construct the Hosp-HHI and our output-based measure of charity volume.

Table 1 presents summary statistics describing the time path of the number of hospitals, beds, utilization, financial information, and charity care provision from 2001 to 2011, separately for nonprofit, for-profit, and government hospitals. Over the sample period, the number of nonprofit hospitals declined by 7.5% and the number of for-profit hospitals declined by 14.6%. The bulk of this decline occurred after 2003 and likely reflects requirements that hospitals complete seismic retrofitting by 2006 (Tom and Jacobson 2008). Average net income among for-profit hospitals began a marked decline in 2004 whereas average net income among nonprofit hospitals increased fairly steadily (save for 2008 and 2009, the years of the financial crisis). With the postrecession recovery, net income for hospitals of all types had improved by 2011.2121 California expanded its Medicaid program in 2010, which may have also improved hospitals' net income (Harbage and King 2012).

Table 1. Summary Data for California Hospitals by Ownership Type and Year, 2001–2011

	Variable	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010	2011
Nonprofit	N	188	181	185	179	174	174	173	172	169	171	168
	Staffed beds	215	219	217	220	224	218	221	219	216	217	211
	Discharges	10,245	10,316	10,639	11,022	11,435	11,390	11,500	11,654	11,729	11,790	11,782
	Gross IP Rev. ($1000s)	$264,565	$312,905	$372,221	$434,502	$498,698	$546,589	$599,348	$657,398	$724,332	$773,276	$825,586
	Net income ($1000s)	$5,486	$7,493	$7,696	$8,662	$11,374	$16,016	$19,765	$11,023	$12,855	$21,398	$26,990
	Charity ($1000s)	$4,124	$4,520	$5,650	$7,365	$9,335	$11,713	$13,617	$15,875	$18,879	$21,849	$24,219
	Charity + Bad Debt ($1000s)	$10,662	$12,239	$14,548	$18,232	$20,963	$24,468	$26,899	$28,940	$33,727	$37,834	$41,601
	Charity: Volume Measure	431	462	517	575	626	647	681	699	725	760	809
	Hospital-HHI: Full sample	4,173	4,258	4,239	4,212	4,232	4,256	4,260	4,269	4,318	4,281	4,262
	Hospital-HHI: Private	4,278	4,298	4,260	4,203	4,187	4,220	4,222	4,225	4,279	4,260	4,252
	Hospital-HHI: Medicare	4,936	4,994	5,048	5,035	5,055	5,055	5,053	5,046	5,095	5,016	4,988
For-profit	N	89	88	87	83	76	78	77	74	78	75	76
	Staffed beds	138	134	144	141	134	135	133	128	127	126	123
	Discharges	6,211	6,339	6,718	6,485	6,720	7,012	6,985	6,950	7,121	6,981	6,961
	Gross IP Rev. ($1000s)	$215,432	$255,041	$323,748	$314,930	$333,097	$351,082	$366,044	$376,114	$398,558	$409,808	$430,973
	Net income ($1000s)	$7,271	$7,339	$8,947	$(923)	$1,851	$935	$450	$3,082	$6,284	$8,105	$8,694
	Charity ($1000s)	$5,145	$4,622	$6,460	$6,134	$6,069	$6,459	$8,042	$7,662	$9,319	$10,003	$9,270
	Charity + Bad Debt ($1000s)	$9,713	$9,151	$12,608	$14,763	$12,516	$13,074	$17,223	$18,311	$22,020	$21,228	$21,790
	Charity: Volume Measure	229	227	257	245	242	250	268	316	337	377	415
	Hospital-HHI: Full sample	3,123	3,102	3,151	3,101	3,031	2,981	2,993	2,898	2,935	2,934	2,939
	Hospital-HHI: Private	3,129	3,095	3,127	3,168	3,039	2,969	3,028	2,954	2,937	2,949	2,945
	Hospital-HHI: Medicare	3,694	3,626	3,700	3,644	3,561	3,538	3,561	3,465	3,528	3,521	3,565
Government	N	62	64	65	63	64	63	64	62	62	60	61
	Staffed beds	151	148	149	155	156	154	153	155	154	149	144
	Discharges	6,941	6,754	6,780	7,042	7,160	7,356	7,275	7,273	7,001	7,254	6,932
	Gross IP Rev. ($1000s)	$141,456	$158,195	$176,217	$197,188	$215,932	$235,677	$246,953	$265,940	$282,339	$309,668	$323,240
	Net income ($1000s)	$21,163	$19,871	$20,977	$22,035	$26,554	$3,991	$4,629	$2,236	$1,712	$752	$15,404
	Charity ($1000s)	$3,711	$3,953	$3,963	$4,378	$5,739	$7,546	$10,607	$12,653	$13,466	$17,463	$17,074
	Charity + Bad Debt ($1000s)	$9,077	$8,796	$9,245	$10,499	$16,395	$19,261	$20,951	$24,066	$26,184	$32,385	$33,957
	Charity: Volume Measure	1,135	1,069	1,025	1,075	1,088	1,100	1,052	1,060	1,026	1,096	1,106
	Hospital-HHI: Full sample	4,289	4,204	4,200	4,179	4,121	4,133	4,071	4,095	4,259	4,128	4,237
	Hospital-HHI: Private	4,407	4,278	4,281	4,273	4,296	4,209	4,155	4,242	4,368	4,426	4,316
	Hospital-HHI: Medicare	5,032	4,881	4,888	4,964	4,852	4,848	4,782	4,821	5,012	4,869	4,903
Medical Care CPI (2001 = 100)		100.0	104.7	108.9	113.7	118.5	123.2	128.7	133.5	137.7	142.4	146.7

Average discharges at surviving hospitals increased over time, as expected given population growth and closures. The average number of beds was largely flat, so growth was primarily the result of higher utilization of existing beds rather than adding new beds.2222 The average exiting hospital had 102 beds while the average surviving hospital had 194 beds.

The middle three rows in each panel contain annual averages of the three charity care measures: charity care, uncompensated care (charity care plus bad debt), and charity volume.2323 Charity volume is the sum of DRG case weights for patients with “County Indigent Programs,” “Other Indigent,” or “Self Pay” as the expected payer.
At nonprofit hospitals, all three measures grew rapidly over the sample period but charity volume grew at a much slower rate than either of the dollar-based measures. As discussed above, this differential likely reflects some combination of increases in list charges; accounting practices that incorporate expenditures not directly related to patient care, such as medical research and teaching, into the reported charity care measures; and provision of charity care outside the inpatient setting. For-profit hospitals likewise show increases in all three measures, with growth in the dollar-denominated measures also outstripping growth in charity volume. Government hospitals also show growing levels of charity care and bad debt, but relatively flat charity volume.

On a per bed basis, spending increases on charity care and uncompensated care are similar for all three ownership types. The pattern is different for the volume-based measure. Government hospitals did not increase charity volume, either on an overall or per bed basis. But they consistently provided a much higher level of charity volume per bed than for-profit and nonprofit hospitals. The average nonprofit hospital provided more charity volume than the average for-profit hospital, but the bulk of this is attributable to their larger size.

Overall, nonprofit hospitals do not provide disproportionately high uncompensated care or charity volume. For example, in 2011, they accounted for about 45% of beds, discharges, and uncompensated care; however, they provided only 35% of charity volume. That same year, for-profit hospitals represented about 26% of beds and discharges and 22% of uncompensated care, but just 18% of charity volume.2424 Comparing just nonprofit and for-profit hospitals in 2011, nonprofits accounted for 63% of beds and discharges and about 66% of uncompensated care and charity volume (versus 37% and 34% on the part of for-profit hospitals).
Government hospitals provide disproportionately high amounts of charity volume: they account for about 30% of beds and discharges, but provide about 47% of charity volume.2525 CBO (2006) reports that uncompensated care accounts for 13% of government hospitals' operating expenses, as compared with 4.7% and 4.2% at nonprofit and for-profit hospitals, respectively.
As suggested by the theoretical section above, the failure of nonprofits to provide high charity volume relative to their scale could reflect financial constraints due to competitive pressures. However, Table 1 also shows that nonprofit hospitals on average face less competition than for-profit hospitals. Moreover, while for-profit hospitals faced slightly greater competition over time, the degree of competition faced by nonprofit hospitals remained roughly unchanged.

4.4 Results

4.4.1 Charity Care Provision and Competition

The baseline regressions model charity care measure m as a function of the competition faced by hospital j and control variables W_j,t, such as patient mix (e.g., uninsured patients living near a given hospital) or local demographics (e.g., income, urban/rural area):

The first term, α_Type(j), is a constant that is specific to each ownership type and α_j is a set of hospital fixed effects. The coefficient on the HHI measure, β_Type(j), captures the extent to which hospitals that face less competition provide more (or less) charity care. We allow this coefficient to vary according to ownership type2626 1[.] equals 1 if the argument is true.
:

Our key test of whether nonprofit hospitals provide more charity care as they face less competition—and, thus, whether the antitrust laws should provide special consideration to nonprofit hospitals—is whether β_FP < 0.2727 For nonprofit hospitals, the coefficient on the market power measure is β_NFP and for for-profit hospitals it is β_NFP + β_FP. Therefore, if an increase in the Hospital HHI (i.e., a decrease in competition) results in a greater increase in charity provision among nonprofit hospitals than among for-profit hospitals, β_NFP must be greater than β_NFP + β_FP (i.e., β_FP < 0).
In addition, our specification allows us to test whether, all else equal, nonprofits provide more charity care than for-profits for any given level of competition. This is a test for whether the tax exemptions afforded to nonprofit hospitals are associated with greater uncompensated care. This amounts to a test of whether α_FP < 0.

Our baseline models are a set of cross-sectional and fixed effect models that are robust to correlations between unobserved time-invariant hospital-specific factors and the error term. We cluster standard errors at the Hospital Referral Region (HRR) level and use robust standard errors.2828 The Dartmouth Atlas partitions the United States into 306 HRRs defined around tertiary hospitals that act as major referral centers.
Cross-sectional results are presented in the upper panel of Table 2. If we omit ownership interactions, the coefficient estimates on hospital-HHI are positive and statistically significant for all charity measures, indicating that each rises with concentration. Adding ownership interactions reveals our key finding that there is no statistically significant positive difference between nonprofit and for-profit hospitals in the relationship between competition and charity care, uncompensated care, or charity volume (top panel of Table 2, row 2). If anything, for-profit hospitals on average provide more uncompensated care than nonprofit hospitals as they face less competition (but only for the dollar-denominated uncompensated care measure). Government hospitals provide less charity volume as they face greater competition, but higher levels of charity and uncompensated care. Especially since charity volume is our preferred measure of charity care, the lower provision of charity volume by government hospitals when concentration increases is a surprising finding that we leave for future research.

Table 2. Cross-Sectional and Fixed Effects Results, Log–Log Specifications

	[1]	[2]	[3]	[4]	[5]	[6]	[7]	[8]	[9]
	No ownership interactionsa^a Also includes year dummies, For-profit and Government hospital dummies, Ln(Total discharges).			Ownership interactionsa^a Also includes year dummies, For-profit and Government hospital dummies, Ln(Total discharges).			Payer mix controlsb^b Also includes year dummies, For-profit and Government dummies, Ln(Total discharges), HSA %Privately insured, HSA %Self-pay.
	Charity Care	Uncomp. Care	Charity Volume	Charity Care	Uncomp. Care	Charity Volume	Charity Care	Uncomp. Care	Charity Volume
Cross-sectional
Ln(Hosp-HHI)	0.872^***	0.264^***	0.316^***	0.456^***	0.076	0.480^***	0.473^***	0.158^***	0.635^***
	(0.116)	(0.0497)	(0.0394)	(0.114)	(0.052)	(0.048)	(0.121)	(0.056)	(0.049)
Ln(Hosp-HHI) x For-profit				−0.210	0.324^***	0.013	−0.214	0.280^**	−0.044
				(0.291)	(0.124)	(0.101)	(0.294)	(0.126)	(0.0977)
Ln(Hosp-HHI) x Government				3.176^***	0.693^***	−1.130^***	3.153^***	0.599^***	−1.328^***
				(0.473)	(0.201)	(0.105)	(0.478)	(0.207)	(0.102)
Observations	3,505	3,502	3,505	3,505	3,502	3,505	3,505	3,502	3,505
R-squared	0.429	0.604	0.708	0.446	0.607	0.716	0.446	0.609	0.728
Hospital fixed effects
Ln(Hosp-HHI)	0.568	0.216	0.44	0.469	0.209	0.420	0.506	0.211	0.304
	(0.607)	(0.235)	(0.298)	(0.617)	(0.232)	(0.302)	(0.638)	(0.236)	(0.230)
Ln(Hosp-HHI) x For-profit				0.065^*	−0.013	0.021	0.063^*	−0.013	0.026
				(0.038)	(0.008)	(0.017)	(0.037)	(0.008)	(0.019)
Ln(Hosp-HHI) x Government				0.098	0.025	0.011	0.099	0.025	0.008
				(0.121)	(0.034)	(0.019)	(0.121)	(0.034)	(0.016)
Observations	3,505	3,502	3,505	3,505	3,502	3,505	3,505	3,502	3,505
R-squared	0.793	0.84	0.928	0.793	0.84	0.928	0.793	0.84	0.934

• Note: Heteroskedasticity-robust standard errors are reported in parentheses.
• ^a Also includes year dummies, For-profit and Government hospital dummies, Ln(Total discharges).
• ^b Also includes year dummies, For-profit and Government dummies, Ln(Total discharges), HSA %Privately insured, HSA %Self-pay.
• *Significance at 10%; **significance at 5%; ***significance at 1%.

When we include hospital fixed-effects (lower panel), the statistical significance of the baseline relationship between the measures of charity care and concentration disappears. That is, charity volume is generally not statistically higher as hospitals have more market power, though the point estimates are positive. Importantly, just as before, there is no evidence that the effect of facing less competition on charity care provision is greater for nonprofit hospitals. As shown in the bottom panel, none of the coefficients on interactions between for-profit ownership and the hospital HHI is negative and significant (the interaction for nonprofit hospitals is the omitted category). Thus, the results in Table 2 do not support the proposition that nonprofit hospitals that face less competition provide more charity care.2929 We also conducted cross-sectional regressions that include a broad set of hospital characteristics (e.g., teaching status, discharges, and rural hospital indicator) and area characteristics (e.g., income, population, poverty, and insurance coverage). Results, available on request, are similar. As highlighted in the theoretical model, cross-subsidization of charity care is premised upon price increases to the privately insured. As a basic check of our market power measure, we replaced our charity dependent variables with price measures. We find that our concentration measure (Hosp-HHI) is positively related to prices. For details on the construction of the price indexes and regression results, see the online appendix.

4.4.2 Provision of Unprofitable Services

Nonprofit hospitals could use their profits to provide important but unprofitable services, such as psychiatric care, rehabilitation, emergency department (ED), trauma services, burn care, and labor and delivery (Dranove, Garthwaite, and Ody 2017; Horwitz 2005; Horwitz and Nichols 2009; Lindrooth et al. 2010; McClellan 1997). And, indeed, compared with for-profit hospitals, nonprofit and government hospitals are more likely to provide these services.3030 For example, 85% of nonprofit hospitals and 89% of government hospitals have an ED, but the corresponding percentage for for-profit hospitals is 74%. Likewise, 17% of nonprofit and 14% of government hospitals, but only 5% of for-profit hospitals, have a trauma center.
However, as the pattern of results from probit models relating the probability of a hospital offering each of these services to concentration and ownership type indicates, nonprofit hospitals are no more likely to offer these services as concentration increases than are for-profit hospitals (Table 3). These services are generally more likely to be provided by hospitals in more concentrated markets, but this is not specific to nonprofit hospitals. For trauma care and burn care, the effect of concentration on the probability of offering these services is stronger among for-profit hospitals. Thus, this analysis also provides no statistical basis for special antitrust treatment for nonprofit hospitals.

Table 3. Probit Regressions for Unprofitable Service Offerings (HRR Fixed Effects, Robust Standard Errors)

	[1]	[2]	[3]	[4]	[5]
Variable	ED	Trauma	Psychiatric	OB	Burn ICU
Ln (Hosp-HHI)	0.496	0.213	0.877^**	1.677^***	−0.293
	(0.848)	(0.637)	(0.362)	(0.455)	(0.665)
Ln (Hosp-HHI) × For-profit	0.395	1.441	−0.0361	0.021	0.891
	(0.640)	(1.000)	(0.565)	(0.582)	(0.877)
Ln (Hosp-HHI) × Government	0.216	0.384	−0.362	0.562	−0.225
	(1.234)	(0.858)	(0.536)	(0.663)	(0.622)
For-profit	−3.845	−12.210	0.093	−0.608	−7.270
	(5.022)	(8.133)	(4.524)	(4.569)	(6.900)
Government	−1.322	−2.866	3.488	−4.580	2.285
	(10.48)	(7.142)	(4.388)	(5.472)	(4.999)
Teaching hospital	1.205^***	2.026^***	1.493^***	0.373	0.941^*
	(0.461)	(0.428)	(0.282)	(0.253)	(0.515)
Rural hospital	−0.249		−2.057^***	−0.857^**
	(0.399)		(0.375)	(0.393)
Case mix index	0.277	0.592^**	−2.196^***	−0.297	0.491^**
	(0.314)	(0.258)	(0.351)	(0.216)	(0.220)
Ln (HSA population)	−0.318^**	0.421^***	0.214^**	0.126	0.082
	(0.126)	(0.138)	(0.096)	(0.101)	(0.137)
HSA: Median income	0.480^**	−0.264	−0.133	0.127	−0.898
	(0.242)	(0.436)	(0.226)	(0.354)	(0.925)
HSA: % Uninsured	−0.829	11.72	−14.16^**	−2.662	−2.829
	(8.079)	(11.19)	(5.647)	(7.090)	(11.61)
HSA: % Privately insured	−3.691^***	3.410^**	1.294	−0.462	3.160
	(1.350)	(1.373)	(0.979)	(1.289)	(1.946)
Constant	−14.60^*	−14.18^*	−13.04^***	−26.23^***	2.76
	(8.107)	(7.433)	(3.617)	(6.464)	(12.75)
Observations	3,450	2,498	3,118	3,450	2,004

• Notes: Specifications include HRR fixed effects and year fixed effects. All fixed effect coefficients are not reported. Heteroskedasticity-robust standard errors are reported in parentheses below the estimated coefficients.
• *Significance at 10%; **significance at 5%; ***significance at 1%.

4.4.3 Sensitivities

We perform several sensitivities.3131 Results available on request.
To address potential endogeneity of our competition measure, we construct a version of the Hosp-HHI using only patients covered by Traditional Medicare. Medicare patients have essentially unfettered choice of hospitals and pay administratively set prices. Therefore, hospitals' shares among Medicare patients are less likely to be affected by hospital market power or pricing. Results are in line with our baseline results.

We also estimate the model using only hospitals in the bottom and top 25% of the distribution of changes in the Hosp-HHI from the beginning to the end of the sample. Consistent with our baseline finding, there is no statistical evidence that nonprofit hospitals provide more charity care as they face less competition.

As a final, distinct check, we examine travel patterns of insured and uninsured patients admitted to hospitals that were in the top 25% of changes in Hosp-HHI. For these hospitals, average travel time among privately insured patients increased by roughly 16% from 2001 to 2007 (from about 20.5 minutes to 24 minutes). This shows that hospitals that faced less competition over time drew insured patients from a broader area. However, there was no corresponding increase in the average travel time of uninsured patients, which remained at 23.5 minutes. That less competition does not lead to a hospital drawing uninsured patients from a broader area suggests that hospitals tend to accept a relatively fixed volume of uninsured patients, irrespective of ownership type or the degree of competition.

5 CONCLUSIONS

Our theoretical model suggests that the welfare implications of less competition among nonprofit hospitals will depend on the link between market power and the provision of uncompensated care. In theory, cross-subsidization facilitated by market power could increase welfare. If borne out empirically, that would suggest an inconsistency between the tax laws, which offer nonprofits favorable treatment in exchange for community benefits, and the antitrust laws, which generally do not.

Our analysis of competition, ownership status, and charity care provision offers no statistical support for special antitrust treatment for nonprofit hospitals. We do not find empirical evidence that nonprofit hospitals are more likely than for-profits to provide more charity care in response to facing less competition. We also examine unprofitable service offerings and find no evidence that nonprofit hospitals that face less competition are more likely to offer these services. Our results therefore provide no empirical justification for applying a different antitrust standard to nonprofit hospitals than to for-profit hospitals.

Regarding the nonprofit hospital tax exemption itself, our results also allow us to test whether, controlling for the level of competition, nonprofit hospitals provide more charity care than for-profit hospitals. We find mixed empirical evidence that nonprofit hospitals provide greater charity volume than for-profits, and strong evidence that government hospitals provide disproportionately high volumes of charity care. Examining the efficacy of the nonprofit tax exemption at promoting charity care is beyond the scope of this article, but our evidence suggests potential roles for enhanced administrative oversight of nonprofit hospitals and greater funding for government hospitals.

1 Under the Affordable Care Act, not all states expanded Medicaid and, even in those did, uninsurance rates remained at 4% or higher. As of 2017, the national rate of uninsurance was between 9% and 10% (Kaiser Family Foundation 2018; Kaiser Family Foundation State Health Facts 2018).

2 For example, in 2010 the Illinois Supreme Court ruled that Provena Covenant Medical Center did not meet the requirements for tax exemption. The court noted that the hospital waived just 0.7% of its revenue, far less than the tax benefits it stood to receive. (Advisory Board 2013; Doyle 2014; Schencker 2015).

3 Government hospitals may be less efficient than private hospitals (Almond, Currie, and Simeonova 2010).

4 Carlton, Bamberger, and Epstein (1995) provide an economic analysis of this case, U.S. v. Brown University (1993). Carlton served as an expert on behalf of the Massachusetts Institute of Technology.

5 FTC v. Butterworth Health (1996). Courts also credited merging hospitals' nonprofit status in FTC v. Freeman (1995) and United States v. Carilion (1989).

6 Tax exemptions for nonprofits are (or used to be) justified by a “bargain” that was “struck between the hospital and the community: a hospital would treat patients who were unable to pay, and the government would grant a tax exemption to the hospital” (Pellegrini 1989). At present, nonprofit hospitals are expected to provide “community benefits.” While the definition of “community benefits” is broader (Nicholson et al. 2000), charity care in the form of free or reduced price services remains at the heart of the justification for nonprofit hospital tax exemptions.

7 Sloan et al. (2001) find that nonprofit and for-profit hospitals provide similar quality but that for-profit hospitals cost Medicare more, all else equal. Because Medicare prices are regulated, this suggests that for-profit hospitals render more expensive services or higher volumes of services than nonprofit hospitals. Miller and Wilson (2018) document differences in exit propensities between nonprofit and for-profit hospitals, with much of the difference due to the two types of hospitals having different underlying characteristics (scale, system membership, local income, etc.).

8 Stensland, Gaumer, and Miller (2010) find that hospitals that face less competition and have greater private payer revenue tend to have higher costs, which can create a misleading impression that private payer profits subsidize unprofitable care provided to government-insured patients.

9 By and large, the need to access charity care is discrete (patients either have insurance or they do not). Our assumption can easily be relaxed and our results continue to apply as long as the altruist places a higher marginal value on healthcare consumption by the poor than the rich.

10 While many of the uninsured are not poor, for simplicity, we use the terms “rich” and “poor” to denote patients who pay above competitive rates and patients who benefit from subsidization.

11 Young et al. (2013) report that about 85% of community benefit expenditures are related to direct patient care (25% for charity care, 15% for subsidized health services, and 45% for unreimbursed costs under non-Medicare government programs).

12 Private insurers commonly negotiate discounts of 40%–60% off of list prices; Medicare payments are typically lower and Medicaid payments are usually lower still. The uninsured may be billed for undiscounted prices, but they rarely pay.

13 Garmon (2009) uses each hospital's cost-to-charge ratio to adjust for this.

14 CMS (2019) explains the construction and usage of DRG weights as follows: “[T]he Secretary shall establish a classification system (referred to as DRGs) for inpatient discharges and adjust payments under the [Inpatient Prospective Payment System, or IPPS] based on appropriate weighting factors assigned to each DRG. Therefore, under the IPPS, we pay for inpatient hospital services on a rate per discharge basis that varies according to the DRG to which a beneficiary's stay is assigned. The formula used to calculate payment for a specific case multiplies an individual hospital's payment rate per case by the weight of the DRG to which the case is assigned. Each DRG weight represents the average resources required to care for cases in that particular DRG, relative to the average resources used to treat cases in all DRGs.”

15 From 2001 to 2008, the set of inpatient services hospitals offer were divided into roughly 550 DRGs. In 2008, CMS changed its DRG system to better account for severity (introducing Medicare Severity-DRGs or MS-DRGs). This resulted in a complete restructuring of the DRG taxonomy, including an expansion to more than 700 MS-DRGs. This change has little effect on our analysis because we focus on the case weights associated with each DRG, and the meaning of those weights did not change (i.e., both before and after the change, a case weight of 2.0 represents a patient with a condition that costs, on average, twice as much to treat as the national average patient).

16 Although nonprofits cannot distribute profits, if profits from excess pricing are dissipated through higher compensation and staff perquisites, a hospital's costs will increase and so will its cost-to-charge ratio. This could misleadingly increase the dollar measures.

17 We exclude MDCs 19 (psychiatric care) and 20 (alcohol and drug admissions), which standalone psychiatric hospitals and treatment centers also provide. Additionally, these services are used disproportionately by the uninsured population and are considered unprofitable; therefore, they are not likely to contribute to market power. We do include patients with these MDCs in our measure of charity volume. To avoid double-counting labor and delivery admissions, we also exclude normal newborns.

18 Our data for the three measures end in 2011, the last year in which California included five-digit zip codes in its discharge data. We present results that use hospital-specific HHIs calculated without Kaiser hospitals, which compete only indirectly with the hospitals we study (Ho and Lee (2017)). Our main results are robust to the inclusion of Kaiser.

19 Compared to Kessler and McClellan, our data encompass roughly 20 times as many patients per year and span 11 years rather than four.

20 Narrower networks have grown somewhat in recent years, but that was nascent as of 2011, the final year of our data.

21 California expanded its Medicaid program in 2010, which may have also improved hospitals' net income (Harbage and King 2012).

22 The average exiting hospital had 102 beds while the average surviving hospital had 194 beds.

23 Charity volume is the sum of DRG case weights for patients with “County Indigent Programs,” “Other Indigent,” or “Self Pay” as the expected payer.

24 Comparing just nonprofit and for-profit hospitals in 2011, nonprofits accounted for 63% of beds and discharges and about 66% of uncompensated care and charity volume (versus 37% and 34% on the part of for-profit hospitals).

25 CBO (2006) reports that uncompensated care accounts for 13% of government hospitals' operating expenses, as compared with 4.7% and 4.2% at nonprofit and for-profit hospitals, respectively.

26 1[.] equals 1 if the argument is true.

27 For nonprofit hospitals, the coefficient on the market power measure is β_NFP and for for-profit hospitals it is β_NFP + β_FP. Therefore, if an increase in the Hospital HHI (i.e., a decrease in competition) results in a greater increase in charity provision among nonprofit hospitals than among for-profit hospitals, β_NFP must be greater than β_NFP + β_FP (i.e., β_FP < 0).

28 The Dartmouth Atlas partitions the United States into 306 HRRs defined around tertiary hospitals that act as major referral centers.

29 We also conducted cross-sectional regressions that include a broad set of hospital characteristics (e.g., teaching status, discharges, and rural hospital indicator) and area characteristics (e.g., income, population, poverty, and insurance coverage). Results, available on request, are similar. As highlighted in the theoretical model, cross-subsidization of charity care is premised upon price increases to the privately insured. As a basic check of our market power measure, we replaced our charity dependent variables with price measures. We find that our concentration measure (Hosp-HHI) is positively related to prices. For details on the construction of the price indexes and regression results, see the online appendix.

30 For example, 85% of nonprofit hospitals and 89% of government hospitals have an ED, but the corresponding percentage for for-profit hospitals is 74%. Likewise, 17% of nonprofit and 14% of government hospitals, but only 5% of for-profit hospitals, have a trauma center.

31 Results available on request.

32 This fits the original IRS definition of charity-care requirements, which state that a nonprofit hospital “must be operated to the extent of its financial ability for those not able to pay for the services rendered” (Seaton and Koob 2009).

33 If nonprofits have no desire to subsidize care for the poor, allowing them the ability to set prices above marginal cost will not aid in achieving this desired social goal.

34 In addition, when c(q) is not homogenous of degree 1 in q, marginal cost pricing may not cover cost.

35 Note that λ > 0 leads to μ < 1, which in turn, means that the nonprofit altruist will charge the rich a price P₁ that exceeds the price set by the social planner.

36 Proof:

APPENDIX A

ANALYSIS OF ALTRUISTIC BEHAVIOR UNDER A FINANCING CONSTRAINT

Consider first the case of for-profit firms when there is no special value attached to the consumption of health care by the poor. Hence if α represents the additional value that society places on each unit of consumption by the poor, then α = 0. In this case there is no reason to have nonprofit firms or grant tax exemptions. Let be a hospital's cost function for providing q₁ units to the rich and q₂ units to the poor. Assume for simplicity that . Suppose ; in this case competition among firms will maximize social welfare, as usual. For , if the firms play Bertrand then, as Sutton has shown, there is no stable equilibrium if there is more than one hospital (Sutton 1991). If instead competition is less intense than Bertrand (e.g., Cournot), there is a stable equilibrium with nonnegative profits for a neighborhood of around 0, for any number of rivals. For this case, a merger that reduces the number of competitors definitely harms consumers because prices rise to both groups with no offsetting benefits. It is precisely for this reason that antitrust policy would block a merger that only reduces competition.

Consider now a nonprofit hospital that receives tax exemptions as a reflection of society's desire to increase health care among the poor (i.e., the involuntarily uninsured).3232 This fits the original IRS definition of charity-care requirements, which state that a nonprofit hospital “must be operated to the extent of its financial ability for those not able to pay for the services rendered” (Seaton and Koob 2009).
Hansmann (1987) finds that “tax exemption offers nonprofit firms a significant advantage in establishing market share vis-à-vis for-profit firms offering similar services.” This is not surprising as the nonprofit status contains a subsidy that can be used to achieve such social goals. Similar to David (2009) we allow nonprofit hospitals access to lower costs than can for-profits: , where , which may be thought of as cost reductions due to exemptions from property and income taxes, respectively. We assume that these cost advantages are sufficient for an “altruist” to always organize as a nonprofit.

We make two changes for the Philipson and Posner model. First, we rule out lump sum taxes and transfers from altruists to the rest of society and impose a financing constraint on nonprofit hospitals. Second, we allow the altruist to value the incremental health consumption of the poor differently than for the rich. With no restrictions on raising funds the social planner could theoretically use nondistortionary lump sum taxes to support hospitals and charity care. In our analysis, the social planner is assumed realistically to face some limits on the use of lump sum taxation and therefore does face a hospital financing constraint, and hence the planner will restrict competition and allow the price to the rich to rise in order to generate profits to provide medical care for the poor.

Suppose that the social welfare function reflects that society values health care consumption at the margin by the poor above what the poor value it for themselves and that the altruist hospital reflects the values of the social welfare function; that is, a social planner and the altruist value the poor's consumption of healthcare by an additional amount, α.3333 If nonprofits have no desire to subsidize care for the poor, allowing them the ability to set prices above marginal cost will not aid in achieving this desired social goal.

Suppose that competition between rival hospitals is Bertrand, with competition of even two hospitals guaranteeing that price reflects marginal social cost. It immediately follows under Bertrand competition between two altruists that

where α is the assumed altruist's (and society's) extra benefit when the poor receive health care, and P₁ and P₂ are prices to the rich and poor, respectively.

Since c(q) = F + c · q and F > 0, there is a natural monopoly element to hospital care as marginal cost pricing will not cover cost.3434 In addition, when c(q) is not homogenous of degree 1 in q, marginal cost pricing may not cover cost.
Moreover, even if F = 0, the optimality conditions related to the altruistic parameter α guarantee that profits are negative at the socially optimal pricing, since the price to the poor is below c. The following proposition states this result formally.

PROPOSITION 1.With Bertrand competition between rival hospitals, the equilibrium cannot produce the socially optimal outcomes in which each hospital remains financially viable for c(q) = F + c · q and F > 0.

If competition is not as strong as Bertrand (e.g., Cournot), there would then be a positive margin earned on the rich and this could provide a source of financing for the healthcare of the poor. But the point that competition for the rich limits the ability to finance healthcare for the poor would remain true.

Now consider a merger to monopoly. The merger allows P₁ to be set above c, generating funds that can be used to subsidize the poor, subject to a zero-profit constraint. The nonprofit altruist monopolist's problem is

(A1)

where α again reflects society's (and the altruist's) valuation of health consumption by the poor. The altruist channels producer surplus generated in the rich market to maximize consumer surplus for the poor. The social planning problem is to maximize the social welfare function subject to non-negative profits for the hospital:

(A2)

The solution is as follows:

This says that the altruist charges the monopoly price to the rich in order to maximize the funds (i.e., (P₁ − c) · q₁ − F) that can subsidize the health consumption of the poor. Note that α does not enter the nonprofit monopolist's pricing rule. A unique price-quantity pair for the poor is determined solely by the funds generated in the rich consumers market. In addition, our analysis is not altered by the presence of insurance, which ultimately weakens the relationship between prices and quantity demanded. The more generous the insurance coverage is, the more inelastic the demand for rich (insured) patients. Insurance, in essence, leads to less distortion from cross-subsidization. In the extreme case, with zero copayment (i.e., full insurance), demand is perfectly inelastic. And while this would, in turn, imply no quantity effects for the insured, a merger to monopoly will still allow the hospital to generate surplus that can then be used to cross-subsidize care for the uninsured. (This extreme case is equivalent to the case where the social planner can use lump sum taxation.)

The solution is a modified Ramsey pricing condition:

where λ is the Lagrange multiplier related to zero profits, μ =  − λ/(1 + λ), and ϵ_i is the elasticity of demand for group i. Note how the social planner pays attention not only to the transfer to the poor (i.e., the amount by which P₂ exceeds c – α), but also to the markup on rich individuals (i.e., the amount by which P₁ exceed c).

Assume for illustration purposes that the demand for health care by the poor is zero at a price of c. Comparing the nonprofit altruist monopoly problem to that of the social planner, it is clear that the (modified) Ramsey pricing solution to (A2) will differ from (A1). In (A1), the nonprofit monopolist pays no attention to the distorting effects of a high p₁ on the health consumption of the rich, and therefore will raise P₁ above the socially optimal level.3535 Note that λ > 0 leads to μ < 1, which in turn, means that the nonprofit altruist will charge the rich a price P₁ that exceeds the price set by the social planner.
For (A1), the optimal solution is to set P₁ at the monopoly price and use all the profits to cover F and the remainder to subsidize consumption of the poor. In contrast, the social planner will trade-off the negative deadweight loss caused by monopoly pricing to the rich against the social external gain associated with each additional unit of consumption by the poor. In general, the “altruist” harms the rich more than is socially desirable in order to serve the poor. This leads directly to proposition .

PROPOSITION 2.When c(q) = F + c · q and F > 0, the exercise of market power is necessary in order for the poor to consume health care. The rich subsidize the poor. The altruist, nonprofit monopolist, however, charges the rich too much and underprices health care to the poor relative to the social optimum.

Figure 1 illustrates this idea. is the single Bertrand duopoly price, which leads to the exclusion of the poor from receiving services. and are the prices set for the rich and the poor under the altruist monopoly. Positive profit margins in Market 1 (i.e., are necessary for cross-subsidization across groups. Independent of any social weights, the altruist behaves as a monopoly in Market 1 and will set below c, which leads to the delivery of services to the poor where the private value of their consumption is below marginal cost (when q > q* in the right hand-side panel of Figure A1).

Figure 2

Graphical Analysis of Cross-Subsidization: Bertrand Duopoly versus Monopoly

The disadvantage of treating nonprofits like for-profits under the antitrust laws is that the poor are underserved if mergers that create market power are not allowed because market power is needed to generate funds to cross-subsidize the poor. The disadvantage of giving nonprofits an exemption from antitrust is that the rich are overcharged even relative to the social optimum, which recognizes the external benefit of consumption by the poor. In the extreme case, merger to monopoly may lead to a decrease in welfare when the loss in consumer surplus resulting from the price increase in Market 1 (area L) coupled with the deadweight loss due to underpricing services for the poor (area D) is greater than the surplus generated for consumers in Market 2 (area G). Since the choice to serve the poor does not by itself constitute a net increase in welfare, in order to justify a merger that suppresses competition from a social stand point, we need sufficiently high value placed by society on consumption by the poor as well as sufficiently inelastic demand for healthcare services for rich patients (which reduces the distortion from cross subsidization).

The simple theme of this theoretical appendix, then, is that competition does not produce the socially desirable outcome even when a nonprofit altruist is structured to follow a “social preferences” to favor the poor. The process of competition limits the ability to price discriminate and to cross-subsidize. Where cross-subsidization is necessary to achieve social optimality, as it typically is when one relies on tax exempt nonprofit organizations to achieve social goals, competition does not necessarily produce the socially desirable outcome.

Next we provides a more detailed welfare analysis for the case of linear demand curves, illustrating how a merger of nonprofit hospitals can increase social welfare by suppressing competition.

Following our analysis above, suppose that a hospital monopolist produces a single service at a total cost of c(q) = F + c · q, and that it is able to divide the aggregate demand into two groups: rich patients (Market 1) and poor (Market 2). These two groups have two distinct downward-sloping demand curves for hospital services, the demand curves are known to the monopolist, and there is no opportunity for arbitrage between groups, as medical care is “nontradable” from the patient prospective. To illustrate our point simply, we assume that under uniform pricing Market 2 is not served profitably when price is set at marginal cost. This is important for our example because when demand curves are linear, price discrimination results in lower welfare and a uniform price is favored (Schmalensee 1981). In our model however, since the poor are excluded under a uniform price, the welfare implications of price discrimination are ambiguous.

The monopolist chooses a price for each group. Let {P₁, P₂} denote the prices in Market 1 and Market 2 respectively. Assume that the demand curve in Market i is q_i = a_i − b_iP_i. Serving Market i is profitable if P_i > c, or a_i > c · b_i for i = 1, 2. If this condition is violated, a for-profit monopoly will not engage in price discrimination. Instead, it will choose a uniform price (i.e., set price in both markets equal to the monopoly price for rich patients), which excludes poor patients from receiving services. On the other hand, a nonprofit monopoly may serve markets in which this condition is violated. By relying on other segments of the population for whom they can price above cost (Market 1), the nonprofit firm will price below cost in Market 2 without violating its nondistribution constraint, which applies to the organization as a whole.

Under the assumption that cb₂ − a₂ > 0, a for-profit monopoly will always choose to exclude the poor. On the other hand, a sufficient condition for a nonprofit monopoly to serve the poor is the ability to set P₁ above c. The monopoly will supply services to the poor even in the extreme case, where the social value of serving the poor (for every level of quantity) is lower than the social cost. As previously discussed, under these conditions, poor patients are served only if the hospital is nonprofit. However, this does not imply that, for example, merger to monopoly will necessarily increase welfare. The change in welfare across groups is given by:

The first term is the surplus generated in Market 2 as a result of such merger to monopoly and the second term is the loss of consumer surplus in Market 1. As expected the desirability of merger (i.e., suppression of competition) increases with α, the additional value that society places on each unit of the poor's consumption. While the nonprofit monopolist does not consider α when choosing the quantity of services to the poor, a greater α will increase the social benefits from eliminating competition. Subsequently this would raise the attractiveness of tax exempt nonprofits as a vehicle for achieving social goals.

Following Proposition , the price for paying consumers chosen by the monopolist (problem (1)) is given by , whereas the price chosen by the social planner (problem (2)) is given by . The profit condition a₁ > cb₁ is necessary and sufficient for .3636 Proof:

Hence, as in the general case, the altruistic nonprofit monopolist overprices healthcare to the rich and overprovides services to the poor.