Journal of Health Economics 7 (1988) 259-284. North-Holland
COMPETITION AMONG HCSPITALS Monica NOETHER Abt Associates, Cambridge, MA 02138-1168, USA
Received March 1987, final version received April 1988 The traditional view of hospital competition has posited that hospitals compete primarily along ‘quality’ dimensions, in the form of fancy equipment to attract admitting physicians and pleasant surroundings to entice patients. Price competition among hospitals is thought to be non-existent. This paper estimates the effects of various hospital market characteristics on hospital prices and expenses in an attempt to determine the form of hospital competition. The results suggest that both price and quality competition are greater in markets that are less concentrated, although the net effect of the two on prices is insignificant. It appears, therefore, that, despite important distortions, hospital markets are not immune to standard competitive forces.
1. Introduction Hospital competition is generally perceived to differ from that observed in other industries where price competition and profit maximization govern firm behavior. Various characteristics of the hospital industry may alter normal incentives and reduce constraints on the prices hospitals may charge. Hospita! competition may instead focus on quality’ in order to attract *This paper was for the most part written while the author was employed by the Federal Trade Commission. It reflects the views of the author and is not intended to represent the position of the Federal Trade Commission or any individual Commissioner. It is condensed from a longer FTC report on the same subject, Noether (1987), which may be requested from the author. Cheryl Asher initiated this project, formulated many of the basic ideas, and collected and compiled most of the data. Her efforts were indispensable to this project. I wish also to thank Jim Ferguson, Mark Frankena, Ted Frech, Scott Harvey, Richard Higgins, Pauline Ippolito, Mike Lynch, Joe Newhouse, Paul Pactler, Dave Scheffman and three referees for many valuable suggestions. John Holsinger and Nellie Liang served as able research assistants. ‘The term quality is used in this paper to suggest all non-price aspects of competition, whether they have positive value or not. Greater quality impiles greater complexity of the bundle of inputs used in a hospital visit. Scott and Flood (1985) summarize many studies that show a relation between changes in medical technology and increases in hospital costs. While, in one sense, such medica! care cm be termed high quality, it is not clear that such care always produces better health outcomes. If some hospital competition is in fact totally wasteful, it presumably results from the risk aversion and paucity of information possessed by consumers. In other situations where quality competition has been important such as the regulated airline industry, the increased services offered, while not welfare maximizing in a first-best sense, at least provided some utility to consumers. This may not be the case for some hospital competition; only physicians and/or hospital administrators may benefit.
0167-6296/88/$3.50 0 1988, Elsevier Science Publishers B.V. (North-Holland)
260
M. Noether, Competition among hospitals
patients and physicians. Since quaiity 1s costly, those factors that increase competition may, by raising the average level of quality, lead to higher prices per unit of output not adjusted for quality.2 In addition, imperfect information about the attributes of competing hospitals may affect the nature of competition. This study attempts to determine the extent, form, and effect of competition among hospitals. Is it true that quality is the only dimension that matters, or does price competition also exist? How important is information? What structural factors of the hospital industry most affect competition? Do concentration and entry barriers, for example, have the same sigcificance in the hospital industry as in other industries? Both hospital price and expense data are used to identify the effects of price and quality competition. The price data should be particularly enlightening. Since they represent charges for specific disease categories, many of the case-mix aggregation problems found in other studies are mitigated here. Moreover, most studies have relied on expense data, which, while useful for analyzing quality competition and efficiency issues, are not helpful in addressing the subject of price competition. Equations describing price, which hold constant exogenous demand and cost conditions, are estimated to determine the effects of various variables describing market structure and the extent of regulation. These price regressions should measure the net effect of any price and quality competition that exists. Since it is impossible to control completely for quality, it is difficult to isolate price competition from these regressions alone. Therefore, equations describing expense are also estimated. These regressions should measure quality, efficiency and scale economy factors, but not the primary effects of price competition. Any differences between the price and expense regressions suggest the exislence of price competition. The next section provides a brief review of the characteristics that are thought to differentiate the hospital industry from many others and outlines the models tested in this paper. Section 3 describes the data and variables used in the empirical work. Section 4 presents results for both price and expense equations, and section 5 provides a conclusion and discusses some implications for the current, apparently more competitive, environment. The regression results described below suggest that a reduction in concentration may lead to an increase in both price and quality competition, implying that price per unit of output fully adjusted for quality may be lower in areas with less concentrated markets. The results of this study imply that, ‘Indeed, in Federal Trade Commission cases challenging hospital acquisitions under the antitrust laws, hospital chains have argued that hospitals do not engage in traditional price competition. The Commission, however, has rejected this argument. @-a1 Corporation of America, 106 F.T.C. 361, 482 (1985), afs 807 F.2d 1381 (7th Cir. 1986) and upheld by U.S. Supreme Court; American Medical International, 104 F.T.C. 1, 202 (1984).]
M. Noaher, Competition among hospitals
261
while hospital markets may have several important distortions, almost a decade ago they were not immune to standard competitive forces, including price competition. Much anecdotal evidence suggests that consumer sensitivity to hospital care prices has increased in recent years. Therefore, since evidence of hospital price competition is found in the 1977-1978 data used in this study, it is reasonable to speculate that even more exists today.
2. Models of hospital behavior 2.1. Special characteristics Many unusual facets of hospital behavior are generally attributed to some special characteristics of the industry. First, since insurance coverage of hospital bills is extensive, consumers (patients) may be relatively insensitive to price. Second, physicians play an important role in determining both the supply of and demand for hospital services through their influence on hospital input and production decisions as well as their role as patients’ agents. The actual consumers, the patients, may themselves possess only poor information about competing hospitals. Since physician objectives may not coincide with hospital profit maximization, their influence may alter hospital behavior, particularly since competition among physicians is itself imperfect.3 Finally, the hospital industry is composed primarily of non-profit firms. The incentives of such firms are not clearly understood, but profit maximization is by definition not the only explicit goal.4 Most models of hospital behavior use these special characteristics of the hospital industry to posit that hospital competition centers around quality, either through the purchase of fancy equipment to attract admitting physicians or through provision of pleasant surroundings to entice patients. Since provision of such quality is costly, it is expected that more competitive hospital markets will evidence higher expenses and, since consumers are presumed insensitive to price, higher prices. Since medical care costs do not consume total GNP, however, some restraint on hospital prices obviously does exist. While increased competition may lead to greater input usage to create a more complex output, it seems likely that some pressure on price is also manifested. Even if prices rise due to ‘quality competition’, margins, the difference between prices and costs, should fall as the degree of competition increases. In other words, competition may have two opposing effects on prices. Production of a more COSQ 3See Kessel (1958), Friedman (1962), Hyde et al. (1954), and Noether (1986a, b). 4Fama and Jensen (1983a, b, 1985) distinguish non-profit firms as those without any residual claimants such as corporation stockholders. This lack of residual claimants assures potential donors that their gifts will not be expropriated. While non-profit hospitals no longer receive much of their funding from private donations, the government still ‘donates’ resources to nonprofit hospitals by granting them tax-exempt status.
262
M. Noether, Competition among hospitals
output bundle pushes upward, while standard competitive pressures lower them. Which effect is stronger is an empirical question. It is difficult, however, to test any models of hospital competition for at least two reasons. First, the definition and measurement of quality or complexity is elusive. There are many unpriced product attributes, such as response time, precautions taken, excess capacity, and amenities. Moreover, to account for quality appropriately, output should be measured rather than inputs, but hospital output is difficult to define. Second, most theories imply that hospital markets are differentiated oligopolies. Such a market structure is notoriously difficult to model. The next section, however, describes the opposing predictions that two models make about the effect of market structure on the nature of competition among hospitals. 2.2. Models of market structure’s effect on hospital competition Different models suggest different effects of market structure on competition among hospitals and on the prices they charge.5 One common assumption is that, as in most markets where firms are not completely differentiated, the elasticity of demand facing each individual firm (hospital) increases with a decline in concentration. For example, firms behave like Cournot oligopolists, and the market becomes continuously more competitive as the number of firms increases. The simple Cournot model, however, considers competition only along price dimensions and posits that prices decrease as concentration falls. If, in the hospital industry quality competition is at least as important, the effect of a decline in concentration on price depends on whether price or quality competition dominates.6 If competition mainly proceeds along costly quality dimensions, then the higher costs resulting from lower concentration and concomitant increased competition may lead to higher prices as well.7 ‘In ah cases, we assume that profit-maximization is one argument (perhaps among others), even in non-profit hospitals* utility functions. Many theories in the literature suggest other motives such as revenue maximization or physician utility maximization. However, it is probably not incorrect to assume that non-profit hospitals are also concerned with profits. See Pauly (1987) for a recent discussion. 6Lehvari and Peles (1973) and Leffier (1982) show that when quality increases the demand for a product, but is not a substitute for quantity, i.e. does not affect durability, then the effect of market structure on quality depends on the shape of the cost function as well as the interaction of the effects of quaiity and quantity on demand. Without specific formulations of demand and cost functions, the relation between quality and market structure is ambiguous. ‘Different predictions about the impact of market structure on price arise when, rather than positing Cournot behavior for hospitals, it is instead assumed that firms are differentiated and that information about their quality and other attributes is costly to obtain and hence imperfect. Using a monopolistic co:npetition framework, Satterthwaite (1979) has suggested that in this situation, as !he number of firms increases, the information available about each diminishes. As a result, the elasticity of demand facing each individual firm decreases as the number of firms increases. Given that framework, decreasing concentration in a market leads to less competition
M. Noether, Competition among hospitds
263
Since it is possible that both price and quality competition exist in the hospital industry, profit or price data alone might be insufficient to determine the effect of market structure on concentration. A positive correlation between profits and concentration could imply either price or quality competition, while a finding of no relation between price and concentration could suggest that concentration has no effect on competition or that the effects of price and quality competition on price offset each other. Therefore, it is necessary also to examine the effects of market structure on expenses. While prices are affected by both price and quality competition, expenses are likely to show primarily the effects of quality competition.* If costly quality competition dominates, decreased competition should result in lower expenses. If price competition is more important, a decline in competition is likely to have no effect on expenses. Predictions from these models can be uniquely identified from coefficients on a market structure variable in price and expense regressions9 Identifying P, E and c as prices, expenses, and concentration, respectively, and using subscripts to denote partial derivatives, the following set of predictions apply to each model: Price competition: PC> 0; F; = 0. Quality competition: P, c 0; E, < 0. Price and quality competition: PC ? 0; E, c 0. No effect of concentration on competition: P, = EC= O.‘O The next section discusses some of the specific variables used in the regressions. In addition to concentration, other factors that directly affect the degree of competition, such as regulation, ownership structure, and HMOs are included. because as information about each firm becomes more costly, consumers choose between a smaller number of competitors. If this is the correct model of hospital competition, then a decline in concentration raises prices if price competition dominates, whereas prices fall if costly quality competition is most important. *It is sometimes thought that non-profit institutions do not operate effciendy unless forced to by competition. In this case, if a growth in competition results in greater efficiency, expenses may fall. It is also possible that the cost decline leads to production of higher quality. Therefore it is not cleat what effect, if any, competition-induced efficiency has on expenses. ‘To the extent that hospital location decisions are endogenous, market concentration is, itself, determined by other competitive conditions and is not exogenous. If this were an important factor, one would expect to see the relative numbers of hospitals in different areas changing over time. In fact, hospital-population ratios appear to be highly correlated over time (the correlation between 1970 and 1977 is 0.97), which suggests that entry is not highly responsive to differences in competitive conditions. “These predictions are for a Cournot model of competition. If the information-cost model were to hold, the first two predicitions would be reversed, i.e., price competition would be suggested by P, ~0; E, =O, while quality competition would derive from P,>O; E,>O.
264
M. Noether, Competition among hospitals
3. The variables, market and data 3.1. The variables Several factors may determine the degree of competition in any hospital market. The number and distribution by size and ownership of hospitals, BS well as substitutes and complements to hospitals may be important* In this study, the market is assumed to include all short-term general non-federal hospitals. Potential substitutes include long-term hospitals, nursing homes, and, indirectly, HMOs that rely on outpatient care.” This section discusses the most important factors that potentially affect competition among hospitals and that are incorporated in the regressions as explanatory variables. Other explanatory variables include management and hospital system afIIliation, public hospitals, and nursing home beds as well as demand and cost variables such as income, public and private insurance coverage, hospital size, wages and occupancy rates. These variables are listed with descriptive statistics in the appendix, and their effects are described more fully in Noether (1987). 3.1.1. Structure A Herfindahl statistic, calculated with beds as the measure of each hospital system’s market share, is used to describe the number and size distribution of firms in an industry. 11 ‘i‘he predictions about the effect of the Herfmdahl statistics were spelled out in the previous section and are not repeated here.13 3.12. Entry regulations State level entry regulations such as Certificate of Need (CON) require approval for construction and for some other major capital expenditures made by hospitals. These laws are currently being phased out by several states, but were strong during the time of the study. The theories predict differing effects of entry regulation. The first assumes that, left unregulated, hospitals compete for physicians and patients through “Freestanding emergency centers and ambulatory care centers, which now compete with at least some traditional inpatient hospital services, did not exist at the time of this study. 12A Herfindahl statistic is calculated by summing the squared market shares of all the firms in the market. It is generally considered a better measure than a simple concentration ratio because it takes account of all firms in the market rather than just the top four or eight. Therefore it provides a measure of firm size dispersion in addition to concentration. The Department of Justice Merger Guidelines released in 1982 and revised in 1984 base their structural criteria on the Herfindahl index. If a hospital system owns more than one hospital in a given SMSA, the multiple hospitals are treated as a single firm in the calculation of the Herfindahl. “Several studies have attempted to measure the effect of hospital concentration on quality, price and costs. Watts (1976) reported in Salkever (1978), Davis (1971 and 1974), Joskow (1980), Farley (1985), and most recently, Luft et al. (1986) and Robinson and Luft (1985) have examined this issue. None has found much evidence of either price competition or eficiencies.
.
M. Noether, Competition among hospitals
265
‘unnecessary’ expenditures on facilities and equipment and posits that entry regulation reduces costs and hence prices. Alternatively, Posner (1974) and Joskow (1981) have suggested that CON regulation fosters cartelizing behavior by hospitals by hindering rent dissipation through quality/amenity competition as well as preventing entry. They predict that entry regulation leads to higher prices with constant or lower costs, i.e., higher margins. Finally, the third hypothesis suggests that regulation of beds and certain large equipment expenditures merely redirects purchases to other furms of capital. Total capital expenditures remain constant while operating costs rise since production using the regulated mix of capital is less efflcient.14 It can be argued that it is inappropriate to draw any inferences about the effects of entry regulation from OLS regressions because the timing of each state’s enactment of the law is endogenous. The data do not support the notion that states with higher per capita health expenditures enacted CON regulations first. There appears to be no significant relation between the level of a state’s health care expenditures, standardized for per capita personal income levels and the timing of CON regulation enactment. This paper therefore assumes that the timing of CON law enactment is exogenous, although it should be recognized that other unmeasured state characteristics may have affected the timing. 3.1.3. Health maintenance organizations HMOs have significantly lower hospital utilization rates than the population at large, even after controlling for differences in the health status of the insured population [see Luft (1981)]. HMOs, because of their utilization constraints,15 may cause hospitals to act more competitively. Their effect on prices and expenses depends on whether they are more price or quality conscious. If, as often suggested, HMOs are more interested in price, their entry should lower prices. It is also possible that HMO location is endogenous, i.e., they entered first into those markets where cost was highest [Frank and Welch (1985)].16 3.1.4. For-profit hospitals While voluntary (non-profit) hospitals still operate the majority of all beds 14Section 1122 (of the Social Security Act) also regulates capital expenditures. Prospective Rate Review regulation has also been common in the last decade. Variables measuring both Section 1122 and prospective rate regulation are included in the regressions. See Noether (1987) for a description of their form and effect. ISIn recent years, some HMOs have become large enough to ctuse hospitals to bid for their contracts by offering price competitive packages. In 1977-1978, however. it is unlikely that HMOs were large enough to elicit much of this behavior. 16Blue Cross plans are also known to extract discounts in areas where their market share is large. However, a variable measuring their share had no effect.
266
M. Noether, Competition
among hospitals
in the United States, the market share controlled by proprietary institutions has increased substantially in recent years. For-profit hospitals could have various effects on competition, depending on their incentives and abilities to compete on quality or price dimensions, to be efficient, and to maximize profits. 3. I .5. Hospital occupancy rates Holding constant a hospital’s occupancy rate, the higher its occupancy rate is relative to the market average, the less it may need to compete. Its higher occupancy may indicate some local market power related to, for example, a beneficial location. i7 If this is true, that market power should be manifested in higher margins, i.e., higher prices for a given level of quality or lower quality for the same price. In this case, a hospital’s relative occupancy rate should be positively correlated with prices or negatively correlated with expenses. On the other hand, a higher relative occupancy may resuit because a hospital is more competitive, i.e. charges lower prices per unit of quality. In this case relative occupancy rates will be negatively associated with prices or positively related to expenses. Both explanations of the role of relative occupancy rate suggests that it may be endogenous. Because of this, regressions were also run without this variable. While total explanatory power was decreased, the effect of other variables was not strongly affected. 3.2. Product and geographic market definitim 3.2.1. Price data
This study analyzes price data for various individual disease categories. While case mix variation still exists within a given disease category, such variation should be smaller than that found in the aggregate hospital revenue or expense measures used in many studies. I* Eleven disease categories are studied. In the price regressions, the dependent variable is the average price (charge) per case for the particular disease category for all hospital services except physicians’ ‘ees. The price data are from Health Care Financing Administration (HCFA) files of average charges for the most frequent diagnoses for Medicare inpatient diagnoses for 1977 and 1978. They are derived from the Medicare “The geographic market used in this study is defined at the SMSA level. As discussed below, in some cases this may be too large. Convenience is often thought to be an important facet of a hospital’s characteristics, so a particular hospital may be able to charge slightly higher prices than others within the SMSA and not lose many patients. ‘*One recent study [Eskoz and Peddecord (19833 uses profit data by individual hospital service (e.g., pharmacy, medical apd surgical, obstetrics, clinical laboratory) to determine the eflect of ownership on margins for different services. The data are subject to case mix problems, however, and the study does not examine the effect of market structure.
M. Noether, Competition among hospitals
267
Provider Analysis and Review (MEDPAR) data which ‘are a 20% sample of all Medicare hospital inpatient bills submitted to HCFA for payment under Title XVIII [Medicare] of the Social Security Act’.19It can be argued that these charge data do not represent the actual prices paid to hospitals since charges are sometimes discounted before payment. If discounts are systematically related to any of the explanatory variables, then the regression results would be biased. It seems most likely that the magnitude of the discounts could be directly related to the degree of competition in a market. If this is true, the effects of price competition measured using the charge data underestimate the true effects. Unfortunately net price data are unavailable to this study so that the actual pattern between charges and net revenues cannot be determined. HCFA consideres the data to provide good proxies for actual price levels.20*21 Both average market prices, where the market area is defined to be a Standard Metropolitan Statistical Area (SMSA), and individual hospita1 level observations are examined. Implicit in the market level analysis is a model of individual hospital behavior that yields equilibrium market prices dependent on various characteristics of the market such as demand, structure, and regulatory environment. Using SMSA-defined markets, it is straightforward to define each hospital’s competitors. Since, however, by aggregating all the individual hospital data, we lose potentially valuable information about differences across hospitals that may affect their pricing decisions, prices at the individual hospital level are also studied. In the market-level regressions, the unit of observation is an SMSA, and the price variables are. weighted (by number of cases) averages across the sample hospitals in a given SMSA. ‘gInformation on the data set is derived InformationLetter No. 81-28, March 31,
from the Bureau of Health Planning’s Program 1981 (U.S. Department of Health and Human Services). Other studies of the data cited in this report are Wennlxrg (1980) and the Program Information Letter 80-38 of August 13, 1980. “HCFA cautions users of the data to remember that they are ‘fir the Medicare enrolled population only’ (emphasis in original), but also notes that persons over the age of 65 accounted for one quarter of all hospital discharges and for one third of all patient days. Other studies have also shown a strong correlation between the pattern of Medicare charges and other prices across hospitals. According to the Program Information Letter 81-28 cited in footnote 19, supra, even though the levels may vary, the data ‘can Le considered reasonable indicators of the overall relative differences in prices charged by hospitals for similar types of diagnoses’. “The data have one major problem. Medicare (and most private insurance plans) divides the price for a hospital visit into two components: a charge representing all of the hospital inputs (e.g., bed, nursing services. operating room, drugs, physical therapy) billed under Part A, while the physician fee, generally billed directly by the physician, is covered under Part B of Medicare. Ideally, the effect of various features of the hospital market on total hospital service price should be studied. Unfortunately, as is true for almost all studies of hospital competition, data for the hospital component of this price cannot be matched with those representing physician charges. The hospital price may not always react in the Sam: fashion as the total charge would to differences in market conditions, particularly if physicians have some control over the production process for hospital services. See Pauly (1978 and 1980) for theory and evidence on the role of physicians in shaping hospital behavior.
268
M. Noether, Competition among hospitals
3.2.2. Geographic market
One disadvantage of defining the geographic market at the SMSA level is that only about 2,800, or less than 50”;/, of all community hospitals, are located in SMSAs. Therefore only half of the available sample of all community hospitals is used. Moreover, hospitals located in SMSAs are not drawn from the distribution applicable to all community hospitals. Hospitals located in SMSAs are more apt to be large: while 44% of the SMSA-hopitals contain more than 200 beds, only 27% of all community hospitals do. Likewise, less than 9% of SMSA-located hospitals have fewer than 50 beds, while 24% of all community hospitals in 1977 were that small. It can also be argued that SMSAs are too large to represent true hospital geographic markets, given the average distances that most patients travei. While it is true that some larger SMSAs probably exaggerate the extent of the geographic market, it is difficult to determine a more appropriate way of defining markets without individually examining each local area. Moreover, use of existing patient flow data is inappropriate for determining the extent of the market for competitive purposes** since such data pertain only to currently prevailing market conditions and say nothing about what would happen if an attempt was made to exercise market power. County boundaries surely are overly narrow in most situations, particularly in metropolitan areas where is is difficult to define appropriate submarkets within the entire area. Recently, an alternative has been proposed [see, e.g., Robinson and Luft (1985)] to include all the hospitals within a 5 or 15 mile radius as a given hospital’s competitors. While this approach has some appeal, it is quite arbitrary. It ignores natural geographic boundaries and existing population centers and neglects the fact that in metropolitan areas the extent of the market is sometimes smaller than in less densely populated areas.23 SMSA boundaries, on the other hand, were constructed with these factors in mind. If the SMSA market definition is overly broad when an SMSA covers a broad geographic area then the SMSA-based Herfindahl underestimates the actual Hertlndahl in those instances. It is true that the Herfindahl tends to be lower in those SMSAs with larger areas. 24There are no obvious differences, 22This sort of test is the hospital version of the Elzinga-Hogarty (1973) LIFO-LIFI test of limited shipments into or out of the proposed market. The LIFO-LOFI test of geographic market boundaries looks for an area where, for the relevant product, shipments can be described as ‘@tie in from outside’ (LIFO) and ‘little out [ram inside’ (LOFI). Recent literature describing antitrust markets has pointed out the flaws of this approach. See, e.g., Sheffman and Spiller (1987) or Ordover and Willig (1982). Shipments data underestimate the market size if an anticompetitive price increase would make shipment from greater distances profitable. Conversely, if some firms in the shipment-defined market are at capacity, they would be unable to counteract an anticompetitive price increase by a subset of firms in a market defined in this way. 23The correlation between population density in an SMSA and its area in square miles is a significant (at the 0.001 level) -0.18. No relation exists between total population in an SMSA and the SMSA’s area (correlation = 0.04), however. 24The correlation between Herfindahl and area is a significant -0.2. The average value for
M. Noether, Competition among hospitals
269
however, in the size or occupancy rates of hospitals in different sized SMSAs. The empirical work outlined below will attempt to test the impact of market size on the relation between concentration and competition. 3.2.3. Expense data and regressions The American Hospital Association’s (AHA) annual survey of United States hospitals provides the expense data. They are matched to the price data by hospital before any aggregation is done to the SMSA level. These data represent total hospital accounting expenses, including salaried personnel, their benefits, professional fees, depreciation, interest, supplies and purchased servicesF5 Tax payments by for-profit hospitals are included.26 The expense data are not disaggregated to the disease category level. This implies that unlike the price data, the expense data may be subject to substantial case mix variation. To the extent possible, however, the various hospital characteristics thought to affect case mix are held constant in the expense regressions to minimize the problem. Teaching hospitals and large urban hospitals which may attract more complicated cases are accounted for by various dummy variables in the regressions. Measures of the special facilities possessed by hospitals (a common indicator used to measure quality) were incorporated in earlier regressions but had insignificant effects on prices and expenses. Since maternity visits tend to be relatively low cost, a variable measuring the ratio of births to admissions is also included in the expense regressions. Also included is a dummy (share in the SMSA level regressions) variable indicating hospitals located in cities with population equal to or exceeding one million to account for the more complicated case mix typically found in urban hospitals. Regressions are run on expense per admission which corresponds most closely to the price per case measure used in the price regressions. Admisthe Herfindahl is 0.36 in those SMSAs with an area of less than 700 square miles (corresponding, roughly, to a 15 mile radius definition), while it averages 0.26 for hospitals in SMSAs with areas greater than 700 but less than 1,000 square miles, and averages 0.12 in those SMSAs encompassing more than 1,000 square miles. The Herfindahl is also, not surprisingly, significantly negatively related to population and population density. These variables are accounted for in the regressions, both through a dummy variable to indicate cities with population of one million or greater and a continuous population density variable. “According to Becker and Sloan (1985, p. 32). system overhead or ‘home ofice’ costs are charged tu individual hospitals, but not on a fee-for-service basis. Percentage of revenue is one method commonly used. It is difficult to determine what bias this may create. 26Data delineating the tax payments of individual hospitals are not available to this study since tax payments are included in an aggregate category of ‘other nonpayroll expenses’ by the American Hospital Association. Becker and Sloan (1985, p. 31) estimate that proprietary hospitals* expenses were on average 3.5 percent higher in 1977 because of tax payments. They also note that higher interest costs payable on corporate debt financing (as opposed to municipal or other tax-exempt) led to a further 1 percent cost disadvantage accruing to proprietary hospitals.
270
M. Noether, Competition among hospitals
sions are adjusted to take account of outpatient visits.*’ Since factor usage is in part determined by the form of competition, variables such as labor input and assets per bed are not exogenously determined in the expense regressions. Therefore the regressions are run omitting such variables. Since the expense data represent accounting rather than economic costs and may be subject to manipulation, some caution must be exercised in drawing conclusions from them. To the extent that they represent prices charged to cost-based insurers, it would not be surprising to find no difference between the price and expense data. On the other hand, if different hospitals manipulate their expenses by different amounts, some noise (similar to that possible in the charge data) may exist. Once again, if less manipulation occurs in more competitive markets the measured results understate the true effects of competition. 3.2.4. Other variables Other variab!es that may affect competition are incorporated as explanatory variables in the regressions. They include hospital system affiliations, management contracts, public hospital status, and the existence of substitutes such as nursing homes. Several variables are also used to control for different demand and conditions across SMSA markets such as income, insurance coverage, percentage of the population on welfare or unemployed, population density and growth., the percentage of the population that is white and the death rate. Also included is a local wage rate. Motivation for usage of all the variables and their results are discussed in Noether (1987). The next section outlines the empirical results. In order to make necessary comparisons, the results from all four sets of regressions, SMSA level price and expense, and individual hospital level price and expense, are described simultaneously. Tables showing coefficients and significance levels for all the included variables can be obtained from the author.
4. Results The estimated coefficients for the effect of concentration on prices and expenses are shown in table 1. Columns 1 and 2 report coefficients from a linear Herlindahl measure (at the SMSA and hospital levels, respectively), while columns 3 and 4 present the comparable coefficients from a dummy variable set equal to one when the Herlindahl exceeds 0.3. These latter results will be discussed later. T-statistics are presented in parentheses below the coefficient estimates. Row 3 shows the percentage effect, evaluated at the “This adjustment is done by the AHA. The AHA expresses outpatient visits as equivalent admissions by multiplying outpatient visits by the ratio of outpatient revenue per visit to inpatient revenue per admission.
M. Noether, Competition among hospitals Table 1 Coefficients on Herhndahl measure of concentration Linear measure of Herfindahl Disease category Diabetes meliitus
Cataract surgery
Acute myocardial infarction Congestive heart failure
SMSA
HOSP
11.46 (0.60) 1.08 ( f 2.0)
- 167.2 ( - 1.84) - 0.63 ( f 0.7)
16.86 (0.47) 1.08 ( + 2.0) 59.36 (1.55) 3.01 ( + 3.3)
- 149.0 (-1.51) -0.31 (f0.4)
-44.36 ( - 0.39) - 0.08 ( f 0.4)
- 143.2 (- 1.21) - 0.46 ( + 0.8)
Fracture of neck and femur Mean % price effect
Expense/admission
HOSP
-21.78 (0.95) 2.33 ( & 4.9) 11.59 (0.22) 0.44 ( + 2.0)
Pneumonia
Hyperplasia of prostate
SMSA
14.57 (0.55) 0.92 ( f 2.0)
- 96.22 (-0.81) - 0.28 ( f 0.7)
Diverticula of intestine
Dummy measure of Herfmdahl
15.01 (0.56) 1.09 ( f 3.9)
- 15.82 (-0.15) - 0.03 ( f 0.3)
Inguinal hernia
for prices and expenses.
Price as dependent variable - 55.22 - 14.33 (-0.20) (-0.70) -0.23 (20.7) -0.04 (kO.3) - 24.38 - 19.37 (-0.31) ( - 0.43) -0.15 (kO.8) - 0.07 ( f 0.4) - 230.7 - 245.6 (- 1.68) (- 1.45) -0.52 (kO.7) -0.31 (kO.4)
Acute, cerebrovascular disease
Respiratory system disease, other
271
- 131.2 (-1.16) -0.24 (kO.4) - 40.24 ( - 0.63) -0.11 (kO.4)
- 152.8 (- 1.29) -0.50 (kO.8) - 112.3 (-1.71) -0.56 (f0.7)
32.45 (0.5 1) 0.11 (kO.4)
33.51 (0.58) 0.20 ( f 0.7)
- 13.70 (-0.15) - 0.03 ( + 0.4)
- 96.26 ( - 0.97) -0.30 (kO.6)
- 197.4 ( - 0.98) -0.18 (kO.4)
- 263.0 (- 1.23) -0.41 (kO.7)
-0.11
-0.35
Expense as dependent variable - 219.9 - 322.3 ( - 3.20) (- 5.30) - 1.13 (kO.7) -0.89 (kO.3)
- 1.325 ( - 0.03) - 1.07 (k4.7) - 23.39 ( - 0.58) - 1.29 ( f4.5)
23.86 (0.45) 0.81 (f2.0) - 1.327 f-0.04) 0.08 (k2.0) 60.69 (1.51) 2.65 ( + 3.6) -9.014 ( - 0.23) - 0.44 ( + 3.9) - 27.25 ( - 0.69) - 1.33 (_+3.8)
24.11 (1-W 2.06 ( + 4.0)
12.74 (0.57) 0.95 ( + 3.4)
17.15 (0.76) 1.72 (k4.5)
18.43 (0.94) 1.64 ( f 3.4)
40.55 (1.18) 1.92 ( f 3.2)
18.00 (0.54) 0.85 jf3.1)
- 4.845 ( - 0.07) -0.13 (k3.7) 1.11
-48.21 (2.08) 4.05 ( + 3.9)
10.82 (0.15) 0.25 ( + 3.2) 0.66
- 46.66 (2.06) - 3.59 ( + 3.5)
“Line 1 of each disease entry shows the coefficient estimate. Line 2 the t-statistic for the coefficient. For the linear measure (columns 1 and 2) line 3 shows the percentage effect of a 0.01 increase in the Herfindahl while for the dummy measures (columns 3 and 4) it shows the percentage effect of a Herfindahl 20.3, both evaluated at mean levels. 95% confidence intervals around the percentage effects are indicated in parentheses in line 3.
272
M. Noether,
Competition among hospitals
mean price, of a 0.01 increase in the linear measure of the Herlindahl and the percentage effect of the dummy Herfindahl measure equalling one. Ninetyfive percent confidence intervals around the percentage estimates are also
reported in the third row of each disease category. Both at the SMSA and individual hospital levels, the t-statistics in the price regressions listed in columns 1 and 2 of table 1 show that the Herfindahl coefficients as calculated are never significantly different from zero. By itself, this could indicate that market structure has no effect on hospital competition. 28 The expense per admission regressions exhibit uniformly negative and significant Herfindahl coefficients, however. The negative correlation between market concentration and expenses implies that competition occurs along costly service and facility dimensions.29 This quality competition finding corroborates common belief. The significance of the Herfindahl in the expense regressions combined with its insignificance in the price equations suggests that while competition does increase the quality or complexity of the bundled output termed a hospital stay, it may not affect price. If price is unaffected by concentration, the price of a quality-adjusted bundle of output (which we cannot measure) falls with increases in the degree of competition, as do hospital margins, and some price competition therefore also exists. Table 1 also presents (in row 3 of each disease category and expense entry) the estimated percentage effects on price and expense of a 0.01 (or 100 points on the familiar l--10,000 scale) increase in the Herfindahl. A 100 point increase is the cutoff under the 1984 Department of Justice guidelines for when a merger must be seriously investigated if the level of the Herfindahl index is in the range of the average market studied here. The measure of the percentage effect of a 0.01 increase in the Herfindahl on prices and expenses resembles an elasticity evaluated at the mean levels, except that rather than using a 1 percent increase in the explanatory variable we use a 0.01 increase in the mean level to make the measure comparable to antitrust considerations of competitive effects. In all cases the 95 percent confidence intervals 2sAlternatively, bed capacity may not be the appropriate way to measure market share. Other measures such as patient days and revenues are highly correlated with the measure used, however, and produce nearly identical resuiis. A squared Herfindahl was also included in an earlier stage of the regressions to test a possible non-linear relation between price or expense and concentration. It did not yield significant results. Similarly, when the Herfindahl was constructed to view hospitals managed by a common company as a single firm, the results did not change. 2q1t is not surprising that the information cost theory is not supported since physicians, who are generally well-informed about different hospital choices, have a strong influence on consumer choice. Areas with a large and rapidly growing population might be thought to have higher prices because information is less available under the information cost theory. It is trt!e that variables that measure population size (a dummy for cities with a million or more people) and growth have significant positive coeficients. These variables, however, are likely to be measuring cost factors.
M. Noether, Competitionamong hospitals
273
reported span negative and positive ranges in the price regressions. On the other hand, the upper bound of the 95 percent coniidence intervals in the expense regressions are less than zero. Unfortunately, since the dependent variables in the price and expense regressions are not directly comparable, the confidence intervals surrounding the effects of an increase in the Herlindahl are also not comparable. It is therefore impossible to draw statistical conclusions about the existence or lack of pr& competition from the data presented. The arguments presented in the paragraphs below suggest that the results appear to be more consistent with the existence of price competition, and the concomitant reduction in quality-adjusted price with reductions in concentration, than with most other theories. It could be argued that the negative relation between concentration and expenses indicates that hospitals in more concentrated markets are larger and have taken advantage of economies of scale and are, as a result, more efficient. This efficiency is perhaps not passed on in the form of lower prices because of the lack of consumer price sensitivity. The correlation between the Herlindahl and hospital size, however, is only -0.03.30 There is no relation between concentration and occupancy rates. Moreover, if the efficiency explanation were correct, hospital size would also be negatively related to expenses. Size actually has a positive effect on expenses, albeit only significant at the SMSA level, perhaps because larger hospitals tend to offer more complex services to sicker patients, and case mix has not been completely held constant in the expense regressions. When a quadratic measure of hospital size is also included it has an insignificant effect on hospital expenses. Interestingly, the effect of size on prices, while generally insignificant, is usually negative. Alternatively, since the expense data may in&de some case mix variation, the observed merely reflect concentrated) people. These
negative relation between concentration and expenses could the fact that urban hospitals (whose markets are generally less tend to offer more sophisticated care and take care of sicker factors have been accounted for, however; population density,
as well as location in a large city, are included as explanatory variables. Moreover, variables measuring the effect of teaching status, the proportion of cases that are births, and, as mentioned above, size are also included to account for case mix variations. Several of these variables are significant. It seems unlikely that remaining unexplained case mix differences are s&iciently larger in the expense regressions than in the price regressions alone to cause the difference between the price and expense results. Robinson and Luft (1985) did not find the addition of elaborate casemix information to affect their results on market structure. “Hospitals in markets with a Herfindahl of 0.5 or greater are somewhat larger than those in less concentrated markets (278 beds versus 252 beds, on average).
274
M. Noether, Competition among hospitals
Looking at the raw data, the simple relation between concentration and individual prices does not appear to differ from that between concentration and expenses. Both prices and expenses are lower in more concentrated markets. In markets where the Herfindahl is 0.5 or greater, expenses are on average 77 percent of those in unconcentrated markets where the Herfindahl is less than 0.3. For the disease code, diabetes, the corresponding percentage is 74 while for congestive heart failure it is 76. These relations suggest that there is nothing fundamentally different between the price and expense data. AS noted above though, the price data are considerably more disaggregate than the expense data. While the disaggregation is an advantage in terms of mitigating case mix problems, it could accentuate any errors-in-variables problems. Such problems could be caused by, for example, mismeasurement of market concentration by defining the geographic market as an SMSA when in fact the market varies according to disease category or by crosssubsidization of charges across disease categories. While the raw data do not suggest it, since the expense data are more aggregate such errors may ‘cancel out’. Therefore, price regressions were run using a more aggregate measure, the average charge for the eleven different disease categories (weighted by number of cases in each category). The resulting Herlindahl coefficients were -23.87 with a t-statistic of -0.35 at the SMSA level and - 119.9 with a tstatistic of r 1.09 at the individual hospital level. These numbers suggest that differing degrees of aggregation in the price and expense regressions are not generating the different effects of concentration. The interaction between are8 and concentration was also tested to attempt to determine whether the SMSA-based geographic market influenced the results. If, in fact, for those SMSAs that are largest, a smaller market definition is appropriate, the Herfmdahl may not be appropriately measuring the relation between concentration and price or expense in those markets. A dumnij variable signifying those markets with areas of greater than 1,000 square miles (corres;-ionding to a radius of about 18 miles) was interacted with the Herfmdahl coefficient, and that interaction variable was included in the regressions in addition to the linear Herfindahl measure. The coefficients on the Herfindahl variable itself are insignificantly different in these regressions, both for prices and expenses. With the exception of one disease category, the coefficients on the interaction variable are positive in the price regressions, albeit for the most part insignificant at normal statistical cutoffs. For the expense regressions, the coefficient is positive at the hospital level and negative at the SMSA level, and both are well within one standard error of zero. These results suggest that concentration as measured may have a greater effect on price competition in those markets with large geographic areas where the extent of the geographic market is most likely to be overstated. They do not suggest that the previously discussed findings result from the SMSA-level geographic market definition.
M. Noether, Competition among hospitals
275
In all cases, the magnitude of the Herfmdahl coefficient suggests that market structure is a minor determinant of price or expense. AS table 1 shows, at the SMSA level, the effect of a 0.01 point increase in the Herfindahl, evaluated at mean levels of Herfindahl and price or expense, ranges from a 0.3 percent decrease in the mean price for acute myocardial infarction and congestive heart failure to a 0.1 percent increase for divertic& of the intestine. While the range of effects measured at the hospital level doubles that at the SMSA level, the effects are all under one percent. The linear specification of concentration is inappropriate if concentration does not matter until some threshold is reached. With this in mind, a dummy variable, set at one when the Herfindahl index equalled at least 0.3 (3,000 on the familiar l-10,000 scale) and 0 otherwise, is substituted for the linear measure, while the remaining variables, apart from a CON dummy discussed below, stay the same. The 0.3 cutoff appears to fit the data best of several levels tried. The coefficients on the Herfindahl dummy and the percentage effects they represent are shown in columns 3 and 4 of table 1. These results mostly confirm those discussed previously. Concentration appears to have no significant net effect on prices but is significantly negatively related to expenses. The magnitude of the estimated percentage effects shown in row 3, however, is generally somewhat larger. The price coefficients suggest that prices are on average 0.7 percent higher at the hospital level and 1.1 percent higher at the SMSA level (albeit insignificantly), while expenses are 3.6 to 4.0 percent lower, when the Herfindahl equals or exceeds 0.3.3l Ninety-five percent confidence intervals once again make any conclusions speculative, but do not contradict the possibility of simultaneous price and expense competition. The regressions using the dummy variables appear to fit the data about as well as the previous linear specifications. Table 2 reports coefficients for the other variables that measure competitive conditions. The effect of CON regulation is initially measured as the number of years a law has been in effect in the state, since CON laws regulate expenditures on durable goods and therefore may only become effective with a lag. This age variable ranges from 0 to 13 for the year, 1977, that is studied. As columns 1 and 2 indicate, Certificate of Need regulation coefftcients are generally significantly positive in the price regressions, particularly at the SMSA level. The SMSA-level expense regression shows a significant positive coefftcient for the CON measures as well. If it is correct to assume, as discussed earlier, that CON laws are exogenous, these results suggest that such regulation does not serve to control expenditures, but rather leads to a costlier reallocation of resources. While the cartel story, which predicts higher prices and/or lower expenses, is not strongly supported “The Herfindahl dummy variables’ coefficients in aggregate (of the eleven disease categories) price regressions suggested very similar results.
4.881 (1.09) 3.506 (0.79)
9.064 (2.10)
3.800 (1.18)
6.008 (1.25) 18.83 (1.78)
5.276 (1.72)
( 1.08)
11.16 (1.93)
47.84 (1.94)
174.0 (2.25)
41.10 (1.14)
14.65 (0.62)
99.12 (2.46)
43.15 (2.44)
66.18 (2.46) 197.5 (3.32)
22.16 (1.33)
47.33 (2.58)
109.3 (3.42) 103.8 (3.31) 90.65 (2.90)
42.92 (2.06) 43.57 (2.79) 183.3 (4.33) 62.17 (2.59)
- 0.7598 ( - 0.34)
1.108 (0.32) 3.095 (0.42)
3.566 (1.59)
2.273 (0.58) 0.3212 (0.08) 0.2841 (0.07) -0.5927 (-0.26)
2.166 (0.80) 0.6195 (0.27) 2.747 (0.52) -0.1151 ( - 0.03)
SMSA 5
__ ._
-
-
-
-
__
. -
(- 1.36)
- 1.722
-3.112 ( - 0.74)
- 3.414 (-1.73)
- 0.2826 (-0.21)
- 1.740 (- 1.32)
- 1.877 ( - 0.78) - 1.139 ( - 0.50)
- 3.448 ( - 1.55)
- 0.0622 (-0.03)
- 0.5828 (-0.44) 0.4345 (0.14)
- 0.8076 (-0.53)
SMSA 7
Change in HMO share
-2.814 ( - 4.42)
-5.109 (-5.57) - 9.863 ( - 4.67)
- 2.936 ( - 4.70) - 1.617 ( - 2.83)
- 3.928 ( - 3.78)
( - 2.80)
- 3.022
-3.112 (-4.11) - 2.239 ( - 4.32) - 3.812 ( - 2.63) - 2.49 (-2.98) - 5.07 1 ( - 4.65)
HOSP 8
62.23 ._ (0.64)
327.3 (2.21) 878.7 (2.72)
230.8 (2.38)
224.0 (2.25)
555.4 (3.18)
505.9 (2.74)
247.8 (2.14) 164.4 (1.67) 248.9 (1.10) 295.4 (1.88) 611.5 (3.69)
SMSA 9
FPSHARE
- 86.90 (-1.14)
-91.99 (-0.77) 351.2 (1.32)
144.9 (1.99)
- 69.85 (-0.85)
76.81 (0.57) 200.7 (1.53)
396.5 (2.80)
- 34.14 (-0.48) - 4.594 ( - 0.02) 11.34 (0.11)
40.23 (0.43)
HOSP 10
182.5 (1.49)
- 17.74 (-0.75)
526.3 (2.85) 953.0 (2.33)
70.69 (0.62)
545.3 (2.53) 862.7 (4.04) 432.2 (2.09) 108.8 (0.89)
329.6 (2.26) 67.56 (0.65) 981.7 (3.46) 389.3 (2.36)
HOSP 12
RELATIVEOCC
106.2 (4.36) 281.9 (7.19) 666.8 (7.40)
135.1 (4.77)
330.5 (7.43) 331.7 (7.67)
179.8 (6.09) 131.6 (5.94) 396.5 (6.45) 216.2 (6.60) 267.1 (5.78)
HOSP 11
FPDUM
-.-
--.
-“_
-
-
r_
.”
is substituted for CONAGE. All other coefficients in the table are from the CONAGE
- 0.0388 (-0.03)
11.89 (3.11)
2.156 (1.24)
2.355 t 2.04) 4.405 (3.36)
2.752 (1.33)
5.707 (2.64) 2.625 (1.27)
1.571 (1.63) 7.062 (2.49) 3.211 (2.02)
3.998 (2.69)
HOSP 6
% popul. in HMO
ToeJRcients on the CON dummy come from a regression where CONDUM regressions.
Expense/admission
Fracture of neck and femur
Hyperplasia of prostate
Diverticula of intestine
7.276 (1.08) 29.36 (2.05)
28.41 (1.18)
3.490
17.81 (2.17) 16.48 (2.10)
Pneumonia
Respiratory system disease, other Jnguinal hernia
129.3 (2.95) 116.7 (2.78)
11.17 (1.95) 8.983 (1.59)
13.80 (1.83)
Acute, cerebrovascular disease
107.6 (2.84)
Acute myocardial infarction Congestive heart failure
5.842 (2.11) 22.12 (2.92) 6.929 (1.58)
5.919 (1.32) 49.00 (2.93) 19.08 (2.72)
72.90 (2.60) 18.99 (0.78) 256.1 (4.59)
Cataract surgery
3.962 (1.05)
11.63 (2.22)
HOSP 4
SMSA 3
HOSP 2
SMSA 1
Diabetes mellitus
Disease category
CON DUMMY*
CONAGE
Table 2 Coefficients on variables affecting competition (t-statistics in parentheses).
M. Noether, Competition among hospitals
277
by these regressions where both prices and expenses appear higher, since non-profit hospitals cannot report monetary profits, their collusion may lead to an increase in expense-generating non-pecuniary benefits. The results are therefore consistent with the cartel theory or the wasteful reallocation scenario. As with concentration, the linear specification of entry regulation age may be inappropriate if the effect of a CON law is not noticeable until it is wellestablished. An entry dummy is defined to equal one only when the CON law was at least three years old (in 1977) Bnd substituted for the linear age of the CON law variable. Its coefficients are shown in columns 3 and 4 of table 2. A CON law that is at least three years old is significantly associated with both higher prices and higher expenses. Since prices are higher by a greater percentage than expenses,32margins appear to be slightly higher, and CON laws may therefore restrict entry. Still, it appears that CON’s strongest effect is that it creates cost-raising inefftciencies which are passed on in higher prices. Interestingly, high concentration and a wellestablished CON law appear to have effects on expenses that are identical but opposite in sign. This suggests that when both conditions exist, the reduction in quality competition just offsets the lessening of efficiency. In this case, expenses are unchanged from the state where neither condition is present. Since prices are higher, however, margins have increased.” A variable that measures the percent of the population belonging to HMOs in 1977 is used along with another that measures the change in HMO market share from 1977 to 1978 to attempt to control for the endogeneity. The positive coefficients on the variable measuring the percent of the population belonging to HMOs, suggest that HMOs may enter costlier markets tirstJ4 On the other hand, the significantly negative effect of the change in HMO market share from 1977 to 1978 implies that their entry 32Prices are on average 4.0 percent higher at the hospital level with a range of 2 percent for diverticula of the intestine to 6.2 percent for acute myocardial infarction, and 4.9 percent higher at the SMSA level with a range of 1.5 to 9.7 percent for the same disease categories producing the extreme values. Expenses are 3.3 percent higher at the hospital level and 4.0 percent higher at the SMSA level. “The effect of a direct interaction between concentration and CON regulation was tested. Various formulations of interaction variables were used to test whether the existence of an entry restriction that would prevent the erosion of supracompetitive prices was necessary for high concentration to have an effect. None of the interaction variables were significant. A CON regulation variable was alternatively defined to equal the actual age of the CON law above the dummy variable cutoff of 3 and 0 below. Since none of these alternative specifications proved superior to those discussed above, their results are not reported. 341”hepositive coefkient may be explained partially by the geographic location of HMOs. Inclusion of regional dummy variables eliminates the significance of the HMO share variable whi!~the west coast dummy variable is highly significant. Since many HMOs are located in California, particularly during the sample period, this is not surprising. While HMOs existedin 64 of 249 SMSAs in 1977 and were spread across over 30 states, their largest shares were found in the West Coast States.
278
M. Noether, Competition among hospitals
subsequently leads to lower costs and prices, perhaps because they engender increased price competition. Two types of variables are used to measure the effects of ownership structure on hospital prices and expenses. First, in all the regressions, the effect of the share of beds operated by for-profit hospitals in the SMSA is measured. In addition, in the individual hospital regressions, dummy variables indicate whether a particular hospital observation is for-profit. The share of proprietary hospitals shows a uniformly positive and often significant coefftcient in the price regressions at the SMSA level and a generally insignificant (but also positive) effect at the individual hospital level. The expense regressions indicate no consistent or significant relation between forprofit market share and hospital costs. These results suggest that increased price competition is not the primary result of for-profit entry. For-profit hospitals appear either to enter into already profitable markets or they are more profit maximization oriented. The ownership dummy variables in the individual hospital regressions provide additional information useful in distinguishing the two theories. For-profit hospital status measured by a dummy variable has a highly significant positive impact on price and an insignificantly negative effect on costs. These results suggest that proprietary hospitals charge prices that are, on average 13 percent higher than those of non-profit hospitals. This result confirms previous work showing that forprofit hospitals generate larger margins. It also implies that the positive coeticient on for-profit market share is attributable not only to the fact that proprietary hospitals enter the most profitable markets, the effect of which shouri be measured entirely by the for-profit market share available, but also to their own pricing policies. The relative occupancy rate of a hospital has the expected positive and significant effect on prices in most cases. Holding constant the hospital’s own occupancy rate which presumably affects costs, the higher its occupancy relative to the average in the local market, the more power it appears to have over price. The relative insignificance and lower magnitude of relative occupancy’s effect on expenses suggests that it is associated with higher margins as well. 5. Conclusion
This research shares with other work in the field the inability to measure quality accurately, and it is therefore difficult to isolate price competition. The empirical work in this study, however, suggests that hospital market structure in 1977-1978 affected both price and quality competition in the hospital industry, although the magnitude of any effect of concentration is small. CON-type entry regulations are important as well. It appears that they are primarily associated with higher costs because of inefficient resource
M. Noether, Competition among hospitals
279
allocation. Since the highest margins seem to be earned in markets with high concentration and well-established CON laws, CON laws may also serve as entry barriers. HMOs appear to enter costlier markets, but their growth is associated with falling prices. This study provides no evidence to suggest that for-profit hospitals have increased price competition in the hospital industry. Instead, for-profit hospitals charge substantially higher prices. Since, until recently, it appears that the only major constraint on prices was that imposed by the necessity to raise private insurance premiums to cover cost increases, it is not surprising that hospitals focused many of their competitive efforts on quality dimensions. Substantial increases occurred during the 1970s as medical care consumed an increasing portion of GNP. In recent years it appears that sensitivity to price is rising, as corporations have taken steps to limit growth in the proportion of their expenses attributable to subsidization of employee health benefits. In efforts to contain costs, many plans have raised their coinsurance rates, and changed the relative reimbursement rates between in-patient and ambulatory surgery to create incentives for greater utilization of the latter, cheaper alternative. Medicare’s Diagnosis Related Group (DRG) system of fixed payment for 470 disease categories has created strong incentive to reduce quality competition. Moreover, new types of arrangements have developed such as Preferred Provider Organizations (PPOs), and HMOs are now much more pervasive and well established. There are various indications that hospitals now compete more on price than they did in the past. The trade press has reported on several instances where urban hospitals have reduced their room and board rates, as business and consumer coalitions in several cities have published information on local hospital prices. Several states have authorized the negotiation of fixed fee contracts between insurance companies and doctors and hospitals [Pierce (1985)]. This evidence suggests that insensitivity to price, due to extensive third party coverage and lack of information, may be diminishing. If this trend continues, it would not be surprising if hospitals increasingly focus their competition on price dimensions and become less concerned with quality, if consumers so desire. This study of hospital competition in 1977-1978 should provide a useful benchmark by which to measure industry changes.
Regulation variable; Age (in 1977) of CON law in state Dummy variable denoting CONAGE 23 Dummy variable indicating state covered by Section 1122 law Dummy variable indicating existence of mandatory prospective reimbursement in state Dummy variable indicating existence of voluntary prospective reimbursement in state
Market structure and ownership variables Herfindahl statistic, computed by hospital bed capacity Dummy variable denoting Herfindahl $0.3 Share of beds owned by for-profit hosprtals Share of beds owned by public hospitala Share of beds in hospitals managed by a third party Share of beds in hospitals that are part of multi-hospital system Dummy variable denoting individual for-profit hospital Dummy variable denoting public hospital Dummy variable denoting outside managed hospital Dummy variable denoting hospital that is part ot system Nursing home bed - population ratio Percentage of SMSA population who are HMO members Change in HMOMEM from 1977 to 1978
Description
Table A.1 Variable list.
4.05 0.56 0.74 0.30 0.16
3.36 0.50 0.42 0.42 0.41 0.21
0.15 0.13 0.10 0.18 0.02 0.25 0.17 0.18 0.03 0.22 0.93 2.73 2.54
0.36
::
3.99 0.50
0.14 0.34 0.14 0.17 0.05 0.18 0.37 0.38 0.18 0.41 3.08 7.41 11.54
Hospital level Mean S.d
3.13 0.48 0.77 0.22
Y75 4.51 7.69
0.73 1.02 1.60
..a
0.20 0.49 0.12 0.26 0.08 0.22
...
Sd.
... ...
.. *
0.30 0.41 0.07 0.21 0.02 0.21
Mean
SMSA level
variables
Nominal wages of hospital industry workers Share of beds in hospitals that are members of the Council of Teaching Hospitals (SMSA regressions); dummy variable denoting hospital membership in COTH (hospital regressions) Average number of beds (SMSA regressions); particular hospital’s number of beds (hospital regressions) Individual hospital size relative to average hospital size in SMSA A dummy variable indicating those SMSAs with population exceeding one million Average occupancy rate of SMSA or hospital (depending on regression) Individual hospital occupancy relative to average hospital occupancy in SMSA Average length of stay for SMSA or hospital (In price regressions, ALS is specific to disease category.)
Cost variables
Per capita income (nominal S) Proportion of population with private insurance coverage, net of duplicate coverage, measured at state level Percentage of population enrolled in Part A (hospital insurance) of Medicare Percentage of population on welfare Percentage of population who are unemployed Proportion of population who are white Population density Percentage increase in population between 1970 and 1978 Death rate
Demand
... 0.34 6.57 ... 1.07
... 0.13 74.45 ... 8.61
7.52
0.97
72.27
0.99 0.48
253
90
250
1.97
0.16
13.32
0.73 0.50
203
2,027 0.31
8.08 1.86 0.10 1,608 11.53 0.19
5.43 7.08 0.83 937 8.70 0.87
9,780 0.11
4.68 2.06 0.10 939 11.49 0.21
3.35 6.83 0.86 424 10.51 0.86
3.04
10.40
1,253 0.20
3.52
10.43
745 0.06
0.78
6,009
9,464 0.13
759 0.06
5,623 0.79
Diabetes mellitus Cataract (w/ surgery) Acute myocardial infarction Congestive heart failure Acute, but ill-defined cerebrovascular disease Pneumonia, unspecified Other diseases of the respiratory system (pulmonary collapse, acute edema) - Inguinal hernia (w/ surgery) - Diverticula of intestine - Hyperplasia of prostate (w/ surgery) - Fracture of neck and femur (w/ surgery) expenses/total adjusted (for outpatient) adrrissions
1,173 996 1,818 3,663 1,191
1,371 933 2,637 1,563 1,975 1,801 1,809
Mean
224 265 362 787 286
352 203 593 391 508 447 462
S.d.
SMSA level
1,341 1,122 2,115 4,249 1,299
1,592 1,060 2,957 1,766 2,294 2,066 2,056
Mean
397 408 659 1,356 532
E 763 766
623 317 998
S.d
Hospital level
“Federal hospitals are not included in the data set because of their restricted clientele. Note: As is apparent from the table, for some variables the SMSA level mean varies considerably from the hospital level mean. For example, the concentration measures are higher at the SMSA level. This is due to different weights placed in calculating grand means. While the SMSA level values are calculated as weighted averages of all the hospitals in the SMSA, in the ‘grand mean’ each SMSA is given equal weight. At the hospital level, since each hospital is weighted equally in calculating the mean, SMSAs with more hospitals receive a greater weight. Since those SMSAs have lower concentration, the average concentration level appears lower. Similarly, average prices at the hospital level appear higher, because hospitals in large urban areas are !ikely to offer a more sophisticated, higher priced product. Since such hospitals are not evenly distributed across SMSAs, they do not receive as much weight in the overall SMSA averages.
Price Price Price Price Total
Price Price Price Price Price Price Price
Dependent variables
Description
Table A.1 (continued)
M. Noether, Competition among hospitais
283
References Becker, Edmond R. and Frank A. Sloan, 1985, Hospital ownership and performance, Economic Inquiry 23,21-36. Davis, Karen, 1971, Relationship of hospital prices to costs, Applied Economics 4, 115-125. Davis, Karen, 1974, The role of technology, demand and labor markets in the determination of hospital costs, in: Mark Perlman, ed., The economics of health and medical care (New York). Elzinga, Kenneth G. and Thomas F. Hogarty, 1973, The problem of geographic market delineation in antimerger suits, Antitrust Bulletin 18,45-81. Eskoz, Robin and Michael K. Peddecord, 1985, The relationship of hospital ownership and service composition to hospital charges, Health Care Financing Review 6,51-58. Fama, Eugene and Michael Jensen, 1983a, Agency problems and residual claims, Journal of Law and Economics 26,327-350. Fama, Eugene and Michael Jensen, 1983b, Separation of ownership and control, Journal of Law and Economics 26,301-326. Fama, Eugene and Michael Jensen, 1985, Organizational forms and investment decisions, Journal of Financial Economics 14, 101-120. Farley, Dean E., 1985, Competition among hospitals: Market structure and its relation to uilization, costs and financial position, Research note, Hospital Studies Program, DHHS pub. no. (PHS)85-3353. U.S. Dept. of Health and Human Services, National Center for Health Services Research and Health Care Technology Assessment. Frank, Richard G. and W.P. Welch, 1985, The competitive effects of HMOs: A review of the evidence, Inquiry 22, 148-161. Friedman, Milton, 1962, Occupational licensure, in: Capitalism and freedom (University of Chicago Press, Chicago) 137-160. Hyde, D.R., P., Wolff, A. Gross and E.L., Hoffman, 1954, The American Medical Association: Power, purpose, and politics in organized medicine, Yale Law Journal 63,938-1022. Joskow, Paul L., 1980, The effects of competition and regulation on hospital bed supply and the reservation quality of the hospital, Bell Journal of Economics 11,421+I7. Joskow, Paul L., 1981, Controlling hospital costs: The role of government regulation (MIT Press, Cambridge, MA). Kessel, Reuben A., 1958, Price discrimination in medicine, Journal of Law and Economics 1, 20-53. Letller, Keith B., 1982, Ambiguous changes in product quality, American Economic Review 72, 956-967. Lehvari, David and Yoram Peles, 1973, Market structure and durability, Bell Journal of Economics 4,235248. Luft, Harold S., 1981, Health maintenance organizations: Dimensions of performance (Wiley, New York). Luft, Harold, James Robinson, Deborah Garnick, Susan Maerki and Stephen McPhee, 1986, The role of specialized clinical services in competition among hospitals, Inquiry 23, 83-94. Noether, Monica, 1986a, The effect of government policy changes on the supply of physicians: Expansion of a competitive fringe, Journal of Law and Economics 29,231-262. Noether, Monica, 1986b, The growing supply of physicians: Has the market become more competitive?, Journal of Labor Economics 4, 503-537. Noether, Monica, 1987, Competition among hospitals (Federal Trade Commission, Washington, DC). Ordover, Janusz and Robert Willig, 1982, The 1982 Department of Justice merger guidelines: An economic assessment, California Law Review 71, 535. Pauly, Mark, 1978, Medical staff characteristics and hospital costs, Journal of Human Resources 13, 77-11. Pauly, Mark, 1980, Doctors and their workshops: Economic models of physician behavior (University of Chicago Press, Chicago, IL). Pauly, Mark, 1987, Nonprofit firms in medical markets, American Economic Review 77, 257-262. Pierce, Bob, 1985, Health insurance on the statehouse floor: 1985 projections, Hospitals 52, no. 3, 52-58.
284
M. Noether, Competition amon,o hospitals
posner, Richard A., 1974, Certificates of need for health care facilities: A dissenting view, in: Regulating health facilities construction, Chirk Havighurst, ed. (American Enterprise Institute, Washington, DC). Robinson, James and Harold Luft, 1985, The impact of hospital market structure on patient volume, average length of stay, and the cost of care, Journal of Health Economics 4, 333-356. Salkever, David, 1978, Competition among hospitals, in: Competition in the health care sector, Warren Greenberg, ed. (Federal Trade Commission, Washington, DC) 191-206. Satterthwaite, Mark A., 1979, Consumer information, equilibrium industry price and the number of sellers, Bell Journal of Economics 10,483S&l. Schetlinan, David and Pablo Spiller, 1987, Geographic market definition under the U.S. Department of Justice merger guidelines, Journal of Law and Economics 30,123-147. Scott, W. Richard and Ann Barry Flood, 1985, Cost and quality of hospital care: A review of the literature, Medical Care Review 23,213-261. Watts, C.A., 1976, A managerial discretion model for hospitals, unpublished Ph.D. dissertation (Johns Hopkins University, Baltimore, MD). Wennberg, Jack, 1980, A small area approach to the analysis of health system performance, in: Health Planning Methods and Technology Series, no. 21, HRP-0102101, Department of Health and Human Services pub. no. (HRA) 80-14012.