Emily L. Wang ‘11

Harvard University, Department of Economics

Health care costs in Massachusetts are increasing much faster than the growth of the state economy. An investigative report by the Attorney General Office finds that prices paid by commercial health insurers to hospitals vary significantly amongst providers offering similar levels of service, and that such price variations are correlated only with market leverage. In this thesis, I examine several theories of why prices vary so much among MA hospitals. I identify five exogenous variables that explain 56 percent of the variation in prices. I discuss policy implications of these findings.


Health care is Massachusetts’s largest industry and a major driver of the state’s economy. Massachusetts is first in hospital access in the nation, providing insurance coverage for 97% of its residents, and ranks in the top ten in quality of care provided. Massachusetts’s health insurers are also consistently rated in the top ten best insurers nationally. Yet Massachusetts hospital spending has always been higher than the national average, reaching 55% per person above the US average in 2008. Acute hospital costs per person in the state are the highest in the nation, at $3,015 per person in 2007. [1]

Proposed reasons for Massachusetts’s high hospital costs are: a large and increasing share of total patient care given by costly teaching hospitals, high surgery rates, high number of MD’s (the nation’s highest physician-to-patient ratio), and the political influence of the large teaching hospitals. Such teaching hospitals are vital to the state’s economy and spend large amounts of money on lobbying.

A new factor

On March 16, 2010, the Massachusetts Attorney General’s Office released a Report for Annual Public Hearing, Examination of Health Care Cost Trends and Cost Drivers, that identified a new factor contributing to high hospital costs—price differentials. The investigation finds that prices paid by commercial health insurers to hospitals and physician groups vary significantly within the same geographic area and amongst providers offering similar levels of service. The AGO claims that price variations are not correlated with (1) quality of care, (2) the sickness of the population served or complexity of the services provided, (3) the extent to which a provider cares for a large portion of patients on Medicare or Medicaid, (4) whether a provider is an academic teaching and research hospital, or (5) differences in hospital costs of delivering similar services at similar facilities. Rather, Attorney General Martha Coakley argues that price variations are only correlated to market leverage as measured by the relative market position of the hospital compared to other hospitals within a geographic area. The AGO defines “leverage” as the ability of the provider to influence the insurer during contract negotiations, calculating leverage as the total revenue paid by an insurer to hospitals within one provider system. Coakley warns that higher-price hospitals are gaining market share at the expense of lower-priced hospitals. She asserts, “The commercial health care marketplace has been distorted by contracting practices that reinforce and perpetuate disparities in pricing.”

The AGO report presents a rare opportunity to view subpoenaed data containing relative payments negotiated between providers and insurers that are normally kept hidden from the public eye. Understanding what drives price differentials among hospitals in Massachusetts is important to check the increase in future health care costs, and is made possible by these data. The legislative mandate for the AGO’s investigation was spurred by the fact that health care costs in the state are increasing much faster than the growth in the economy, GDP, and wages. If unchecked, this increase would threaten the future viability of universal health care access and Massachusetts’s role as a national leader in health care reform.

Explaining price differentials in the Massachusetts commercial insurance market will also be beneficial towards understanding price differentials among hospitals in other states across the US. In an interview, Thomas Glynn, Chief Operating Officer at Partners HealthCare in Boston from 1996 to 2010, states, “When you read the Attorney General’s report, you get the impression that [price variations] are only in Massachusetts, where in fact they are in other states too.” He notes that major price disparities among hospitals in the New Hampshire market rebuts the notion that the issue of rate differentials is somehow peculiar to Massachusetts. Just as hospitals within the dominant Partners system charge the highest prices for many services, hospitals within the dominant Dartmouth-Hitchcock system in New Hampshire also charge the highest prices for many services in the state. [2]

How prices are set

The AGO report brings some transparency to the public as to how prices for health care are set. Prices paid to providers are the result of many discrete negotiations. Each insurer negotiates a price for every service with each hospital. These negotiations result in disparate prices paid for the same or similar type of service between hospitals.

Effect of demand elasticity on equilibrium price. (Image edited from www.bized.co.uk.)

There has been extensive debate over whether prices in the health care market can be modeled by the intersection of supply and demand curves, and if so, how they are set. In this paper, I will address possible factors that can affect the price equilibrium by altering the intersection of these curves. These factors can either work by changing the elasticity of the curves, or by causing shifts in the curves.

For example, we can trace the higher price negotiated by a provider with an insurer to the higher demand by consumers for that same provider (Figure 1). Initially, the demand curve D for the provider intersects with the supply curve S1 at equilibrium price P0. Because of the heightened reputation of the hospital, consumers demand its services more, and so the demand curve becomes more inelastic, and is now represented by D1. Since there is more demand for its services, a provider may now want to increase costs to build more high-technology services or to train more staff. These increases in cost are represented by a shift back in the in the supply curve from S1 to S2, which in turn translates into a higher equilibrium price P2 in the marketplace.

Figure 1 can also be used to compare a hospital with a more elastic curve D to a hospital with a more inelastic curve D1. If costs increase and there is a upward shift in the supply curve, the hospital with the more inelastic demand curve will have a higher equilibrium price P2 than the hospital with the more elastic demand curve and equilibrium price P1.

Rebuttals to the AGO report

The AGO’s finding that high and increasing hospital costs in Massachusetts are driven largely by hospitals’ market dominance, not by high utilization or the quality or cost of providing care, is contested in a report commissioned by Partners HealthCare, and a second report commissioned by the American Hospital Association.

Partners’ rebuttal finds (1) general conceptual and methodological issues in the report, (2) problems with its market share analyses, and (3) questions in its treatment of quality measures.  It states that the AGO uses simplistic methodology to reach its conclusions, and that the major reason that hospital costs have escalated in Massachusetts is because hospital quality has improved, not because powerful providers have demanded higher prices. It cites as a major methodological issue the report’s use of univariate analyses to examine relationships of price to quality, case mix, teaching status, Medicare/Medicaid census, and costs.

The AHA rebuttal focuses on the Attorney General’s claim that negotiated commercial rates for hospital services at large provider organizations are substantially higher than those for other organizations, and that observed differences are due predominantly to market power/leverage. They note that “market power” should be defined to specifically mean the anticompetitive exercise of power to set prices that reduces consumer welfare, as the term is used by antitrust agencies and defined by the courts. They find that the AGO report uses market power claims without regard to market area or geographic analyses, including assessment of alternatives available. They stress that anticompetitive market power should not be confused with consumer preferences for certain providers, highly differentiated services, or specialized services. A hospital can become a highly desired provider simply by being perceived to provide excellent and state-of-the-art services, and strong consumer preferences for specific hospitals provide an incentive for them to improve services, enhance quality, or expand output of services in great demand, and to expect an appropriate return on the investments required to provide these services.

Theoretical framework

In this paper, I examine why prices for similar services vary so much among hospitals in Massachusetts. There are several plausible reasons, and it is important to test them in light of the ongoing debate about the causes of high hospital costs in this state. The Attorney General Office’s report, while exposing a surprising disparity in prices paid by commercial insurers to different hospitals for the same services, does not present a rigorous economic framework to explain differential pricing among providers. The critiques of the AGO report by Partners and the AHA, while illuminating of major hospitals’ justifications for their higher prices, are likely to be biased since they were commissioned by the hospitals themselves.

I will use more extensive multivariate modeling than the Partners and AHA reports (2010). Like Cutler and Sheiner (1999), I will take into account the effects of demographic and hospital characteristics in determining price. I note that use of total amount of revenue paid by an insurer to a provider system as a measurement of market leverage is endogenous to price, and thus cannot be used to explain price differentials. Instead, I will explore the effects of exogenous variables on price in my regression model. In particular, I will propose measurements of market competition that are based on total admissions. By using such predicted flows, in the form of an estimated demand model, rather than actual flows, I will attempt to avoid the endogeneity problem.

The following are several different theories that I propose to explain these price differentials, and that I will test in my models.

Market competition—two competing hypotheses:

A) Negative correlation between market competition and prices

For commercial insurers, the higher-spending markets are those that are the least competitive on the provider side. Small markets with only one or a few big hospital systems negotiate higher prices.

B) Positive correlation between market competition and prices

When a market has a lot of hospitals, an “arms race” mentality can drive up costs as they vie to buy the best new technology, attract doctors, and win patients, who because of insurance are indifferent to costs. The failure of an insurer to contract with a large provider organization would cause serious network disruption, not only because a large percentage of their members would be forced to seek care elsewhere, but because employers and others are less interested in purchasing products that do not include the dominant providers. This is because larger providers who have built up greater facilities, services, and “brand name” than other smaller hospitals, become “must-haves.” As Berenson et al. note, “Must-have hospitals have market leverage over health plans, because plans cannot plausibly threaten to exclude them….[They] must be included in a plan’s provider network to make the plan acceptable to customers.” The demand curve thus becomes more inelastic.

Characteristic market power

Teaching or community hospital status will affect the prices negotiated by a hospital. Previous literature says that teaching hospitals are likely to negotiate higher prices than their community counterparts, as they must bear the cost of training programs, they have more staff to pay salaries to, more overhead costs, and they pay physicians money for teaching, not only for practice. This would cause the supply curve to shift upward, and thus for the supply and demand curve to intersect at a higher price.

Quality of care

One would expect to see a positive correlation between price and quality of care in a rational, value-based health market. Patients demand better quality of care so there is an outward shift in the demand curve. This may also have the effect of making the demand curve become more inelastic, as large movements in quantity would only cause small movements in price. Both situations have the effect of causing the supply and demand curves intersect at a higher price.  Furthermore, better care indicates higher costs in the form of more physicians, more nurses, more specialists, and more testing.

Leverage theory/essential provider

A hospital with a larger array of unique services offered will be more likely to charge higher prices. This is because it will be one of only a few hospitals that patients can go to for a specific treatment.  Market demand will become more inelastic.

Brand name

A positive correlation between prices and brand name would be expected. If consumers subjectively believe that a hospital is a “superior” hospital to go to, demand and thus prices will rise. Patient preference reduces the elasticity of demand and keeps prices up. Reputation could also lead to an outward shift in the demand curve, which would cause the equilibrium price to rise as will.

Income of area

If a hospital is located in an area that has well-insured patients, they may be able to charge more because they know that their patients will be able to afford the bill. Conversely, an area with higher income would have more patients demanding health care since they can afford it. Thus, higher income represents an outward shift in the demand curve, and hence a rise in prices. We would expect prices to be positively correlated with the income of the hospital’s surrounding area.


One would expect that prices would be positively correlated with the percentage of patient charges attributed to Medicare, Medicaid, other government payers, and/or free care due to cost-shifting. Disproportionate share hospitals may charge higher prices since Medicare and Medicaid pay less than private insurance, and the hospital needs to get reimbursed for more than simply what the government pays.


The data comes from a variety of sources. The Massachusetts Division of Health Care Finance and Policy, in collaboration with the Attorney General’s Office, collected written testimony from Tufts Health Plan, Blue Cross Blue Shield, and Harvard Pilgrim Health Care. All three insurers provided data on relative differences in hospital prices in 2008. All three indices are created by taking a market basket of inpatient and outpatient services, and computing the total charge for these services at each hospital. These measures of relative payments are all adjusted by the Case Mix Index (CMI), or the severity of the cases. The sample size was 198. [3]

The National Bureau of Economic Research provided American Hospital Association (AHA) data on services offered by each hospital in 2008. The AHA 2008 Annual Survey of Hospitals lists hospital system affiliation. [4] The Massachusetts Division of Health Care Finance and Policy provided data on total admissions, teaching or community status, and the number of Medicare and Medicaid admissions at each hospital.

Quality of care data comes from the US Department of Health and Human Services Medicare Hospital Compare Website. Hospital “Outcome of Care Measures” show what happened after patients with certain conditions received hospital care. The death rates are the 30-day risk-adjusted death rates for heart attack, heart failure, and pneumonia; in other words, whether patients died within 30 days of hospitalization. The rates of readmission show how often patients are readmitted within 30 days of discharge from a previous hospital stay for heart attack, heart failure or pneumonia. Death rates and rates of readmission show whether a hospital is doing its best to prevent complications, teach patients about their condition at discharge, and ensure that patients make a smooth transition to their home or another setting such as a nursing home.

The Current Population Survey provided data on the median income of each zip code in Massachusetts in 2008.


I will test the effect of various hospital and market characteristics on price. Typically, major health plans and hospitals negotiate prices for inpatient health care services using a base case rate. The base case rate is controlled for the severity and complexity of each case. Additional prices are negotiated for a limited set of other inpatient services such as very high-cost or experimental procedures. For outpatient services, health plans have set standard fee schedules for all outpatient services (e.g. standard fees are set for radiology, laboratory work, observation, behavioral health, etc).  The plans and hospitals negotiate a specific multiplier to each of these standard fees; for example, a provider with a 1.2 multiplier for radiology services would be paid 120% of the standard fee schedule rate for covered radiology services.

PRICE is a variable that measures the relative price paid to each hospital for a comprehensive market basket of both inpatient and outpatient services. In other words, relative price is the variation in the contractually negotiated amount that an insurer agrees to pay each provider in their network, as compared to the network-wide average. The health plans calculated this “payment relativity factor” controlling for differentiating factors such as volume, product mix, service mix (complexity), and case mix (acuity). It is thus possible to compare the pure “price” that insurers negotiate with different hospitals for all hospital inpatient and outpatient services.

I begin with a set of univariate models that model the effect of BRAND, SERVICES, and TEACHING on PRICE. These independent variables are reasonable to start with since it is likely that the “name brand” of the hospital would heavily influence demand, as would the number of specialty services provided at the hospital, since people would have nowhere else to go for such services. The teaching or community hospital status of a provider should also affect price since teaching hospitals attempt to differentiate themselves from community hospitals by attracting nationally renowned faculty specialists who can foster recognition for the hospital, leading to higher demand by consumers. Teaching hospitals are also costlier to operate, and consumers must bear the cost of the training programs.

The models are as follows:




BRAND is a variable that describes the subjective consumer “brand” perceptions of the hospital. This variable is constructed by taking the number of times a hospital is ranked in the top 50 hospitals in the US News and World Report for any of 16 specialties. These specialties include Cancer, Diabetes & Endocrinology, Ear, Nose & Throat, Gastroenterology, Geriatrics, Gynecology, Heart & Heart Surgery, Kidney Disorders, Neurology & Neurosurgery, Ophthalmology, Orthopedics, Psychiatry, Pulmonology, Rehabilitation, Rheumatology, and Urology.

SERVICES describes the specialty service lines offered by the hospital. I constructed SERVICES by counting the number of services a hospital offers from a selection of 14 different services. The services were chosen because they are advanced high-technology services that only large hospitals are likely to have. They include: Cardiac Catheterization Lab, Extracorporeal Shock-Wave Lithotripter, Full-Field Digital Mammography, Image-Guided Radiation Therapy Hospital, Intensity-Modulated Radiation Therapy, MRI, Open Heart/Cardiac Surgery, Proton Emission Therapy, Heart Transplant, Liver Transplant, Tissue Transplant, Kidney Transplant, Lung Transplant, and Other Transplant.

The above services represent a group of the most advanced, expensive, resource-intensive services offered. They require highly trained doctors, nurses and technicians and would not be economically feasible unless there was a large volume of patients using the services. Heart transplantation, for example, is only available in a few hospitals in Massachusetts. Such advanced and expensive services are used by hospitals to distinguish themselves from other hospitals in a competitive market.

The sum of services was represented by the z-statistic, SPECIALTIES. In addition, I looked at the number of neonatal intensive care beds (NEOINTENSIVE), neonatal intermediate care beds (NEOINTERMEDIATE), pediatric cardiac beds (PEDCARD), and cardiac intensive care beds (CARDINTENSIVE) that each hospital had. I converted these variables into z-statistics and then averaged across the z-statistics to compute SPECIALBEDS. I averaged the above two z-statistics, SPECIALTIES and SPECIALBEDS to arrive at SERVICES.


SERVICES thus captures the leverage of the hospital as measured by whether the hospital offers certain specialty service lines. I used two ways of measuring TEACHING that have comparable results: [5]

a) By setting the independent variable to be 1 if the hospital is a teaching hospital and 0 if the hospital is a community hospital.  The following 15 hospitals are assigned a “1” as they are listed by the DHCFP as being teaching hospitals: [6] Baystate Medical Center, Beth Israel Deaconess, Boston Medical Center, Brigham and Women’s, Cambridge, Caritas St Elizabeth’s, Children’s Hospital, Dana Farber Cancer Institute, Lahey Clinic Medical Center, MEEI, Mass General Hospital, Mount Auburn, St. Vincent’s, Tufts Medical Center, and UMass Memorial Medical Center.

b) By assigning the variable to be the sum of the total number of interns and fellows in each hospital. [7]

THP and BCBS are the insurer dummy variables. For THP, I assigned a value of 1 for the Tufts Health Plan relative payments and a 0 for Blue Cross Blue Shield and Harvard Pilgrim Health Care relative payments. For BCBS, I assigned a value of 1 for Blue Cross Blue Shield relative payments and a 0 for Tufts Health Plan and Harvard Pilgrim Health Care relative payments.

In each of the initial univariate regressions (Table 1 of Supplemental Figures), the individual variables BRAND, TEACHING, and SERVICES are significant predictors of price. The p-value is less than 0.01 for BRAND, TEACHING (fellows and residents), and SERVICES, indicating statistical significance to the 1% level.  The model with TEACHING (0/1 coding) has a p-value of <0.1, indicating statistical significance to the 10% level.

It is imperative, however, to check the independent variables for multicollinearity (Table 2 in the Supplemental Figures). The results draw attention to the fairly high correlations between the regressors BRAND, TEACHING, and SERVICES.

1) BRAND and SERVICES (r = .580)

2) TEACHING and SERVICES (r = 0.666)

3) BRAND and TEACHING (r = 0.480)

This suggests that the theories they measure are capturing the same effects in the regression models.  To further confirm this, I paired the variables off to test what effect combinations of two of these three variables would have on price.

The next models tested are:




I then construct another model in which all three variables were included:


See Table 3 in the Supplemental Figures for tests of multicollinearity. In each of the regressions, one or both of the variables become insignificant. This strongly suggests that the three variables are highly correlated, and that it is thus necessary to drop two out of these three independent variables from the final regression. Because SERVICES remains consistently significant across all regressions, it is kept in the final model while BRAND and TEACHING are dropped.

The percentage of patients on Medicare and Medicaid are also likely to be factors of interest as they may determine price through cost-shifting. I tested the effect of these percentages on PRICE:


The Massachusetts Division of Healthcare Finance and Policy provided information on the number of Medicare and Medicaid Managed Care Admissions and Medicare and Medicaid Non-Managed Care Admissions at each hospital.

MEDICARE is equal to the share of total admissions that belong to total Medicare admissions. MEDICAID is equal to the share of total admissions that belong to total Medicaid admissions. See Table 4 in the Supplemental Figures for the effect of the percentage of Medicare and Medicaid patients on price. Since MEDICARE and MEDICAID are statistically significant predictors of price (p < 0.01 for MEDICARE and p < 0.05 for MEDICAID), I reasoned that these also should be put in the final regression model.

We would also expect Quality of Care to have a significant impact on prices. The six quality of care measures reported by the Medicare Compare website are: the Hospital 30-Day Mortality Rates for Heart Attack (MORTHEART), Hospital 30-Day Readmission Rates for Heart Attack (READMHEART), Hospital 30-Day Mortality Rates for Heart Failure (MORTFAIL), Hospital 30-Day Readmission Rates for Heart Failure (READMFAIL), Hospital 30-Day Mortality Rates for Pneumonia (MORTPNEU), Hospital 30-Day Readmission Rates for Pneumonia (READMPNEU). Note that a higher QOC value indicates a worse score.

The diagram depicts the target hospital surrounded by its competitors.

  • Let our target hospital (square) have 400 beds.
  • Let a hospital with more than 500 beds be a pentagon.
  • Let a hospital with between 300-500 beds be indicated by a right triangle.
  • Let a hospital with less than 300 beds be indicated by a diamond.
  • Let a hospital outside the 15-mile radius of influence be indicated by an X.
  • All hospitals within the target’s own system are removed.

The QOC variable is constructed by taking each individual quality of care variable and transforming it into a z-statistic, and then averaging across the z-statistics for the overall measure to be used in the regressions. QOCDUMMY is a dummy variable that identifies those hospitals that had no reported quality of care measures by the Medicare Compare website.

Finally, we would also like to see what the effect of market competition has on prices, as this was the entire premise of the Attorney General’s Report. I formed several measures of market competition for the purposes of analysis.

BASIC Measure of Market Competition

The Target Hospital faces hospitals outside its own system as competitors. In the diagram above, all the shapes surrounding the target hospital are considered competitors. Competition is measured by taking the total number of weighted average beds the target hospital faces within a 15-mile radius. A 15-mile radius is commonly used in literature, as Horwitz (2007) notes. Gresenz (2004) found that 10.4 miles is the mean distance radius that captures 75 percent of discharges and 21.5 is the mean distance radius that captures 90 percent of discharges from acute care hospitals in non-rural settings, while 14.2 miles is the mean distance radius that captures 75 percent of discharges and 25.2 is the mean distance radius that captures 90 percent of discharges from acute care hospitals in rural settings. Thus, it is reasonable that a 15-mile radius could be used in both rural and non-rural settings as it captures the majority of discharges in the target hospital’s area.

Hospital location is defined by the hospital’s latitude and longitude. The distances between the hospitals are calculated with Google Earth. Weighted Average Beds is defined as the average number of beds available for immediate patient use, excluding beds not immediately available because of renovation, maintenance, or physical plant problems. The Weighted Average Bed measurement is calculated by taking the sum of the number of calendar days each bed is available, divided by 365 (366 leap year). [8]

LARGEST measure of market competition

The Target Hospital only competes with large hospitals with more than 500 weighted average beds that lie within a 15-mile radius.  The target faces the pentagons; the other shapes have dropped out as competitors.

SAME SIZE measure of market competition

I divide the sample into three tiers. The first tier are those hospitals that have more than 500 weighted average beds, the second tier are those hospitals that have between 300 and 500 weighted average beds, and the third tier are those hospitals that have less than 300 weighted average beds. I then find the hospitals within 15-miles of the target hospital that are either within its tier or in the tier above it. Market competition is defined as the sum of weighted average beds that the target hospital faces according to these criteria. The idea behind this strategy is that hospitals will only be competitive with hospitals about its size or larger than it. The Target Hospital thus only faces hospitals in its tier and in the tier above within a 15-mile radius. In the diagram, the target faces the pentagons and right triangles, but not the diamonds.

These different measures of market competition yield comparable results. [9] I reason that the best measurement to keep in the final regression model is the BASIC measurement of market competition in which all hospitals are considered within the 15-mile radius, as a hospital will compete with all hospitals outside its system, not only with the largest ones.

I now arrive at the primary econometric model (Table 5):


An expanded model with INCOME and TOTALADM can also be considered (Table 5):


INCOME is the average income of the top 40 zip codes that travel to the hospital. This measurement for the income of the patients was chosen instead of taking the average incomes of the zip codes within a fixed radius of the hospital because the fixed radius measurement does not take into account the fact that some hospitals are referral hospitals. For example, most of MGH’s patients do not come from the immediate surrounding area. [10]

TOTALADM is the total admissions at each hospital, which is also a proxy for the size of the hospital.

It should be noted that the coefficients of SERVICES, MEDICARE, MEDICAID, QUALITY OF CARE, and

MARKET COMPETITION in the two models are very similar.  These are the coefficients of greatest interest, as they have statistically significant effects on relative price. Therefore, I disregard the effects of income and total admissions since they do not contribute appreciably to the primary economic model. Adding total admissions to the regression did not increase the correlation suggesting that hospital size was already accounted for by the services intensity. For the purposes of quantifying the results, we will consider the primary economic model as our final regression. The R2 value indicates that 56 percent of the variation in prices can be explained by the model, suggesting that it is a good fit for the data.

Table 6 lists the summary statistics.  It is clear that there are some hospitals offering many specialty services, but that most hospitals do not.  Furthermore, the mean and standard deviation for the “brand” variable show that only a few hospitals dominated the U.S. News and World Report, while the others did not make a single top 50 list at all.  This is characteristic of a marketplace with a few, very dominant hospitals in the marketplace.  It is also noteworthy that the range of the quality of care offered at the different hospitals is quite large, suggesting that Coakley’s assertion that the same quality of care is being provided among all hospitals must be questioned.

Specification tests

The following are robustness checks are performed to support the result of the above primary econometric model.

In Table 7 of the Supplemental Figures, I use the BASIC measurement of market competition that treats all hospitals surrounding the target hospital within a 15-mile radius as competitors. The variable SERVICES is replaced in the primary econometric model with BRAND or either form of TEACHING. The models produce comparable results.

To further evaluate robustness, I replace the BASIC measure of market competition with the other two proposed measures of market competition (Tables 8 and 9).  The regressions also showed similar results.

Since the predictors remain significant even after substitution with comparable independent variables, and the coefficients remain similar in magnitude, I conclude that the results are robust, and that my final regression model is appropriate for the data.


I propose that a five percent increase or decrease in prices would be considered a significant movement in prices. This shift is a good approximation for the range that private individuals are willing to pay for a service offered at several different hospitals, and is consistent with the merger guidelines notion of a significant price increase by the Department of Justice. From the angle of public policy, any price shift above this barline would be cause for scrutiny.

From Table 5, we can see a clearer picture of what exogenous factors Massachusetts policymakers can focus on as drivers of price variations among providers. A one standard deviation increase in the quantity of services leads to a 7.8 percent increase in prices. A one standard deviation rise in quality of care leads to a 12.9 percent rise in prices. These two variables are in terms of z-statistics. For the other variables that are not in terms of z-statistics, I multiplied their coefficients by their standard deviations to get the percent price shift from the mean = 1. A one standard deviation rise in Medicare leads to a 12 percent decrease in prices. A one standard deviation rise in Medicaid leads to an 8 percent decrease in prices. A one standard deviation rise in market competition leads to a 5 percent fall in prices.

We see from the results that all of the coefficients represent a major deviation in price from the mean. The predictors cause large shifts in price that are similar in magnitude. This suggests that all the hospital and market characteristics listed above are important drivers of price differentials, and furthermore affect price to approximately the same magnitude.

My results call into question the Attorney General’s report, which states that price differentials are solely determined by market leverage, and by no other theories. Clearly, the other theories are just as important, if not more so. Moreover, market competition causes a smaller shift in prices in my model in comparison with the effects of the other theories.

My models and regression results show that MEDICARE and MEDICAID are among the most significant predictors of price (to the 1% level, with p < 0.01). This is not what the AGO found. My results yield a negative coefficient on Medicare and Medicaid, suggesting that hospitals that want to negotiate higher prices need to decrease their reliance on Medicare and Medicaid patients. A possible explanation for this result is that private prices are affected negatively since hospitals with a greater share of Medicare and Medicaid patients will have a lower average price, because the government is free to set whatever prices it wants, and the hospitals have little bargaining power over the federal and state governments.  Specifically, if a hospital has a lot of Medicare and Medicaid patients its average price will be less. This is because Medicare and Medicaid patients do not pay the provider as much as the private patients for a given service. For example, when an insurance company comes to negotiate with the hospital over prices, it will ask how much the hospital was paid to take care of the average pneumonia patient the previous year. A hospital with many Medicare and Medicaid patients will have a lower average price for its pneumonia patients than a hospital with many privately insured patients, and thus be at a disadvantage in negotiating a price for the next year.

Services are, as I show, highly correlated with size, teaching status, and brand, and are a marker of large academic teaching hospitals, which have the clout to negotiate higher rates with commercial insurers. Hospital executives cite this as a likely reason for price differentials in their testimony at the March 2010 DHCFP hearing. Berenson et al. (2010) state, “Another source of ‘must-have’ status comes from providing unique, specialized services, which the hospital then uses to demand and win higher rates for all services.”

Glynn says, “There is a confusion about prices. They are not based on the good or service provided but rather based upon an existing insurance model instituted in the 1950s.” In the 1950s, insurers had to decide how to ration specialty beds among the Massachusetts hospitals. For example, to prepare for the event of a plane crash at Logan Airport, a set number of burn units had to be built in the state. Blue Cross Blue Shield thus decided how many beds for burn patients would be allocated to each hospital. Since Massachusetts General Hospital was an established teaching hospital at the time, and had the most highly trained personnel and facilities, it was reasonable for MGH to receive the most number of burn units.  However, rather than charging each individual burn victim the full price of the burn treatment service, MGH covered costs by raising prices for other services, such as the price for a cholecystectomy.  Thus, the cost of a more basic procedure (such as removal of a gallbladder) offered at MGH “is never based on its true costs, but rather based on the insurance model, as this is how the system was designed,” says Glynn. More specifically, the cost of the service can be approximated by the sum of its true cost and a portion of the cost of expensive specialty services offered at MGH.

It’s interesting that QOC correlates with price. This is the opposite of what Coakley argues in the AGO report. However, she used different measures of QOC, which shows that QOC is a contentious variable to assess. Insurance companies tend to use check lists to monitor quality of care (e.g. if the hospitals are following published best practices guidelines, such as giving aspirin to heart attack patients, and immunizing elderly patients against the flu), whereas Medicare Compare looks at actual results in patients (mortality rates, readmission rates).

I find that the more competition a hospital faces, the lower its prices are. This is reasonable since a hospital that dominates a geographic area would likely be able to set prices. However, this negative correlation is probably largely due to the medium size hospitals. I do not think it applies to the large teaching hospitals in Boston, since they are mostly fully occupied because of trimming of excess beds in the last twenty years, so they do not need to compete with each other for patients.

My models explain 56% of the variation in prices. The testimony from the hospital and commercial insurance executives at the March 2010 DHCFP Hearing would suggest that much of the remaining variation in prices is due to bargaining that goes on in contract negotiations each year. Jack Dutzer, President and CEO of Fallon Clinic, stated in oral testimony to the Attorney General’s Office, “We believe both that [Fallon Clinic is] fundamentally more efficient than most of the healthcare systems around us and that we have much, much more to do in order to become even more efficient. The opaque nature of the relationship between payers and providers, however, keeps us from knowing just how efficient we are and to what degree we may be part of the problem.” Such behind the scenes negotiations are not measured econometrically.

Policy proposals

The Attorney General Report claims that leverage theory is the only theory that causes the demand curve to be inelastic. One must be cautious, however, in deciding which policies to pursue from the results of the AGO investigation. Antitrust laws against collusion will have less effect if market competition is not the only factor affecting price, as I have shown in my thesis. Heavy regulation of market leverage will not work if there are many other variables affecting the elasticity of the demand curve. Policies aimed to curb market dominance (like in the case of MGH) to make the demand curve more elastic, and thus lower price, may be a waste of resources if reputation, teaching status, and high quality of care will keep the demand curve very inelastic regardless.

I discuss some policies that could be pursued. Insurers in Massachusetts have traditionally offered wide-open networks, meaning members can use the hospital or provider of their choice. But as health care costs have continued to increase at 7.5% a year, “tiered” and “limited” networks have been promoted as an immediate way to control costs. Massachusetts insurers report brisk business in plans that offer lower premiums in exchange for limits on use of popular but expensive hospitals or practices. However, executives at Partners, one of the high cost providers targeted, warned that when people became seriously ill and found that they would have to pay more to go to their first choice hospital, they would be angry at their insurers. They also warned that their hospitals might not be able to continue to use their profit margins to subsidize unprofitable services such as rehabilitation and mental health.

Another initiative that the Commonwealth is promoting is the Patient-Centered Medical Home. This is focused on better treatment and follow-up of chronic medical conditions, such as congestive heart failure and diabetes, and emphasizes patient-centered care delivered by teams of primary care providers, including physicians and nurses, to coordinate health needs, visits to specialists, hospital admissions, and reminders for checkups and tests. But it should be noted that there is as yet no hard data that demonstrate that hospital costs for treatment of a group of PCMHI patients would be lowered compared to a comparable group of patients given their usual care. [11]

The Commonwealth could also examine the type and number of services provided in each hospital. In this thesis, I showed in my final model that a selected group of expensive, high technology services correlated with hospital price disparities. If hospitals could avoid unnecessary duplication of these and similar expensive services, they might be able to bring down overall costs.

My model shows that the percent of patients on Medicare and Medicaid correlates negatively with price. Hospitals claim that Medicare and Medicaid are underpaying the actual costs of taking care of their patients. This gives a strong incentive for the financially strong hospitals to develop services aimed at younger, employed, commercially insured patients, since their insurers will pay at a higher rate. [12] As a consequence, the financially strong hospitals receive higher payment levels, and become even stronger financially than those hospitals that have a large percentage of Medicare/Medicaid patients. The simple solution to this driver of price disparities would be for the federal and state governments to increase reimbursement for Medicare and Medicaid patients to the relatively generous levels of twenty years ago, but this solution is impossible with current federal and state budget deficits.

The AGO should continue to annually publicize rankings of provider prices for similar services. The bad publicity gives higher-cost providers an incentive to lower their prices. Children’s Hospital, which is at the top of costly providers, announced at the end of last year after the AGO report that they were lowering their prices for selected surgical services by 20% and prices for radiology services by 17%. They are also sending their specialists to several community hospitals to deliver care in more convenient and lower cost settings.

The AGO should investigate and prohibit insurer/provider contract provisions that perpetuate market disparities, such as payment parity provisions, which lock in payment levels and reduce providers’ incentive to offer lower prices to insurers, product participation provisions, which limit the ability of insurers to offer new products, supplemental payments, which are separately negotiated and contribute to the lack of transparency in payment rates, and growth caps, which prevent smaller providers from expanding to compete with the larger providers. [13]


I have found that market leverage is not the only determining factor of price differentials among hospitals in Massachusetts. It appears that differences in prices are in fact value based.  In other words, higher prices are explained by something that consumers or society value, such as better quality, increased complexity of services provided, reputation of the hospital, and training new doctors. But controlling market leverage should not be completely disregarded from a policy standpoint. Dutzer of Fallon Community Health Plan adds in his oral testimony at the DHCFP hearing, “Continuing an imbalance of reimbursement for intangible values of reputation and/or market dominance will hurt our efforts to bring affordability and stability into the nation-leading efforts of our state to provide quality and access to all residents of the Commonwealth.”

Finally, I acknowledge that all empirical studies have some variability that is not explained by the model. No model can capture all the idiosyncratic factors that affect each hospital’s costs and rate negotiation process. I should also note that an econometric model does not prove causality, only that dependent and independent variables are correlated. In addition, the factors affecting hospital costs are constantly changing as the payer environment changes.


I thank David Cutler of the Harvard Department of Economics for his guidance, insight and encouragement in this research.  I also thank Stan Veuger of the Department of Economics, Thomas Glynn of the Harvard Kennedy School, and Jean Roth of the National Bureau of Economic Research.


Attorney General Office’s (AGO) Report for Annual Public Hearing on March 16, 2010, Examination of Health Care Cost Trends and Cost Drivers.

Berenson, R.A., P. B. Ginsburg, P.B., Kemper, N.  Unchecked provider clout in California foreshadows challenges to Health Reform.  Health Affairs 29: 4699-4705 (2010)

Boston Globe Spotlight Team. A healthcare system badly out of balance: the Partners effect.  Boston Globe, November 16, 2008.

Commonwealth of Massachusetts DHCFP website: Health Care Cost Trends March 16, 2010 Hearing.  Written Testimony from Witnesses.  www.mass.gov/dhcfp/costtrends/written testimony

Cutler, D. and Sheiner, L. The Geography of Medicare, American Economic Review, 89:228-233 (1999)

DHCFP Factsheet February 2010, www.mass.gov/dhcfp/costtrends

DHCFP website: Patient-Centered Medical Home Initiative, www.mass.gov/dhcfp/pcmhi

Dranove, D.and White, W. D.  Recent theory and evidence on competition in hospital markets.  J. Economics and Management Strategy  3:169-209 (1994).

Dreyer, P.  Analysis of the Attorney General’s Report Titled “Examination of Health Care Cost Trends and Cost Drivers”

Graduate Medical Education Directory 2010-2011, American Medical Association, 2010

Gresenz, C.R., Rogowski, J., Escarce, J. J.  Updated variable-radius measures of hospital competition.  Health Services Research 39:417-430 (2004)

Guerin-Calvert, M. E., and Israilevich, G.  A Critique of Recent Publications on Provider Market Power. October 4, 2010

Horwitz, J., Nichols, A.  What do nonprofits maximize? Nonprofit hospital service provision and market ownership mix.  National Bureau of Economic Research, Working paper No. 13246 (2007)

Kowalczyk, L.  Partners HealthCare rebuts AG findings on cost.  Boston Globe, June 17, 2010.

Sager, A., and Soloway, D. Written Expert Testimony at the DHCFP March 16, 2010 Hearing.  www.mass.gov/dhcfp/costtrends/written testimony

  1. [1] DHCFP Factsheet 2010. 
  2. [2] From a 2010 Partners Healthcare presentation given in response to the AGO Report.
  3. [3] See appendix for complete list of 66 hospitals included in this study. N = 198 since there are three insurers.
  4. [4] For example, Mass General and Brigham and Women’s Hospital are both members of Partners HealthCare.
  5. [5] See section on specification tests.
  6. [6] The DHCFP defines teaching hospitals using the Medicare Payment Advisory Commission’s definition of major teaching hospital as having “At least 25 full-time equivalent medical school residents per one hundred inpatient beds.”
  7. [7] From the Graduate Medical Education Directory, 2010-2011.
  8. [8] See: http://my.mass.gov/?pageID=eohhs2subtopic&L=6&L0=Home&L1=Researcher&L2=Physical+Health+and+Treatment&L3=Health+Care+Delivery+System&L4=DHCFP+Data+Resources&L5=Hospital+Summary+Utilization+Data&sid=Eeohhs2
  9. [9] See specification tests section.
  10. [10] See DHCFP data.
  11. [11] Reference: DHCFP website on Patient-Centered Medical Home Initiative.
  12. [12] Alan Sager and D. Soloway, Written Expert Testimony at the DHCFP March 2010 Hearing.
  13. [13] Testimonies of BCBS, HPHP, THP at the March 2010 DHCFP Hearing.