## Abstract

Do information frictions in health care markets lead to higher prices and price dispersion? Focusing on medical imaging procedures, this paper examines the equilibrium effect of a unique statewide price transparency website. Price information leads to a shift to lower-cost providers, especially for patients subject to a deductible. Furthermore, supply-side effects play a significant role in the long run, benefiting all insured individuals. Supply-side effects reduce price dispersion and are especially relevant in concentrated markets. These effects are important given that high prices are thought to be a primary cause of high private health care spending.

## I. Introduction

WHILE the price of health care procedures varies widely across medical providers, these prices are often difficult for patients to observe. Consequently, individuals often choose providers without comparing prices.1 A large theoretical literature, beginning with Stigler (1961) and Diamond (1971), argues that information friction can impede competition and lead to higher prices. Technological innovations have made it easier for consumers to compare prices in a number of markets, potentially increasing price competition.2 At the same time, information about prices may facilitate collusion, potentially decreasing price competition.3

In this paper, I ask how information about health care prices affects the market for health care services. Exploiting the introduction of a unique website that provided market-wide information for a subset of procedures in a state, I provide evidence on the long-run equilibrium effect of information about health care prices. I emphasize that the supply-side response to price transparency is particularly important. By observing detailed information on copay, coinsurance, deductible, and insurer payments, I also provide evidence about how insurers and patients split the savings that result from price transparency.

While we expect consumers to benefit from price transparency by choosing low-cost providers, price transparency may also allow insurers to negotiate lower prices with health care providers. This is because price transparency effectively increases residual demand elasticity, potentially incentivizing high-cost providers to lower prices. These supply-side effects could benefit all consumers, including those who do not use the information. The presence of this externality may imply a role for the public provision of information. Effects on negotiated health care prices are especially important given that high health care expenditure in the United States is often attributed to high prices, and there is currently limited evidence on policies that can reduce these prices.4

I exploit the introduction of a publicly provided website that allows individuals to access information about their insurer-specific out-of-pocket price for certain medical procedures. While previously studied price transparency efforts have primarily been conducted by specific employers, the website provided information that could be used by all privately insured individuals in the state. Since the intervention was market-wide, it potentially generated significant supply-side effects in addition to demand-side effects. In the main specification, I employ a difference-in-difference methodology that takes advantage of two sources of variation: the timing of the website introduction and variation among procedures available on the website. I also show that results are robust to exploiting cross-state variation. I focus on the universe of outpatient medical imaging visits, which account for over 9 million claims. I argue that the medical imaging procedures that were on the website are quite similar to the medical imaging procedures that were not on the website, allowing for a useful comparison. I also provide empirical support for the assumption that procedures on the website are unlikely to be differentially affected by time-varying demand and cost factors that affect prices in other ways.

First, I examine transaction prices, which include both demand- and supply-side effects. Over the five-year period after the website started, there is a 3% reduction in total visit cost for medical imaging visits on the website relative to medical imaging procedures not on the website. Much of this savings goes to consumers, reducing out-of-pocket costs by 5%. This effect increases over time, and by the fifth year, out-of-pocket prices are 11% lower relative to the control group. Individuals with the most to gain from using the website—those under their deductible—see almost double the savings over the period. These results are highly significant and robust to the inclusion of individual fixed effects and detailed insurer and individual controls.

Next, I examine the mechanisms driving the reduction in transaction prices. On the demand side, individuals with access to the website are more likely to choose a low-cost provider. Next, I analyze the supply-side effects using a difference-in-difference specification that controls for demand-side effects. Specifically, I include provider-procedure-insurer fixed effects that control for transaction price differences due to switching across providers. Since providers and insurers negotiate a new price schedule at most once a year, the supply-side effect may take time to materialize. The estimates imply that providers reduce their prices in the long run, defined as more than two years after the introduction of the website. Visit prices decline by 2%. The effect is greater for providers operating in concentrated markets that are likely to have the highest margins in the absence of the website. Although these effects are relatively modest, consistent with the fact that price transparency tools currently receive modest use, they suggest that supply-side effects could be quite large if more consumers were informed about prices. One concern is that prices for procedures not on the website are indirectly affected. I show that these results are robust to an alternative identification strategy using cross-state variation that helps address this concern.

A theoretical literature has found that price dispersion can result from information frictions in the market.5 In order to test whether price dispersion in the market for health care services is due in part to information frictions, I use a difference-in-difference methodology to directly examine the effect on price dispersion. The estimates imply that the website reduces price dispersion for affected procedures, as measured by the interquartile range of negotiated prices, by $159 on average relative to the price dispersion of the control procedures. Prior research has focused on the demand-side response to health care price transparency efforts by individual employers and has found relatively small effects in the short run (Lieber, 2017; Whaley, 2015a; Desai et al., 2016). In particular, Desai et al. (2016) finds no reduction in spending in the year after an employer offers a price transparency tool. However, the previous literature has focused on price transparency efforts that are available to a small subset of consumers. In contrast, the New Hampshire website was publicly available to all consumers in the state; therefore, as I argue, supply-side effects may be important in the long run. In addition, New Hampshire's price transparency tool differed from other tools due to the quality of the underlying price information and the ease of use. There is little evidence on the supply-side effects of market-wide price transparency. Whaley (2015b) focuses on a website providing information to specific employers and finds a reduction in the price of laboratory tests using an event-study methodology. While the previous literature has focused on price transparency tools for individual employers, I examine the introduction of a first-of-its-kind website that was available to all insured individuals in a state.6 In a discussion of price transparency, Sinaiko and Rosenthal (2011) note that tools like New Hampshire's website may generate supply-side effects; however, the effects may take time to materialize.7 In addition, the effect of price transparency on prices is theoretically ambiguous given the possibility that price transparency may facilitate provider collusion (Cutler & Dafny, 2011). Using individual-level data on outpatient medical imaging visits by all insured individuals in New Hampshire, this paper provides the first evidence quantifying the overall equilibrium effects of price transparency for both individuals and insurers in a state. Understanding the equilibrium effect is particularly relevant given that many states are currently considering price transparency legislation.8 The remainder of this paper is as follows. Section II provides additional background on the website and health care pricing, and Section III, describes the data. Section IV describes the main empirical strategy and discusses the demand-side and supply-side effects on prices. Section V concludes. ## II. Institutional Details and Background Recent research has documented a large degree of price dispersion in health care, especially in the private sector (Philipson et al., 2010; Newhouse et al., 2013).9 Even relatively homogeneous medical services vary in price. For instance, Cooper et al. (2018) find that MRI prices vary by a factor of 12 across the country. There is a large degree of price dispersion even within a geographically constrained area. For example, the total price of a back MRI in New Hampshire for individuals covered by Anthem, the largest insurer in the state, varies widely, with an upper and lower quartile of$1,085 to $2,472, respectively. Consequently, the out-of-pocket price that individuals pay ranges as well, especially for those under their deductible (the out-of-pocket interquartile range is$143). More generally, table 1 shows the potential savings if all consumers switched to a low-cost provider, defined as a provider in the first quartile of the price distribution in the state.10 Often these lower-cost providers are outpatient facilities, such as medical imaging centers, rather than hospitals. Across a range of procedure categories, savings would be between 44% and 73%. Even if individuals switched to the provider with the median price, they would save 16% to 58% on average.

Table 1.
Potential Cost Savings If Consumers Switched to Low-Price Providers
Consumers Switch to 1st Quartile ProviderConsumers Switch to Median Provider
Procedure ClassMean Total Visit PriceMean% SavingsMean% Savings
Computed tomography (CT) 1,604 659 58.9% 995 37.9%
Magnetic resonance Imaging (MRI) 1,767 989 44.0% 1,283 27.4%
X-ray 593 152 74.3% 240 59.5%
Consumers Switch to 1st Quartile ProviderConsumers Switch to Median Provider
Procedure ClassMean Total Visit PriceMean% SavingsMean% Savings
Computed tomography (CT) 1,604 659 58.9% 995 37.9%
Magnetic resonance Imaging (MRI) 1,767 989 44.0% 1,283 27.4%
X-ray 593 152 74.3% 240 59.5%

The table shows the average transaction price paid in 2006, along with the potential savings if every patient paid at most the 25th or 50th percentile of visit price in New Hampshire for each procedure given the patient's insurance company and insurance type. All prices in 2010 dollars. Figures reflect the potential demand-side savings (e.g., hold negotiated prices fixed).

One explanation for why these price differences persist even for relatively homogeneous products is that patients lack information about prices. Health care prices are determined through bargaining between insurers and providers, and insurers often agree not to publicly disclose the negotiated contracts. Perhaps for this reason, surveys show that the majority of individuals do not compare prices before receiving medical care.11

In order to allow health care consumers to find low-cost options, the state of New Hampshire began requiring health insurers operating in the state to submit medical claims to a centralized database in 2005. These data were then used to calculate the median bundled out-of-pocket prices for various medical procedures. In March 2007, New Hampshire launched its HealthCost website.12 Individuals enter the procedure, their insurance information (including remaining deductible), their postal code, and search radius and obtain information about each provider's expected out-of-pocket price, insurer price, and total price. The site automatically takes into account copayment and coinsurance levels given their insurance. Results are sorted by out-of-pocket price, making it easy to select the least expensive provider from the point of view of the patient. More recently, the tool has provided additional information.13 Although other states have since started price transparency websites of their own, including California, Maryland, Florida, Oregon, and New Jersey, New Hampshire's price transparency efforts are the most comprehensive.14 It should be noted that although the New Hampshire tool is relatively easy to use, compared to other tools, it still requires individuals to understand basic information about their health insurance.15 There may be additional scope to lessen the burden for patients using these tools.

At the time it was introduced, the website had price information for about 35 procedures. The website focuses on outpatient procedures since patients often schedule these appointments ahead of time and may have more scope for choosing among providers.

The HealthCost website has received significant attention in the state, with over forty articles in the local public press. In addition, the New Hampshire Insurance Department promoted the website by encouraging primary care doctors to tell patients about it. Insurers were also encouraged to inform their enrollees of the website.

Among individuals who could have benefited from the website, there was meaningful take-up. I construct a measure of website usage with monthly website traffic logs provided by the New Hampshire Insurance Department. Figure 1 shows the number of price searches on the website since 2005. When the website began, there were roughly 1,000 searches per month for the price of medical imaging procedures, which grew over time. Searches for the price of medical imaging procedures make up about half of all searches using the website.

Figure 1.

Website Searches for Health Care Prices, by Month

The figure shows the number of times the price transparency tool is used to acquire price information in each month. Includes all searches using “Health Costs for Insured Patients” wizard on either nhhealthcost.nh.gov or nhhealthcost.org. Website traffic data is not available for the period after 2010 due to a change in the website host.

Figure 1.

Website Searches for Health Care Prices, by Month

The figure shows the number of times the price transparency tool is used to acquire price information in each month. Includes all searches using “Health Costs for Insured Patients” wizard on either nhhealthcost.nh.gov or nhhealthcost.org. Website traffic data is not available for the period after 2010 due to a change in the website host.

It has been noted that the use of price transparency tools, including the New Hampshire tool, is low relative to the number of total patients (Mehrotra, Brannen, & Sinaiko, 2014; Sinaiko & Rosenthal, 2016). In contrast to the previous literature, I use website traffic logs to examine take-up.16 The website has price information for only a limited number of procedures, and individuals are unlikely to use it if they find that their procedure is not listed.17 They are also unlikely to use it if they are receiving inpatient care. Focusing on privately insured individuals in New Hampshire receiving an outpatient medical imaging procedure that is listed on the website, I find that take-up is 8%. Take-up is lower for other procedures available on the website, perhaps because medical imaging procedures tend to be more standardized. Although I focus on medical imaging procedures, I argue that understanding the high prices for these procedures is particularly important given that medical technology, especially related to medical imaging, is often cited as one of the key drivers of health care cost growth.18

In addition to immediate demand-side effects from the website, there is anecdotal evidence of supply-side effects. Analysts have noted that “the balance of plan-provider negotiating power began shifting significantly in New Hampshire, a result in large part of public transparency efforts.”19 For instance, Exeter Hospital and Anthem, the largest insurer in New Hampshire, had a public dispute over contract terms in 2010. Anthem argued that prices at Exeter Hospital were too high, pointing to the website as evidence, and it was eventually able to negotiate rate cuts.20

Why might provider prices respond to price information? First, consider the case in which providers have market power and are able to unilaterally set price. If consumers become more price sensitive due to better information about prices, the profit-maximizing price will decline. In the market for private health care, prices are determined through bilateral negotiations between providers and insurers rather than set unilaterally. In this case, a similar mechanism applies. However, equilibrium-negotiated prices may also depend on insurer incentives (Ho, 2009; Gowrisankaran, Nevo, & Town, 2015; Ho & Lee, 2017). To the extent that the website affects either provider or insurer gains from trade, negotiated prices may be affected in equilibrium.21

## III. Data

The main data set covers the universe of private insurance enrollment and medical claims in the state of New Hampshire from 2005 to 2011. These data were collected as part of the New Hampshire Comprehensive Health Care Information System, which assembled data from all commercial insurers with enrollees who were state residents or receive services under a policy issued in the state. These are the same data used to construct prices for the website.

Each outpatient claim has a CPT/HCPCS code to identify procedures.22 These codes are very specific (e.g., code 72120 is “X-ray examination, spine, lumbosacral; bending views only, 2 or 3 views”). I limit the sample to the universe of outpatient medical imaging claims, which includes 289 procedures related to X-rays, computerized tomography (CT) scans, and magnetic resonance imaging (MRI) scans. These procedures all use imaging to diagnose internal conditions. The number of procedures in each category are listed in table 2.

Table 2.
Availability of Outpatient Medical Imaging Procedure Price Information on Website
Number or Unique Procedures
Procedure CategoryOn WebsiteNot on Website
Computed tomography (CT) 15 47
Magnetic resonance imaging (MRI) 21 65
X-ray 34 107
Total 70 219
Number or Unique Procedures
Procedure CategoryOn WebsiteNot on Website
Computed tomography (CT) 15 47
Magnetic resonance imaging (MRI) 21 65
X-ray 34 107
Total 70 219

Table shows the number of unique outpatient medical imaging procedures, as identified using CPT/HCPCS codes, on the price transparency website versus not on the website. Procedure codes with updated descriptions are considered separate procedures.

Inpatient medical imaging procedures, such as those that are part of major surgeries, are excluded from the analysis. Since individuals have little ability to choose a provider when medical imaging procedures are part of an inpatient episode, the website includes information only about outpatient medical imaging procedures.23

In addition to the principal medical imaging procedure, there are often supplemental procedures such as contrast agents that are billed along with the main procedure. The quantity and price of these supplemental procedures may also vary across providers. When comparing the cost across medical providers, the relevant price is determined by the entire bundle of procedures. For this reason, the website has information about the cost of an entire visit. For the same reason, my analysis focuses on the price of the visit. I also examine the principal procedure price alone. The construction of the visit price and principal procedure price is described in more detail in the online appendix, section A. Conducting analysis at the visit level has important implications for the interpretation of the results and comparisons with the large literature that conduct analysis at the claim level.

There are 811,553 individuals under age 65 with at least one medical imaging claim between 2005 and 2011. Using individuals' postal code, I merge on additional demographic information, including income and education, using the 2007–2011 American Community Survey. I also construct each individual's Charlson Comorbidity Index using International Classification of Diseases (ICD) codes in the claims data set (Charlson et al., 1987; Stagg, 2006). The Charlson Comorbidity Index is an integer score that summarizes comorbid conditions that predict mortality. Individual demographics are summarized in the first panel of table 3.

Table 3.
Summary of Privately Insured Individuals with Medical Imaging Claims
MeanSDMinimumMaximum
Male 0.48 0.50
Age 36.9 17.6 0.0 64.0
Charlson Comorbidity Index 0.5 0.7
Zip income (1000s) 68.5 21.2 4.9 240.8
Zip more than B.A. degree 33.8 13.8 0.0 100.0
Insurance type
PPO 0.32 0.47
POS 0.14 0.34
HMO 0.39 0.49
EPO 0.07 0.25
Other 0.09 0.29
Insurance company
Anthem 0.45 0.50
Cigna 0.24 0.43
Harvard Pilgrim 0.13 0.33
Other 0.18 0.38
Plan characteristics
Plan has deductible 0.45 0.50
Plan has copay 0.83 0.38
Plan has coinsurance 0.24 0.43
Number of individuals 811,549
MeanSDMinimumMaximum
Male 0.48 0.50
Age 36.9 17.6 0.0 64.0
Charlson Comorbidity Index 0.5 0.7
Zip income (1000s) 68.5 21.2 4.9 240.8
Zip more than B.A. degree 33.8 13.8 0.0 100.0
Insurance type
PPO 0.32 0.47
POS 0.14 0.34
HMO 0.39 0.49
EPO 0.07 0.25
Other 0.09 0.29
Insurance company
Anthem 0.45 0.50
Cigna 0.24 0.43
Harvard Pilgrim 0.13 0.33
Other 0.18 0.38
Plan characteristics
Plan has deductible 0.45 0.50
Plan has copay 0.83 0.38
Plan has coinsurance 0.24 0.43
Number of individuals 811,549

Summary statistics are for all unique privately insured individuals in New Hampshire over the period 2005 to 2011 with at least one outpatient medical imaging visit.

Each medical claim is also associated with an anonymized provider identifier that can be linked to additional information such as provider postal code and whether the provider is a hospital or a nonhospital facility. This information is used to construct provider concentration in each county.

The vast majority of individuals in the sample are covered by a managed care organization, either a health maintenance organization (HMO) plan, preferred provider organization (PPO) plan, point-of-service (POS) plan, or an exclusive provider organization (EPO) plan. The defining feature of managed care plans is that insurers negotiate lower prices with a selected network of providers. The plan types differ according to the standards used when individuals select providers within the network. Only 3% of individuals have an indemnity (fee-for-service) plan. Plan type is summarized in the second panel of table 3. Three main insurers are operating in New Hampshire: Anthem, Cigna, and Harvard Pilgrim. Less than a fifth of individuals are enrolled in another plan (see the third panel of table 3).24

The plans offered in New Hampshire over the period differ in their cost-sharing characteristics. In particular, 45% of individuals pay a deductible at some point over the period (see the last panel of table 3). In general, individuals are responsible for all health care costs under the deductible amount in a given year. Although I do not observe the deductible amount associated with each plan, I do observe the deductible paid on each visit. Using observed deductible payments, I construct an indicator for whether each individual is under or over her deductible in a given year in order to test whether individuals benefit more from the website when they are subject to a deductible.25

Over the period, there are 9.2 million claims that constitute 2.1 million medical imaging visits (i.e., there are about three supplemental procedures on average per medical imaging visit). For each health claim, I observe the copayment, coinsurance, and deductible paid by the individual, which together make up the out-of-pocket price. In addition, I observe the insurer-paid amount. Together, the out-of-pocket price and insurer-paid amount constitute the total price received by the provider, often called the allowed amount.26

The average price paid by individuals and insurers is presented in table 4. Insurers pay the majority of the cost for medical imaging procedures. Although out-of-pocket prices are low on average, there is high variance, and some individuals are fully exposed to the total price.

Table 4.
Summary of Outpatient Medical Imaging Visit Price
Visits on WebsiteVisits Not on Website
PrewebsitePostwebsitePrewebsitePostwebsite
MeanSDMeanSDMeanSDMeanSD
Patient cost
Copayment 15.6 41.4 19.2 39.0 15.4 53.2 18.6 40.4
Coinsurance 12.5 81.8 21.0 111.9 14.6 87.1 25.6 132.7
Deductible 46.1 181.7 84.3 298.0 58.1 206.4 103.4 331.2
Total out-of-pocket cost 76.6 226.5 124.8 331.4 90.8 253.8 148.0 371.9
Insurance cost
Paid amount 634.4 1,381.4 793.5 1,737.6 740.8 1,730.7 970.9 2,141.8
Total
Allowed amount 846.1 1,716.7 942.9 1,848.3 989.2 2,113.1 1,149.6 2,269.7
Charge amount 1,236.6 2,861.4 1,602.5 3,393.9 1,471.4 3,331.5 1,947.2 3,976.6
Observations (visits) 501,358 1,176,476 124,017 301,902
Total procedures 2,018,224 5,376,584 464,574 1,325,082
Visits on WebsiteVisits Not on Website
PrewebsitePostwebsitePrewebsitePostwebsite
MeanSDMeanSDMeanSDMeanSD
Patient cost
Copayment 15.6 41.4 19.2 39.0 15.4 53.2 18.6 40.4
Coinsurance 12.5 81.8 21.0 111.9 14.6 87.1 25.6 132.7
Deductible 46.1 181.7 84.3 298.0 58.1 206.4 103.4 331.2
Total out-of-pocket cost 76.6 226.5 124.8 331.4 90.8 253.8 148.0 371.9
Insurance cost
Paid amount 634.4 1,381.4 793.5 1,737.6 740.8 1,730.7 970.9 2,141.8
Total
Allowed amount 846.1 1,716.7 942.9 1,848.3 989.2 2,113.1 1,149.6 2,269.7
Charge amount 1,236.6 2,861.4 1,602.5 3,393.9 1,471.4 3,331.5 1,947.2 3,976.6
Observations (visits) 501,358 1,176,476 124,017 301,902
Total procedures 2,018,224 5,376,584 464,574 1,325,082

Table shows summary statistics related to transaction prices. Note that prices are lower when the website is available. Includes all outpatient medical imaging visits for privately insured individuals in New Hampshire from 2005 to 2011. All prices in 2010 dollars.

The summary statistics presented in table 4 preview the results. Although the price of all procedures is increasing over time, the simple difference-in-difference estimate using the average total price implies that the price of procedures on the website declined by $64 relative to the price of procedures not on the website. However, it is important to control for changes in the composition of procedures and changes across time affecting the control group. ## IV. Effect on Prices I begin by examining the overall effect on transaction prices, including total visit amount, out-of-pocket amount, and insurer amount. I examine the heterogeneous effects and show that results are robust to a number of specifications. Using a similar identification strategy, I show that this effect is due to both demand-side and supply-side factors. ### A. Empirical Strategy In order to estimate the causal effect of price transparency on prices, I exploit two sources of plausible exogenous variation: the timing of the website introduction and the availability of medical imaging procedures on the website. In particular, I construct $OnWebm$, which indicates whether procedure $m$ is ever available on the website. I also construct $Postt$, which indicates if the website is available at month $t$. This takes the value of 1 if the date of admission is March 2007 or later. The baseline difference-in-difference specification is given by $log(1+pimjkt)=β(OnWebm×Postt)+αXit+λm+λk+λt+ɛimjkt.$ (1) The outcome of interest is $pimjkt$, the price of a visit for individual $i$ with insurance $k$ obtaining procedure $m$ from provider $j$ at time $t$. I consider both the patient's out-of-pocket cost, as well as the cost to the insurer. The baseline specification controls for individual covariates $Xit$, which includes age, gender, Charlson comorbidity index, income, education, rural classification, and member plan characteristics (deductible, coinsurance, and copay). I also include procedure fixed effects, $λm$, and insurer fixed effects, $λk$, which control for time-invariant factors that may be correlated with prices and the availability of the website.27 I include as well month fixed effects, $λt$, which control for time-varying factors that may be correlated with prices and website availability. Finally, $ɛimjkt$ is a vector of idiosyncratic random errors. Prices are highly correlated within each month since individuals tend to be subject to a deductible in the beginning of the year but not at the end of the year. To account for correlation within a month, standard errors are clustered at the month level. The unit of analysis is an individual medical imaging visit. The dependent variable is transformed using $log(1+y)$ since the out-of-pocket price and the insurer price can be 0, making $log(y)$ undefined. An alternative to using OLS with a log-transformed dependent variable is to use GLM with a log-link function. I discuss this alternative model along with robustness results in section IVC. The coefficient of interest, $β$, is interpreted as the change in prices due to the presence of the website in log points. This should be interpreted as the intent-to-treat effect, keeping in mind that take-up was 8% and individuals who did not use the website may be indirectly affected due to supply-side effects. The main identifying assumption is that in the absence of the website, the procedures on the website and the procedures not on it would follow common trends. I use a number of methods to examine the validity of this assumption, including examining trends prior to the introduction of the website and a falsification test. In order to isolate the supply-side effect of price transparency, I use a similar identification strategy but control for the demand-side effects. In particular, I include fixed effects that control for the variation in price of each procedure across providers and insurers. This approach is similar to that of Christensen et al. (2015). The specification is now $log(1+pimjkt)=β(OnWebm×Postt)+αXit+λjmk+λt+ɛimjkt,$ (2) where2 the vector $λjmk$ includes an indicator for each combination of provider, procedure, and insurer. Any variation in transaction prices due to the fact that individuals switch to lower-cost providers after the introduction of the website is absorbed by these fixed effects. Another way to see this is to note that conditional on going to the same provider, with the same insurance, and receiving the same procedure, a change in transaction prices must be due to a supply-side effect. Note that these fixed effects also absorb changes in prices due to the entry of providers over the period. The remaining variation in price identifies $β$, which can now be interpreted as the reduction in transaction prices due to lower negotiated prices for a given provider. Due to the fact that the regression is run at the individual level, $β$ can be interpreted as the supply-side effect weighted by quantity demanded after taking into account demand-side effects. For instance, if only rarely chosen providers reduce their price, $β$ will be smaller than the average unweighted effect on provider prices.28 I focus on the weighted results since from a policy perspective, it is important to understand the effect the average patient experiences. It is important to note that these supply-side factors could be driven by a number of factors. One mechanism is that insurers negotiate lower prices for specific procedures; however, this need not be the case. Providers may reduce list prices (i.e., Chargemaster prices), which then get passed on to insurers if contracts are negotiated as a discount relative to list prices. Alternatively, providers could bill less aggressively (either less up-coding or fewer supplemental procedures). Distinguishing between these explanations is difficult given the secrecy around provider-insurer negotiations; however, I provide some evidence on this issue by examining the effect on the entire visit price (which includes all supplemental procedures), as well as the effect on the principal procedure price alone. An identifying assumption is that the website affected prices only for procedures on the website; there are no spillover effects to procedures not featured on the website. This would be violated if, for instance, hospitals responded to the website by cutting prices for all medical imaging procedures. This assumption would also be violated if providers raised prices for procedures not on the website to compensate for lower prices for procedures on the website. I address these concerns by exploiting cross-state variation. In particular, the specification is $log(1+pimjkt)=β(InNHj×Postt)+αXit+λjmk+λt+ɛimjkt,$ (3) where $InNHj$ indicates if provider $j$ is in New Hampshire, and thus had information available on the website. The identifying assumption is that prices for providers outside New Hampshire that are in the NHCHIS database do not change in response to the website given that New Hampshire patients make such a small part of their demand. I discuss this in more detail in section IVG. I also examine how the supply-side effect varies by the degree of local competition among providers. In particular, I use the Herfindahl index in each county for each procedure category, which is defined as $HHIcl=∑jsjcl2$, where $sjcl2$ is the market share of provider $j$ in county $c$ among all procedures in procedure category $l$. The period prior to the introduction of the website is used to calculate $HHIcl$ in order to address concerns that the market structure may have been endogenously affected by the website. Finally, I examine price dispersion directly as measured by the interquartile range of prices. I exploit the same sources of variation but aggregate to the procedure-month level and use the interquartile range of prices, $IQRmt$, as the outcome variable. To examine the dispersion in transaction prices, $IQRmt$ is defined as the difference between the third and first quartile of transaction prices for each procedure in each month. To examine the dispersion in provider prices (or negotiated prices), $IQRmt$ is defined as the difference between the third and first quartile of prices for each procedure in each month after aggregating to the provider-procedure-month level.29 Under the same assumptions as previous specifications, $β$ can be interpreted as the dollar change in the interquartile range of prices due to the website. ### B. Effect on Transaction Prices Figure 2 presents the main results for transaction prices by half-year with the full set of controls and fixed effects.30 In the periods before the website, there is no significant price effect for procedures that were eventually on the website and those that were never on it. This provides evidence that the procedures on the website had similar trends in the preperiod as the procedures that were not on the website, supporting the common trends assumption. Once the website launched, the total visit amount (i.e., allowed amount) declined for procedures on the website relative to procedures not on the website. The effect becomes significant in the second year. This effect grows over time, eventually leading to a 4% reduction in the total visit amount (figure 2a) and 11% reduction in patients' out-of-pocket amount (figure 2b) five years after the introduction of the website. As I discuss in subsequent analysis, this is partially due to the fact that supply-side effects take time to materialize because prices are renegotiated infrequently. In addition, the website was being used more often in the later period (see figure 1). Figure 2. Equilibrium Effect of Price Transparency Website on Spending, by Time from Website Introduction The charts show point estimates for each half-year using the difference-in-difference baseline specification as described in section IVB. The estimates reflect the overall equilibrium effect, including both demand-side and supply-side effects. The omitted period is the half-year prior to the start of the price transparency website. Error bars indicate 95% confidence interval using standard errors clustered at the month-year level. Figure 2. Equilibrium Effect of Price Transparency Website on Spending, by Time from Website Introduction The charts show point estimates for each half-year using the difference-in-difference baseline specification as described in section IVB. The estimates reflect the overall equilibrium effect, including both demand-side and supply-side effects. The omitted period is the half-year prior to the start of the price transparency website. Error bars indicate 95% confidence interval using standard errors clustered at the month-year level. The first panel of table 5 presents the average effect on visit amount over the five-year period. The estimates imply a 3.1% reduction from a mean of$950. This can be interpreted as the reduction in spending per visit due to both demand- and supply-side factors. For patients, the estimates imply a 5.4% reduction in out-of-pocket prices due to the website from a mean of $115.45 (see second panel of table 5). These results are significant at the 1% level. Next, I examine how the effect varies by deductible status. Individuals who are not past their deductible are potentially exposed to the full cost of the procedure, and thus have the most to gain from choosing a low-cost provider.31 Consistent with this fact, I find that individuals subject to a deductible see a much larger reduction in out-of-pocket cost: a 10.3% decline (from a mean of$295).32

Table 5.
Effect of Price Transparency Website on Visit Price Baseline Difference-in-Difference Estimates
Deductible
AllNo DeductibleNot PastPast
Dependent Variable: Log(1 $+$ Total Visit Amount)
$OnWebm×Postt$ −0.031*** −0.029*** −0.044*** 0.014
(0.004) (0.006) (0.007) (0.010)
Mean level 950.47 835.65 1038.61 1121.35
Adjusted $R2$ 0.368 0.367 0.360 0.371
Observations 1,984,798 1,004,200 633,716 346,843
Dependent Variable: Log(1 $+$ Patient Out-of-Pocket Amount)
$OnWebm×Postt$ −0.055*** −0.043*** −0.109*** −0.032
(0.009) (0.011) (0.015) (0.020)
Mean level 115.26 23.27 295.91 51.54
Adjusted $R2$ 0.323 0.168 0.200 0.091
Observations 1,984,798 1,004,200 633,716 346,843
Dependent Variable: Log(1 $+$ Insurer Paid Amount)
$OnWebm×Postt$ −0.038*** −0.030*** −0.026* 0.012
(0.005) (0.006) (0.014) (0.010)
Mean level 777.11 756.81 680.18 1012.49
Adjusted $R2$ 0.305 0.380 0.202 0.383
Observations 1,984,798 1,004,200 633,716 346,843
Individual controls Yes Yes Yes Yes
Insurance FE Yes Yes Yes Yes
Month-year FE Yes Yes Yes Yes
Procedure FE Yes Yes Yes Yes
Deductible
AllNo DeductibleNot PastPast
Dependent Variable: Log(1 $+$ Total Visit Amount)
$OnWebm×Postt$ −0.031*** −0.029*** −0.044*** 0.014
(0.004) (0.006) (0.007) (0.010)
Mean level 950.47 835.65 1038.61 1121.35
Adjusted $R2$ 0.368 0.367 0.360 0.371
Observations 1,984,798 1,004,200 633,716 346,843
Dependent Variable: Log(1 $+$ Patient Out-of-Pocket Amount)
$OnWebm×Postt$ −0.055*** −0.043*** −0.109*** −0.032
(0.009) (0.011) (0.015) (0.020)
Mean level 115.26 23.27 295.91 51.54
Adjusted $R2$ 0.323 0.168 0.200 0.091
Observations 1,984,798 1,004,200 633,716 346,843
Dependent Variable: Log(1 $+$ Insurer Paid Amount)
$OnWebm×Postt$ −0.038*** −0.030*** −0.026* 0.012
(0.005) (0.006) (0.014) (0.010)
Mean level 777.11 756.81 680.18 1012.49
Adjusted $R2$ 0.305 0.380 0.202 0.383
Observations 1,984,798 1,004,200 633,716 346,843
Individual controls Yes Yes Yes Yes
Insurance FE Yes Yes Yes Yes
Month-year FE Yes Yes Yes Yes
Procedure FE Yes Yes Yes Yes

Estimates from the baseline difference-in-difference specification presented in equation (1). The unit of observation is a patient visit, which may contain multiple medical claims. The sample consists of all commercial claims related to outpatient medical imaging procedures in New Hampshire from 2005 to 2011. OLS regression standard errors clustered at the month-year level in parentheses. $*$$p<0.10$, $**$$p<0.05$, and $***$$p<0.01$.

Individuals who are subject to a coinsurance payment may still have an incentive to find a low-cost provider and benefit from the website.33 In addition, individuals who do not use the website and find a low-cost provider may still benefit if providers reduce their prices. This is the supply-side effect that I explore in detail in following sections. The results imply that individuals who do not have a plan with a deductible see a significant reduction in prices of about 4% (see column 2 in table 5). I find no evidence of an effect for individuals past their deductible. These individuals often have little exposure to prices.

Although individuals are not likely to internalize the cost to the insurer when choosing a provider from the website, there may be an indirect benefit to the insurer. In particular, there is a mechanical correlation between the individual's out-of-pocket price and the insurer price when the individual is subject to a coinsurance payment. To test whether insurers also benefit from the website, I use the same specification but with the insurer-paid amount as the dependent variable. The third panel of table 5 presents the results. The insurers save 3.7% on average over the period (from a mean of $777), a highly statistically significant amount. Much of this savings is due to individuals without a deductible, consistent with the fact that most of the savings when there is a deductible accrue to the individual. The magnitude of the effect is larger for patients and insurers than for the total visit amount. Although it may initially seem that the total effect should be between the effect for patients and the effect for insurers, this is not necessarily the case given that the effect is measured in percentage terms and the fraction of the cost paid by the insurer is not constant. It also should be noted that although the percent change in prices is larger for out-of-pocket amount than insurer amount, the insurers actually benefit more from the website in absolute terms. This is because the insurer covers the majority of the cost: 88% of the total price on average. ### C. Robustness of Demand-Side Effects Online appendix table A-2 shows that out-of-pocket price and insurer-paid amount results are robust to a number of specifications. In particular, the results are not driven by changes in observable characteristics of individuals. In addition, the fact that the results are robust to insurance fixed effects implies that effects are not driven by changes in insurance enrollment over the period. In the online appendix, section B, I provide additional evidence that the website did not affect insurance enrollment. Another concern is unobservable individual characteristics. In particular, individuals who obtain private health insurance after the introduction of the website could be different on unobservable dimensions. In the online appendix, table A-2, column 5, I control for individual fixed effects. Identification now comes from the same individuals who received medical imaging procedures before and after the introduction of the website. Although this results in a much smaller sample size, the results are largely robust to this specification; however, the effect on insurer-paid amount become insignificant. A related concern is that the website changed the complexity of procedures due to an increase (or decrease) in the probability that an individual has a procedure when information is available. This concern is somewhat mitigated by the fact that medical imaging procedure codes are quite specific and are standardized across providers. In addition, in the online appendix, section C, I show that the website did not have a meaningful effect on the probability of receiving any medical imaging procedure or the number of medical imaging procedures. The out-of-pocket price may be 0 if the insurer pays the full cost due to full insurance. Similarly, the insurer price may be 0 if the individual pays the full cost because the individual is under the deductible. Due to this issue, the dependent variable is transformed using $log(1+y)$ to avoid undefined values. One alternative to this transformation, which generates point estimates with a similar interpretation, is to use GLM with a log-link function and an untransformed dependent variable (Manning & Mullahy, 2001). For computational tractability, I aggregate to the procedure-month level and use a weighted GLM approach. Column 6 in the online appendix, table A-2, shows that results are similar to the baseline case, although standard errors are slightly larger. For completeness, column 5 shows the weighted OLS results, which are also similar. In table A-3 in the online appendix, I conduct a falsification exercise in which I test whether there was an effect on prices in the one-year period before the website actually existed. Consistent with the assumption that results are not driven by differential price trends, all of the estimates are quite close to 0. Eleven of the twelve estimates are not statistically significant, and the one that is significant implies that there was an increase in transaction prices for procedures eventually on the website. ### D. Heterogeneous Effects I examine results by demographic characteristics to examine which groups benefit the most from the price information. These results are presented in table 6. First, I examine individuals who received a medical imaging procedure immediately after having an emergency. Note these are relatively minor emergency visits since I exclude inpatient admissions. I find no statistically significant effect for outpatient emergency visits. The difference between the effect for emergency and nonemergency visits is statistically significant, consistent with the idea that nonemergency procedures could potentially be scheduled further in advance, allowing time to use the website. The third and fourth columns of table 6 examine the effect for rural and urban patients, who may benefit differently from price information due to the availability of providers. The magnitude of the effect is larger for urban patients, although the difference is statistically significant only at the 10% level. Table 6. Effect of Price Transparency Website on Visit Out-of-Pocket Price, by Patient Characteristics Emergency VisitUrbanicity YesNoRuralUrban $OnWebm×Postt$ −0.018 −0.057*** −0.007 −0.060*** (0.014) (0.009) (0.030) (0.009) F statistic of difference 5.862** 2.960* Adjusted $R2$ 0.399 0.325 0.358 0.320 Observations 198,041 1,786,758 189,240 1,795,559 Age Income $≤35$ $>35$ $≤$ Mean $>$ Mean $OnWebm×Postt$ −0.060*** −0.051*** −0.049*** −0.066*** (0.018) (0.011) (0.011) (0.011) F statistic of difference 0.163 1.103 Adjusted $R2$ 0.349 0.315 0.324 0.322 Observations 325,523 1,305,073 1,179,840 804,959 Full controls Yes Yes Yes Yes Month-year FE Yes Yes Yes Yes Procedure FE Yes Yes Yes Yes Emergency VisitUrbanicity YesNoRuralUrban $OnWebm×Postt$ −0.018 −0.057*** −0.007 −0.060*** (0.014) (0.009) (0.030) (0.009) F statistic of difference 5.862** 2.960* Adjusted $R2$ 0.399 0.325 0.358 0.320 Observations 198,041 1,786,758 189,240 1,795,559 Age Income $≤35$ $>35$ $≤$ Mean $>$ Mean $OnWebm×Postt$ −0.060*** −0.051*** −0.049*** −0.066*** (0.018) (0.011) (0.011) (0.011) F statistic of difference 0.163 1.103 Adjusted $R2$ 0.349 0.315 0.324 0.322 Observations 325,523 1,305,073 1,179,840 804,959 Full controls Yes Yes Yes Yes Month-year FE Yes Yes Yes Yes Procedure FE Yes Yes Yes Yes Estimates from baseline difference-in-difference specification for various subpopulations. The dependent variable is log(1 $+$ patient out-of-pocket price). OLS regression standard errors clustered at the month-year level in parentheses. $*$$p<0.10$, $**$$p<0.05$, and $***$$p<0.01$. At the time of the website launch, younger and higher-income individuals were more likely to have broadband Internet.34 This may have increased access to the price transparency website. The bottom row of table 6 shows that the magnitude of the effect is larger for younger and higher-income individuals; however, the differences are not statistically significant. Taken together, these results imply that price transparency provides benefits across a range of demographic groups. However, the benefits accrue most to groups that had the highest ability to shop around, potentially exacerbating health inequality. ### E. Demand-Side Effects The primary motivation for the website was to allow individuals to shop around for medical care. In table 7, I directly examine provider choice outcomes using the same difference-in-difference specification to provide evidence that price effects are mediated by changes in demand. Table 7. Effect of Price Transparency Website on Search Behavior and Provider Type Dependent Variable: Same Provider as Last Medical Imaging VisitLow-Cost ProviderProvider in New HampshireDistance to Provider (miles) $OnWebm×Postt$ −0.0209** 0.0652*** 0.0036*** 0.1756** (0.0095) (0.0077) (0.0012) (0.0739) Full controls Yes Yes Yes Yes Month-year FE Yes Yes Yes Yes Procedure FE Yes Yes Yes Yes Adjusted $R2$ 0.044 0.038 0.458 0.426 Observations 806,294 1,642,953 1,984,799 1,984,799 Dependent Variable: Same Provider as Last Medical Imaging VisitLow-Cost ProviderProvider in New HampshireDistance to Provider (miles) $OnWebm×Postt$ −0.0209** 0.0652*** 0.0036*** 0.1756** (0.0095) (0.0077) (0.0012) (0.0739) Full controls Yes Yes Yes Yes Month-year FE Yes Yes Yes Yes Procedure FE Yes Yes Yes Yes Adjusted $R2$ 0.044 0.038 0.458 0.426 Observations 806,294 1,642,953 1,984,799 1,984,799 Estimates from a linear probability model using the same controls as the baseline difference-in-difference specification presented in equation (1). A low-cost provider is defined as a provider with an average out-of-pocket cost in the lowest decile in each county conditional on procedure, insurer, and year. OLS regression standard errors clustered at the month-year level in parentheses. $*$$p<0.10$, $**$$p<0.05$, and $***$$p<0.01$. Over the period, 34% of individuals with repeat visits went to the same provider as their prior visit. Using a difference-in-difference linear probability model where the dependent variable is an indicator for whether the chosen provider is the same as the prior visit, I examine whether the website made individuals more likely to switch providers. Column 1 in table 7 implies that the website significantly reduced the probability of going to the same provider as the previous visit by about 6% (2 percentage point reduction from the mean). New Hampshire is a relatively small state, and many residents work in surrounding states, particularly in Massachusetts.35 Almost a third of the individuals in the sample go to medical imaging providers outside the state; however, the website provides only information about the price of providers within the state. The third column of table 7 shows that the website significantly increased the probability that individuals went to a provider within New Hampshire. Table 7, column 4, presents the results using distance between the individual and provider zip code as the dependent variable. When individuals lack price information, they may choose a provider close to their home since distance is a known characteristic. Conversely, if prices are known, they may be willing to drive farther.36 I find evidence for this hypothesis: when the website is available, individuals drive farther on average. One limitation of the data is that there is no information about patient referrals. The website may have affected individuals' choice of provider, or it may have affected physicians' referrals. Since I do not observe referrals, I do not differentiate between these mechanisms. ### F. Supply-Side Effects Given that individuals switch to lower-cost providers, did this put downward pressure on prices? I examine supply-side effects by estimating a model with controls for demand-side factors, namely, provider-procedure-insurer fixed effects. The supply-side effect for each time period is shown in figure 3.37 Panel a shows the effect on the total visit price, which includes all supplemental procedures. Compared to the equilibrium effect presented in figure 2a, the point estimates are smaller in magnitude, especially in the period right after the introduction of the website. However, the estimates are still highly significant, especially in the later period, implying that provider prices decline in the long run. Figure 3. Supply-Side Effect of Price Transparency Website, by Time Period The charts show point estimates for each half-year using the difference-in-difference specifications described in section IVF. The specification controls for demand-side changes; therefore point estimates reflect only supply-side effects. The omitted period is the half-year prior to the start of the price transparency website. The chart on the left shows estimates using the entire visit price as the dependent variable, while the chart on the right shows the effect when the dependent variable does not include supplementary procedures. Error bars indicate 95% confidence interval using standard errors clustered at the month-year level. Figure 3. Supply-Side Effect of Price Transparency Website, by Time Period The charts show point estimates for each half-year using the difference-in-difference specifications described in section IVF. The specification controls for demand-side changes; therefore point estimates reflect only supply-side effects. The omitted period is the half-year prior to the start of the price transparency website. The chart on the left shows estimates using the entire visit price as the dependent variable, while the chart on the right shows the effect when the dependent variable does not include supplementary procedures. Error bars indicate 95% confidence interval using standard errors clustered at the month-year level. Recall that the visit price is determined by the prices of a bundle of procedures. The primary way that the provider can change the price of the bundle is by changing the price of the principal medical imaging procedure, which makes up more than half of the cost of the bundle on average. Figure 3b examines the effect on the principal procedure price. There is a large supply-side effect three to five years after the website, which I interpret as evidence that the website reduced negotiated prices in the long run. Columns 1 and 2 in table 8 formalize these results. When isolating the supply side, the short-run effect is quite small. Visit prices declined by 1.0% while the principal procedure price declined is negligible. The long-run effects are larger: a 1.7% reduction in visit prices and 3.0% reduction in principal procedure price. These results are statistically significant. These results should be interpreted as the supply-side effect experienced for the average consumer in the sample. Table 8. Supply-Side Effect of Price Transparency Website Total Visit Effect by County HHITotal Visit Effect by Provider Type Total Visit PricePrincipal Procedure Price$≤$ First Quartile HHI$>$ Forth Quartile HHIHospitalnon-hospital $OnWebm×PostShortRunt$ −0.010* −0.005 0.007 −0.042*** −0.009 −0.009 (0.006) (0.006) (0.012) (0.014) (0.011) (0.007) $OnWebm×PostLongRunt$ −0.017*** −0.030*** 0.000 −0.048*** 0.016 −0.024*** (0.006) (0.006) (0.012) (0.015) (0.013) (0.006) Individual controls Yes Yes Yes Yes Yes Yes Provider $×$ Procedure $×$ Insurer FE Yes Yes Yes Yes Yes Yes Month-Year FE Yes Yes Yes Yes Yes Yes F-statistic of difference (SR) 7.355*** 0.002 F-statistic of difference (LR) 6.723** 7.393*** Mean level 950.38 450.01 6.01 5.99 6.60 5.95 Adjusted $R2$ 0.497 0.807 0.475 0.467 0.559 0.464 Observations 1,967,086 1,967,083 609,793 441,664 338,478 1,628,608 Total Visit Effect by County HHITotal Visit Effect by Provider Type Total Visit PricePrincipal Procedure Price$≤$ First Quartile HHI$>$ Forth Quartile HHIHospitalnon-hospital $OnWebm×PostShortRunt$ −0.010* −0.005 0.007 −0.042*** −0.009 −0.009 (0.006) (0.006) (0.012) (0.014) (0.011) (0.007) $OnWebm×PostLongRunt$ −0.017*** −0.030*** 0.000 −0.048*** 0.016 −0.024*** (0.006) (0.006) (0.012) (0.015) (0.013) (0.006) Individual controls Yes Yes Yes Yes Yes Yes Provider $×$ Procedure $×$ Insurer FE Yes Yes Yes Yes Yes Yes Month-Year FE Yes Yes Yes Yes Yes Yes F-statistic of difference (SR) 7.355*** 0.002 F-statistic of difference (LR) 6.723** 7.393*** Mean level 950.38 450.01 6.01 5.99 6.60 5.95 Adjusted $R2$ 0.497 0.807 0.475 0.467 0.559 0.464 Observations 1,967,086 1,967,083 609,793 441,664 338,478 1,628,608 Estimates from the difference-in-difference specification that controls for demand-side factors presented in equation (2). The dependent variable is $log(1+y)$, where $y$ is either the visit price or principal procedure price. For visit price, the unit of observation is a patient visit, which may contain multiple medical claims. For principal procedure price, the unit of observation is the primary medical imaging procedure within each visit. HHI is calculated for individuals' county for each procedure class in the period before the website. The sample consists of all commercial claims related to outpatient medical imaging procedures in New Hampshire from 2005 to 2011. OLS regression standard errors clustered at the month-year level in parentheses. $*$$p<0.10$, $**$$p<0.05$, and $***$$p<0.01$. Overall, this is evidence of a significant reduction in negotiated prices, especially in the long run. The fact that the principal procedure price is most affected in the long run is consistent with the fact that prices are renegotiated infrequently. One caveat is that the changes in demand caused providers to reduce the price of supplemental procedures in addition to principal procedure prices. Since supplemental procedures, such as contrast agents and examinations related to medical imaging procedures, are common across procedures on the website and procedures not on the website, these reductions in prices would be differenced-out. This would lead to an underestimate of the supply-side effects of price transparency. Thus, table 8's estimates of the savings from the website are conservative. Providers operating in concentrated markets may be able to negotiate higher prices with insurers (Dranove, Shanley, & White, 1993; Town & Vistnes, 2001; Gowrisankaran et al., 2015). I define the Herfindahl index by county and procedure category in the period prior to the introduction of the website. There is significant variation in competition: some counties have a single provider in the market for certain procedure categories, while others are relatively unconcentrated. Columns 3 and 4 of table 8 present the supply-side effect on total visit price by market concentration. In both the short run and long run, there is a larger effect of the website in high-concentration markets (those with Herfindahl index above the fourth quartile) compared to low-concentration markets (those at or below the first quartile). This difference is statistically significant for both the short and long runs. This is evidence that price transparency put the most downward pressure on prices in markets where price cost margins were likely the highest. Finally, columns 5 and 6 of table 8 compare results for hospitals and nonhospital providers, which are likely free-standing outpatient facilities. The results are insignificant for hospitals, although there is a smaller sample. The supply-side effects are primarily driven by nonhospital providers, which decrease prices by 2.4% in the long run. The difference between the effect for hospitals and nonhospitals is statistically significant at the 1% level. I also examine a similar specification after aggregating to the provider-procedure-insurer level, thus capturing the unweighted effect across all providers regardless of market share. The magnitude of the supply-side effect is slightly larger, indicating that the providers with low market share decreased prices even more.38 ### G. Robustness of Supply-Side Effects One important concern is that the supply-side effects may be biased due to the fact that procedures not on the website are affected due to providers' reducing all prices in response to the website or due to increasing prices for procedures that are not on the website to compensate for lost profits from procedures on the website. Although there is little theoretical justification for why profit-maximizing providers would not optimize prices for each procedure individually (see Frakt, 2011), I address this concern by exploiting cross-state variation in price information.39 Although the NHCHIS data set includes only individuals insured in New Hampshire, many of these individuals live or work outside New Hampshire and go to providers outside New Hampshire, largely in Massachusetts and Vermont. Therefore, prices for medical imaging providers outside New Hampshire can be used as a control group since the website listed prices only for providers within New Hampshire. Unlike prices for medical imaging procedures that are not on the website at providers in New Hampshire, the prices for medical imaging procedures outside the state are not affected by cost shifting.40 I limit the sample to procedures available on the website, exploiting variation across time and across state borders. The specification is presented in section IVA. The results, which can be found in table A-4 are qualitatively similar as the baseline specification. In the short run, the estimates do not imply a statistically significant effect on negotiated prices. However, there is a negative and significant effect in the long run. The magnitude of the effect is larger than the specification presented in table 8, implying almost a 5% reduction in prices in the long run. Overall, these estimates provide evidence that the supply-side results presented in table 8 do not reflect providers that raised the price of procedures not on the website to offset lower negotiated prices for procedures on the website. The downside of this approach is that there is a smaller sample. ### H. Price Dispersion Theoretical work has emphasized that information frictions can give rise to price dispersion in equilibrium even when products are homogeneous (Salop & Stiglitz, 1977; Burdett & Judd, 1983). At the same time, recent work has documented the large degree of price variation for privately provided health care, even for relatievely standardized procedures (Cooper et al., 2018), raising questions about the underlying cause of this variation. In this section, I examine whether a reduction in information frictions can reduce price dispersion, as measured by the interquartile range of prices. I find evidence that the website reduced the range of transaction prices, as well as the range of negotiated provider prices. Columns 1 and 2 in table 9 present the results for transaction prices. There is a$231 reduction in the interquartile range of visit prices on the website relative to those not on the website (a 19% reduction from the mean of $1,184). There is also a$104 reduction in the interquartile range of principal procedure prices. Both results are significant at the 1% level.

Table 9.
Effect of Price Transparency Website on Price Dispersion
Interquartile Range of Transaction PricesInterquartile Range of Provider Prices
Total Visit PricePrincipal Procedure PriceTotal Visit PricePrincipal Procedure Price
$OnWebm×Postt$ −231.0*** −103.7*** −158.9** −96.3***
(63.8) (24.0) (68.7) (20.7)
Individual controls Yes Yes Yes Yes
Procedure FE Yes Yes Yes Yes
Month-year FE Yes Yes Yes Yes
Mean IQR 1183.5 658.5 992.4 541.4
Adjusted $R2$ 0.307 0.447 0.271 0.465
Observations 13,572 13,572 13,572 13,572
Interquartile Range of Transaction PricesInterquartile Range of Provider Prices
Total Visit PricePrincipal Procedure PriceTotal Visit PricePrincipal Procedure Price
$OnWebm×Postt$ −231.0*** −103.7*** −158.9** −96.3***
(63.8) (24.0) (68.7) (20.7)
Individual controls Yes Yes Yes Yes
Procedure FE Yes Yes Yes Yes
Month-year FE Yes Yes Yes Yes
Mean IQR 1183.5 658.5 992.4 541.4
Adjusted $R2$ 0.307 0.447 0.271 0.465
Observations 13,572 13,572 13,572 13,572

Estimates from the difference-in-difference specification described in section IVA. The unit of observation is a procedure-month. Interquartile range is defined as the difference between 75th and 25th percentiles price for each procedure-month. OLS regression standard errors clustered at the month-year level in parentheses. $*$$p<0.10$, $**$$p<0.05$, and $***$$p<0.01$.

More important, I examine the effect on the interquartile range of negotiated provider prices (see columns 3 and 4 in table 9). The interquartile range of visit prices declines by $159, a 16% reduction from the mean. Focusing on the principal procedure price, which was likely most affected by the website, the interquartile range of provider prices declined by$96, or 17.8% of the mean. These results are also statistically significant.

Together with the previous results, these results imply that the website decreased both the mean and variance of the distribution of transaction prices. This was due in large part to a shift in the distribution of negotiated prices. Consistent with the theoretical literature on information frictions and price dispersion, lower information friction allows patients and insurers greater ability to discipline the market.

## V. Discussion and Conclusion

The health care system can be complicated to navigate, and information frictions are thought to be pervasive (Reinhardt, 2012). In this paper, I examine how a publicly available website providing price information affected the market for medical imaging procedures. While previous research has focused on the demand-side effect of information supplied by employers, I examine the equilibrium effects of a statewide initiative.

Overall, the HealthCost website reduced the cost of medical imaging procedures by 5% for patients and 4% for insurers. A simple calculation implies that individuals saved around $7.9 million and insurers saved$36.0 million on x-ray, CT scan, and MRI scans over the five-year period.41 While demand-side effects are important, there are significant supply-side effects in the long run when information is available to all individuals. In other words, this is evidence that price opacity softens provider competition, leading to higher prices. This is particularly important given that the average price of medical imaging procedures in the United States is roughly double that of other OECD countries.42

Previous research has found modest effects of price transparency initiatives at the employer level. However, this paper provides evidence that price transparency can be effective in the long run, especially when it is available to the entire market. The presence of supply-side effects also implies that usage of the price transparency tool generates positive spillovers for other consumers due to lower prices, motivating the public provision of price information. Given that a sizable portion of benefits accrues to individuals under their deductible, it also implies that insurers may not have a strong incentive to provide these tools.

While this paper focuses on x-rays, CT scans, and MRI scans, it is important to consider whether the results apply more broadly. It has been estimated that 30% to 40% of spending on medical services is for procedures that are shoppable, and price transparency may be able to generate equilibrium effects for these procedures as well.43 Price transparency tools are unlikely to have an effect for other procedures, such as complicated surgeries with prices that are determined on a case-by-case basis.

Finally, this paper does not examine other margins of adjustment such as the entry and exit of providers. All specifications implicitly control for changes in the set of providers; however, it may be that the website caused new, low-cost medical imaging providers to enter the market.44 In fact, there was entry of free-standing outpatient medical imaging facilities after the website started, although it is not clear that it was due to the website.45 At the same time, price transparency could negatively affect the profitability of more expensive providers, such as hospitals, potentially leading to exit. Future work should examine these additional margins.

Given that website traffic logs reveal that only a small fraction of individuals receiving medical imaging procedures in New Hampshire use the website, the supply-side effects may be quite large if all consumers were informed about prices. This is particularly important because policies that lower prices are seen as key for lowering the cost of privately provided health care in the United States. I examine these issues in future work.

## Notes

1

Surveys show that consumers rarely compare prices for medical services. See Public Agenda (2015).

2

See Clay et al. (2001), Morton, Zettelmeyer, and Silva-Risso (2001), Brown and Goolsbee (2002), and Goldmanis et al. (2010).

3

For example, see Cutler and Dafny (2011).

4

For instance, see Anderson et al. (2003), Koechlin, Lorenzoni, and Schreyer (2010), and Cooper et al. (2018).

5

Even markets with homogeneous products can exhibit price dispersion in equilibrium in the presence of information frictions. This is true with heterogeneous consumers (Salop & Stiglitz, 1977) as well as homogeneous consumers (Burdett & Judd, 1983).

6

Note that Christensen, Floyd, and Maffett (2015) examine the effect of information about list prices (rather than out-of-pocket prices) and find little evidence of effects on negotiated prices.

7

The authors hypothesize that price transparency could either lower or raise prices, but note that “it is too early to tell what the outcome of experiments with increased transparency will be.”

8

A majority of states and the federal government have proposed some form of price transparency. See Nicholson (2015).

9

Note that a large literature also focuses on variation in Medicare spending (Fisher et al., 2003; Fisher, Bynum, & Skinner, 2009).

10

I calculate the first quartile of the price distribution conditional on individuals' insurance and procedure. I consider the case in which all individuals paying about this price switch to the provider charging the first quartile price.

11

According to a nationally representative survey, 79% of individuals stated that they could not compare prices (or did not even try) before receiving medical care (Public Agenda, 2015).

12

Originally nhhealthcost.org; the website can now be found at nhhealthcost.nh.gov.

13

In early 2016, after my period of analysis, the website added information about provider quality and a guide to health insurance. The website also has a separate feature providing price information for uninsured individuals. I do not observe uninsured individuals and therefore do not examine the effect of this information.

14

New Hampshire was the only state to receive a grade of A from Catalyst for Payment Reform's 2015 Report Card on State Price Transparency Laws.

15

For instance, the website requires individuals to know their remaining deductible. However, there is evidence that some people may not know this information (Cunningham, Denk, & Sinclair, 2001; Handel & Kolstad, 2015).

16

Mehrotra et al. (2014) examine usage of the New Hampshire HealthCost tool from 2011 to 2013 using Google Analytics data and find small take-up relative to the state population. I use server traffic logs starting in 2007, a period in which Google Analytics data were not available. In conversations with the administrators of the HealthCost website, there was concern that Google Analytics data were not capturing all of the ways in which individuals accessed the HealthCost website.

17

If individuals visit the website but do not use the search tool because their procedure is not listed, they are not counted in the website traffic data.

18

See, for example, Newhouse (1992).

19

See Tu and Gourevitch (2014).

20

For more information, see “Exeter Hospital Says Costs Being Used as Negotiating Tactic,” Seacoastonline.com, November 14, 2010.

21

I explore these mechanisms in more detail in a follow-up work.

22

Current Procedural Terminology (CPT) are codes developed and maintained by the American Medical Association. Healthcare Common Procedure Coding System (HCPCS) codes are an extension of CPT codes that include additional procedures, such as nonphysician services.

23

The website does include prices for a few inpatient procedures that are not related to medical imaging (e.g., newborn delivery).

24

For more detail on the construction of demographic covariates, see online appendix, section A.

25

Individuals who know they will fulfill their deductible over the course of the year should not be price sensitive. However, to the extent that they have uncertainty about their future health care use or are myopic, individuals will be price sensitive even if they are close to hitting their deductible. For this reason, I consider all individuals who have not passed their deductible.

26

Capitation payments for medical imaging procedures were negligible during the period.

27

For the insurer fixed effects, I define an insurance plan as a unique combination of insurance firm and insurance type (e.g., Anthem HMO).

28

I do in fact find that the magnitude of estimated effects is larger when examining the unweighted supply-side effect using a similar specification after aggregating the data to the provider-procedure-insurer level.

29

Providers that perform a given procedures five or fewer times over the period are removed when calculating the interquartile range.

30

The specification used for figure 2 is $log(1+pimjkt)=∑tβt(OnWebm×SemiYeart)+αXit+λm+λk+ɛimjkt$. The interaction with the time period before the introduction of the website is omitted.

31

Depending on the specifics of the plan design, some procedures may not count toward an individual's deductible. In addition, some individuals classified as under their deductible hit their deductible on the current visit. For these reasons, the out-of-pocket cost is often less than the full price when an individual is not past the deductible.

32

I conduct a Wald test and determine the difference between the effect for an individual subject to a deductible and those not subject to a deductible is significant.

33

Coinsurance payments are a set percentage of the total price, often between 5% and 25%, that the individual pays.

34

Individuals age 18 to 29 were 23 percentage points more likely to have broadband Internet than those age 50 to 64. Those with income over $75,000 were 46 percentage points more likely than those with an income under$30,000. See Pew Internet and American Life Project, Home Broadband Adoption, July 2007.

35

About 17% of workers living in New Hampshire work in a different state, one of the highest rates in the nation. See Out-of-State and Long Commutes, Census 2011.

36

In follow-up work, I present a model that formalizes this intuition.

37

The specification used for figure 2 is $log(1+pimjkt)=∑tβt(OnWebm×SemiYeart)+αXit+λmjk+λt+ɛimjkt$. The interaction with the time period before the introduction of the website is omitted.

38

Results available upon request.

39

Note that the cross-price elasticity between different procedures is likely 0 (e.g., there is no substitution between arm X-rays and leg X-rays). Therefore, there is little theoretical justification why a change in demand for one procedure would generate spillover effects for other procedures when firms are profit maximizing.

40

In particular, any changes in demand due to the website are unlikely to affect prices outside New Hampshire, given that New Hampshire patients likely make up a very small fraction of patients in these states.

41

All figures in 2010 dollars.

42

The average price of an MRI scan is $1,200 in the United States but only$569 in other OECD countries with available data. The average price of CT scan is $228 in the United States but only$98 in other OECD countries with available data. See Squires (2011).

43

See White and Eguchi (2014) and Health Care Cost Institute (2016).

44

In particular, month fixed effects absorb entry and exit of providers.

45

Analyzing these issues would require a different identification strategy beyond the scope of this paper.

## REFERENCES

Anderson
,
Gerard F.
,
Uwe E.
Reinhardt
,
Peter S.
Hussey
, and
Varduhi
Petrosyan
, “
It's the Prices, Stupid: Why the United States Is So Different from Other Countries,
Health Affairs
22
(
2003
),
89
105
.
Brown
,
Jeffrey R.
, and
Austan
Goolsbee
, “
Does the Internet Make Markets More Competitive? Evidence from the Life Insurance Industry,
Journal of Political Economy
110
(
2002
),
481
507
.
Burdett
,
Kenneth
, and
Kenneth L.
Judd
, “
Equilibrium Price Dispersion,
Econometrica
51
(
1983
),
955
969
.
Charlson
,
Mary E.
,
Peter
Pompei
,
Kathy L.
Ales
, and
C. Ronald
MacKenzie
, “
A New Method of Classifying Prognostic Comorbidity in Longitudinal Studies: Development and Validation,
Journal of Chronic Diseases
40
(
1987
),
373
383
.
Christensen
,
Hans Bonde
,
Eric
Floyd
, and
Mark G.
Maffett
, “
The Effects of Price Transparency Regulation on Prices in the Healthcare Industry
,”
Chicago Booth research paper
14-33
(
2015
).
Clay
,
Karen
,
Ramayya
Krishnan
, and
Eric
Wolff
, “
Prices and Price Dispersion on the Web: Evidence from the Online Book Industry,
Journal of Industrial Economics
49
(
2001
),
521
539
.
Cooper
,
Zack
,
Stuart V.
Craig
,
Martin
Gaynor
, and
John Van
Reenen
, “
The Price Ain't Right? Hospital Prices and Health Spending on the Privately Insured,
Quarterly Journal of Economics
134
(
2018
),
51
2007
.
Cunningham
,
Peter J.
,
Charles
Denk
, and
Michael
Sinclair
, “
Do Consumers Know How Their Health Plan Works?
Health Affairs
20
(
2001
),
159
166
.
Cutler
,
David
, and
Leemore
Dafny
, “
Designing Transparency Systems for Medical Care Prices,
New England Journal of Medicine
364
(
2011
),
894
895
.
Desai
,
Sunita
,
Laura
Hatfield
,
Andrew
Hicks
,
Michael
Chernew
, and
Ateev
Mehrotra
, “
Association between Availability of a Price Transparency Tool and Outpatient Spending,
JAMA
315
(
2016
),
1842
1843
.
Diamond
,
Peter A.
, “
Journal of Economic Theory
3
(
1971
),
156
168
.
Dranove
,
David
,
Mark
Shanley
, and
William D.
White
, “
Price and Concentration in Hospital Markets: The Switch from Patient-Driven to Payer-Driven Competition,
Journal of Law and Economics
36
(
1993
),
179
204
.
Fisher
,
Elliott S.
,
David E.
Wennberg
,
Thrse A.
Stukel
,
Daniel J.
Gottlieb
,
F. Lee
Lucas
, and
Etoile L.
Pinder
, “
The Implications of Regional Variations in Medicare Spending. Part 1: The Content, Quality, and Accessibility of Care,
Annals of Internal Medicine
138
(
2003
),
273
287
.
Fisher
,
Elliott S.
,
Julie P.
Bynum
, and
Jonathan S.
Skinner
, “
Slowing the Growth of Health Care Costs—Lessons from Regional Variation,
New England Journal of Medicine
360
(
2009
),
849
852
.
Frakt
,
Austin B.
, “
How Much Do Hospitals Cost Shift? A Review of the Evidence,
Milbank Quarterly
89
(
2011
),
90
130
.
Goldmanis
,
Maris
,
Ali
Hortaçsu
,
Syverson
, and
Onsel
Emre
, “
E-Commerce and the Market Structure of Retail Industries,
Economic Journal
120
(
2010
),
651
682
.
Gowrisankaran
,
Gautam
,
Aviv
Nevo
, and
Robert
Town
, “
Mergers When Prices Are Negotiated: Evidence from the Hospital Industry,
American Economic Review
105
(
2015
),
172
203
.
Gudiksen
,
Katie
, “
Spotlight on 2018 State Legislation Summary: The Year in Review,
Source on Healthcare Price and Competition
(
December 17
,
2018
).
Handel
,
Benjamin R.
, and
Jonathan T.
, “
Health Insurance for ‘Humans’: Information Frictions, Plan Choice, and Consumer Welfare,
American Economic Review
105
(
2015
),
2449
2500
. doi:10.1257/aer.20131126.
Health Care Cost Institute
,
Spending on Shoppable Services in Health Care,
issue brief 1
(
2016
).
Ho
,
Katherine
, “
Insurer-Provider Networks in the Medical Care Market,
American Economic Review
99
(
2009
),
393
430
.
Ho
,
Katherine
, and
Robin S.
Lee
, “
Insurer Competition in Health Care Markets,
Econometrica
85
(
2017
),
379
417
. doi:10.3982/ECTA13570.
Koechlin
,
Francette
,
Luca
Lorenzoni
, and
Paul
Schreyer
, “
Comparing Price Levels of Hospital Services across Countries,
OECD iLibrary
(
2010
).
Lieber
,
Ethan M. J.
, “
Does It Pay to Know Prices in Health Care?
American Economic Journal: Economic Policy
9
(
2017
),
154
179
.
Manning
,
Willard G.
, and
John
Mullahy
, “
Estimating Log Models: To Transform or Not to Transform?
Journal of Health Economics
20
(
2001
),
461
494
.
Mehrotra
,
Ateev
,
Tyler
Brannen
, and
Anna
Sinaiko
, “
Use Patterns of a State Health Care Price Transparency Web Site: What Do Patients Shop for?
Inquiry: Journal of Health Care Organization, Provision, and Financing
51
(
2014
),
1
3
.
Morton
,
Fiona Scott
,
Florian
Zettelmeyer
, and
Jorge
Silva-Risso
, “
Internet Car Retailing,
Journal of Industrial Economics
49
(
2001
),
501
519
.
Newhouse
,
Joseph P.
, “
Medical Care Costs: How Much Welfare Loss?
Journal of Economic Perspectives
6
:
3
(
1992
),
3
21
.
Newhouse
,
Joseph P.
,
Alan M.
Garber
,
Robin P.
Graham
,
Margaret A.
McCoy
,
Michelle
Mancher
,
Ashna
Kibria
et al
, et al.,
Variation in Health Care Spending: Target Decision Making, Not Geography
(
Washington, DC
:
,
2013
).
Nicholson
,
Elizabeth
, “
2015 Price Transparency Initiative State Survey: Price Disclosures,
Source on Healthcare Price and Competition Technical Report
(
2015
).
Philipson
,
Tomas J.
,
Seth A.
Seabury
,
Lee M.
Lockwood
,
Dana P.
Goldman
,
Darius N.
Lakdawalla
, and
David M.
Cutler
, “
Geographic Variation in Health Care: The Role of Private Markets,
Brookings Papers on Economic Activity
(
2010
)
325
361
.
Public Agenda
, “
How Much Will It Cost? How Americans Use Prices in Health Care
” (March
2015
), https://www.publicagenda.org/media/how-much-will-in-cost
Reinhardt
,
Uwe
,
Divide et Impera: Protecting the Growth of Health Care Incomes (Costs)
,
Health Economics
21
:
1
(
2012
)
41
54
.
Salop
,
Steven
, and
Joseph
Stiglitz
, “
Bargains and Ripoffs: A Model of Monopolistically Competitive Price Dispersion,
Review of Economic Studies
44
(
1977
),
493
510
.
Sinaiko
,
Anna D.
, and
Meredith B.
Rosenthal
, “
Increased Price Transparency in Health Care—Challenges and Potential Effects,
New England Journal of Medicine
364
(
2011
),
891
894
.
Sinaiko
,
Anna D.
, and
Meredith B.
Rosenthal
, “
Examining a Health Care Price Transparency Tool: Who Uses it, and How They Shop for Care,
Health Affairs
35
(
2016
),
662
670
.
Squires
,
David A.
, “
The US Health System in Perspective: A Comparison of Twelve Industrialized Nations,
Issue Brief (Commonwealth Fund)
16
(
2011
)
1
14
.
Stagg
,
Vicki
, “
Charlson: Stata Module to Calculate Charlson Index of Comorbidity
Statistical Software Components,” Boston College Department of Economics
(April
2006
).
Stigler
,
George J.
, “
The Economics of Information,
Journal of Political Economy
69
(
1961
)
213
.
Town
,
Robert
, and
Gregory
Vistnes
, “
Hospital Competition in HMO Networks,
Journal of Health Economics
20
(
2001
):
733
753
.
Tu
,
Ha
, and
Rebecca
Gourevitch
, “
Moving Markets: Lessons from New Hampshire's Health Care Price Transparency Experiment,
California HealthCare Foundation
, April
2014
.
Whaley
,
Christopher M.
, “
Searching for Health: The Effects of Online Price Transparency,
University of California, Berkeley Working Paper
(
2015a
).
Whaley
,
Christopher M.
Provider Responses to Online Price Transparency,
University of California, Berkeley Working Paper
(
2015b
).
White
,
Chapin
, and
Megan
Eguchi
, “
Reference Pricing: A Small Piece of the Health Care Price and Quality Puzzle,
Mathematica Policy research reports
(
2014
).

## Author notes

I am grateful to Kate Ho, Chris Conlon, Michael Riordan, Tal Gross, and Doug Almond for helpful comments and suggestions. I also thank Chris Hansman, Evan Riehl, Maria Micaela Sviatschi, many seminar participants, and this journal's editor and referees. In addition, I thank the New Hampshire Department of Health and Human Services, along with Tyler Brannen and Maureen Mustard, for providing data and background. I gratefully acknowledge support from the National Science Foundation and the National Institute on Aging (T32-AG000186). Views expressed in the paper, as well as any errors, are my own.

A supplemental appendix is available online at http://www.mitpressjournals.org/doi/suppl/10.1162/rest_a_00765.