Price Transparency and Healthcare Costs: The Case of the New Hampshire Healthcare Market. Rui Wang Tulane University.

Transcription

1 Price Transparency and Healthcare Costs: The Case of the New Hampshire Healthcare Market Rui Wang Tulane University November 2, 2018 Abstract There is a longstanding debate in health economics and health policy concerning whether price transparency reduces healthcare spending. A common argument is that price information reduces healthcare costs, in part by allowing consumers to engage in more cost-conscious shopping through the selecting of lower-cost providers, in addition to increasing the ability of insurers to negotiate lower prices. However, an alternative argument is that price transparency may increase the level of healthcare prices by facilitating tacit collusion or by misdirecting patients to high-cost providers. This paper examines plausibly exogenous variation in the total payments of imaging services generated by the introduction of New Hampshire HealthCost (NHHC), a public website that discloses the bundle prices of healthcare procedures. I use rich visit-level information from the New Hampshire commercial claims database, which includes all claims related to imaging services delivered in New Hampshire from 2005 to I find that, due to supply-side responses, NHHC reduced total payments for procedures with new price information by 1.3%. Given that the average total payments of these procedures was $586 prior to NHHC and there were 439,454 related visits after NHHC was launched, this is an approximate $7.40 reduction per visit and $3,241,788 in total savings. I also find that the reduction was more pronounced for insurers who paid within the top quintile, relative to their competitors, before the new information was available. Finally, I show that the reduction in prices was largely attributable to renegotiation frictions. 1. Introduction 1. Introduction Over 61% of the U.S. population has private health insurance with healthcare prices set through a complex negotiation process between insurers and providers (Cooper et al., 2018). Insurers negotiate different contracts with different providers, which causes considerable price variation. Meanwhile, nearly all insurance contracts contain a gag clause that prevents both insurers and providers from disclosing their negotiated fees to anyone outside the contract (Whaley 2015). Thus, the healthcare market often features a lack of price transparency, and significant price variation is difficult to observe. Without price information, patients are unaware of price variation and are also unable to choose cheaper providers. Insurers are limited in their ability to negotiate low prices through threatening to steer their patients elsewhere, and instead negotiate lower prices by leveraging their ability to select networks. As a result, providers can increase their bargaining ability by focusing on non-price attributes, and thus charge more with little pressure of losing patients because of the high prices (Whaley 2015). Along with improvements in technology that make it easier to collect and distribute data, price transparency initiatives have emerged as a potential way to reduce price variation and healthcare spending. In March 2007, New Hampshire (NH) launched a public website called HealthCost, which 1

2 discloses prices for 31 common outpatient medical procedures. This website allows users to input their own health plans and search related estimated prices of procedures across providers in NH. Employing the difference-in-differences method, Brown (2017a) finds HealthCost narrowed price variation and reduced healthcare spending on imaging services by 3%. Brown (2017b) develops an empirical model of demand and supply for healthcare services and shows that the equilibrium prices of imaging services in NH decrease if more consumers use HealthCost. Although Brown(2017a) finds that the reduction of healthcare spending was due in part to lower negotiated prices from the supply side by controlling for the consumer response, he does not move forward to study the insurerprovider contracting behavior that directly results in the lower negotiated prices. This paper extends Brown (2017a) analysis by in depth investigating how New Hampshire HealthCost (NHHC) affected insurers contracting behavior with providers, and how the related renegotiation friction affected healthcare costs. From an insurer s perspective, I develop a theoretical model in which insurers face uncertainty about the provider s costs or bargaining parameters, so that price transparency reduces uncertainty and permits insurers to negotiate lower prices with providers they previously inaccurately perceived as high-cost. Thus, the model shows the potential heterogeneity in the renegotiated prices by whether the insurers paid more to the same providers than their competitors before the price is disclosed. To test the prediction from the theoretical model, I examine the heterogenous effect of NHHC on negotiated prices caused through insurer-provider renegotiations. Price transparency reduces healthcare costs through two mechanisms. On the demand side, price transparency allows consumers to engage in more cost-conscious shopping and select lower-cost providers or cheaper procedures. On the supply side, price transparency can increase the ability of insurers to bargain with providers for lower negotiated prices. Given that this paper focuses on how price transparency influences healthcare costs through changing the insurer-provider contracting behavior and the related negotiated prices, I conduct a baseline evaluation of the average supply-side effect of NHHC on healthcare costs before studying heterogeneous contracting behaviors. Similar to Brown (2017a), this paper uses two sources of variation to provide plausibly causal identification---the observed pre- and post-negotiated prices between insurers and providers and the variation of price disclosure across procedures due to NHHC s inclusion of only a few procedures. By comparing prices of the procedures with and without disclosed price information, before and after the introduction of NHHC, I identify the effect of NHHC on healthcare costs. Inspired by Brown (2017a), I provide evidence that the inconsistency between Current Procedural Terminology (CPT) codes and procedure descriptions on the website lead consumers and insurers to not only respond to the procedures exactly disclosed on the website (defined as Tier 1 procedures), but also to those procedures that were not disclosed on the website but had shared either CPT codes or procedure descriptions with Tier 1 procedures (defined as Tier 2 procedures). Thus, I regard Tier 1 and Tier 2 procedures as the procedures with disclosed price information and the effect of NHHC includes the effect on both Tier 1 and Tier 2 procedures. 1 Additionally, the effect identified here is the overall effect driven by both consumers switching to lower-cost providers or cheaper procedures and insurerprovider contracting with lower negotiated prices. I isolate the supply-side effect by introducing provider-procedure-insurer fixed effects to control for any heterogeneity caused by consumers 1 Although Brown(2017a) also define Tier1 and Tier 2 procedures as the procedures with disclosed price information, he does not provide the formal evidence that Tier 2 should be included in the treatment procedures. 2

3 choosing different providers, procedures, or insurers. I find that NHHC reduced negotiated prices by 1.3% via supply-side responses over the period Given that the average total payments of these procedures was $586 prior to NHHC and there were 439,454 related visits after NHHC was launched, this is an approximate $7.40 reduction per visit and $3,241,788 in total savings. This result is robust when emergency visits are used instead of full sample, and with the introduction of crossstate variation. To extend Brown (2017a), this paper further studies insurer-provider contracting behaviors under price transparency. The baseline supply-side effect results from at least two types of insurer contracting. First, insurers renegotiate with the existing network providers for lower negotiated prices based on the available price information. Second, insurers reselect network providers to contract for lower negotiated prices motivated by lower search costs which are facilitated by price information. Given the data limitations, this paper focuses on insurer-provider renegotiations. Following the prediction of the theoretical model, this paper mainly investigates the heterogeneous effects on negotiated prices resulted from insurer-provider renegotiations by introducing a third variation in insurer payments prior to NHHC---the quintile of the median payment for a procedure that an insurer paid to a given provider relative to payments paid by other insurers to the same provider. Based on the sample where an insurer-provider contract existed in each fiscal year, I find little average treatment effects on negotiated prices caused by insurer-provider renegotiation under price transparency, but a significant effect for insurers that were paying within the top quintile of prices relative to their competitors at the time NHHC was introduced. Specifically, this effect is -3 percentage points or -5.66% of the baseline prices for treatment procedures. Notably, the reduction of total payments was passed on to insurers instead of consumers with 4.5% less in insurer payments. These results are robust to using alternative quintiles of the median prices that insurers paid in the last year instead of prior to NHHC. It s noteworthy that the reduction of the total payment for insurers paid in the top quintile to a given provider is different from that of Brown(2017a) who finds the interquartile range of negotiated prices decreased. The prices that insurers paid in the top quintile given a specific provider are not necessarily in the top quintile or quartile of negotiated prices across providers. To check this difference, I also use the quintile of the median price an insurer paid to a provider for a procedure relative to the median prices that other insurers paid for the same procedure, but not necessarily to the same provider, to conduct a triple difference regression. I do not find significant heterogeneous effect. This result combined with the heterogeneous effect mentioned above indicates that insurers may not consider how much they pay across the entire market, but consider whether they pay more than their actual competitors to the same providers. To check whether price transparency facilitated providers to collude in raising prices, I conduct triple difference specifications. I generate the quintile indicators of the total payments that providers received from a given insurer relative to the total payments received by other providers from the same insurer. I do not find consistent significant effects on negotiated prices. However, the event study shows the negotiated prices for insurers paid in the top quintile prior to NHHC gradually increased after a decline over time during the first 9 quarters of the post period. This indicates the potential for providers to conduct consequent renegotiation for higher prices after insurers obtained lower prices and the related placement of the prices changed. 2 This result is similar to Brown (2017) but with slightly smaller magnitude. The difference might be due to the different sample base. 3

4 To attribute the heterogenous effect to insurer-provider renegotiation, I examine the effects on insurer-provider renegotiation. I find that NHHC increased the probability of renegotiation only for insurers that paid in the top quintile, which is consistent with the heterogeneous effect on negotiated prices. In addition, I run a similar triple difference regression based on a subsample where the procedures in the post-nhhc period have been renegotiated. I find that there was a larger effect on procedures that were considered in renegotiation. Prior research has focused on consumer responses to price transparency initiatives (Wu et al., 2014; Whaley et al., 2014; Whaley, 2015; Desai et al., 2016), while less attention has been paid on the response from the supply side. Whaley (2017) examines how providers change their prices due to patients access to an online employer-provided price transparency platform, and he finds a 1-4% reduction in prices for laboratory tests and no price response for physician visits. Brown (2017a) isolates the supply-side effect from the overall effect of NHHC on healthcare spending, and finds providers in a lower concentration market reduced negotiated prices more than those in a higher concentration market. Brown (2017b) simulates the change in insurer-provider negotiated prices due to more consumers using the price information. For healthcare spending reduction, the supply-side response to price transparency is more important than consumer responses because a relatively small share of consumers benefit from searching for low-cost providers, while all consumers in the market benefit from negotiated price changes. This paper contributes to the literature by extending the understanding of the supply-side responses to price transparency in three aspects. First, this paper provides explicit evidence that due to inconsistent procedure information, insurers respond to procedures that are not specifically disclosed on the website, but retain similar features with other disclosed procedures. This finding indicates the importance of accurately describing procedures in price transparency initiatives. Additionally, it provides a warning for following studies to be careful in defining the disclosed procedures, while also considering the potential spillover effect caused by imprecise information when evaluating the effect of price transparency. Second, this is the first paper studying insurer-provider contracting behaviors under price transparency from an insurer s perspective. Compared to consumers who also care about healthcare quality and providers who are the beneficiary from higher prices, insurers, as the profitmaximizing agents and direct payer of healthcare, may have more incentives to use price information to lower prices. With more frequency, insurers are creating price transparency tools with the purpose of disclosing prices to their members. However, prior studies pay little attention to insurer s roles and their efforts to lower prices through these price transparency initiatives. This paper fills in this blank and explains how, when informed of the negotiated prices of their competitors, insurers will renegotiate with providers for lower prices. Third, this paper first shows that price transparency affects the probability of insurer-provider renegotiation and directly links the renegotiation with the reduction of total payments. Additionally, this paper addresses an important issue in economics and health policies, by contributing to the broader literature on the theory of asymmetric information in health and other markets, and how new information spurs price competition. High healthcare spending has become a national concern in the U.S. In 2016, the U.S. spent $3.3 trillion in healthcare, accounting for 17.9% of the GDP. 3 3 This was nearly twice as much as 10 other high-income countries spending on healthcare, where the average spending was only 11.9% of the GDP. The 10 high-income countries are Canada, Germany, 4

5 Since the end of 2007, the price of healthcare has continuously grown faster than prices in the general economy. As the Peterson Center on Healthcare executive director Jay Want stated: It is not because Americans are getting older, sicker or using more services. It is because we are being charged more for stuff. 4 The lack of price transparency in the U.S. healthcare market may explain much of the high spending on healthcare. However, there is a longstanding concern with using price transparency to reduce the prices. Traditional economic theory has long recognized that information frictions can facilitate collusion and lead to higher prices (Stigler,1961; Arrow,1963; Diamond, 1971). Meanwhile, within the context of healthcare, in the absence of information about healthcare quality, price information may mislead patients to high-cost providers if higher prices are regarded as a proxy for better quality (Sinaiko et al., 2011; Cutler, 2011; Whaley, 2015). 5 This paper empirically demonstrates that price transparency could be a potential way to reduce prices with few instances of unintended collusion with providers. Following the introduction to NHHC and the healthcare market in New Hampshire in Section 2, I describe my data in Section 3. Section 4 presents the empirical methods and discusses baseline findings about supply-side effects. Section 5 investigates insurer-provider renegotiation and the related heterogeneous effect on negotiated prices. Section 6 concludes. 2. Healthcare Price Transparency in New Hampshire New Hampshire was one of the first states to create an All-Payer Claims Database (APCD). In an effort to use price transparency to reduce price variation and healthcare spending, in March 2007 NH launched HealthCost, a public website listing prices for 31 common outpatient medical procedures. Unlike other public websites, HealthCost tailors information to the consumer s specific needs through interactivity, allowing them to input their own insurance information and compare costs across providers in NH. The cost information is specific enough for consumers to assess what their treatment will cost by providing estimated total payments, insurer payments and out-of-pocket costs in addition to provider list prices, and procedure-specific costs, instead of overall hospital rates. The website is maintained and updated by the NH Insurance Department on a quarterly basis. Considerable public attention had been paid to the website since its introduction. Over 40 articles in the local public press reported on it. In the meantime, primary care doctors and insurers were encouraged to inform their patients or members about the website (2017a). However, in its early years, few consumers used it. As Australia, the U.K., Japan, Sweden, France, the Netherlands, Switzerland, and Denmark (Kaiser Family Foundation, 2018). 4 For additional information, see Healthcare Forum: Why are healthcare prices so high, and what can be done about them? May 9, 2018, Kaiser Family Foundation. 5 From the supply side, price transparency can facilitate tacit collusion. With price information, lowcost providers can obtain the maximum price an insurer is willing to pay, and this can lead to an increase in prices. Providers that have signed a most-favored nation (MFN) clause with insurers cannot charge lower prices to any other insurers. With this information available, providers have to increase prices for other insurers. From the demand side, if quality information is unavailable, patients can rely on higher prices as a proxy for better quality, which shifts patients to high-cost providers and also indirectly decreases the pressure on providers to reduce their prices. 5

6 Brown (2017b) notes, consumers used the website for about 8% of medical imaging visits. This might be why a June 2009 study did not find any immediate effect of NHHC on price variation for medical procedures (Tu et al. 2009). Tu et al. (2009) also attributes the lack of effect to weak provider competition. NH has 26 acute hospitals, of which 13 are critical access hospitals. Critical access hospitals are usually in rural areas and are characterized by geographic segmentation. Meanwhile, these hospitals receive cost-based reimbursement from Medicare, so they do not compete with other providers in a meaningful sense. Figure 1 maps the Herfindahl Index of imaging services in each county, in which darker blue means higher market concentration. The map on the left shows the Herfindahl Index of each county was larger than 0.25 prior to NHHC. Although some counties in Southeast (non-rural areas) had decreased Herfindahl Index after NHHC was launched, the average Herfindahl Index of counties was 0.27, indicating the market was still uncompetitive in general (see Figure 1b). Although Tu et al. (2009) does not find any effect of NHHC, it highlights the wide variation in hospital prices, which has promoted insurers to take actions over time. A follow-up 2014 report on NHHC documents insurers responses, including launching their own price transparency initiatives from 2011 and using the price information to renegotiate for lower prices. For instance, Anthem, the dominant insurer in NH, threatened to terminate its contract with Exeter Hospital in 2010, citing Exeter s payment rates, which were 50% higher than competitors. This well-publicized dispute resulted in a three-year contract with overall cuts of about $10 million. As Tu et.al (2014) noted, the balance in plan-provider negotiating power began to shift in NH. Before reducing healthcare spending, information friction is expected to reduce price variation. Since Brown (2017a) has found that NHHC reduced price dispersion measured by the interquartile range of prices. My paper does not include price variation in the main analysis, but presents some visual evidence about the reduced price variation from the sample used in this paper. Because the prices for Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and X-RAY procedures are different, I split the sample into CT and MRI scans, and X-RAY observations. For each of the two samples I compare the distribution of total payment related to disclosed procedures to that related to nondisclosed procedures before and after NHHC. Figure 2 shows distributions of total payments for procedures with price information shift to the left and become slightly concentrated after NHHC was launched. This indicates the primary effect of price transparency: procedures with price information are more likely to have lower prices with smaller variation after NHHC. 3. Data The primary data are the restricted commercial medical claims in NH obtained from NH Comprehensive Healthcare Information System, which assembles all private medical claims for state residents or those who received services under a health plan issued in the state. The study period is restricted to 2005 to 2010, as Anthem and Harvard Pilgrim launched their own price transparency initiatives after 2010 and NHHC briefly closed in 2014, making the period after 2010 extremely noisy. Because the website provided price information only for providers in NH during our study period, I focus on claims that are associated with services delivered in NH. These data are the same data that the website uses to construct price information, and provide rich information about health plan members, procedures, providers, payments, and health plans. 6

7 Each claim contains enrollee characteristics, including the type of health plan, the name of insurers, age, gender, resident city, county and zip code, and their relationship to the subscriber of the health plan. Because enrollees aged 65 and older might have different healthcare utilization because they likely have Medicare in addition to their private insurance, this paper considers claims for individuals under age 65. To capture an individual s health risk, I calculate two Charlson Comorbidity Indexes for each individual based on primary and secondary International Classification of Disease (ICD) codes in the claims dataset (Charlson et al., 1987; Stagg, 2006; Brown, 2017a). There are 420,944 unique individuals in the sample, around 45% are males, and the average age is 40. Considering that consumers, insurers, and providers might be more likely to respond to imaging services because these services are not substantially differentiated and are relatively price elastic to consumers, this paper focuses on imaging services with CPT codes from to There are 305 unique procedures, including 49 CT scans, 59 MRI scans, and 197 X-RAYs. Although NHHC initially only disclosed 13 imaging services with CPT codes between and 76499, inconsistent CPT codes and procedure descriptions on the website might direct users to other very similar procedures. Considering this potential spillover effect, I regard 60 procedures as procedures disclosed on the website in main analyses. Section 4 discusses the appropriate way to define treatment procedures in the NHHC case. Of the disclosed procedures, 72% are X-RAYs, while CT and MRI scans together make up less than 30%. Non-disclosed procedures have the same pattern. The claim status has three types: processed, denied, and reversed. I only consider processed claims because they refer to the final actual payment for the visits. Because the website only discloses price information for outpatient procedures, inpatient visits are excluded. In terms of providers, I can identify their names, locations, specialties, and the types. 7 Because different insurers might use slightly different names for the same providers, and a provider might thus have multiple IDs in the raw data set, I standardize the names of providers and assign each provider to a unique ID. However, for providers that have multiple locations in different settings, and each location bills differently, I assign each location to a unique ID because the prices might differ among those locations. For example, Southern New Hampshire Health runs Southern New Hampshire Medical Center and Foundation Medical Partners, which bill separately and differently. Unlike CT and MRI scans, some X-RAYs can be delivered in a physician s office and billed under the physician s name. Because bargaining between physicians and insurers might be different to bargaining between facility providers and insurers, I focus on visits delivered in facilities. There are about 300 unique providers and 65.26% of our observations are delivered in hospitals. 8 In terms of procedure prices, these data provide detailed information about hospital charges, insurer payments, prepaid, copay, coinsurance and deductibles for each claim, which allows me to calculate total payments (the negotiated price), out-of-pocket costs, and negotiated discount rates. Because NHHC discloses a median bundled price for visits with the same imaging procedures delivered by the same facilities, I aggregate prices for claims of an individual related to the same procedures delivered in the same facility at the same day to calculate the price of a visit. An imaging visit includes facility 6 Myelography uses contrast material and X-RAYs, or CT, or MRI to look for problems. Because the CPT codes do not provide information about the specific myelography method used, five myelography procedures are excluded. 7 There is a variable indicating whether the provider is an individual or facility. 8 This is different from the sample in Brown (2017a), which comprises fewer hospital visits than nonhospital visits. 7

8 and professional services that can be billed together or separately. In the dataset, some claims are incomplete because providers have not billed for all the services performed during the encounter, or the insurance carrier has not processed all the claims submitted for the visit. To calculate the price of visit as close as possible to the actual price, I remove visits only billed with physician fees because physician fees are much lower than the actual prices for imaging visits, especially for CT and MRI scans. In addition, some imaging services delivered during the surgeries have extremely high prices. Similar to the website, I remove the highest first percentile to reduce bias caused by the extremes. The sample comprises 1,033,517 visits involving 62 unique health plans issued by 32 insurers. Anthem and Harvard Pilgrim are together involved in more than half of the observations. Over 90% of observations involve Managed Care plans, including Health Maintenance Organization (HMO) plans, Exclusive Provider Organization (EPO) plans, Point-of-Service (POS) plans, and Preferred Provider Organization (PPO) plans with 52.76%, 3.40%, 15.05% and 24.22% of observations, respectively. Table 1 summarizes the characteristics of visits according to whether procedures were available on the website before or after NHHC was launched. In general, the difference in characteristics between the two groups of procedures remained after NHHC was launched. Compared to procedures related to non-disclosed procedures, disclosed procedures involve older and sicker patients, fewer CT and MRI scans, more X-Rays and related providers, and fewer hospital visits and visits with HMO, EPO, or PPO plans in both pre- and post-treatment periods. 9 Two noteworthy changes for both groups of procedures occurred after the introduction of NHHC. First, providers and health plans increased, indicating that the market became more competitive from both the insurer and provider sides. This is consistent with the decreased Herfindahl Index for procedures within a county. In addition, Anthem s market share decreased about 10 percentage points, and Harvard Pilgrim s share increased by more than 15 percentage points. This change reflected the competition between the two largest insurers in New Hampshire. Table 2 presents each type of payments of interest according to different factors. All prices are inflation-adjusted to 2015 dollars using the Medical Care Services CPI from the U.S. Bureau of Labor Statistics. Payments for hospital visits were double those for non-hospital visits. This difference could be due to higher prices of hospitals or that hospitals delivered more CT or MRI scans than nonhospitals. The total payment for X-rays was about 10% of that of CT and MRI scans. Total payment was around 50% of the discounted provider charge for both hospital and non-hospital visits. The insurer payment took up 85% of total payments, while out-of-pocket costs only accounted for 15%. In general, provider charges, total payments, insurer payments, and out-of-pocket costs increased after NHHC was introduced. However, there were many variations across insurers. All insurers had lower provider charges for both groups of procedures in the post-nhhc period, except Cigna whose provider charge increased for non-disclosed procedures. Total payments for disclosed procedures decreased for Aetna, Anthem, and Cigna, while those for Harvard Pilgrim and other insurers increased. Although Aetna made a lower total payment for disclosed procedures, it paid higher prices for non-disclosed procedures. Again, insurer payments from Aetna and Anthem decreased, while payments from Harvard Pilgrim and other insurers increased. 4. Supply-Side Effect 9 According to Charlson Comorbidity Index (CCI), the individuals in this paper are healthier than those in Brown (2017a), where the average CCI is

9 In this section, I isolate supply-side effects from the overall effect of price transparency. 4.1 Empirical Methodology I first design a baseline empirical specification to examine the overall effect of price transparency. Equation (1) presents the model:, (1) where is the price of procedure k delivered by provider h to individual i with insurance j at time t. The price considered here includes total payments, insurer payments, out-of-pocket costs, and hospital charges. Because insurer payments and out-of-pocket costs can be zero, the dependent variable is transformed into log(1+ 10 is the indicator for procedures that are listed on the website. is the time dummy, taking the value of 1 if the visit is in March 2007 or after, and 0 otherwise. is a vector of visit covariates, including individual characteristics (age, gender, Charlson comorbidity), member plan characteristics (EPO, HMO, POS, PPO, deductibles, coinsurance, and copays), the type of providers, the number of claims, the indicators of CPT modifiers and emergency visits. The outcomes here are the bundled prices of a specific procedure visit. A visit with only one procedure can have more than one claim. Claims can refer to more than one payment. I control for the number of claims to capture the heterogeneity in the prices caused by a different number of claims. Because CPT modifiers describe more details of the procedures that might be correlated to the prices, I generate a set of dummies for CPT modifiers. 11 Because the amounts of deductibles, coinsurance, and copays are parts of dependent variables and are also affected by price transparency, I generate a set of dummies to indicate them rather than controlling for the amounts to avoid over-control and endogeneity. 12 Considering coinsurance rates are directly related to insurer payments and out-of-pocket costs, I include the coinsurance rate as an independent variable when the dependent variable is insurer payment or out-of-pocket cost. Patient resident county fixed effects ( ), provider s county fixed effects ( ), procedure fixed effects ( ) and insurer fixed effects ( ) are added to control for time-invariant differences that can be correlated with the availability of the website. denotes year fixed effects and month fixed effects, which control for the time varying factors. represents an unobserved disturbance, which is clustered at the provider s city level. is the coefficient of interest, which captures the average percent changes of prices for procedures disclosed on the website relative to procedures without disclosed prices across providers. The changes can be due in part to consumer switching to cheaper providers or procedures, and in part to providers reducing prices. 10 When examining the effect on total payment, I also use log ( ) and as the dependent variables to do the regression. The results are similar. 11 For example, CPT modifier 26 refers to physician services, which are much cheaper than modifier TC, which refers to facility services. 12 Whether a health plan has deductibles, coinsurance and copays is less likely to be affected by price transparency. 9

10 To control for consumer switching to cheaper providers, I add provider fixed effects into Equation (1) as follows: where denotes provider fixed effects that absorb variations in payments due to consumer switching to cheaper providers after the introduction of the website. Now the coefficient of interest isolates the within-providers variation in payments over time. Aside from switching to cheaper providers, consumers might also choose cheaper procedures from the same providers. For example, consumers can substitute CT scans with MRI scans if MRI scans are cheaper in the same hospital. To control for this part of the consumer s response, following Brown (2017a), I modify Equation (2) using provider-procedure-insurer fixed effects to replace provider fixed effects, procedure fixed effects, and insurer fixed effects in Equation (1). The specification becomes where represents the fixed effects of the intersections of providers, procedures and insurers. now can be interpreted as the changes in payments for the same procedures delivered by the same providers, which can only be attributed to that providers and insurers change the negotiated prices. There are different ways for insurers to obtain lower negotiated prices from providers. One is by asking for lower list prices if negotiated prices are a discount from list prices. Insurers can also ask for higher discounts from the list prices. To further investigate how the negotiated prices change, I also generate discount rate and use it as the dependent variable instead of. 16 For a causal interpretation of, the underlying assumption in Equation (3) is that there are no time-varying unobserved characteristics that differentially affect dependent variables of disclosed procedures relative to non-disclosed procedures, an assumption that I address in the next subsection. 4.2 Effect on Payments NHHC discloses price information for a procedure with a relevant CPT code and a procedure description. Initially, some CPT codes and procedure descriptions on the website were not consistent. For example, Code was listed on the website as MRI-Knee, but the same code was used for hip and ankle MRIs. In addition, for some X-RAYs CPT codes varied with the number of views, the procedure descriptions on the website may not differentiate the number of views. 13 This inconsistency 16 Discount rate=1-(total payment/list prices). 13 For example, Code refers to the X-RAY exam of wrist with two views, while Code refers to the X-RAY exam of wrist with three views. 10

11 might have directed consumers to procedures that did not match both CPT codes and procedure descriptions on the website. Additionally, this inconsistency might have prompted insurers to renegotiate all potentially related procedures, even if they did not match CPT codes and procedure descriptions. In the meantime, even without inconsistent information, insurers might still put the same type of imaging procedures related to the same region into the same packages to negotiate prices, given that these procedures are so similar (e.g. Spine X-RAY with two or three views and Spine X-RAY with at least four views). Thus, I argue that there might be a spillover effect on procedures that do not match both CPT codes and procedure descriptions, but match either of them. To identify this spillover effect, I define treatment and control procedures in three ways. First, I restrict treatment procedures to be procedures that match both CPT codes and procedure descriptions on the website. The rest of procedures belong to the control group. Panel A of Table 3 shows the results. Column 1 presents the overall effect estimated from Equation (1). Except for discount rates, the effect on total payments, insurer payments, out-of-pocket costs, and hospital charges are all insignificant, but the signs for total payments, insurer payments, and out-of-pocket expenses are negative. The results shown in Column 2 are similar when provider fixed effects are added. In Column 3, where I control for the same procedures, providers, and insurers, the only significant effect is on out-of-pocket costs. The introduction of NHHC reduced out-of-pocket costs by 1.1%. There is no significant effect on total payments, insurer payments, hospital charges, and discount rates. However, this almost null effect could be biased by the potential spillover effect on some control procedures that match either CPT codes or procedure descriptions on the website. To check whether the spillover effect biased the estimates toward to zero, I remove the procedures that match either CPT codes or procedure descriptions from the control group. The results are shown in Panel B of Table 3. Column 1 shows that the introduction of NHHC reduced total payments by 2.3%, insurer payments by 1.6%, and out-of-pocket costs by 1.9%, and increased the discount rate by 1.7%. There was no effect on hospital charges. When controlling for consumer switching to cheaper providers, the effect reduced but was still significant for total payments, out-of-pocket costs, and discount rates. Results in Column 2 suggest that about 22% of the mean reduction in total payments, 42% of the mean reduction in out-of-pocket costs, and 6% of the mean increase in discount rate were due to consumer switching to cheaper providers. After further control for consumer switching to cheaper procedures within the same providers, the effect on insurer payments and out-of-pocket costs declined, and discount rates declined to zero. The only significant effect was on total payment. In contrast, hospital charges conditional on the same procedures within hospitals decreased by 1%. Column 3 indicates that the introduction of NHHC reduced negotiated prices by 1% and this reduction was mainly due to providers lowering their list prices rather than insurers obtaining higher discounts. Taken together, the results in Table 3 indicate that it is important to provide clear descriptions of procedures for price transparency policies. Results in Panel B indicate that the introduction of NHHC also negatively affected procedures that do not match both CPT codes and procedure descriptions, but match either of them. Inspired by the results in Panel B and Brown (2017a), I redefine the treatment procedures and control procedures by putting the procedures that match either CPT codes or procedure descriptions on the website into the treatment group. Panel C of Table 3 provides the related results. Unsurprisingly, the results in the three columns for all the dependent variables are similar to the results in Panel B. Because the results estimated from individual data are weighted, it is normal that the effect is larger in Panel C than Panel B, due to more treatment procedures and relatively fewer control procedures. From Column 3 in Panel C, I find NHHC reduced negotiated prices of imaging procedures by 1.3% over the period

12 Given that the average total payments of these procedures was $586 prior to NHHC and there were 439,454 related visits after NHHC was launched, this represents a reduction of approximately $7.40 per person and $3,241,788 total savings in New Hampshire. Because private insurers and providers renegotiate prices every one to three years, the average effect shown in column 3 of Panel C is expected to average over an escalating effect of price transparency on negotiated prices. To investigate, I estimate event studies by defining treatment procedures as those that match both CPT codes and procedure descriptions on the website, or match either of them. I slightly modify Equation (3) by using a family of intersections of quarter dummies and the indicator of treatment procedures to replace, and quarter dummies to replace year and month fixed effects. Figure 3 presents event study estimates for each quarter prior to and after the introduction of NHHC. The omitted category is the quarter just prior to the start of NHHC. As shown in Figure 3a, in the pre-treatment period there is no significant effect on negotiated prices for procedures disclosed on the website. This demonstrates a statistically insignificant pre-trend in negotiated prices, indicating that the DID estimate for negotiated prices is appropriate. After the introduction of NHHC, the negotiated prices do not decline until the nine quarters later. This is consistent with the stickiness of insurer-provider contracts. Figure 3b shows a similar pattern for provider charges, but with a slightly significant pre-trend. To address the bias caused by the trend, I allow the trend in provider charges to vary by whether the procedures are disclosed on the website. Differential trends conditional on treatment status and other controls would be suggestive of time-varying unobserved heterogeneity. The result is shown in Table A1 in Appendix. Column 2 shows that the effect of NHHC on provider charges becomes insignificant, indicating that the effect on provider charges shown in Column 3 of Table 3 is more likely to be driven by the downward trend. In contrast, with the specific trend, the effect on negotiated prices is still significant and even larger than before. Because previous evidence shows no pre-trend in negotiated prices, I prefer to interpret the effect on negotiated prices by using the results without controlling for trend in case the trend introduces more disturbance, as Wolfers (2003) illustrates. It should be noted that the effect on the discount rate becomes significant with controlling for the trend. In addition, Figure 2c shows that the discount rate increases over time in the post period and the effect becomes significant from the ninth quarter after the introduction of NHHC. This is consistent with the decreasing pattern for negotiated prices shown in Figure 2a. Taken together, results from Table 3, Table A1 and Figure 3 suggest a meaningful decrease in negotiated prices for private payers for procedures with price information after the introduction of NHHC, and this decrease is attributed to decreasing provider charges and increasing negotiated discount rates. I also estimate event studies by defining treatment and control procedures in the other two ways mentioned above and I find very similar results if I remove the procedures that match either CPT codes or procedure descriptions from the control group. Figure A1-A3 in the Appendix presents the results, which indicates that it is feasible to regard procedures that match either CPT codes or procedure descriptions in the treatment group as treatment procedures. Henceforth, procedures available on the website include both procedures that match CPT codes and procedure descriptions, and those that match either CPT codes or procedure descriptions. 4.3 Robustness of Supply-side Effect 12

13 To check whether Equation (3) sufficiently eliminates consumer response, I use emergency visits of imaging procedures to estimate Equations (1) and (2) by assuming that consumers are less likely to search for low-emergency providers and procedures. 14 Columns 1 and 2 of Table A2 in the Appendix show that without salient control for consumer response, NHHC reduced the negotiated prices for emergency visits by 1.9% and 1.4% respectively. These effects were less than the 3.3% and 2.5% reductions in the full sample shown in Columns 1 and 2 of Table 3. The smaller effects on emergency visits could be due in part to reduced consumer response and in part to the higher bargaining power of providers for emergency services. Although it is unclear which part causes the smaller effect, the results for emergency visits suggest that the model I use does eliminate consumer response. There is a real concern that the control procedures could also be affected by price transparency. On the one hand, given that insurers might bargain over a category of procedures rather than each specific one, control procedures that belong to the same category of treatment procedures can also have lower prices, causing results to be biased toward to zero. On the other hand, providers might agree to lower their prices for treatment procedures, but increase their prices for control procedures, which causes an overestimation in results. To address this concern, I introduce out-of-state observations to account for cross-state variations for both treatment and control procedures. The dataset contains claims submitted by all health plans used in NH, which potentially includes some claims for out of state services. Residents living close to the boundary with the NH health plan might travel to neighboring states for imaging procedures. Companies whose headquarters are in NH could purchase a NH health plan for their employees who live out of state. In addition, students who attend university out of state might be covered under their parents plan in NH. NHHC did not disclose price information for out of state providers during the study period. 15 These out of state services take up a small fraction of the local market, and might not cause a supply-side response. Therefore, prices for the procedures outside of the state are not affected by NHHC and thus can be used as benchmarks to compare the negotiated prices for both treatment and control services in NH. I first select the observations with procedures disclosed on the website. Using the specification of Equation (3) but with standard errors clustered at state level, I compare the negotiated price in NH to that in comparison states before and after NHHC. Column 3 of Table A2 shows that NHHC decreased the negotiated prices by 15.3% and this result is robust to the specific state trends added in Column 4. This effect is much larger than the main estimates in Panel C and column 3 of Table 3. This may be due to fewer out-of-state observations. In addition, to check whether providers increase the prices for procedures not available on the website, I also generate a cross-state sample featuring only procedures not available on the website. Using the same method, I find that after NHHC there was no significant increase in prices for procedures not available on the website. This result is also robust to specific state trends. 5. Insurer-Provider Contracting 5.1 Average Effect of Contracting with the Same Providers 14 Indeed, the literature that shows consumers are more likely to search for elastic healthcare. 15 It does now, though. 13

14 With price information, insurers have two choices to obtain lower negotiated prices from providers. One is to continue to contract with the same providers but bargain for lower prices based on the available price information. Alternatively, insurers can switch to cheaper providers for lower prices motivated by lower search costs facilitated by price information. The supply-side effect established in Section 4 is driven by at least these two approaches. 16 The data allow me to track prices for a group of insurer-provider-procedures in the entire local market over time, which provides an ideal sample to study insurers renegotiation with the same providers for the same procedures. However, about 50% of observations with combined insurer-provider-procedures only exist in some study periods. Because this group of insurer-provider-procedures normally involves very low patient volumes within a year, it is hard to distinguish whether their absence in certain years was caused by insurers switching to new providers or just because there were no visits in those years. Therefore, this section only focuses on the first choice of insurers, identifying how transparency affects insurers contracting with the same providers. Table A3 in Appendix shows the average effect on negotiated prices estimated by the DID method in Section 4, but using the observations with the combinations of insurer-provider-procedure that can be observed in each study year. 17 Column 1 shows that there is little average effect on the negotiated prices if insurers chose to contract with the same providers for the same procedures. However, Columns 2 and 3 suggest that the effect is significant within a provider or within a given insurer and a given provider, indicating that there might be heterogeneous effects across procedures or insurers. The following subsection specifically assesses this heterogeneity. 5.2 Heterogeneity in Insurer-Provider Renegotiation Theory Motivated by Grennan (2016), who models bargaining between providers and medical suppliers with information shock, I follow Rubinstein (1985) to model uncertainty of insurers about the bargaining parameters of a given provider. Although patients are the ultimate consumers of healthcare, insurers are direct payers for healthcare. It is reasonable to analyze bargaining between insurers and providers by regarding insurers as the buyer of healthcare. Suppose a single insurer negotiates with a single provider over a per-unit healthcare surplus V = wtp c equal to the insurer s willingness-to-pay (wtp) for a unit of the provider s service, minus the provider s marginal cost (c) of providing a unit of healthcare. Beginning with the insurer, each player in turn makes a proposal for the division of the surplus. After one insurer has made an offer, the provider must decide to accept or reject it and make a counteroffer in the next round. Each player has a discount factor that parameterizes bargaining strength and can be thought of more generally as a proxy for a host of factors that might affect a realworld negotiation such as market share, quality of healthcare, and reputation. The insurer has a discount factor, while the provider is either strong type with discount factor or high-cost type with discount factor. With incomplete information, the provider knows his own 16 Other factors include new entry of procedures, or providers, or insurers. 17 I exclude out-of-network visits under managed health plans because there is no contract between insurers and out-of-network providers. Out-of-network visits are defined as visits under a managed health plan with zero discount from the provider list price, and the insurer pays only some of the total payment. 14

15 type, but the insurer has only a subjective prior belief weak type. of the probability that the provider is the As Rubinstein(1985) shows, the equilibrium split of this surplus depends on both the prior belief of the insurer and the type of the provider. There exists a cutoff prior belief so that the insurer is indifferent between believing and not believing the provider is a weak type: (4) If the insurer is sufficiently pessimistic about the provider being the weak type, then the insurer offers what he would offer the strong type in a complete information game of Rubinstein (1982): (5) and both provider types accept this offer. However, if the provider is more optimistic about the probability that the provider is the weak type, then the insurer offers: (6) which the weak provider type accepts. The strong provider type will reject this offer, and counteroffer with a price ( that would make a weak provider no better off than, but make the strong provider prefers. Now suppose there is a new information shock that fully reveals a provider s type, then the equilibrium prices would move from the type of asymmetric information to complete information, where the type of the provider is common knowledge to the insurer. Then the equilibrium offer for the weak provider is:, (7) and that for the strong provider is:, (8) where = (. This equation indicates that the impact of new information on the provider s price is heterogeneous, which depends on the type of the provider and the prior belief of the insurer. If the provider is a weak type, both pessimistic and optimistic insurers would like to renegotiate with the provider, lowing price to. Prediction If the price transparency reduces the information searching cost to 0, both pessimistic and optimistic insurers will be informed. When the informed pessimistic insurer find the price he paid to provider is at high level in the market, he would like to renegotiate for lowing the price down to. Such renegotiation will cause high prices to fall in cases where the provider was in fact the weak type. 15

16 For the optimistic insurer, because the price he paid is already low in the market, the information should have less effect on price reduction when the provider is the weak type. When the provider is the strong type, the optimistic insurer may be subjected to renegotiation requested by the strong provider for increasing the price to. The NHHC disclose median bundled prices for procedures, which might not be thorough enough to fully reveal the type of the provider. However, such disclosed information can shift the insurer s prior belief toward the truth for specific procedures. By comparing the prices that other insurers paid, the insurer can verify or change the prior belief and make relevant responses. 5.3 Heterogeneous Effect To empirically investigate the theoretical prediction, I extend Equation (3) by allowing outcomes to vary by the insurer s prior belief on each procedure. I attempt to proxy for the insurer s prior belief by constructing the placement quintiles of the insurer-provider pair s median price for a procedure in the full distribution of the median price paid to the same provider for the same procedure at the time NHHC was introduced. A higher quintile means that the insurer is more likely to be a pessimistic insurer. Each quintile indicator denoted by, is intersected with in Equation (3) as follows: where is the treatment effect of NHHC for each quintile of the pre-information price distribution. Unlike, estimates heterogeneous treatment effects depending on the preinformation price the insurer paid to a provider relative to other competitors. Because October 1 is the beginning of the fiscal year for many NH hospitals and the time when price changes take place, I generate a new time frame for each year from October 1- September 30 and let equal 1 from October 1, This analysis is similar to that of Grennan (2016), who provides evidence that a hospital that pays higher prices for the product at the time of information might negotiate lower prices after joining price transparency initiatives. The results are presented in Table 4 and suggest that the NHHC effect exhibits substantial heterogeneity depending on the pre-information price the insurer was paying for a procedure relative to competitors. The effects on negotiated price are statistically zero in all but the top quintile of the pre-nhhc price distribution, where the effect is -3%. This effect is higher than the average effect (- 1.3%) estimated in Section 4. This evidence is consistent with theoretical predictions that in the absence of information, pessimistic insurers pay providers high prices regardless of those providers true bargaining parameters, so information would lead those insurers to negotiate lower prices after price transparency. It is also worth noting that the lower parts of the distribution do not shift upward significantly, but their signs are positive. A similar heterogeneous effect is also found for insurer 18 For more information, see The Impact of Price Transparency on HealthCost Services in New Hampshire. 16

17 payments and discount rates. Insurers that were paying within the top quintile paid 4.5% less with a larger discount after the introduction of NHHC. In contrast to insurer payments, out of pocket costs increased 4.3% for consumers whose health plans were paying in the top quintile, but they decreased 4.4% for consumers whose health plans were paying within the fourth quintile. There is little effect on the provider charge, indicating that the reduction of negotiated prices came from higher bargaining discount rate. Given that discount rate is regarded as a proxy of bargaining power in the literature (Sorensen, 2003), the results in Table 4 show that NHHC increased the bargaining power of insurers who were paying higher rates before price transparency. I also perform event studies separately for each quintile of the price distribution. The results for the top quintile of the pre-nhhc price distribution are shown in Figure 4 and 5. Figure 4a presents event study estimates on negotiated prices for each quarter prior to and after the introduction of NHHC. The pre-trends in the ten quarters pre-nhhc were essentially zero for all the outcomes, while there was a steady decline after price information access: about one and half years after NHHC was launched, the treatment effect was -5.5 percentage points relative to the introduction date. However, from ten quarters after the introduction of NHHC onwards, the estimate for negotiated prices became positive, though not significantly. This makes sense because after obtaining lower prices, the placement of the prices for an insurer might change and thus would be followed by consequential renegotiations for higher prices by the provider. The lack of pre-trends can be also seen for provider charges, discount rates, insurer payments and out-of-pocket costs in Figures 4b and 4c, and Figure 4, respectively. There was no significant change to provider charges, as shown in Figure 4b, and Figure 4c shows that the decrease in negotiated prices was driven by the steady increase in discount rates. Figures 5a and 5b indicate that the reduction of negotiated prices passed to insurer payments instead of out-of-pocket costs. Taken together, Figures 4 and 5 are consistent with Table 4. It s noteworthy that the reduction of the total payment for insurers paid in the top quintile to a given provider is different from that of Brown(2017a) who finds the interquartile range of negotiated prices decreased. The prices that insurers paid in the top quintile given a specific provider are not necessarily in the top quintile or quartile of negotiated prices across providers. To check this difference, I also use the quintile of the median price an insurer paid to a provider for a procedure relative to the median prices that other insurers paid for the same procedure, but not necessarily to the same provider, to conduct a triple difference regression. Table A4 in Appendix shows there was little significant effect for insurers that paid in the top quintiles across the providers rather than within a provider. This result combined with the heterogeneous effect mentioned above indicates that insurers may not consider how much they pay across the entire market, but consider whether they pay more than their actual competitors to the same providers. However, because NHHC updates price information quarterly, the initial placement of payments might be different from later placement, so it might not differentiate insurers contracting. In this case, the estimates will be biased. To address this concern, I conduct a new price indicator using the quintile of the insurer-provider pair s median price in the latest year. Apparently, this indicator is endogenously affected by NHHC. However, because this endogeneity causes underestimated results, it is acceptable as long as an effect is found. The results are shown in Table 5. Compared to Table 4, Column 5 of Table 5 shows similar effects on insurer payments and out-of-pocket costs, but a higher effect on negotiated prices. NHHC reduced negotiated prices by 4.7% instead of 3% in Table 4. For negotiated discounts, the effect disappeared while provider charges reduced significantly by 2.8%. For other lower quintiles, the results are quite similar to the main model we use, except for the second quintile, where negotiated prices, insurer payments, and provider charges increased by 2%, 2.8%, and 2.5% 17

18 respectively. These results make sense given that the latest price placement can affect the negotiation more directly than an older one. In the meantime, this significant effect for the lower quintile also indicates that cheaper providers might charge higher prices, which is one of the main concerns about the intended consequences of price transparency. To directly check whether there is any backfire effect from the provider side, I use a similar method by conducting placement quintiles of total payments that a provider receives from an insurer relative to other providers. Quintiles at the time of introduction of NHHC and in the latest year are used separately in two separate specifications, as I do for insurers. Higher quintile means the provider received more from the insurers than their competitors. The results are presented in Table 6. Column 1 shows that providers that received payments within the lower quintile did not increase their negotiated prices after NHHC was launched. The only significant effect was that providers increased their prices when the price in the last year was within the second quintile and the relevant increase rate was 3.8%. It is also worth noting that two specifications consistently show that the negotiated prices decreased by 2.2% for providers that received the top quintile of payment and this reduction was not from providers reducing their list prices but from insurers obtaining a 1.1% greater negotiated discount. Taken together, the results in Tables 4, 5, and 6 suggest that NHHC increased the bargaining power of insurers that paid more before new information was available, and there is no consistent significant evidence of collusion between providers that received lower payments. The effects for each quintile estimated above are the average effect on procedures disclosed on the website. However, renegotiation does not necessarily occur for each treatment procedure. Thus, if the heterogeneous effects were attributed to insurer-provider renegotiation, we would expect a larger effect on procedures that are considered in the renegotiation, or a significant effect of the price transparency on the likelihood of renegotiation. I consider these two effects separately by flagging provider-insurer-procedure-year observations in which renegotiation is observed. Although I do not have data on when renegotiations took place, I do observe whether the negotiated price of providerinsurer-procedures changed in each year and the group visits with the same price together within year. I define the occurrence of renegotiation to a provider-insurer-procedure-year if I observe that prices are different from those in the last year and that the related visits are greater than one. 19 To investigate the effect of price transparency on renegotiated procedures, I run the same regressions but base them on a subsample where the procedures in the post-nhhc period have been renegotiated. The results in Table 7 suggest that prices conditional on renegotiation reduced by 3.8% for insurers that paid in the top quintile before NHHC, and by 5.5% for insurers who paid in the top quintile in the last year. These reductions are larger than those for all the treatment procedures estimated above. To investigate whether price transparency increase the rate of renegotiation, I estimate the same quintile specification but with the indicator of renegotiation as the dependent variable. The results in 19 This definition is based on the assumption that if the price change was driven by renegotiation rather than the extraordinary complexity of the procedures, the change should be related to a group of visits with similar complexity. If there is only one visit with a new price, the changed price is more likely to be driven by the extraordinary complexity of the procedures. So, including the restriction on the number of visits with the same prices could exclude the effect on price changes driven by the extraordinary complexity of the procedures. I also use a less conservative standard to define the renegotiation, where the visits related to the changed prices should be greater than three, five, or ten, and the results are similar. 18

19 Table 7 suggest that the effect of price transparency on the likelihood of renegotiation is statistically significant at the 1% level in the top quintile of price for both quintile specifications, where price transparency increased the probability of renegotiation by percentage points, or about % of the mean probability of renegotiation for treatment procedures. Point estimates of two specifications in other quintiles are not significant, are negatively significant at conventional levels, or are not consistent even with larger positive significant effects. Taken together, the results in Tables 4-7 provide evidence that price transparency changes insurer-provider contracting behavior heterogeneously. 6. Conclusion This paper uses unique market-wide payment data from a large multi-payer database to investigate the extent to which private insurers, faced with new price information, pay less by renegotiating for lower prices from providers. I use variation in total payments generated by price transparency initiatives in New Hampshire to estimate the role of new price information on average provider payments from the supply side. The results support the argument that price transparency increases the bargaining ability of insurers and forces high-cost providers to lower their prices. The change in payments for procedures disclosed on the website was significantly higher than that for non-disclosed procedures. This qualitative finding is robust when emergency visits are used instead of the full sample, and with the introduction of cross-state variation. Motivated by Grennan (2016), I extend my empirical analysis to consider the underlying heterogeneous insurer-provider contracting through which price transparency affects payments. First, the theoretical model suggests that information will cause prices to fall for insurers that paid more before the new information became available. Consistent with this prediction, the magnitude of the effect for insurers that were paying within the top quintile is much larger than that for the full sample, and for insurers that were paying within the lower quintile. This finding also holds from the provider s perspective. The result suggests that the price of procedures delivered by providers who received more than their competitors reduced by 2.2%. I then link the reduction with renegotiation by flagging provider-insurer-procedure-year observations of renegotiation. Using a similar regression, I find that price transparency increased the probability of renegotiation and the effect on procedures conditional on renegotiation is larger than the effect on all the treatment procedures. This paper provides evidence about the potential of price transparency as a general strategy to reduce healthcare expending. Future studies could build on this research in a few respects. First, while the results in this paper suggest that the introduction of price transparency can indeed reduce healthcare expenditure, the analysis only covers imaging services. It is important to investigate whether the results apply more broadly. Given the variations in price elasticity to consumers and differences between providers across procedures, there are rich opportunities for future empirical analysis of the effect of price transparency on broader procedures. Second, this paper lacks reliable information about the entry of new providers, insurers, or procedures, leaving an unresolved question about how price transparency reduces negotiated prices through the mechanism of new entry. Given that APCD is expanding, there is a potential for studies to fill this gap. Third, the results cover the three years after the introduction of NHHC, and it is unclear that whether the spillover effect on non-disclosed procedures will be increased by the growing bargaining power of insurers. The New Hampshire commercial database also contains data after 2010, providing a chance to study the dynamic effect over the long term. 19

20 References Arrow, Kenneth J Uncertainty and the Welfare Economics of Medical Care. The American Economic Review, 53(5), Brown, Zack. (2018). An Empirical Model of Price Transparency and Markups in Health Care. Working paper Brown, Zack. (2018). Equilibrium Effects of Health Care Price Information. Review of Economics and Statistics. Forthcoming. Cooper, Zack, Stuart Craig, Martin Gaynor, and John Van Reenen. (2018). The Price Ain t Right? Hosital Prices and Health Spending on the Privately Insured. NBER working paper. Christensen, Hans B., Eric Floyd, and Mark Maffett. (2015). The Effect of Price Transparency Regulation on Prices in the Healthcare Industry. Working paper Christensen, Hans B., Eric Floyd, and Mark Maffett. (2017). The Only Prescription is Transparency: The Effect of Charge-Price-Transparency Regulation on Healthcare Prices. Working paper Cutler, David and Leemore Dafny. (2011). Designing Transparency Systems for Medical Care Prices. The New England Journal of Medicine. 364: Darden, Michael, Ian McCarthy, and Eric Barrette. (2018). Hospital Pricing and Public Payments. NBER working paper. Desai,Sunita, Laura A. Hatfield, Andrew L. Hicks, Michael E. Chernew, and Ateev Mehrotra. (2016). Association Between Availability of a Price Transparency Tool and Outpatient Spending. Journal of the American Medical Association.315(17): Diamond, A. Peter. (1971). A model of price adjustment. Journal of Economic Theory. 3(2): Dranove, David, Mark Shanley, and William D. White. (1993). Price and Concentration in Hospital Markets: The Switch from Patient-Driven to Payer-Driven Competition. Journal of Law and Economics. 36(1): Greene, Jessica, Judith H. Hibbard, and Rebecca M. Sacks.(2016). Summarized Costs, Placement of Qaulity Stars, and Other Online Displays Can Help Consumers Select High-Value Health Plans. Health Affairs. 35:4. Grennan, Matthew and Ashley Swanson. (2018). Transparency and Negotiated Prices: The Value of Information in Hospital-Supplier Bargaining. NBER working paper. Hibbard, Judith H., Jessica Greene, Shoshanna Sofaer, Kirsten Firminger and Judith Hirsh. (2012). An Experiment Shows that a Well-Designed Report on costs and Quality Can Help Consumers Choose High-Value Health Care. Health Affairs. 31:3. 20

21 Kim, Michelle. (2011). The effect of Hospital Price Transparency in Healthcare Markets. Ph.D. dissertation in Health Care Management and Economics, University of Pennsylvania. Lieber, Ethan M.J.(2017). Does it Pay to Know the Prices in Health Care? American Economic Journal: Economic Policy. 9(1): Papanicolas, Irene, Liana R. Woskie, MSc, and Ashish K.Jha. (2018). Health Care Spending in the United States and Other High-income Countires. Journal of American Medical Association. 319(10): Rubinstein, A. (1985). A Bargaining Model with Incomplete Information about Time Preferences. Econometrica. 53(5): Russell, Anna. (2015). Moving the Needle: Howe Transparency Could Lower Costs and Improve Quality in United States Hospitals. Yale University EliScholar. Sinaiko, Anna D., and Meredith B. Rosenthal. (2011). Increased Price Transparency in Health Care--- Challenges and Potential Effects. The New England Journal of Medicine. 364: Sinaiko, Anna D., and Meredith B. Rosenthal. (2016). Examining a Health Care Price Transparency Tool: Who Uses it, and How They Shop for Care. Health Affairs. 35(4): Sorensen, Alan T. (2003). Insurer-Hospital Bargaining: Negotiated Discounts in Post-Deregulation Connecticut. The Journal of Industrial Economics. 51(4): Stigler, George J. (1961). The Economics of Information. Journal of Political Economy.69:213. Tu, Ha, and Johanna R. Lauer.(2009). Impact of Health Care Price Transparency on Price Variation: The New Hampshire Experience. Center for Studying Health System Change. November Tu, Ha, and Rebecca Gourevitch.(2014). Moving Market: Moving Market: Lessons from New Hampshire s Health Care Price Transparency Experiment. California HealthCare Foundation. Whaley, Christopher, Jennifer Schneider Chafen, Sophie Pinkard, Gabriella Kellerman, Dena Bravata, Robert Kocher, and Neeraj Sood. Association Between Availability of Health Service Prices and Payments for These Services. Journal of the American Medical Association. 312(16): Whaley, Christopher. (2017). Provider Responses to Online Price Transparency. Working paper. Wolfers, Justin. (2003). Did Unilateral Divorce Law Raise Divorce Rates? A Reconciliation and New Results. American Economic Review. 96(5): Wu, Sze-jung, Gosia Sylwestrzak, Christiane Shah, and Andrea DeVries. (2014 ). Price Transparency for MRIs Increased Use of Less Costly Providers and Triggered Provider Competition. Health Affairs. 33:8. 21

22 Figure 1: The Herfindahl Index of Imaging Services across Counties in New Hampshire Notes: Purple points refer to 26 hospitals. Herfindahl Index is calculated based on market shares of healthcare providers, including hospitals and non-hospital facilities. 22