Institutions For Mental Diseases Medicaid Waivers: Impact On Payments For Substance Use Treatment Facilities

Abstract

The Institutions for Mental Diseases (IMD) exclusion prohibits use of federal Medicaid funds to treat enrollees ages 21–64 in psychiatric residential treatment facilities that have more than sixteen beds. In 2015, the federal government created a streamlined application pathway for state waivers of this rule to allow Medicaid coverage for substance use disorder treatment in residential facilities. From 2015 to 2018, nine states received IMD waivers. Using the 2010–18 National Survey of Substance Abuse Treatment Services, we examined changes in residential and outpatient SUD treatment facilities’ acceptance of Medicaid and other types of health coverage, as well as self-pay arrangements and provision of charity care, after state adoption of IMD waivers. Acceptance of Medicaid increased 34 percent at residential SUD treatment facilities and 9 percent at intensive outpatient facilities two years after waiver implementation. Delivery of medications for opioid use disorder did not increase in residential facilities postwaiver but did increase to some extent in outpatient facilities. Our findings suggest that IMD waivers may be an important tool for advancing access to a full continuum of SUD treatment for Medicaid enrollees.

The Institutions for Mental Diseases (IMD) exclusion prohibits federal Medicaid funds from paying for services for enrollees ages 21–64 residing in an inpatient mental health or substance use disorder treatment (SUD) facility with more than sixteen beds.¹ The intent of the exclusion was to shield the federal government from the cost of covering care for enrollees residing in state psychiatric hospitals, prevent the inappropriate institutionalization of enrollees, and promote the creation and use of community-based behavioral health services.²

Since the 1980s, Medicaid has become an important source of payment for SUD treatment.³ This change has accelerated as more individuals with SUD have become covered by Medicaid under the Affordable Care Act (ACA), which defines SUD treatment as an essential health benefit.⁴ Amid an unprecedented opioid crisis, policy makers have argued in the last several years that the IMD exclusion is antiquated and prevents Medicaid from covering the full continuum of evidence-based SUD treatment, including residential treatment programs.^5,6

In 2015, the Centers for Medicare and Medicaid Services (CMS) issued guidance to state Medicaid directors providing a streamlined approval process for granting Section 1115 IMD waivers for SUD treatment.^7,8 This guidance authorized the first national demonstration project for states to include IMD care in their Medicaid benefits packages. Guidance was expanded in 2017 and 2018.^9,10 The 2017 guidance stated that to be granted an IMD waiver, states had to meet several criteria, which included covering a continuum of care for SUD treatment in Medicaid benefits packages.¹⁰

In addition to increasing SUD treatment options for enrollees, IMD waivers are intended to increase the availability of medications for opioid use disorder.^7–10 For example, with its IMD waiver, West Virginia Medicaid began to cover methadone maintenance for the first time.¹¹

By September 2020, IMD waivers for SUD treatment had been approved in twenty-eight states with waivers pending in eight states.⁸ To establish whether the waivers adopted to date increased enrollees’ treatment options in terms of settings and specific treatments, we drew national data on SUD treatment facilities and examined whether the types of payment arrangements and the provision of specified opioid treatment modalities at residential facilities and nonresidential facilities changed within the first two years of waiver adoption. Examining whether this Medicaid policy change had spillover effects to other payment arrangements and provision of medications for opioid use disorder to patients other than Medicaid enrollees can help identify the broader market-level implications.

We hypothesized that IMD waivers would increase Medicaid acceptance among residential and nonresidential facilities. We also hypothesized that IMD waivers would affect acceptance of other forms of coverage, although the direction of this impact was uncertain. Finally, we hypothesized that IMD waivers would increase availability of medications for opioid use disorder and opioid treatment programs as part of state efforts to augment the continuum of care among all SUD treatment providers.

Study Data And Methods

We used the National Survey of Substance Abuse Treatment Services 2010–18. These data are collected by the Substance Abuse and Mental Health Services Administration (SAMHSA) to monitor SUD treatment facilities typically subject to regulation by state authorities. The data set includes for-profit, nonprofit, and government-run facilities. The National Survey of Substance Abuse Treatment Services provides a snapshot of each facility’s operations on the survey day near the end of March each year. Response rates were 86–92 percent during our study period.¹²

We examined changes in acceptance of Medicaid and other health coverage and self-pay arrangements and provision of charity care at residential, intensive outpatient, and nonintensive outpatient treatment facilities. Our operational definition of an institution for mental diseases was a residential facility that offered overnight care. IMD were a subset of the residential facilities we examined as we could not consistently restrict our sample to facilities that had sixteen or more beds allocated to SUD treatment, as the National Survey of Substance Abuse Treatment Services did not collect information on beds in all years. In 2011, 2013, 2015, and 2017 (when the National Survey of Substance Abuse Treatment Services collected information on beds allocated to SUD treatment), IMD with more than sixteen beds made up 65 percent of all residential facilities. Thus, 65 percent of residential facilities in our sample meet the definition of an institution for mental diseases.

Each observation in our data represents a unique facility in a survey year; we could not track a facility over time because the National Survey of Substance Abuse Treatment Services does not include longitudinal facility identifiers. Observation counts were as follows: 33,630 residential, 46,358 intensive outpatient, and 41,538 nonintensive outpatient facilities. Online appendix exhibit A1 lists the number of facilities in each year for all samples.¹³

We created separate indicators capturing whether a facility accepted Medicaid, private insurance, other public insurance (Medicare, military insurance, or other state financed insurance), and self-payment. We examined whether the facility provided charity care (that is, a sliding fee scale [patients pay a share of the overall treatment bill], free care to all patients, or some other form of financial assistance to patients who could not pay). We also identified whether the facility had either a certified opioid treatment program or offered medications for opioid use disorder treatment (that is, buprenorphine, methadone, and naltrexone).

To identify the states that implemented IMD waivers, we used data from the Kaiser Family Foundation Medicaid waiver tracker.⁸ We excluded Massachusetts and Vermont from our analysis because these states adopted partial IMD waivers in the 1990s.

We estimated a modified difference-in-differences model to examine changes in acceptance of different types of health coverage, self-pay arrangements, and provision of charity care among SUD treatment facilities operating in states that adopted an IMD waiver and state that did not adopt a waiver. Our main specification was an event-study that includes separate indicators for each year relative to IMD passage in adopting states. This model allowed us to estimate intent-to-treat effects. We constructed indicators for five to ten years before the IMD waiver through two years after the waiver. The years before implementation were coded as negative years (for example, one year before was −1), the year of the implementation was year zero, and the years after implementation were coded as positive years (for example, one year after was +1). The sample of facilities within states that were observed in each lead and lag indicator varied because we did not observe all IMD waiver-adopting states for up to ten years before the event and two years after IMD waiver adoption. Appendix exhibit A2 reports the states that populated each lead and lag variable.¹³

Our outcomes data end in March 2018, by which time nine states (in addition to Massachusetts and Vermont, which were excluded from our sample) had adopted IMD waivers and forty states had not. Of the forty states, sixteen adopted waivers between April 2018 and July 2020. We included those states in our analysis, recording the fact that facilities in these states were observed during our study period in their prewaiver period.^14–16

We adjusted our regression models for state and year fixed effects. We adjusted for facility-level ownership: for-profit versus nonprofit or government. We adjusted for time-varying state-level characteristics: ACA Medicaid expansion status,¹⁷ having a Medicaid health home program targeting SUD,¹⁸ the share of a state’s Medicaid enrollees covered by a comprehensive risk-based managed care organization,^19,20 the presence of a recreational marijuana law,²¹ a prescription drug monitoring program,²² and time-varying demographics (age, sex, race/ethnicity, foreign birth, veteran status, education, unemployment rate, and poverty rate).²³ A 2016 CMS regulation allowed Medicaid managed care plans to start covering SUD IMD services under certain conditions.²⁴ Our inclusion of state and year fixed effects in regression models should reduce the influence of this national regulation on our estimates. Finally, we controlled for state population.²⁰ Regression models were estimated using linear probability models. We reported 95 percent confidence intervals that accounted for within-state clustering.²⁵

We conducted several sensitivity analyses. We estimated a standard difference-in-differences model that collapsed the event-study indicators into one pre- and one posttreatment period; this method offers a summary estimate of overall treatment effects. The IMD exclusion does not apply to general hospitals, but IMD located in general hospitals were not separately identified in the 2014 National Survey of Substance Abuse Treatment Services. We therefore dropped the 2014 data and estimated models excluding general hospitals from the residential group. Because IMD waivers could have led more facilities to open, we estimated waiver impacts on the total number of residential facilities in a state and stratified by ownership status (for-profit and nonprofit or government).

Our study had several limitations. First, our observational study design may not have fully accounted for factors otherwise causing facilities in waiver states to change payment acceptance or medication provision. Although our event-study should identify any state-level trend differences that preceded the waivers, state Medicaid programs may have self-selected to apply for waivers based on unique factors arising in those states (for example, a severe opioid crisis) that independently influenced the outcomes we studied. Second, the National Survey of Substance Abuse Treatment Services relies on information reported by facility employees, which is subject to error. The survey also does not capture outcomes of treatment received or whether care received was effective (for example, indicators for successful discharge from treatment). Third, we lacked information on bed size in some years of our data and could not definitively identify which facilities would be classified as IMD. This limitation could have induced measurement error.²⁶ Finally, the National Survey of Substance Abuse Treatment Services database captured publicly funded facilities subject to oversight from state agencies that regulate public sector SUD treatment, but likely missed many facilities that accepted only private health insurance and that did not opt to be listed in the SAMHSA treatment locator.

Study Results

Exhibit 1 displays the nine states that adopted IMD waivers for SUD treatment from 2010 to March 2018: California, Indiana, Kentucky, Louisiana, Maryland, New Jersey, Utah, Virginia, and West Virginia. In the years before the IMD waiver was adopted, residential facilities operating in eventual waiver states were significantly less likely to accept all forms of public insurance than nonwaiver states in 2010–15 (that is, the years of our study period in which no states had adopted an IMD waiver), with the greatest difference being in Medicaid acceptance (32 percent; exhibit 2). Residential facilities in waiver states were slightly more likely than those in nonwaiver states to be for-profit (appendix exhibit A3).¹³

Exhibit 1:

States that have adopted substance use disorder treatment Institutions for Mental Diseases (IMD) exclusion waivers January 2010–March 2018

IMD waiver adopting state	IMD waiver adoption date
California	January 1, 2016
Indiana	February 1, 2018
Kentucky	January 12, 2018
Louisiana	February 1, 2018
Maryland	January 1, 2017
New Jersey	July 1, 2017
Utah	November 1, 2017
Virginia	April 1, 2017
West Virginia	January 1, 2018

SOURCE Authors’ analysis of the years of adoption of IMD waivers collected from the Centers for Medicare and Medicaid Services. NOTES Massachusetts and Vermont adopted partial IMD waivers before March 2018, but these states were excluded from the analysis sample. The following states adopted a waiver between April 2018 and July 2020: Alaska (January 1, 2019), Delaware (August 1, 2019), District of Columbia (January 1, 2020), Idaho (April 17, 2020), Illinois (July 1, 2018), Kansas (January 1, 2019), Michigan (April 5, 2019), Minnesota (July 1, 2019), North Carolina (January 1, 2019), Nebraska (July 9, 2019), New Hampshire (July 10, 2018), New Mexico (January 1, 2019), Ohio (October 1, 2019), Pennsylvania (July 1, 2018), Rhode Island (January 1, 2019), Washington (July 17, 2018), and Wisconsin (October 31, 2018).

Exhibit 2:

Effect of Institutions for Mental Diseases (IMD) waivers on the average percentage of residential facilities accepting different types of payment and providing charity care, 2010–18

Outcome	Medicaid	Private	Other public	Self-payment	Charity care
Percent of facilities accepting different types of payment and providing charity care in IMD waiver states, averaged over years before waiver adoption	32	59	40	91	70
Years relative to IMD waiver adoption year a
−10 to −5	−0.01	0.02	0.02	0.00	0.00
−4	0.00	0.02	0.01	0.00	−0.01
−3	0.02	0.01	0.01	0.00	−0.01
−2	0.01	0.01	0.00	−0.00	−0.01
−1	—b	—b	—b	—b	—b
0 (year of adoption)	0.01	−0.00	−0.02	−0.03***	0.00
+1	0.06****	−0.01	0.01	−0.02*	−0.03**
+2	0.11***	0.08****	0.07***	−0.04**	−0.02
n (facilities in a year)	32,814	32,937	33,209	33,270	33,587

SOURCE Authors’ analysis of the National Survey of Substance Abuse Treatment Services linked to data on years of adoption of IMD waivers collected from the Centers for Medicare and Medicaid Services. NOTES Other public is Medicare, other state-financed, or military coverage. Charity care is a sliding fee scale (patients pay a share of the overall treatment bill), free care to all patients, or some other form of financial assistance to patients who cannot pay.

^aRegression coefficient estimates of a covariate-adjusted linear probability regression model. The numbers reflect percentage-point changes in the outcome variables. See online appendix exhibit A5 for details (see note 13 in text).

^bReference period for interpreting the event-time indicators.

^*p < 0.1

^**p < 0.05

^***p < 0.01

^****p < 0.001

Appendix exhibit A4 reports our outcome variables averaged over the years 2010–15 and 2016–18 for states adopting and not adopting an IMD waiver.¹³ The percentage of residential facilities accepting Medicaid, private insurance, and other public insurance increased in both groups of states over time. The percentage of residential facilities accepting self-payment was roughly stable over time for both groups of states. The percentage of residential facilities providing charity care decreased over time in both states that adopted an IMD waiver and states that did not adopt an IMD waiver.

Exhibit 2 shows how the average percentage of residential facilities accepting different forms of insurance and self-payment and providing charity care changed in IMD waiver states after waiver adoption compared with in nonwaiver states, as well as the covariate-adjusted average percentage-point changes in the proportion of residential facilities accepting each payment type and providing charity care in states with IMD waivers compared with in nonwaiver states. These coefficient estimates report the interaction terms between the indicator for being an IMD waiver state and time relative to the waiver (for example, −1). In the years before IMD waiver implementation, adjusted trends in accepting different types of payment and providing charity care were not statistically different between the two groups of states, thus providing evidence that there were no major factors affecting facilities’ acceptance of insurance, self-payment, and provision of charity care before the waivers were implemented. Medicaid acceptance by residential treatment facilities increased by 6 percentage points in waiver states relative to nonwaiver states in year +1 of the waiver and 11 percentage points in year +2. Relative to the baseline proportion Medicaid acceptance rate of 32.0 percent, these represent increases of 19 percent and 34 percent, respectively.

Acceptance of private insurance at residential facilities did not significantly change in waiver states relative to nonwaiver states in the year of waiver adoption or in year +1, but by year +2 it increased by 8 percentage points, representing a 14 percent increase over baseline (exhibit 2). Acceptance of other public insurance also increased by year +2 by 7 percentage points (18 percent over baseline) at residential facilities in waiver states compared with nonwaiver states. Conversely, in year +2, acceptance of self-payment decreased by 4 percentage points compared with baseline (4 percent). For charity care, there was a significant decrease of 3 percentage points in year +1 (4 percent), but there was no significant difference by year +2.

Exhibit 3 shows changes in insurance acceptance, self-payment acceptance, and charity care provision at intensive outpatient facilities after IMD waiver adoption compared with changes in nonwaiver states. Intensive outpatient treatment facilities’ Medicaid acceptance increased by 5 to 7 percentage points in the year of waiver adoption through year +2 (9–12 percent over baseline) after waiver adoption. Intensive outpatient facilities’ acceptance of other public insurance increased by 4 percentage points in waiver states by year +2 compared with in nonwaiver states (7 percent over baseline). There were no statistically significant changes in private insurance or self-pay acceptance at intensive outpatient facilities after waiver adoption. Charity care provision declined by 2 percentage points (3 percent below baseline) in these facilities in the year of waiver adoption, but coefficient estimates were not statistically significant in year +1 or year +2. In nonintensive outpatient facilities, we observed no change in the probability that a facility accepted Medicaid payments after waiver adoption; however, the probability that a facility accepted private coverage payments increased by as much as 6 percentage points (11 percent above baseline), whereas the probability that a facility accepted self-payments decreased by 10 percentage points (11 percent below baseline) in year +2 (exhibit 4). Within the sample of nonintensive outpatient facilities, the probability that a facility offered charity care increased by 6 percentage points (8 percent over baseline) 2 years after waiver adoption. Appendix exhibits A5–A7 provide additional details on the analysis reported in exhibits 2–4.¹³

Exhibit 3:

Effect of Institutions for Mental Diseases (IMD) waivers on the average percentage of intensive outpatient facilities accepting different types of payment and providing charity care, 2010–18

Outcome	Medicaid	Private	Other public	Self-payment	Charity care
Percent of facilities accepting different types of payment and providing charity care in IMD waiver states, averaged over years before waiver adoption	57	68	59	93	80
Years relative to IMD waiver adoption year a
−10 to −5	−0.01	0.03**	0.01	0.02**	0.04*
−4	−0.00	0.02	0.01	0.01	0.03**
−3	0.00	0.01	0.02	0.01	0.02
−2	0.00	0.00	0.01	0.00	0.01
−1	—b	—b	—b	—b	—b
0 (year of adoption)	0.05**	0.00	−0.01	0.00	−0.02***
+1	0.07****	−0.01	0.02*	−0.01	−0.05
+2	0.05**	0.01	0.04**	0.00	−0.02
n (facilities in a year)	45,498	45,712	45,984	46,077	46,317

SOURCE Authors’ analysis of the National Survey of Substance Abuse Treatment Services linked to data on years of adoption of IMD waivers collected from the Centers for Medicare and Medicaid Services. NOTES Other public is Medicare, other state-financed, or military coverage. Charity care is a sliding fee scale (patients pay a share of the overall treatment bill), free care to all patients, or some other form of financial assistance to patients who cannot pay.

^aRegression coefficient estimates of a covariate-adjusted linear probability regression model. The numbers reflect percentage-point changes in the outcome variables. See online appendix exhibit A6 for details (see note 13 in text).

^bReference period for interpreting the event-time indicators.

^*p < 0.1

^**p < 0.05

^***p < 0.01

^****p < 0.001

Exhibit 4:

Effect of Institutions for Mental Diseases (IMD) waivers on the average percentage of nonintensive outpatient treatment facilities accepting different types of payment and providing charity care, 2010–18

Outcome	Medicaid	Private	Other public	Self-payment	Charity care
Percentage of facilities accepting different types of payment and providing charity care in IMD waiver states, averaged over years before waiver adoption	53	54	54	89	77
Years relative to IMD waiver adoption year a
−10 to −5	−0.02	−0.01	−0.02	−0.00	0.02*
−4	−0.01	−0.01	−0.02*	−0.00	0.01
−3	−0.01	0.01	−0.01	0.01	0.01
−2	−0.01	0.00	−0.01	0.00	−0.00
−1	—b	—b	—b	—b	—b
0 (year of adoption)	0.01	0.02**	−0.02*	−0.01	−0.01
+1	0.02	0.04	−0.01	−0.00	−0.00
+2	0.04	0.06****	−0.03	−0.10****	0.06***
n (facilities in a year)	40,862	40,762	41,193	41,174	41,507

SOURCE Authors’ analysis of the National Survey of Substance Abuse Treatment Services linked to data on years of adoption of IMD waivers collected from the Centers for Medicare and Medicaid Services. NOTES Other public is Medicare, other state-financed, or military coverage. Charity care is a sliding fee scale (patients pay a share of the overall treatment bill), free care to all patients, or some other form of financial assistance to patients who cannot pay.

^aRegression coefficient estimates of a covariate-adjusted linear probability regression model. The numbers reflect percentage point changes in the outcome variables. See online appendix exhibit A7 for details (see note 13 in text).

^bReference period for interpreting the event-time indicators.

^*p < 0.1

^**p < 0.05

^***p < 0.01

^****p < 0.001

Appendix exhibits A8, A9, and A10 show the changes in the probability that a facility had an opioid treatment program or provided any medications for opioid use disorder in states adopting the waiver compared with those that did not.¹³ We did not find evidence that IMD waivers affected these outcomes among residential treatment facilities. The percentage of nonintensive outpatient facilities with opioid treatment programs and the proportion of those providing medications for opioid use disorder increased in the first year after waiver adoption but did not persist in year +2. In intensive outpatient facilities, the probability of having an opioid treatment program declined in year +2 after IMD waiver adoption, but the probability of providing medications for opioid use disorder increased. We hypothesize that the medications for opioid use disorder finding is driven by the provision of buprenorphine and naltrexone.

We did not observe a clear pattern for the overall number of residential facilities in the states postwaiver (appendix exhibit A19).¹³ There was a shift toward a greater number of for-profits.

Discussion

The IMD exclusion has prohibited states from using federal Medicaid funds for services provided to nonelderly adult enrollees residing in psychiatric institutions, but IMD waivers have granted states greater flexibility to cover residential SUD treatment in these facilities. States adopting IMD waivers are required to provide Medicaid coverage to enrollees that supports delivery of a continuum of services for SUD treatment. In the two years after state adoption of IMD waivers during 2010–18, we found that residential facilities’ acceptance of Medicaid increased more in IMD waiver states than in states not implementing IMD waivers. This change occurred one year after the waiver implementation and continued in the subsequent year. The increased acceptance of Medicaid by SUD treatment facilities has also been observed as a response to public insurance expansions²⁷ and to the federal mental health parity law.²⁸ Our findings underscore that facilities operating in the SUD treatment sector are likely to change the types of payments they accept in response to insurance policy reforms that alter how SUD treatment is financed.

We also found evidence that Medicaid acceptance increased at intensive outpatient facilities in states implementing the IMD waiver more than at similar facilities in nonwaiver states. The magnitude of these changes was substantial and suggests that IMD waivers broadened the range of SUD treatment services available to nonelderly adult Medicaid enrollees.

More surprisingly, residential facilities’ acceptance of private insurance and non-Medicaid public insurance programs increased more in states implementing IMD waivers compared with in states not implementing waivers. One possible explanation is that participation in other insurance programs became easier for facilities once they began accepting Medicaid. This change could be related to investments such as hiring billing staff to file claims and purchasing computer systems; that is, once facilities increased staff and purchased computer systems required for Medicaid participation, the facilities had the capacity necessary to participate in private insurance or other public insurance programs.²⁹ There were not clear trends in acceptance of public or private insurance by outpatient facilities after states’ IMD waiver adoption. Increased acceptance of Medicaid by intensive outpatient facilities could reflect the requirement in IMD waivers that states broaden the continuum of care in Medicaid benefits packages to include services such as partial hospitalization.

Less beneficially for patients, there was reduced charity care at residential facilities in the first year postwaiver. There were also significant declines in self-payment at residential facilities in years +1 and +2. If the IMD waivers increased the volume of Medicaid enrollees in residential treatment facilities, those facilities may have been less likely to accept clients that could not pay for care as they have limited treatment slots. This change may be associated with less access to SUD treatment for such populations.

Although IMD waivers were intended to increase access to medications for opioid use disorder, we did not find evidence of such changes at residential treatment facilities after waiver implementation. However, we found some evidence that provision of medications for opioid use disorder increased at outpatient treatment facilities postwaiver, with apparently longer-lasting effects in intensive outpatient settings.

These findings should be considered in the context of two related challenges for SUD treatment. First, there is a general underprovision of medications for opioid use disorder, which is recognized as a clinically effective treatment.^30,31 Second, IMD waiver critics have raised the concern that by enabling Medicaid financing for residential care, the waivers will promote the inappropriate use of residential treatment over other modalities, such as outpatient treatment with medication, which may be more appropriate for many patients.³² Thus, policy makers may wish to consider ways to ensure that Medicaid enrollees, with their now-better access to residential treatment, do not disproportionately rely on this modality of treatment if other modalities are more appropriate.

Conclusion

Since 2015, IMD waivers for specialty SUD treatment have become an increasingly common approach for states to expand their coverage of SUD treatment. This expansion is expected to grow further after the implementation of the Substance Use–Disorder Prevention that Promotes Opioid Recovery and Treatment for Patients and Communities Act of 2019, which allows states that did not receive IMD waivers to receive short-term amendments to their Medicaid state plans to cover services received in IMD. Residential treatment will thus likely contribute to the growth of Medicaid expenditures on SUD treatment and further expand the role of Medicaid as a payer for SUD services while reducing the burden on SUD treatment providers who provide free or heavily discounted care. IMD waivers may be an important step in advancing access to a full continuum of SUD treatment for Medicaid enrollees. Although the IMD waivers are intended to broaden the continuum of care in Medicaid benefit packages, their ultimate impact on the health of Medicaid enrollees will depend on concurrent efforts to improve the quality of care in every setting.

Biographies

Johanna Catherine Maclean (catherine.maclean@temple.edu) is an associate professor of economics at Temple University, in Philadelphia, Pennsylvania, and a research associate at the National Bureau of Economic Research in Cambridge, Massachusetts, and a research affiliate at the Institute of Labor Economics.

Hefei Wen is an assistant professor in the Division of Health Policy and Insurance Research, Department of Population Medicine, Harvard Medical School and the Harvard Pilgrim Health Care Institute, in Boston, Massachusetts.

Kosali Simon is the Herman B Wells Endowed Professor at the Paul H. O’Neill School of Public and Environmental Affairs, and associate vice provost for health sciences, Indiana University, in Bloomington.

Brendan Saloner is an associate professor in the Department of Health Policy and Management at the Johns Hopkins Bloomberg School of Public Health, in Baltimore, Maryland.