Just What the Nurse Practitioner Ordered: Independent Prescriptive Authority and Population Mental Health

We examine whether relaxing occupational licensing to allow nurse practitioners (NPs)â€�?registered nurses with advanced degreesâ€�?to prescribe medication without physician oversight is associated with improved population mental health. Exploiting time-series variation in independent prescriptive authority for NPs from 1990â€“2014, we find that broadening prescriptive authority is associated with improvements in self-reported mental health and decreases in mental-health-related mortality, including suicides. These improvements are concentrated in areas underserved by psychiatrists and among populations traditionally underserved by mental health providers. Our results demonstrate that extending prescriptive authority to NPs can help mitigate physician shortages and extend care to disadvantaged populations.


Introduction
Limited access to mental health care services in the United States is a major public health concern.
While one in five Americans suffers from a mental illness-including generalized anxiety and major depressive disorders-nearly one third of the U.S. population lives in areas that are underserved by mental health care providers (Bureau of Health Workforce, 2016). 1 As mental disorders tend to develop early in life and persist over the lifecycle (Merikangas et al., 2010), the costs of not receiving treatment can be substantial. In addition to direct medical costs, untreated mental illness is associated with lower human capital accumulation, worse labor market participation and performance, and greater criminal activity (see, for example, Currie and Stabile, 2006;Ettner et al., 1997;Greenberg and Rosenheck, 2008;. Recent evidence demonstrates that increases in all-cause mortality among non-Hispanic whites are being driven by "deaths of despair" (deaths from suicides, drug and alcohol poisonings, and alcoholic liver disease and cirrhosis ;Deaton, 2015, 2017), adding energy and urgency to the search for policies that can be used to improve population mental health.
In this paper, we examine whether allowing nurse practitioners (NPs)-a class of registered nurses with advanced degrees in nursing-to prescribe medication without physician supervision or collaboration leads to improvements in measures of population mental health. Leveraging a novel dataset that documents legislative changes granting NPs independent prescriptive authority over 25 years, we find that states that broaden prescriptive authority experience improvements in population mental health. These improvements are concentrated among disadvantaged populations, suggesting that extending independent prescriptive authority to NPs is an important policy tool that can be used to improve the mental health of populations with limited access to care.
Despite a burgeoning literature demonstrating that NPs can safely and efficiently provide a variety of services, including an endorsement of the skills of NPs by the Institute of Medicine (IOM, 2011), efforts to extend prescriptive authority beyond physicians are controversial. Opponents 1 According to the 2016 National Survey on Drug Use and Health, only 43.1% of U.S. adults aged 18 and older with a mental illness received counseling (inpatient or outpatient) or used a prescription medication for problems with mental health in the past year (SAMHSA, 2017). worry that allowing NPs to prescribe medication will put patients in danger since NPs receive fewer years of training, are held to different legal standards, and go through a different process of licensing than medical doctors. 2 Critics further note that extending prescriptive authority beyond physicians need not expand overall use of pharmacological treatment, as the prescriptions written by non-physician providers may simply crowd out the prescriptions previously written by physicians. The American Medical Association (AMA), a national professional organization representing physicians and medical students in the Unites States, has been particularly vocal in opposing the expansion of state-level scope of practice legislation (AMA, 2010).
To measure the impact of extending independent prescriptive authority to NPs on population mental health, we exploit time-series variation in state-level scope of practice legislation and mental health outcomes from 1990 to 2014 using a difference-in-difference framework. We use two complementary categories of health outcomes: (1) self-reported mental health at the individualyear level from the Behavioral Risk Factor Surveillance System and (2) mental-health-related mortality at the county-quarter level from the U.S. Mortality Files. 3 Our results demonstrate that extending prescriptive authority to NPs leads to significant reductions in the number of days spent in poor mental health (0.17 days per month, or 5% of the mean). We further find a negative relationship between independent prescriptive authority and mental-health-related mortality, although the effect is imprecisely estimated.
Notably, improvements in mental health outcomes resulting from independent prescriptive authority for NPs are larger and more precisely estimated in areas that are underserved by physicians and among disadvantaged populations. In particular, areas that are underserved by psychiatrists see statistically significant improvements in self-reported mental health and mental-health-related mortality that are at least twice as large as those experienced on average: respondents in underserved states see an additional reduction of 0.17 days per month in poor mental health (5% of the 2 It has been estimated that NPs can safely provide 70-80% of the care provided by physicians in primary care (Scheffler et al., 1996). Furthermore, evidence suggests that there are no differences in health outcomes between patients treated by NPs rather than MDs, and patient satisfaction is, if anything, higher among patients seen by NPs (Horrocks et al., 2002;Laurant et al., 2008;Lenz et al., 2004;Mundinger et al., 2000;Naylor and Kurtzman, 2010). 3 As described in Section 3, we define "mental-health-related mortality" as suicides, injuries of undetermined intent, and accidental deaths involving poisonings, drownings, firearms, and trains. mean) and underserved counties see additional reductions of over five mental-health-related deaths (11% of the mean) and almost two suicides (7% of the mean) per quarter. Individuals with a high school degree or less also see strong improvements in both measures of mental health, with the largest benefits in percentage terms accruing to individuals with low levels of education residing in undeserved areas.
Allowing NPs to prescribe independently should disproportionately affect disadvantaged populations for two reasons. First, since psychiatrists and other physicians are more likely to locate in urban and suburban areas, populations in rural areas have the most limited access to psychotropic treatment (Hartley et al., 2004). Second, psychiatrists are less likely than all other physician specialties to accept insurance, with acceptance rates being lowest for Medicaid beneficiaries (Bishop et al., 2014). Therefore, even in areas with a sufficient number of physicians, access to psychotropic medications may still be limited for certain populations. Since NPs are more likely than physicians to locate in rural and inner-city locations and to accept public insurance (Buerhaus et al., 2015;Everett et al., 2009;Grumbach et al., 2003;Larson et al., 2003), granting independent prescriptive authority to NPs has the potential to address physician shortages and extend care to disadvantaged populations.
Finally, using detailed prescription data from 2006 to 2014, we find evidence that extending independent prescriptive authority to NPs increases the use of psychotropic medications among disadvantaged populations. Similar to our findings for mental-health-related mortality and selfreported mental health, these effects are concentrated in areas that are underserved for mental health care resources. We find the largest increases in antidepressant and antipsychotic use among Medicaid beneficiaries-a low-income population that we expect to be particularly affected by scope of practice legislation for NPs. Despite having a much shorter time frame, which greatly limits our statistical power over our analyses of mental health outcomes, we nonetheless find evidence that the use of psychotropic medications among disadvantaged populations increases when more providers can prescribe independently.
In light of rising rates of abuse of prescription pain medication (SAMHSA, 2014), one concern with broadening prescriptive authority for NPs is that such legislation could lead to a greater number of opioid analgesics available for misuse. Again using prescription data from 2006-2014, we find that broadening prescriptive authority leads to general increases in opioid prescriptions. While there is a strong, positive relationship between opioid prescriptions per capita and deaths involving drugs within locations over time (Schnell and Currie, 2018), we nevertheless find that allowing NPs to prescribe independently leads to fewer mental-health-related deaths, a figure that includes overdoses. Therefore, if anything, our main results underestimate the effects of only extending prescriptive authority for non-controlled substances such as antidepressants and antipsychotics.
Our work contributes to a growing literature in economics that empirically examines the implications of occupational licensing, most of which measures the effects of such legislation on wages, employment, and prices across related occupations and services. 4,5 We depart from this literature by focusing on outcomes of the production process-self-reported mental health and mental-health-related mortality-rather than the organization and division of resources across actors in the production process itself.
The two most closely related studies to our work are Stange (2014) and Traczynski and Udalova (2018). Stange (2014) finds that allowing NPs to prescribe controlled substances with or without physician supervision only leads to modest increases in healthcare utilization, whereas Traczynski and Udalova (2018) find that allowing NPs to both practice and prescribe independently leads to increases in utilization of primary care services. Our paper departs from this previous literature in four important dimensions. First, given well-documented shortages in access to mental health care services in the United States, we focus on the impact of broadening scope of practice legislation on the use of psychotropic medications and mental health outcomes. 6 Second, we use a time 4 A type of occupational licensing, scope of practice restrictions for NPs are often justified as the state protecting the consumer from receiving substandard care. If consumers are more confident in the services provided as a result of this legal reassurance, restrictive scope of practice legislation will increase demand. However, since restrictive scope of practice legislation limits the number of providers who can perform a given service, these increases in demand may be offset by decreases in supply. While restrictive scope of practice legislation should weakly increase the quality of services, the theoretical effects on quantity are ambiguous. 5 For example, see Kleiner and Park (2010) and Marier and Wing (2011) for the case of dentists and dental hygienists and Dueker et al. (2005), Kleiner et al. (2016), Stange (2014), and Xue et al. (2016) for the case of physicians and non-physician providers. 6 Our focus on mental health motivates the law changes that we consider. Since access to a provider who can horizon of 25 years, which allows us to consider the effects of a larger number of law changes than previous papers. Third, we use larger and more representative datasets than previous work; combined with the long time horizon, this gives us the power to look for heterogenous effects of broadened scope of practice legislation. 7 We find that extending independent prescriptive authority to NPs has larger effects among disadvantaged populations, making it a particularly attractive policy instrument for reducing inequality. Finally, given our unique prescription data, we are able to document a "first stage" that helps us understand the mechanisms through which broadened scope of practice legislation affects population mental health.
More broadly, our work contributes to the literature that examines how mental health is affected by policy interventions. Previous studies have focused primarily on policy efforts to improve access to physician-provided care-by increasing access to health insurance, mandating parity in reimbursement for mental health care services, or altering incentives for graduating physicians to enter either psychiatry or primary care-and find mixed results (see, for example, Cunningham, 2009;Finkelstein et al., 2012;Rabinowitz et al., 2008). In contrast to this line of work, we focus on a policy that can increase the accessibility of medical care for disadvantaged populations immediately and at a low cost: there are currently over 234,000 NPs already licensed in the United States who could prescribe independently if legislation permitted them to do so (AANP, 2017).
Taken together, our results demonstrate that relaxing occupational licensing for non-physician providers can help mitigate the negative consequences of limited access to physician-provided health care. In particular, states that are underserved by physicians can grant independent prescriptive authority to NPs to improve the mental health of their residents. The potential for such prescribe psychotropic medications is a significant barrier to mental health care in the United States, we focus on legislation that allows NPs to prescribe medication without the supervision or collaboration of a physician. In contrast to Stange (2014), we do not separately consider legislation that allows NPs to prescribe controlled substances in addition to unscheduled drugs. Since the majority of antidepressants and antipsychotics are not scheduled, we believe that the relevant legislation is whether NPs can prescribe at least non-controlled substances independently. We note, however, that only one state (Kentucky) allows NPs to prescribe unscheduled drugs but not controlled substances independently (see Footnote 10 for additional information on independent prescriptive authority for controlled substances). 7 While Traczynski and Udalova (2018) find no evidence of heterogenous effects, they are likely underpowered given both the small sample size of the Medical Expenditure Panel Survey (MEPS) and the more limited time period that they consider. We further note that the MEPS is only representative at the national level, whereas the data we use is either representative at the state level (BRFSS), covers the universe of deaths (U.S. Mortality Files), or covers the universe of prescriptions filled at U.S. retail pharmacies for certain drug classes (IQVIA). legislative action remains large: as of January, 2015, only 24 states and the District of Columbia had granted independent prescriptive authority to NPs. Noticeably, no state in the South has yet to allow NPs to independently prescribe. This paper proceeds as follows. We begin by providing additional background on NPs and scope of practice legislation in Section 2. We then introduce our data in Section 3. In Section 4, we examine how mental-health-related mortality and self-reported mental health respond when independent prescriptive authority is extended to NPs. In Section 5, we examine how the number of prescriptions for antidepressants, antipsychotics, and opioids change when NPs can prescribe independently. Section 6 concludes.

Background
The number of NPs in the United States has grown rapidly in recent decades, with the number of licensed NPs more than doubling from 120,000 in 2007 to over 234,000 today (AANP, 2017).
To become an NP, registered nurses must complete a master's or doctoral program that provides advanced clinical training beyond their undergraduate nursing education and complete local licensure and national certification requirements. NPs practice in a wide range of settings, including physician practices, hospitals, community health centers, and private NP practices (AANP, 2014).
While the training requirements for NPs are similar across the United States, states have the authority to dictate what NPs are able to do. In states with liberal scope of practice legislation, NPs have the authority to evaluate, diagnose, and treat patients-which includes ordering and interpreting diagnostic tests, initiating and managing treatments, and prescribing medication-under the licensure authority of the state board of nursing. In states with more restrictive scope of practice legislation, NPs are required to undergo career-long supervision, delegation, or team-management by another health provider in order to provide patient care. 8 Such legislation can be very costly 8 As is common in the literature, we refer to these mandatory professional relationships as "physician supervision or collaboration." The legal requirements for such relationships, both for practice authority and prescriptive authority, differ across states with restrictive scope of practice legislation. For example, NPs in Georgia can only prescribe if a physician delegates the authority pursuant to a written protocol agreement, NPs in Virginia can only prescribe as part for NPs: anecdotal evidence suggests that NPs often have difficulty finding or affording physicians who are willing to supervise or work in collaboration, and it is not uncommon for NPs to have to move or stop practicing when the physician with whom they contract has moved, retired, or died (Sadeghi, 2017). Broadened scope of practice legislation therefore lowers the cost of practice and may increase access by increasing both the number of providers and the effective labor supply of each provider.
We focus specifically on scope of practice legislation that grants NPs the authority to prescribe medication independently. Given significant disparities in access to providers who can prescribe psychotropic medications (Bishop et al., 2014;Hartley et al., 2004), extending independent prescriptive authority to NPs may be particularly relevant for improving the provision of mental health care in the United States. Broadly speaking, there are two types of treatment for mental illness: psychotherapy and psychotropic medication. A complementarity between the two has been well documented, and in most cases it is recommended that a patient receive a combination of both (SAMHSA, 2015). Despite this ideal of psychotherapy in conjunction with psychotropic medication, it is often much easier to find consistent access to therapy than to medication (Thomas et al., 2009). While all mental health professionals can offer some degree of counseling services, traditionally only psychiatrists and other medical doctors have the legislative authority to prescribe medications independently.
Although some NPs specialize in psychiatric and mental health, these providers make up less than four percent of the total NP workforce (AANP, 2014). Rather, the vast majority of NPs are trained in primary care programs and focus on adult, family, and pediatric health; gerontology; and women's health. Despite their generalist training, most primary care NPs diagnose and treat mental illness on a regular basis. According to a recent survey, 66 (63) percent of general practice NPs treat anxiety (depression) in their practice, with the numbers being even more pronounced among NPs in family practice (76 and 74 percent, respectively) (AANP, 2012). This pattern is similar among physicians: while psychiatrists are the only MDs that specifically focus on mental health, many of a patient care team, and NPs in South Carolina must maintain a written collaborative agreement with a physician in order to prescribe. general practitioners also provide mental health services (Kessler and Stafford, 2008). Therefore, increased access to NPs in general, rather than just those specializing in psychiatric medicine, may have the potential to improve population mental health.

Data
We combine data from seven sources to measure how extending prescriptive authority to NPs affects population mental health. In particular, we merge a new dataset detailing independent prescriptive authority for NPs with administrative data on mental-health-related mortality from the U.S. Mortality Files, survey data on self-reported mental health from the Behavioral Risk Factor Surveillance System survey, and detailed data on prescriptions filled at U.S. retail pharmacies from IQVIA's Xponent database. These data are supplemented with information on the provision of local medical resources and population demographics from the Area Resource Files, the American Community Survey, and the U.S. Census. Each dataset is described in detail below.

Independent Prescriptive Authority
Our first dataset documents whether NPs had the legislative authority to independently prescribe medication in each month from 1990 to 2014 in each state and the District of Columbia. This dataset was constructed by the authors and combines information from the The Nurse Practitioner's "Annual Legislative Update," correspondences with state nursing boards, and readings of primary source legislation. 9 As discussed in Section 2, the language of scope of practice legislation is particular to each state. We define independent prescriptive authority as the ability to prescribe medication with-9 The Nurse Practitioner is a journal addressing clinical issues relevant to NPs and other primary care providers. Every January since 1989, the journal has published the "Annual Legislative Update" which summarizes both the practice environment and the level of prescriptive authority for NPs in each state. While informative, these overviews do not consistently include dates of legislative action nor comprehensive coverage of the precise changes made to a state's legislation. Therefore, the information provided by the journal alone is not sufficient for a quantitative analysis of independent prescriptive authority. out physician collaboration or supervision. 10,11 As of January 1 st , 1990, six states and the District of Columbia had already granted NPs statutory authority to independently prescribe medication.
Between 1990 and 2014, 18 states changed their scope of practice legislation to allow NPs to prescribe without physician involvement. This geographic and temporal variation in scope of practice legislation is displayed in Figure 1; the exact dates of the law changes used in our analysis are provided in Table A.1. Figure 1: Changes in Independent Prescriptive Authority for NPs: 1990NPs: -2014 Prescriptive authority since 1990 Gained prescriptive authority (1990−2014) No prescriptive authority Notes: We define a state as having independent prescriptive authority if NPs registered in the state have the statutory authority to prescribe medications without physician collaboration or supervision. The exact dates of the law changes used in our analysis are provided in Table A.1.
It is difficult to say how states decide when to grant NPs independent prescriptive authority. If states broaden scope of practice legislation for NPs in response to declining economic conditions 10 Except for Kentucky, in which NPs are required to have a collaborative practice agreement with a physician in order to prescribe controlled substances, all states that allows NPs to prescribe unscheduled drugs independently also allow NPs to prescribe controlled substances independently. While independent prescriptive authority for both unscheduled drugs and controlled substances is included with licensure in many states with liberal scope of practice legislation, in some states with independent prescriptive authority, such as Connecticut and the District of Columbia, NPs must apply for a controlled substance registration to independently prescribe Schedule II-V drugs. 11 Eight of the 24 states that allow NPs to prescribe independently require a post-licensure or certification period of supervision or collaboration before an NP is granted independent prescriptive authority. For example, a prescribing mentorship with a physician or another advanced practice nurse is required for the first 1,000 hours of practice in Colorado; in Maine, an NP must be supervised by either a physician or an NP for the first 24 month of the NP's practice. Since these restrictions are temporary, rather than career-long, we consider states that require a preliminary period of supervision or collaboration as having independent prescriptive authority. or worsening health, or at a similar time as other laws that independently impact population mental health, then our results will be subject to omitted variable bias. Pei et al. (ming) demonstrate that balancing regressions-which use candidate controls as dependent, rather than independent, variables-provide a powerful test of orthogonality. In Table A.3, we estimate balancing regressions to examine whether the timing of laws granting NPs independent prescriptive authority are correlated with various individual-level, county-level, and state-level outcomes (e.g., individual health insurance status, county-level provider counts, and state-level unemployment rates). Reassuringly, we find no evidence that the law changes we consider are driven by these measures.

Health Resources
Increasing the supply of providers who can prescribe should have greater impacts in areas with an insufficient supply of such providers. According to the Health Resources and Services Administration (HRSA), an area is "underserved" for mental health care services if there is fewer than one psychiatrist for every 30,000 people. Using this definition, we identify underserved counties by combining county-level psychiatrist counts from the HRSA's Area Resource File in 1990 with county population estimates from the 1990 Census. We use the measure from the beginning of our sample to avoid introducing bias from changes to the supply of medical providers driven by changing scope of practice legislation (Xue et al., 2016). 12 Over our sample period, approximately 20% of the U.S. population lived in counties that were underserved for mental health care services.
The survey data outlined in Section 3.3.2 only provides us with the state, not the county, of respondents. We therefore also need a measure of how well-equipped each state is for mental health care services. To take into account the geographic distribution of resources within a state, we define a state as being "underserved" for mental health care services if the population-weighted average of binary, underserved categorizations across all counties in the state is less than the median across all states in 1990. 13 Figure 2 shows the geographic variation in underserved counties and 12 Despite this concern, all of our results are robust to instead using a time-varying measure of the provision of mental health care resources (see Tables A.6 and A.15). 13 We could alternatively define underserved at the state-level by applying the HRSA's definition to state-level Notes: Following the definition provided by the HRSA, a county is "underserved" for mental health care services if the county has fewer than one psychiatrist per 30,000 people. We identify underserved counties by combining information on the number of psychiatrists per county from the HRSA's 1990 Area Resource File with county population estimates from the 1990 Census. A state is "underserved" for mental health care services if the population-weighted average of binary, underserved categorizations across all counties in that state is less than the median across all states in 1990.

states.
While we define an area as being underserved for mental health care resources by whether they have enough psychiatrists to serve the population, this characterization is an over-simplification of how mental health care is delivered in the United States. Psychiatrists are the only MDs specifically trained to treat mental illness, although in practice many general practitioners also provide treatment for mental illness (Kessler and Stafford, 2008). Despite overlap in the services provided by psychiatrists and general practitioners, we define areas as underserved for mental health care services using psychiatrist-to-population ratios for two reasons. First, the HRSA only defines mental health care shortage areas using the number of psychiatrists, so it is not clear which threshold would be appropriate to use if we were to consider counts of both psychiatrists and general practitioners. Second, in practice, defining shortage areas based on psychiatrist-to-population ratios also captures areas that are underserved by general practitioners; that is, areas defined as underserved by psychiatrists also have fewer general practitioners per capita (see Table 1). Therefore, we identify meaningful variation in the availability of mental health care providers by focusing on areas underserved by psychiatrists.

Mental Health Outcomes
We identify the impact of independent prescriptive authority for NPs on population mental health using two complementary outcomes: mental-health-related mortality and self-reported days in poor mental health. For each outcome, we consider how extending independent prescriptive authority to NPs impacts both the local population as a whole and disadvantaged subpopulations who may find it particularly difficult to access physician-provided care.
psychiatrist-to-population ratios. Note that this is mathematically equivalent to applying the HRSA's definition to the population-weighted average of county-level psychiatrist-to-population ratios within a state. However, since cities tend be far from the HRSA's definition of underserved, and since cities have large populations, no state is considered underserved according to this measure. We therefore use the cut-off of above and below median to create a binary measure, although we show in Panel B of Table A.15 that our results are robust to instead using a continuous measure of underserved (population-weighted average of binary, underserved categorizations across all counties in the state; between zero and one).
We believe that the broad measure of mental-health-related mortality provides a more accurate picture of mortality caused by poor mental health rather than suicides alone for two reasons. First, geographic variation in reported suicides reflects both differences in true suicides and differences in cause-of-death reporting (Bjorkenstam et al., 2014;Cooper and Milroy, 1995;Rockett et al., 2006Rockett et al., , 2010Tollefsen et al., 2010). When someone dies from an overdose of oxycodone, for example, the local coroner decides whether to label the death as a suicide or an accidental poisoning. Our broad measure of mental-health-related deaths captures both causes of death, whereas "suicides" only captures the former. Second, drug addiction is an increasingly important category of mental illness, and thus we are interested in drug-related deaths even if suicide was not the individual's intent. 15 The mortality files contain demographic information for the deceased individual. In particular, the deceased's county of residence, sex, race, age, and level of education are recorded. We use this information to determine both the total number of deaths at the county level as well as the number of deaths among subpopulations of interest. As the mortality files contain no information on the deceased's income, we use education as a proxy for socioeconomic status.
While the mortality files tell us the number of people who died, they provide us with no in-14 A large literature documents that injuries of undetermined intent and accidental deaths commonly obscure true suicide rates (see, for example, Bjorkenstam et al., 2014;Cooper and Milroy, 1995;Hilkevitch, 2005;Rockett et al., 2006;Rockett et al., 2010). On average over our sample period, 53.4% of mental-health-related deaths were suicides; 31.9% were accidental poisonings; 6.6% were injuries of undetermined intent; and 5.8%, 1.4%, and 1.1% were accidental deaths involving drownings, firearms, and trains, respectively. 15 There is an extensive body of literature in medicine and psychiatry discussing the feedback between substance abuse disorders and other types of mental illness. Most prominently, the self-medication hypothesis posits that substance abuse is often related to other underlying mental illness via self-medication (Barkus and Murray, 2010;Khantzian, 1985Khantzian, , 1997Nock et al., 2010;Regier et al., 1990). Notes: Observations are at the county-quarter level. Statistics are weighted by population. "Mental-health-related deaths" include suicides, injuries of undetermined intent, and accidental deaths involving poisonings, drownings, firearms, and trains. "Ever (Never) Indep. Rx" includes counties that had independent prescriptive authority for NPs at some point (at no point) during our sample. "(Not) Underserved" reflects whether a county was underserved for mental health care services in 1990. "Low educ." is defined as having a high school degree or less. Mortality statistics come from the U.S. Mortality Files, provider counts come from the HRSA's Area Resource Files, and all other variables come from the Census and the five-year pooled American Community Surveys (ACS). Census and ACS variables are linearly interpolated at the county-year level for years not available.
formation on the size of the population base. When one area reports having more deaths than another, we cannot determine from the mortality files alone whether this is because the population is larger and the death rates are the same, or whether the location experienced a disproportionate number of deaths. To take into account the size of the relevant population, we combine the number of deaths at the county-quarter level with county-year population counts from the Census and As shown in Table 1, counties in states that allow NPs to prescribe independently at some point during our sample period tend to be less densely populated and less racially diverse. However, both groups of counties have very similar unemployment rates and age profiles. Perhaps unsurprisingly, underserved counties are on average less densely populated, less educated, and more white than counties with adequate mental health care resources.
Before proceeding to a formal difference-in-difference analysis, we first examine mortality patterns in the raw data. Figure 3 shows unadjusted mental-health-related mortality rates at the county-year level in event time around the year in which states grant NPs independent prescriptive authority. We plot death rates separately for counties that are and are not underserved for mental heath care services and separately for the whole population and among individuals with a high school degree or less. Suicide rates are fairly noisy, likely due both to low incidence and to differences in how suspected suicides are reported across counties and over time (Rockett et al., 2006(Rockett et al., , 2010. However, there is a clear pattern of decreased mental-health-related deaths after states allow NPs to prescribe independently. These decreases are more pronounced in underserved counties and 16 The Census only provides intercensal county-level population estimates across age, sex, race, and ethnic groups, while our analysis requires county-level population estimates for all of the subgroups that we consider. For example, we consider mental-health-related mortality among individuals with a high school degree or less, and the Census does not provide intercensal population estimates by education. We have compared our linearly interpolated estimates to the intercensal estimates from the Census where available, and the numbers are nearly identical. Our results are therefore robust to using the intercensal estimates when possible. are the largest among individuals with a high school degree or less who live in these underserved areas. Furthermore, there is no evidence of pre-trends in mortality before the law changes.

Underserved
Not underserved Notes: Observations are at the county-year level and are population weighted. "Mental-Health-Related Deaths" includes suicides, injuries of undetermined intent, and accidental deaths involving poisonings, drownings, firearms, and trains. A county is "underserved" if it had fewer than one psychiatrist per 30,000 people in 1990. "Low educ." is defined as having a high school degree or less.

Self-Reported Mental Health
Our second outcome is the number of days in the past month that a person reports being in poor This question is not designed to draw a particular mental health diagnosis, but rather to indicate whether a respondent experiences any symptoms associated with a wide range of mental health conditions. Importantly, responses are elicited from those with both diagnosed and undiagnosed mental illnesses, as respondents are not asked whether they have ever been diagnosed with a mental illness by a doctor.
We consider as outcome variables both the number of days reported in poor mental health and an indicator for whether the respondent reported having spent at least 21 of the past 30 days in poor mental health. According to the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5), to be diagnosed with a major depressive episode a patient must have either "a depressed mood most of the day, nearly every day" or "a markedly diminished interest or pleasure in all, or almost all, activities most of the day, nearly every day" for two consecutive weeks. In addition to major depressive disorder, the diagnostic criteria for many mental health conditions include extended time periods over which symptoms must be experienced in order for the diagnosis to apply. Thus, a binary variable denoting individuals who experienced prolonged symptoms allows us to identify people who may be suffering from a more severe mental illness.
In addition to self-reported mental health, the BRFSS includes information on each respondent's sex, race, ethnicity, age, education, income, and employment and health insurance status.
These variables allow us to separately consider disadvantaged populations and to control for underlying differences across respondents in our analysis. Table 2, BRFSS respondents report spending an average of 3.36 days in the past month in poor mental health, with 66% of respondents reporting no days in poor mental health and 6% of respondents reporting at least 21 days in poor mental health. Similar to the pattern observed in Table 1, survey respondents in states that ever had independent prescriptive the population. The missing state-years do not correspond to the year before, the year of, or the year after a relevant law change for any state, and thus our identification is not directly affected. Notes: Observations are at the individual level. Statistics are weighted using BRFSS sample weights. "Ever (Never) Indep. Rx" includes respondents living in states that had independent prescriptive authority for NPs at some point (at no point) during our sample period. "(Not) Underserved" reflects whether respondents are living in states that were underserved for mental health care services in 1990. "Low educ." is defined as having a high school degree or less. Some categorical variables do not sum to one; the difference reflects the percentage of missings. authority during our sample have similar age and employment profiles, although states that grant independent prescriptive authority over our sample period are less racially diverse, more educated, and have higher income profiles. As with counties, states that are underserved for mental health care services are more white and less educated than states with a sufficient provision of mental health care resources.

Underserved
Not underserved Notes: Observations are at the individual level and are weighted using the BRFSS sample weights. A state is "underserved" if the population-weighted average of binary, underserved categorizations across all counties in the state is less than the median across all states in 1990. "Low educ." is defined as having a high school degree or less.
As in Section 3.3.1, we examine trends in self-reported health in the raw data before proceeding to a formal difference-in-difference analysis. Figure 4 shows unadjusted self-reported mental health at the individual level in event time around the year in which states grant NPs independent prescriptive authority. We plot self-reported mental health separately for states that are and are not underserved for mental heath care services and separately for the whole population and among individuals with a high school degree or less. The figures suggest that self-reported mental health improves when states grant NPs independent prescriptive authority. The patterns in these improvements follow those observed for mental-health-related mortality in Figure 3: across both measures of self-reported mental health, those living in underserved states report both higher initial levels of poor mental health and larger decreases following broadened scope of practice legislation.
Notably, individuals with low levels of education have worse average baseline mental health and experience the largest improvements when independent prescriptive authority is extended to NPs.

Prescription Data
Finally, to examine how extending independent prescriptive authority to NPs influences the number of prescriptions, we use the IQVIA Xponent database. Our version of the database contains the universe of antidepressant, antipsychotic, and opioid prescriptions filled at U.S. retail pharmacies between 2006 and 2014 and allows us to construct prescription measures at the county-year level.
The database further contains information on how a patient paid for their prescription, allowing us to stratify by payment type: Medicaid, third party, and cash. This is particularly important given that the predicted impacts of expanded scope of practice legislation are largest for populations that traditionally find it difficult to access physician-provided care, such as Medicaid beneficiaries (Buerhaus et al., 2015;Everett et al., 2009;Grumbach et al., 2003). We focus on antidepressants and antipsychotics as two representative classes of pharmaceuticals used to treat mental illness. 18 These medications are widely used: as shown in Table 3, 0.67 antidepressants and 0.12 antipsychotics per capita were filled on average in each year between 2006 and 2014. 19 These medications are also commonly prescribed by NPs, with 8% of antidepressant and 10% of antipsychotic prescriptions being written by NPs.
In addition to antidepressants and antipsychotics, we further examine whether extending independent prescriptive authority to NPs leads to increases in the number of opioid prescriptions-a class of drugs with a high potential for abuse and addiction (SAMHSA, 2014). We note, however, that the market for opioids experienced many changes between 2006 and 2014, including a range of state-level policies aimed at limiting prescribing and curbing abuse (Meara et al., 2016). As these changes may coincide with the changes in scope of practice legislation that we consider, we interpret the results involving opioids with caution. 20 While the data is sufficiently detailed to allow us to examine the number of prescriptions written by physicians and NPs separately, we believe that the total number of prescriptions most accurately reflects changes in prescription patterns associated with expanded prescriptive authority for two reasons. First, we want to capture the net effect of broadened scope of practice legislation on prescription patterns. If people switch from a physician to an NP when NPs are granted independent prescriptive authority but still receive the same prescription, we would not want to claim that the law change resulted in improved access to pharmaceuticals. Second, there are technical issues that arise when attributing prescriptions to different types of providers. When NPs have a collaborative or supervisory relationship with a physician, the prescription pad used by the NP may bear either the NP's name and national provider identifier (NPI) or a combination of the affiliated physician and NP's information. 21 If NPs obtain their own prescription pads when they gain independent prescriptive authority, we would observe a mechanical shift in the number of prescriptions from MDs to NPs in the absence of any true change in the providers writing prescriptions. all three drug classes. If there are decreasing marginal returns to prescriptions, these level differences will result in the marginal patient who receives a prescription in an underserved county to experience larger health benefits than the marginal patient who receives a prescription in a county with a sufficient provision of mental health care resources. 21 In states that require NPs to maintain collaborative or supervisory agreements in order to prescribe, legislation dictating whose information must be on prescription orders and labels differs from state to state. For example, while NPs in Florida are required to maintain a written collaborative agreement for prescriptive authority, prescribing NPs must use their own prescription pad and their name must appear on the prescription label. In contrast, while Kansas requires NPs to maintain a written protocol authorized by a physician in order to prescribe, both the name of NP and the name of the collaborating physician must appear on both the prescription pad and the prescription label. Skillman et al. (2012) estimate that only 76% of NPs had an NPI in 2010, providing an upper bound for the percent of NPs who could have a prescription pad bearing only their name.

Prescriptive Authority and Mental Health Outcomes
To identify whether extending independent prescriptive authority to NPs improves population mental health, we exploit time-series variation in state-level scope of practice legislation and mental health outcomes using a difference-in-difference framework. As described in Section 3.3, we consider two categories of mental health outcomes: mental-health-related mortality and self-reported days in poor mental health. The impact of independent prescriptive authority on each category of outcomes is considered in turn below.

Mental-Health-Related Mortality
Does the prevalence of suicides and other mental-health-related deaths change when NPs are allowed to independently prescribe? Letting Deaths cqy denote either of these outcomes in county c in quarter q of year y, we estimate the following equation: Deaths cqy = β 0 + β 1 Indep. Rx sqy + β 2 P op cy + β 3 X cy + γ c + γ q + γ y + cqy (1) where Indep.Rx sqy is an indicator denoting whether NPs had independent prescriptive authority in state s in quarter q of year y; P op cy is the population of county c in year y; X cy is a vector of other county-year controls; 22 and λ c , λ q , and λ y are county, quarter, and year fixed effects, respectively.
In all of our analyses in this section, standard errors are clustered by state and observations are weighted by population. 23 To avoid introducing measurement error into the outcome, our preferred specification uses the number of deaths in a county-quarter as the dependent variable and includes a control for the corresponding population estimate on the right-hand side. While one could use county-level death rates as the dependent variable, death rates are very sensitive to population counts, and precise countylevel population estimates are only available every ten years. 24 Whereas measurement error from population estimates on the right-hand side will attenuate the estimated coefficient on population, it will not affect the precision of our estimated coefficients of interest. On the other hand, measurement error in the outcome would serve to attenuate all estimated coefficients, including those of key policy relevance.
Increasing the supply of providers who can prescribe medication should impact mental-healthrelated mortality most for populations living in areas with an insufficient supply of prescribing providers and for populations who find it more difficult to access physician-provided care. To allow the impact of changing scope of practice legislation to differentially influence mental health in counties with an under-provision of mental health care services, we estimate the following 22 County-year controls include population density, percent male, percent black, percent age 18 and under, percent age 65 and over, percent with a high school degree or less, a quadratic in median income, percent in poverty, percent unemployed, and the number of practicing psychiatrists and primary care physicians per 100,000. See Table A
To look specifically at disadvantaged populations, we further estimate Equation (2)   Estimates from Equations (1) and (2) are provided in Table 4. As shown in Columns (1) and (4), on average there is no statistically significant effect of granting independent prescriptive authority to NPs on mental-health-related mortality across all counties. However, counties that are underserved for mental health care services experience larger and more precisely estimated decreases in mortality when NPs can prescribe independently. As shown in Column (2), underserved counties experience a reduction of 1.88 suicides per quarter, or 7% of the mean, relative to counties with an adequate provision of mental health care resources when independent prescriptive authority is extended to NPs. Considering all mental-health-related deaths in Column (5), we see that underserved counties experience an overall reduction of 5.83 deaths per quarter, or 11% of the mean, when NPs can prescribe independently.
Looking to Columns (3) and (6), we see that individuals with low levels of education living in underserved counties see even greater reductions in suicides and mental-health-related deaths when NPs are granted independent prescriptive authority. Relative to individuals with a high school degree or less who reside in counties with a sufficient provision of mental health care resources, suicides and mental-health-related deaths are reduced by 10% and 12%, respectively, among individuals with low education who reside in underserved counties when independent prescriptive authority is extended to NPs (1.42 fewer suicides and 3.40 fewer mental-health-related deaths per county-quarter). Given that suicides and mental-health-related deaths are quite rare, there is likely more noise when we restrict our sample to individuals with low education. It is therefore no- The results presented in Table 4 are robust to a wide range of alternative specifications and variable definitions. While a county is considered underserved in Table 4 if it had fewer than one psychiatrist per 100,000 in 1990, and Table A.10 shows that the results are robust to the inclusion of state-level linear time trends.
Finally, the results presented in Table 4 are not driven by any one particular state. Tables A.11 and A.12 show that the estimates are robust to separately dropping each state that extended independent prescriptive authority to NPs over our sample period.
To verify the parallel trends assumption and to examine the time path of effects, we estimate event study specifications. In particular, we estimate a version of Equation (1)  Deaths cy = β 0 + α n n∈{−3,5}\−1 1 {y * s + n = y} + β 1 P op cy + β 2 X cy + γ c + γ y + cy (3) where y * s denotes the year of the law change extending independent prescriptive authority to NPs in state s, 1 {y * s + n = y} is an indicator denoting whether an observation is n years from the law change, and all other variables are defined as in Equation (1). Figure 6 presents the α coefficients and 95% confidence intervals from estimation of Equation (3) with mental-health-related mortality as the dependent variable. The left panel includes mental-health-related mortality among the entire population, while the right panel focusses only on individuals with low levels of education. In both panels, we see that there is no statistically significant difference in mental-health-related mortality in the years before independent prescriptive authority is granted to NPs. Following the law change, however, we see that mental-health-related mortality significantly decreases. These effects appear in the first year that is fully treated (the first year following the law change) and persist over time. 25 Weighting by population has a greater effect on the estimates when the outcome variable is in rates than in levels. We note, however, that the pattern of effects is very comparable across the unweighted specifications in Table A.8 (mortality levels) and Table A.9 (mortality rates per 100,000). Notes: Observations are at the county-year level and are population weighted. Standard errors are clustered by state. All regressions include county and year fixed effects; additional controls include total population (or subgroup population), population density, percent male, percent black, percent age 18 and under, percent age 65 and over, percent with a high school degree or less, a quadratic in median income, percent in poverty, percent unemployed, and the number of practicing psychiatrists and primary care physicians per 100,000. "Low educ." is defined as having a high school degree or less. The panel is unbalanced, years before and after the event time window are included using separate indicators, and zero denotes the year of the law change.

Self-Reported Mental Health
In Section 4.1 we found that allowing NPs to prescribe independently leads to significant reductions in mental-health-related mortality. As mortality is an extreme outcome, we next ask whether extending independent prescriptive authority to NPs leads to improvements in the mental health of individuals on a day-to-day basis. As described in Section 3.3.2, we consider both the number of days in the past month individuals report being in poor mental health as well as a binary variable which equals one if the individual reports having spent at least three weeks in poor mental health and zero otherwise. Letting P oor M ental Health isy denote either of these outcomes for individual i in state s in year y, we estimate the following equation: P oor M ental Health isy = β 0 + β 1 Indep. Rx sy + β 2 X isy + γ s + γ y + isy where Indep. Rx sy is an indicator denoting whether NPs had independent prescriptive authority in state s in year y; X isy is a vector of individual-level controls; 26 and γ s and γ y are state and year fixed effects, respectively. We define a state as having independent prescriptive authority in a given year if NPs had the legislative authority to prescribe independently at any point within the year; all of our results are robust to alternative timing assumptions. In all of our analyses in this section, standard errors are clustered by state and observations are weighted using the BRFSS sample weights. 27 As before, we examine whether extending independent prescriptive authority to NPs has differential impacts on the mental health of individuals living in states with an under-provision of mental health care services and among disadvantaged populations. Letting U nderserved s be an indicator that equals one if state s was less equipped for mental health care services in 1990 and zero otherwise, we estimate the following equation: P oor M ental Health isy = β 0 + β 1 Indep. Rx sy + β 2 Indep. Rx sy · U nderserved s +β 3 X isy + γ s + γ y + isy where all other variables are defined as in Equation (4). To look specifically at disadvantaged populations, we further estimate Equation (5) separately for different subpopulations of interest.
Results for individuals with low levels of education are provided with the main results below; refer to Table A.14 for results for other subpopulations.  (4) and (5). Looking first to Column (1), we see that respondents report having spent 0.17 fewer days in poor mental health when NPs are allowed to prescribe-a reduction of 5% relative to the mean. As with mental-health-related mortality, Column (2) demonstrates that the benefits are concentrated among respondents in areas that are less equipped for mental health care services. For respondents in underserved states, allowing NPs 26 Individual-level controls include indicators for age groups, education groups, income quintiles, and employment status and indicators denoting whether the respondent is male, white, black, Hispanic, married, and has health insurance. See Table A.16 for results from specifications that include alternative sets of controls. 27 Refer to Table A.17 for results using BRFSS sample weights that have been adjusted to account for the 2011 redesign of the survey (Simon et al., 2017). to prescribe independently leads to an overall reduction of 0.29 days in poor mental health, or an additional reduction of 0.17 fewer days in poor mental health relative to states with adequate mental health care resources. Finally, consistent with the mortality results, we find that the most disadvantaged populations-individuals with low levels of education living in areas underserved by psychiatrists-see the greatest reductions in poor mental health on a day-to-day basis. As seen in Column (3), independent prescriptive authority leads to a reduction of almost 10.5% in poor mental health days (0.41 days in poor mental health) relative to the average for low-education respondents who live in underserved states. Allowing NPs to independently prescribe further leads to reductions in the likelihood that individuals spend at least three weeks in poor mental health. As shown in Columns (4) and (5) of Table 5, independent prescriptive authority leads to a 5-8% reduction in this measure of more severe mental illness among the population as a whole. As before, the effects are larger among the most vulnerable populations: we see in Column (6) that individuals who are both living in under-served states and have low levels of education experience a 20% reduction in the probability of reporting 21+ days in poor mental health when NPs can prescribe independently.
Taken together, the results in Table 5 indicate that individuals with both minor and more severe mental illnesses benefit from the expansion of prescriptive authority. As with our mortality results, we provide a variety of additional analyses to probe the robustness of these results. While a state is considered underserved in Table 5 if the population-weighted average of binary, underserved categorizations across all counties in the state is less than the median across all states in 1990, Finally, as our estimates are limited in precision, we do not have the power to estimate individual event-time coefficients from an augmented version of Equation (4). Recall that the BRFSS is a survey of a subsample of the U.S. population, while the mortality and prescription data cover the universe of outcomes in the United States, so we have less statistical power when examining 28 We follow Simon et al. (2017) and reconstruct each individual's sample weight as the fraction of their assigned BRFSS sample weight over the sum of all individuals' sample weights for that year. 29 It is less of a concern to add state-level linear time trends to the mortality regressions, as the outcome there varies at the county-quarter level. Additionally, there is within-year variation in law changes and within-state variation in health resources. self-reported mental health.

Prescriptive Authority and the Number of Prescriptions
We found in the previous section that extending prescriptive authority to NPs leads to improvements in population mental health. Allowing NPs to prescribe medication without physician oversight could affect population mental health in two ways. The first and most direct route is that removing restrictions on NP's ability to prescribe may allow more people to access medication that affects their mental health. A second and less direct route is that granting NPs independent prescriptive authority may attract more individuals to the profession or redirect already licensed NPs to areas with more liberal scope of practice legislation, thereby expanding access to health care services more broadly (Xue et al., 2016). If increased access to health care leads to improved well-being, then extending prescriptive authority to NPs could improve population mental health independently of the number of prescriptions written.
Using prescription data from 2006-2014, we examine whether there is evidence of increased use of prescription medications when NPs are allowed to prescribe independently. Letting P rescriptions cy denote antidepressant, antipsychotic, or opioid prescriptions per capita in county c in year y, we estimate the following equation: P rescriptions cy = β 0 + β 1 Indep. Rx sy + β 2 X cy + γ c + γ y + cy (6) where all other variables are defined as in Equation (1) As before, we expect that extending prescriptive authority to NPs will lead to larger increases in prescriptions in counties with an under-provision of mental health care services and among populations who are traditionally disadvantaged. Letting U nderserved c be an indicator that equals one if county c was less equipped for mental health care services in 2000 and zero otherwise, we estimate the following equation: P rescriptions cy = β 0 + β 1 Indep. Rx sy + β 2 Indep. Rx sy · U nderserved c +β 3 X cy + γ c + γ y + cy (7) where all other variables are defined as in Equation (6). To look specifically at disadvantaged populations, we further estimate Equation (7) separately for prescriptions paid for by Medicaid. 30 Estimates from Equations (6) and (7) are provided in Table 6. As shown in Columns (1)   Compared to the mortality effects documented in Section 4.1, these estimates imply elasticities of mental-health-related deaths with respect to antidepressant and antipsychotic use among disadvantaged populations of 0.16 and 0.34, respectively. Although there is general consensus in the medical literature that antidepressant and antipsychotic use is associated with suicide prevention (Zalsman et al., 2016), the magnitude of these effects is not well understood: while a study comparing the relationship between antidepressant use and suicides in Sweden found an elasticity of less than 0.10 (Carlsten et al., 2001), work in the United States found an elasticity among adolescents of 3.5 (Olfson et al., 2003). We note that the intervention we study will increase the prescribing of different classes of medication simultaneously. If multiple types of drugs contribute to the reductions in mental-health-related mortality that we find, then our implied elasticities per drug class will be biased upwards.
The final two columns of Table 6 examine how independent prescriptive authority for NPs affects per capita opioid prescriptions. Columns (5) and (6)  As the scope of practice legislation we consider is not limited to psychotropic medications, it is not surprising that prescriptions for opioids also increase when NPs can prescribe independently. 31 Given that rates of opioid abuse in the United States have reached epidemic levels, however, any policy that increases the provision of opioids could be troublesome (SAMHSA, 2014). Despite the strong association between opioid prescriptions per capita and deaths involving drugs within counties over time (Schnell and Currie, 2018), we found in Section 4.1 that extending independent prescriptive authority to NPs leads to decreases in mental-health-related mortality, a measure that includes overdose deaths. Given the concurrent rise in opioid prescriptions, it is noteworthy that we find significant reductions in mental-health-related mortality when NPs are allowed to prescribe independently.
As in our mortality analysis, we use an event study specification to verify the parallel trends assumption and to examine the time path of effects. In particular, we estimate a version of Equation 1 {y * s + n = y} + β 1 X cy + γ c + γ y + cy (8) where all other variables are defined as in Equation (3). Figure 7 presents the α coefficients and 95% confidence intervals from estimation of Equation (8)  Notes: Observations are at the county-year level and are weighted by Medicaid eligibles. The dependent variables are county-level prescriptions paid for by Medicaid per Medicaid eligible. Standard errors are clustered by state. All regressions include county and year fixed effects; additional controls include population density, percent male, percent black, percent age 18 and under, percent age 65 and over, percent with a high school degree or less, a quadratic in median income, percent in poverty, percent unemployed, and the number of practicing psychiatrists and primary care physicians per 100,000. The panel is unbalanced, years before and after the event time window are included using separate indicators, and zero denotes the year of the law change.

Conclusion
Taken together, our results indicate that granting independent prescriptive authority to NPs is an important policy tool that can be used to improve population mental health. Policies that increase the number of providers who can prescribe medication may be particularly important in the United States, where the supply of physicians has not kept pace with rising demand for health care services. Although the discussions surrounding independent prescriptive authority for NPs focus primarily on shortages of primary care providers who can prescribe medication, we show that these laws also have important implications for mental health. In particular, states that grant independent prescriptive authority to NPs see improvements in self-reported mental health and reductions in the prevalence of mental-health-related deaths, including suicides. Improvements are greatest for individuals who live in areas that are underserved by psychiatrists and among populations who have been shown to have more difficulty accessing physician-provided care.
A back-of-the-envelope calculation suggests that 1,596 mental-health-related deaths were averted in underserved counties in 2014 alone by states allowing NPs to prescribe independently. If all states granted NPs independent prescriptive authority, the number of deaths averted yearly in underserved counties would rise to over 5,000. In addition, many more lives would be saved among disadvantaged populations who live in counties with adequate mental health resources but who have difficulty accessing physician-provided care. Furthermore, any estimate of deaths averted underestimates the full effect of the policy: for every person who commits suicide, there are over a thousand struggling with mental illness (Bureau of Health Workforce, 2016).
The effects of granting NPs independent prescriptive authority are similar in magnitude to the effects of gaining health insurance. For example, when Medicaid was extended to low-income adults using a lottery in Oregon, those who gained insurance reported spending 11% fewer days per month in poor mental health and were 10% less likely to screen positive for depression (Finkelstein et al., 2012). Furthermore, use of medication to treat depression increased by 33% relative to the control mean (Baicker et al., 2013). Our results demonstrate that the benefits of having health insurance will be mediated by the availability of providers who are able to fully treat patients.
It is noteworthy that we observe a consistent pattern of effects across two very different measures of mental health. Self-reported "days in poor mental health" allows us to examine whether populations suffering from mental illnesses of varying severity, including minor mental illness, notice improvements when NPs are allowed to prescribe. On the other hand, mental-healthrelated mortality allows us to examine whether populations suffering from very severe mental illnesses-that is, mental illness that may result in death-see improvements when independent prescriptive authority is extended beyond physicians. Even if extending prescriptive authority to NPs impacts one of these outcomes, it is not clear ex ante that prescriptive authority should also impact the other. In particular, since suicides and other deaths caused by poor mental health are relatively rare, it is possible that population mental health could improve without measurable effects on such extreme outcomes. The consistency of our results across these two categories of outcome measures indicates that prescriptive authority for NPs is associated with improved mental health across a spectrum of severity.
When independent prescriptive authority is extended to NPs, all NPs, not just those who specialize in mental health, have the statutory authority to prescribe. Just like physicians, however, some NPs specialize in psychiatric medicine. Psychiatric NPs with prescriptive authority traditionally provide psychotherapy in addition to psychotropic treatment, in contrast to the current movement among psychiatrists to only prescribe medications. It is therefore possible that the improvements in mental health that we observe are at least partly driven by an increase in "fullservice" mental health providers-that is, specialists that provide both psychotherapy and psychotropic treatment. However, it is also possible that our results are driven by an increase in the overall supply of general practitioners who can prescribe. It remains an open question whether extending independent prescriptive authority to NPs results in improved mental health because such laws increase the number of general health care providers who can prescribe psychotropic treatment or because they increase the number of providers who provide psychotherapy in conjunction with psychotropic treatment. Answering this question is a promising area for future research.

1990-2014
Month granted Notes: As outlined in Section 3.1, the law changes in this table are derived from information from the The Nurse Practitioner's "Annual Legislative Update," correspondences with state nursing boards, and readings of primary source legislation. We define "independent prescriptive authority" as the ability to prescribe medication without physician collaboration or supervision. More information on the relevant bills and legislative processes are available from the authors by request.  On average over our sample period, 53.4% of mental-health-related deaths were suicides; 31.9% were accidental poisonings; 6.6% were injuries of undetermined intent; and 5.8%, 1.4%, and 1.1% were accidental deaths involving drownings, firearms, and trains, respectively.  (1)   Notes: Observations are at the county-quarter level and are population weighted. Standard errors are clustered by state. All regressions include county, quarter, and year fixed effects; additional controls include total population (or subgroup population), population density, percent male, percent black, percent age 18 and under, percent age 65 and over, percent with a high school degree or less, a quadratic in median income, percent in poverty, percent unemployed, and the number of practicing psychiatrists and primary care physicians per 100,000. A county is "underserved" if it had fewer than one psychiatrist per 30,000 people in 1990. "Low educ." is defined as having a high school degree or less. Notes: Observations are at the county-quarter level and are population weighted. Standard errors are clustered by state. All regressions include county, quarter, and year fixed effects; additional controls include total population (or subgroup population), population density, percent male, percent black, percent age 18 and under, percent age 65 and over, percent with a high school degree or less, a quadratic in median income, percent in poverty, percent unemployed, and the number of practicing psychiatrists and primary care physicians per 100,000. A county is "underserved" if it had fewer than one psychiatrist per 30,000 people in a given year. "Low educ." is defined as having a high school degree or less.   additional controls include population density, percent male, percent black, percent age 18 and under, percent age 65 and over, percent with a high school degree or less, a quadratic in median income, percent in poverty, percent unemployed, and the number of practicing psychiatrists and primary care physicians per 100,000. A county is "underserved" if it had fewer than one psychiatrist per 30,000 people in 1990. "Low educ." is defined as having a high school degree or less. Notes: Observations are at the county-quarter level and are population weighted. Standard errors are clustered by state. All regressions include county, quarter, and year fixed effects and state-specific linear time trends; additional controls include total population (or subgroup population), population density, percent male, percent black, percent age 18 and under, percent age 65 and over, percent with a high school degree or less, a quadratic in median income, percent in poverty, percent unemployed, and the number of practicing psychiatrists and primary care physicians per 100,000. A county is "underserved" if it had fewer than one psychiatrist per 30,000 people in 1990. "Low educ." is defined as having a high school degree or less.  A.2 Self-Reported Mental Health   Notes: Observations are at the individual level with BRFSS sample weights. Standard errors are clustered by state. All regressions include state and year fixed effects; additional controls include indicators for age groups, education groups, income quintiles, and employment status and indicators denoting whether the respondent is male, white, black, Hispanic, married, and has health insurance. In Panel A, a state is "underserved" if the population-weighted average of binary, underserved categorizations across all counties in the state is less than the median across all states in a given year (binary). In Panel B, "underserved" is the population-weighted average of binary, underserved categorizations across all counties in the state in 1990 (continuous between zero and one). "Low educ." is defined as having a high school degree or less. Notes: Observations are at the individual level with BRFSS sample weights. Standard errors are clustered by state. All regressions include state and year fixed effects. Panel B also includes indicators for age groups and indicators denoting whether the respondent is male, white, black, or Hispanic. A state is "underserved" if the population-weighted average of binary, underserved categorizations across all counties in the state is less than the median across all states in 1990. "Low educ." is defined as having a high school degree or less.  Simon et al. (2017), we reconstruct each individual's sample weight as the fraction of their assigned BRFSS sample weight over the sum of all individuals' sample weights for that year. Standard errors are clustered by state. All regressions include state and year fixed effects; additional controls include indicators for age groups, education groups, income quintiles, and employment status and indicators denoting whether the respondent is male, white, black, Hispanic, married, and has health insurance. A state is "underserved" if the population-weighted average of binary, underserved categorizations across all counties in the state is less than the median across all states in 1990. "Low educ." is defined as having a high school degree or less. All regressions include state and year fixed effects and state-specific time trends; additional controls include indicators for age groups, education groups, income quintiles, and employment status and indicators denoting whether the respondent is male, white, black, Hispanic, married, and has health insurance. A state is "underserved" if the population-weighted average of binary, underserved categorizations across all counties in the state is less than the median across all states in 1990. "Low educ." is defined as having a high school degree or less.  Notes: Observations are at the individual level with BRFSS sample weights. Standard errors are clustered by state. All regressions include state and year fixed effects; additional controls include indicators for age groups, education groups, income quintiles, and employment status and indicators denoting whether the respondent is male, white, black, Hispanic, married, and has health insurance. A state is "underserved" if the population-weighted average of binary, underserved categorizations across all counties in the state is less than the median across all states in 1990.

Working Paper Series
A series of research studies on regional economic issues relating to the Seventh Federal Reserve District, and on financial and economic topics.
The Effects of the Massachusetts Health Reform on Financial Distress