State Legislators’ Support for Behavioral Health Parity Laws: The Influence of Mutable and Fixed Factors at Multiple Levels

Abstract

Policy Points.

• When communicating with state legislators, advocates for state behavioral health parity laws should emphasize that the laws do not increase insurance premiums.

• Legislators’ opinions about the impacts of state behavioral health parity laws and the effectiveness of behavioral health treatment have more influence on support for the laws than do their political party affiliation or state‐level contextual factors.

• Reducing legislators’ stigma toward people with mental illness could increase their support for state behavioral health parity laws

Context

Comprehensive state behavioral health parity legislation (C‐SBHPL) is an evidence‐based policy that improves access and adherence to behavioral health treatments. However, adoption of C‐SBHPL by state legislators is low. Little is known about how C‐SBHPL evidence might be most effectively disseminated to legislators or how legislators’ fixed characteristics (eg, ideology), mutable characteristics (eg, beliefs about the policy's impact), and state‐level contextual factors might influence their support for behavioral health policies. The purpose of our study is (1) to describe the associations between legislators’ fixed and mutable characteristics, state‐level contextual factors, and support for C‐SBHPL; and (2) to identify the mutable characteristics of legislators independently associated with C‐SBHPL support.

Methods

We conducted a multimodal (post mail, email, telephone) survey of US state legislators in 2017 (N = 475). The dependent variable was strong support for C‐SBHPL, and the independent variables included legislators’ fixed and mutable characteristics and state‐level contextual factors. We conducted multivariable, multilevel (legislator, state) logistic regression.

Findings

Thirty‐nine percent of the legislators strongly supported C‐SBHPL. After adjustment, the strongest predictors of C‐SBHPL support were beliefs that C‐SBHPL increases access to behavioral health treatments (aOR = 5.85; 95% CI = 2.41, 14.20) and does not increase insurance premiums (aOR = 2.70; 95% CI = 1.24, 5.90). Stigma toward people with mental illness was inversely associated with support (aOR = 0.86; 95% CI = 0.78, 0.95). After adjustment, ideology was the only fixed characteristic significantly associated with support for C‐SBHPL. State‐level contextual factors did not moderate associations between mutable characteristics and support for C‐SBHPL.

Conclusions

Legislators’ mutable characteristics are stronger predictors of C‐SBHPL support than are most of their fixed characteristics and all state‐level contextual factors, and thus should be targeted by dissemination efforts.

Laws passed by state legislators shape the context in which treatments for mental health and substance use disorder (ie, behavioral health) are sought and delivered.1, 2 Comprehensive state behavioral health parity legislation (C‐SBHPL) is one evidence‐based law that state legislators can adopt to improve access and adherence to behavioral health treatments.3 C‐SBHPL mandates that health insurance companies provide, without discrepancy, the same level of coverage for all behavioral health and medical/surgical benefits (eg, copayments, limits on visits). In 2012, the US Community Preventive Services Task Force found strong evidence supporting C‐SBHPL and recommended its widespread implementation. The recommendation was based on a review of 30 studies published between 1965 and 2011 that found that C‐SBHPL improved the utilization of behavioral health treatments (eg, receiving the proper treatment for an appropriate duration) and reduced out‐of‐pocket spending for behavioral health services.3 An accompanying economic review of 12 studies published between 1950 and 2011 found that C‐SBHPL did not meaningfully increase insurance premiums and, in some cases, was associated with lower premiums because health insurers used managed‐care practices to restrict utilization and thus contain costs.4

While C‐SBHPL was arguably more critical before the federal Mental Health Parity and Addiction Equity Act of 2008 (MHPAEA) and the Patient Protection and Affordable Care Act (ACA)—which extended the MHPAEA's reach by requiring that plans operating on state insurance exchanges and Medicaid benchmark plans include behavioral health coverage as an “essential health benefit”5—C‐SBHPL remains important because it can supplement federal parity laws and improve compliance.6, 7, 8 An analysis by the Kennedy Forum in 2018 found that most states lack strong laws to support the implementation of the MHPAEA and identified two ways in which C‐SBHPL can supplement federal laws.8 First, C‐SBHPL can require insurers and Medicaid managed‐care organizations (MCOs) to cover all conditions in the Diagnostic and Statistical Manual of Mental Disorders, as well as specific treatments. These requirements are significant because neither the MHPAEA nor the ACA specifies the types of behavioral health conditions or treatments that insurers or Medicaid MCOs are required to cover.9 As a result, many insurers and Medicaid MCOs do not cover some conditions (eg, autism)10 and evidence‐based treatments (eg, medication‐assisted treatment for opioid use disorder).11, 12, 13, 14 Second, C‐SBHPL can direct state agencies (eg, insurance departments) to perform market conduct examinations and other activities to monitor the MHPAEA's compliance. These directives are important because state agencies are largely responsible for enforcing the MHPAEA and evidence suggests that compliance is weak in many states.11, 12, 13, 14 For example, in March 2019 a federal judge in northern California ruled that the practices of United Behavioral Health of the UnitedHealth Group—one of the largest managed‐care companies in the United States—were not in compliance with the MPHAEA.15 In addition, efforts to repeal the ACA and dismantle essential health benefit requirements threaten the MHPAEA's reach and would further elevate the importance of C‐SBHPL.16

Even though C‐SBHPL is an effective and cost‐neutral policy that most US adults support,17 only 19 states have implemented it,18 owing to widespread political opposition. As with federal behavioral health parity laws,9, 19 C‐SBHPL proposals have been met with strong opposition from the health insurance industry.20 Industry arguments against C‐SBHPL have revolved around claims that these laws would result in the excessive use of behavioral health services, which would lead to higher premiums and possibly the layoffs of insurance company employees.20, 21 While empirical evidence does not support these claims,4 an analysis between 1980 and 2005 of state‐level factors associated with the passage of C‐SBHPL found that high state unemployment rates, which indicate economic pressure, were associated with a lower probability of passing C‐SBHPL.21

The Potential of Policy Dissemination Research

The uptake of C‐SBHPL could be increased by investigating how evidence regarding the law is disseminated to state legislators, the decision makers with the exclusive authority to adopt it. Despite the numerous barriers to evidence‐informed policymaking,22, 23, 24, 25, 26 research on public opinion suggests that support for evidence‐based policies (eg, C‐SBHPL) can be cultivated by disseminating evidence in ways that are compelling and tailored to individual characteristics and contexts.27, 28, 29, 30, 31, 32, 33, 34, 35, 36 Reports by the US National Academies have highlighted the importance of effectively disseminating evidence to state legislators,37, 38, 39, 40 although we have little empirical guidance regarding the design of these dissemination strategies.41

This knowledge gap reflects the fact that health policy dissemination research—defined as the study of the targeted distribution of health evidence to policymakers42—is underdeveloped relative to other areas in the field of dissemination and implementation (D&I) health. A review of D&I research funded by the National Institutes of Health between 2007 and 2014 found that only 8.2% of projects were focused on policy contexts, just 2.7% of which were dissemination studies.43 Policy dissemination research has also lacked the conceptual sophistication of other areas of D&I research, such as studies of clinical and health care organizations.

Conceptual shortcomings are particularly pronounced in at least two areas. First, D&I studies typically account for outer‐contextual factors, such as features of the sociopolitical environment, because they influence how interventions are perceived and whether they are adopted.44, 45 Research on policy dissemination, however, has almost exclusively focused on associations between legislators’ individual characteristics and their support for specific health policies46, 47, 48, 49, 50, 51, 52, 53 and has not examined how outer‐contextual factors might affect the support of such policies. Identifying those features of outer‐context associated with legislators’ support for evidence‐based policies is important because it can inform how dissemination activities are tailored for legislators in different sociopolitical environments.

Second, studies of associations between legislators’ characteristics and their support for policies have not differentiated between legislators’ fixed and mutable characteristics. Fixed characteristics are those that are not likely to be changed by the dissemination of information, whereas mutable characteristics are those that could be changed by the dissemination of information. If an association is found between a fixed characteristic (eg, conservative ideology) and policy support, this finding can inform how dissemination materials are tailored so that they might persuade legislators with that characteristic. For example, evidence summaries could be framed for conservative legislators showing that the evidence resonates with conservative values like authority and sanctity.54 Alternatively, if an association is detected between a mutable characteristic (eg, beliefs about a policy's economic impacts) and policy support, this finding can be used to select the information emphasized in dissemination materials to change the characteristic. For example, evidence summaries that stress cost‐effectiveness might alter legislators’ beliefs about cost‐effectiveness and subsequently augment their support of such policies.

Study Purpose

The objectives of our study were to conceptually advance the field of policy dissemination research in these two areas and, more narrowly, to generate knowledge that can improve how evidence regarding C‐SBHPL is disseminated to state legislators. Our study's aims were (1) to describe associations between legislators’ fixed characteristics, mutable characteristics, state‐level contextual factors, and their support of C‐SBHPL; and (2) to identify legislators’ mutable characteristics that are independently associated with their support of C‐SBHPL.

Methods

Study Design and Data

Between March and September 2017, we conducted a multimodal (post mail, email, telephone) survey of US state legislators, using the National Conference of State Legislatures’ (NCSL) contact database for the sample frame.55 We randomly sampled up to 60 legislators from every state who were in office as of January 15, 2017, excluding legislators who had newly entered office on January 1, 2017, because they had limited legislative experience at the time of the survey. Sixty legislators were sampled in 33 states, 59 legislators in nine states, between 58 and 55 legislators in five states, and between 39 and 31 legislators in three states. The variation in the number of legislators sampled in each state reflected differences in the number of seats in each state legislature and the proportion of seats occupied by legislators who had newly entered office in January 2017. We also removed legislators for whom valid email contact information was unavailable and those who had left office before March 7, 2017, when our data collection began. The result was a sample frame of 2,902 legislators whom we contacted to complete the survey.

We post‐mailed each legislator two paper versions of the survey with self‐addressed postage‐paid return envelopes, post‐mailed one invitation to complete a web‐based version of the survey, emailed 10 invitations to complete the web‐based survey, and made as many as 15 phone calls to complete the survey over the telephone. In total, a legislator who did not complete the survey was contacted 29 times. Our recruitment materials specified that only the legislator could complete the survey, not their staff. SSRS, a survey research firm, collected the data, and we obtained approval of the Institutional Review Board.

The survey was completed by 475 legislators, which is 6.4% of the total population of 7,383 state legislators in the United States. The response rate was 16.4%, which is considered reasonable for state legislators56 and was higher than the response rates of recent surveys of state legislators.46, 57, 58 Respondents were significantly more likely than nonrespondents to be female (32.6% vs 23.0%, χ² = 19.73, p < .001), from the Midwest US Census Region (30.5% vs 22.5%, χ² = 14.27, p < .001), and Democrats (48.8% vs 42.4%, χ² = 10.19, p = .001). We calculated the nonresponse weights and applied them to adjust for these differences using a sample poststratification approach in which weighting classes were based on the full sample frame.59 The weights accounted for differences between the respondents and nonrespondents in gender, geographic region, and political party and have been used in previous analyses of the survey data set.60, 61

Measures

The survey instrument consisted primarily of items adapted from public opinion surveys about behavioral health and legislator surveys about the use of research in policymaking (complete details can be found in the published study protocol).55 The survey instrument was vetted by the NCSL. We also conducted cognitive interviews with five former state legislators to inform our final revisions before fielding the interviews.62

Dependent Variable

The dependent variable was support for C‐SBHPL, defined as “state laws that require health insurance companies to provide the same level of coverage for all mental health/substance use disorder and physical health benefits (eg, identical deductibles, copayments, visit limits) with no discrepancy.”18, 63 Based on this definition, legislators rated the extent to which they supported C‐SBHPL on a five‐point scale (1 = strongly oppose, 5 = strongly support). This variable was dichotomized as “strongly support” (yes/no) because strong support is most proximal to a legislator adopting C‐SBHPL (ie, introducing or voting in favor of a C‐SBHPL bill).

Independent Variables

There were three categories of independent variables: legislators’ fixed characteristics, their mutable characteristics, and state‐level contextual factors. We based our selection of the characteristic variables on McGinty and colleagues’ review of communication strategies to generate support for behavioral health policies64 and on Watson and Corrigan's framework of policymakers’ behavioral health decision making.65 Our selection of state‐level variables was based on Hernandez and Uggen's study of state‐level factors associated with the adoption of C‐SBHPL22 and on literature about the politics of behavioral health policymaking.9, 66, 67, 68

Legislators’ Fixed Characteristics

The NCSL's contact database provided information on legislators’ gender (male/female) and political party affiliation (Democrat/Republican/other). The legislators’ highest level of education was obtained by self‐report at the end of the survey. Their ideology was assessed by two separate items asking the legislators to indicate “how [they] usually think of [themselves] when it comes to …,” “social” and “fiscal’ issues” on a seven‐point scale (1 = extremely liberal, 7 = extremely conservative). These items were adapted from American National Election Studies’ questionnaires69 that have been used in prior surveys of state legislators.70 For each legislator, we summed the social and fiscal ideology scores to create an aggregate score (range = 2 to 14), which we transformed into a categorical variable in which scores of 2 to 6 were coded as liberal, 7 to 9 as moderate, and 10 to 14 as conservative.

Legislators’ Mutable Characteristics

Beliefs about C‐SBHPL impact were assessed by two separate items in which legislators rated the extent to which they believed that C‐SBHPL “increases access to mental health/substance use disorder services” and “increases health insurance premium costs” on five‐point scales (1 = strongly disagree, 5 = strongly agree). The items were developed for the survey and dichotomized as “agree/strongly agree” (yes/no) based on strong evidence that C‐SBHPL does increase access to behavioral health services via expanded insurance coverage and does not increase insurance premiums.3, 4

Beliefs about the effectiveness of treatments for mental health and substance use disorders were assessed by two separate items in which legislators rated the extent to which they believed that “mental health treatments can help people with mental illness lead normal lives” and “substance disorder treatments can help people with a substance use disorder recover” on five‐point scales (1 = strongly disagree, 5 = strongly agree). We adapted the wording of a question about the effectiveness of mental health treatments that has been used with the general public71 in order to assess the legislators’ beliefs about the effectiveness of substance use disorder treatments. These variables were dichotomized as “strongly agree” (yes/no) because the evidence unequivocally supports that behavioral health treatments can be effective.

Stigma toward people with mental illness was assessed by four items previously used to characterize the stigma of mental illness among the general public.18 Two items assessed attitudes about the dangerousness of people with mental illness (eg, the belief that people with serious mental illness are far more dangerous than the general public), and two items assessed preferences for social distance from people with mental illness (eg, willingness to work closely with someone with a serious mental illness). Cronbach's alpha was .80 for the four items, similar to that when the items were used with the general public (Cronbach's alpha = .77).18 Each item was scored on a five‐point scale, and the two social distance items were reverse coded so that higher scores corresponded to more stigma. We then summed the scores of the four items to create a composite mental illness stigma score for each legislator (range = 0 to 16).

State‐Level Contextual Factors

A recent mass shooting in each legislator's state was assessed because these events often generate public support for policies to address mental illness.68, 72, 73, 74 This information was obtained from the Mother Jones’ Guide to Mass Shootings in America Database, which tracks mass shootings via media reports (1982 to present).75 We chose the Mother Jones database among numerous mass shooting databases (eg, the Stanford Mass Shooting Database) because it had the most up‐to‐date information and used the narrowest definition of a mass shooting. This definition requires that the shooting resulted in ≥ 4 deaths (excluding the perpetrator), was carried out in a public place, and was not linked primarily to gang violence. We calculated the number of days since a mass shooting had occurred in each legislator's state using the date of survey completion as the index date and then classified the legislators according to whether or not a shooting had occurred in the past 1,095 days (ie, 3 years). We chose three years for a cut‐off point assuming that to be recent enough that the event could still influence a legislator's opinions. Fifteen states had not had a mass shooting since 1981, so legislators from these states were classified as > 1,095 days.

We obtained the opioid overdose death rate per 100,000 population and the annual percentage change in opioid overdose death rate in each legislator's state from the Kaiser Family Foundation,76 which produced age‐adjusted 2014 and 2015 estimates using data from the National Vital Statistics System Multiple Cause‐of‐Death Mortality Files.

We determined the C‐SBHPL implementation status in each legislator's state using information from the NCSL.19 Each legislator's state was classified as C‐SBHPL implemented, limited SBHPL implemented, or no SBHPL implemented.

We based the partisan composition of each state legislature using information from the NCSL.77 Each legislator's state was classified as Republican if the party had majority control of both chambers of the legislature, Democratic if the party had majority control of both chambers, and split if one party controlled each chamber or if the legislature was unicameral.

We used the unemployment rate in each legislator's state in the most recent full month before the date of the survey's completion because it is an indicator of economic pressure against C‐SBHPL.22 This information was obtained from seasonally adjusted unemployment estimates provided by the US Bureau of Labor Statistics.78

We assessed the proportion of employer‐sponsored health benefit plans fully self‐insured in each legislator's state, weighted for the number of plan participants, in 2015 because these plans are exempt from compliance with C‐SBHPL under provisions of the Employee Retirement Income Security Act of 1974 (ERISA).79 We obtained this information from Form 5500 filing data compiled for a US Department of Labor annual report.80

Analysis

We created a de‐identified data set of the legislators’ survey responses, individual characteristics, and state‐level contextual factors. We retained the data on the legislators in those states that had implemented C‐SBHPL because these laws were assumed to have typically passed before the legislators took office—an assumption supported by a post hoc analysis revealing that legislators in states with C‐SBHPL or limited‐SBHPL were no more likely to have heard of C‐SBHPL than were legislators in states that had not implemented SBHPL (56.9% vs 51.6%, p = .357). We used descriptive statistics to characterize the sample and bivariate analyses (ie, χ² and t‐tests) to compare differences between legislators who did and did not strongly support C‐SBHPL. We used multilevel (legislators nested in states) random‐intercept binary logistic regression models to estimate the influence of mutable characteristics on strong support for C‐SBHPL, adjusting for fixed characteristics and state‐level contextual factors. The multilevel models accounted for the clustering of strong support for C‐SBHPL among legislators in the same state (intraclass correlation coefficient = 0.28). We used likelihood‐ratio tests to determine whether a model's fit was significantly improved by the addition of mutable characteristics to a model that included only fixed characteristics, and then whether the model's fit was improved by the subsequent addition of state‐level contextual factors. This produced adjusted odds ratios (aORs) and 95% confidence intervals (CIs).

Results

At least one legislator completed the survey in all 50 states (number of legislators per state: range = 1–29, mean = 9.5, mode = 9). Thirty‐nine percent of the legislators strongly supported C‐SBHPL (Table 1). Only 16.1% believed that C‐SBHPL did not increase insurance premiums, but 72.8% believed that it did increase access to behavioral health treatments. Approximately half the legislators believed that treatments for mental health (54.1%) and substance‐use disorder (49.1%) could be effective, and the mean stigma score was 5.6 (standard deviation [SD] = 3.4). Forty‐five percent of the legislators were in states that had implemented C‐SBHPL; 78.1% were in Republican‐controlled legislatures; and 21.9% were in states that had experienced a mass shooting in the past three years.

Table 1.

Characteristics of Respondents to State Legislator Survey and of State‐Level Contextual Factors, 2017 (N = 475)

	%
Legislators’ Fixed Characteristics
Gender
Female	24.6
Male	75.4
Highest level of education
College degree or less	51.3
Postgraduate degree or more	48.7
Political party
Democrat	43.5
Republican	54.1
Other	2.4
Ideology
Liberal	28.0
Moderate	23.7
Conservative	48.4
Legislators’ Mutable Characteristics
Strongly supports C‐SBHPL
Yes	38.7
No	61.3
Beliefs about impact of C‐SBHPL
Believes that C‐SBHPL increases access to behavioral health treatments
Yes	72.8
No	27.2
Believes that C‐SBHPL does not increase insurance premiums
Yes	16.1
No	84.0
Beliefs about treatment effectiveness
Strongly agrees that mental health treatments can be effective
Yes	54.1
No	45.9
Strongly agrees that substance use disorder treatments can be effective
Yes	49.1
No	50.9
Stigma toward people with mental illness
Stigma score (mean, SD)a	5.6 (3.4)
State‐Level Contextual Factors
Days since mass shooting
> 1,095 days	78.1
≤ 1,095 days	21.9
Opioid overdose death rate
2015 opioid overdose death rate per 100,000 state residents (mean, SD)	11.9 (7.5)
2014/2015 percent change in opioid overdose death rate per 100,000 state residents (mean, SD)	9.3 (14.7)
C‐SBHPL implementation status
C‐SBHPL implemented	43.06
Limited SBHPL implemented	37.02
No SBHPL implemented	19.93
Partisan composition of legislature
Democrat	22.7
Republican	70.2
Split/unicameral	7.1
Unemployment
State unemployment rate (mean, SD)	3.9 (0.9)
Magnitude of C‐SBHPL exemption for employer self‐insurance
Proportion of employer‐sponsored health benefit plans fully self‐insured in state, participant weighted (mean, SD)	21.5 (9.6)

Abbreviation: C‐SBHPL = comprehensive state behavioral health parity legislation. Data are weighted between respondents and nonrespondents for differences in gender, geographic region, and political party.

^aRange = 0 to 14.

Associations Between Legislators’ Fixed Characteristics and C‐SBHPL Support

The proportion of legislators who strongly supported C‐SBHPL was significantly higher among women than men (54.1% vs 37.7%, p < .0001), Democrats than Republicans (66.3% vs 15.4%, p < .0001), and those who were ideologically liberal (76.7%) compared with those who were moderate (48.7%) or conservative (11.3%) (p < .0001) (Table 2). Highest level of education was not significantly associated with support for C‐SBHPL.

Table 2.

Unadjusted Associations Between Characteristics of Legislators, State‐Level Contextual Factors, and Strong Support for Comprehensive State Behavioral Health Parity Laws, 2017 (N = 475)

	Legislator Strongly Supports C‐SBHPL (%)	Legislator Does Not Strongly Support C‐SBHPL (%)	P‐Valuea
Legislator's Fixed Characteristics
Gender
Female	54.1	45.9	< .0001
Male	33.7	66.3
Highest level of education
College degree or less	36.6	63.4	.35
Postgraduate degree or more	40.8	59.2
Political party
Democrat	66.3	33.7	< .0001
Republican	15.4	84.6
Other	62.7	37.3
Ideology
Liberal	76.7	23.3	< .0001
Moderate	48.7	51.3
Conservative	11.3	88.7
Legislator's Mutable Characteristics
Beliefs about C‐SBHPL impact
Believes that C‐SBHPL increases access to behavioral health treatments
Yes	49.7	50.3	< .0001
No	8.4	91.6
Believes that C‐SBHPL does not increase insurance premiums
Yes	76.1	24.0	< .0001
No	31.5	68.5
Beliefs about treatment effectiveness
Strongly agrees that mental health treatments can be effective
Yes	57.9	42.1	< .0001
No	16.8	83.2
Strongly agrees that substance use disorder treatments can be effective
Yes	59.6	40.4	< .0001
No	19.0	81.1
Stigma toward people with mental illness
Stigma score (mean, SD)b	3.8 (3.0)	6.8 (3.2)	< .0001
State‐Level Contextual Factors
Days since mass shooting
> 1,095 days	35.7	64.3	.0107
≤ 1,095 days	49.6	50.4
Opioid overdose death rate
2015 opioid overdose death rate per 100,000 state residents (mean, SD)	12.3 (6.2)	11.7 (8.3)	.3094
2014/2015% change in opioid overdose death rate per 100,000 state residents (mean, SD)	11.1 (13.7)	8.0 (15.3)	.03
C‐SBHPL implementation status
C‐SBHPL implemented	34.2	65.8	.03
Limited SBHPL implemented	46.7	53.3
No SBHPL implemented	34.0	66.0
Partisan composition of legislature
Democrat	60.3	39.7	<.0001
Republican	30.5	69.5
Split/unicameral	52.7	47.3
Unemployment
State unemployment rate (mean, SD)	4.0 (0.9)	3.9 (0.9)	.37
Magnitude of C‐SBHPL exemption for employer self‐insurance
Proportion of employer‐sponsored health benefit plans fully self‐insured in state, participant weighted (mean, SD)	20.3 (7.9)	22.2 (10.7)	.02

Abbreviation: C‐SBHPL = comprehensive state behavioral health parity legislation.

^aχ² test or t‐test. Data are weighted between respondents and nonrespondents for differences in gender, geographic region, and political party.

^bRange = 0 to 14 among legislators who do not strongly support C‐SBHPL and 0 to 12 among legislators who do strongly support C‐SBHPL.

Associations Between Legislators’ Mutable Characteristics and C‐SBHPL Support

Beliefs about the impact of C‐SBHPL were associated with support for C‐SBHPL (Table 2). The proportion of legislators who strongly supported C‐SBHPL was substantially higher among those who believed that C‐SBHPL increased access to treatments for behavioral health and those who believed that C‐SBHPL did not increase insurance premiums, compared with those who did not believe that the laws had these impacts (49.7% vs 8.4% and 76.1% vs 31.5%, respectively, both p < .0001). The legislators were also more likely to strongly support C‐SBHPL if they believed that treatments for behavioral health could be effective. For example, 59.6% of legislators who strongly agreed that substance‐use disorder treatments could be effective strongly supported C‐SBHPL, compared with only 19.0% of those who did not strongly agree (p < .0001). Legislators who strongly supported C‐SBHPL also carried less stigma toward people with mental illness than legislators who did not strongly support C‐SBHPL (mean stigma score = 3.8 vs 6.8, t = 10.22, p < .0001).

Associations Between State‐Level Contextual Factors and Support for C‐SBHPL

A recent mass shooting was associated with support for C‐SBHPL, with 49.6% of the legislators in those states having experienced a mass shooting in the past three years strongly supporting C‐SBHPL, compared with 35.7% of the legislators in states that did not experience a shooting (p = .011) (Table 2). The 2015 opioid overdose death rate was not associated with legislator support for C‐SBHPL. The percent change in opioid overdose death rate between 2014 and 2015 was, however, slightly higher in those states in which legislators supported C‐SBHPL than in states in which legislators did not support C‐SBHPL (average percent change = +11.1% vs +8.0%, p = .026). The legislators in states that had implemented C‐SBHPL were not more likely to strongly support C‐SBHPL than were those in states that had not implemented any behavioral health parity law (34.2% vs 34.0%). Strong support for C‐SBHPL was higher among legislators in states with Democrat‐controlled rather than Republican‐controlled legislatures (60.3% vs 30.5%, p < .0001). The mean proportion of employer‐sponsored health benefit plans that were fully self‐insured was slightly lower in states where legislators strongly supported C‐SBHPL than in states where the legislators did not strongly support C‐SBHPL (20.3 vs 22.2, p = .01). State unemployment rate was not associated with legislators’ support for C‐SBHPL.

Adjusted Associations Among Legislators’ Characteristics, State‐Level Contextual Factors, and C‐SBHPL Support

Legislators’ mutable characteristics related to beliefs about the impact of C‐SBHPL, the effectiveness of behavioral health treatments, and the stigma of mental illness remained significantly associated with strong support for C‐SBHPL after adjusting for fixed characteristics (Table 3) and state‐level contextual factors (Table 4). After adjusting for the fixed characteristics of gender, political party, and ideology (Table 3, Model 3), the belief that C‐SBHPL increases access to behavioral health treatments was independently associated with six‐fold higher odds of strong support for C‐SBHPL (aOR = 6.29; 95% CI = 2.70, 14.67). The belief that C‐SBHPL does not increase insurance premiums was associated with three‐fold higher odds of strong support (aOR = 3.11; 95% CI = 1.48, 6.55). The strength of these associations changed only minimally in the full multilevel model that accounted for state‐level clustering and adjusted for state C‐SBHPL status, legislature's partisan composition, 2014/2015 percent change in opioid overdose death rate, a mass shooting occurring in the past three years, and the state's proportion of fully self‐insured, employer‐sponsored health benefit plans (Table 4, Model 6).

Table 3.

Adjusted Associations Between Legislators’ Mutable Characteristics and Strong Support for Comprehensive State Behavioral Health Parity Laws, Controlling for Legislators’ Fixed Characteristics, 2017 (N = 475)

Independent Variable: Legislators’ Mutable Characteristics	Model 1a aOR (95% CI)	Model 2b aOR (95% CI)	Model 3c aOR (95% CI)
Beliefs about impact of C‐SBHPL
Believes that C‐SBHPL increases access to behavioral health treatments	7.89 (3.53, 17.63)	7.07 (3.06, 16.29)	6.29 (2.70, 14.67)
Believes that C‐SBHPL does not increase insurance premiums	4.65 (2.26, 9.57)	3.25 (1.55, 6.83)	3.11 (1.48, 6.55)
Beliefs about treatment effectiveness
Strongly agrees that mental health treatments can be effective	2.08 (1.15, 3.75)	2.05 (1.09, 3.86)	1.85 (0.95, 3.58)
Strongly agrees that substance use disorder treatments can be effective	2.52 (1.43, 4.43)	2.13 (1.17, 3.89)	2.17 (1.17, 4.02)
Stigma toward people with mental illness
Stigma score	0.81 (0.74, 0.88)	0.86 (0.78, 0.93)	0.87 (0.80, 0.96)

Abbreviations: aOR = adjusted odds ratio, CI = confidence interval, and C‐SBHPL = comprehensive state behavioral health parity legislation. Binary logistic regression models. All models are mutually adjusted (ie, controlling for legislators’ other mutable characteristics that are not serving as the primary independent variable). Data are weighted between respondents and nonrespondents for differences in gender, geographic region, and political party.

^aModel 1 controls for gender and legislators’ other mutable characteristics

^bModel 2 controls for variables included in model 1 plus political party

^cModel 3 controls for variables included in model 2 plus ideology

Table 4.

Adjusted Associations Between Legislators’ Mutable Characteristics and Strong Support for Comprehensive State Behavioral Health Parity Laws, Controlling for Legislators’ Fixed Characteristics and State‐Level Contextual Factors, 2017 (N = 475)

Independent Variable: Legislator's Mutable Opinions	State‐Level Model 1a aOR (95% CI)	State‐Level Model 2b aOR (95% CI)	State‐Level Model 3c aOR (95% CI)	State‐Level Model 4d aOR (95% CI)	State‐Level Model 5e aOR (95% CI)	State‐Level Model 6f aOR (95% CI)
Beliefs about impact of C‐SBHPL
Believes that C‐SBHPL increases access to behavioral health treatments	6.47 (2.70, 15.50)	6.38 (2.66, 15.29)	6.27 (2.60, 15.11)	6.24 (2.58, 15.09)	6.25 (2.58, 15.13)	5.85 (2.41, 14.20)
Believes that C‐SBHPL does not increase insurance premiums	3.20 (1.48, 6.92)	3.18 (1.47, 6.88)	2.97 (1.36, 6.46)	2.85 (1.30, 6.23)	2.84 (1.30, 6.21)	2.70 (1.24, 5.90)
Beliefs about treatment effectiveness
Strongly agrees that mental health treatments can be effective	1.82 (0.91, 3.65)	1.82 (0.91, 3.65)	1.75 (0.86, 3.55)	1.68 (0.82, 3.44)	1.69 (0.83, 3.46)	1.69 (0.83, 3.41)
Strongly agrees that substance use disorder treatments can be effective	2.34 (1.21, 4.53)	2.35 (1.21, 4.53)	2.39 (1.23, 4.64)	2.43 (1.25, 4.75)	2.41 (1.23, 4.72)	2.41 (1.24, 4.66)
Stigma toward people with mental illness
Stigma score	0.87 (0.79, 0.96)	0.87 (0.79, 0.96)	0.87 (0.79, 0.96)	0.86 (0.78, 0.95)	0.86 (0.78, 0.95)	0.86 (0.78, 0.95)

Abbreviations: aOR = adjusted odds ratio, CI = confidence interval, and C‐SBHPL = comprehensive state behavioral health parity legislation. Binary multilevel (legislator, state) logistic regression models. All models are mutually adjusted (ie, controlling for legislators’ other mutable characteristics that are not serving as the primary independent variable). Data are weighted between respondents and nonrespondents for differences in gender, geographic region, and political party.

^aModel 1 controls for state‐level clustering (with state‐level random intercept), legislators’ fixed characteristics (gender, party, political party, and ideology), and legislators’ other mutable characteristics

^bModel 2 controls for variables in model 1 plus implementation status of C‐SBHPL

^cModel 3 controls for variables in model 2 plus partisan composition of legislature

^dModel 4 controls for variables in model 3 plus mass shooting < last 1,095 days

^eModel 5 controls for variables in model 4 plus percent change in opioid overdose death rate from prior year

^fModel 6 controls for variables in model 5 plus proportion of employer‐sponsored health benefit plans fully self‐insured in state

The belief that treatments for substance‐use disorders could be effective remained significantly associated with strong support for C‐SBHPL in the fully adjusted multilevel model (aOR = 2.41; 95% CI = 1.24, 4.66; Table 4, Model 6). The association between the belief that mental health treatments could be effective and C‐SBHPL support was attenuated and lost statistical significance. A low score for mental illness stigma remained significantly associated with C‐SBHPL support. A mental illness stigma score that was 1 point higher, on a 14‐point scale, was associated with 14% lower odds of strong support for C‐SBHPL (aOR = 0.86; 95% CI = 0.78, 0.95).

Ideology was the only fixed characteristic that was significantly associated with C‐SBHPL in the fully adjusted multilevel model (Appendix A). Compared with ideologically conservative legislators, the odds of strong support for C‐SBHPL were two times higher among moderate legislators (aOR = 2.14; 95% CI = 0.95, 4.84) and four times higher among liberal legislators (aOR = 4.25; 95% CI = 1.55, 11.66). Likelihood‐ratio tests showed that the addition of mutable characteristics significantly improved the model's fit, compared with a model containing only the fixed characteristics of gender, political party, and ideology (p < .0001). In contrast, adding state‐level contextual factor variables did not significantly improve the model's fit above and beyond mutable and fixed characteristics (p = .08).

Discussion

We found that legislators’ mutable characteristics were strongly associated with strong support for C‐SBHPL and that these associations remained strong after adjusting for fixed characteristics and state‐level contextual factors. Although both fixed legislator characteristics and state‐level contextual factors were associated with C‐SBHPL support in bivariate analyses, almost none of these associations were significant in regression models adjusting for mutable characteristics. A legislator's ideology was the only fixed characteristic that remained significantly associated with C‐SBHPL support. These findings suggest that the dissemination of evidence concerning C‐SBHPL, and potentially other evidence‐based behavioral health policies, should target mutable characteristics, as they might precede changes in policy support and, ultimately, policy adoption. Tailoring materials to be disseminated to ideologically conservative, moderate, and liberal legislators might also be beneficial.61

The belief that C‐SBHPL increases access to treatments for behavioral health was the strongest predictor of strong support for C‐SBHPL. This is consistent with D&I studies in clinical and organizational settings, which suggest that perceptions of the effectiveness of interventions influence decisions about adopting interventions.44, 45 This finding also implies that legislators who support C‐SBHPL perceive behavioral health treatments as possibly being effective. However, beliefs about the effectiveness of behavioral treatments were weaker predictors of C‐SBHPL support than was the belief that C‐SBHPL increases access to behavioral health treatments. This suggests that information about the proximal impacts of C‐SBHPL (ie, increasing access to behavioral health treatments via expanded insurance coverage) might have more influence on support than does information about distal policy impacts (ie, people being able to lead normal lives because they can access effective behavioral health treatments).

This finding is also consistent with a recent communications experiment,81 which found that people randomized to read narratives about system‐level barriers to mental health treatment were more willing to pay additional taxes for mental health system improvements than were people randomized to read narratives about effective treatment and recovery. Taken together, these findings suggest that dissemination materials should emphasize how C‐SBHPL can address the system‐level problem of insurance companies restricting access to behavioral health treatments instead of emphasizing that behavioral health treatments can be an individual‐level solution to behavioral health problems.

The belief that C‐SBHPL does not increase insurance premiums was also a strong predictor of C‐SBHPL support. Although increases in insurance costs would be borne by the legislators’ constituents and not state government, this finding is consistent with previous research showing that considerations related to the budget's impact have substantial influence on state legislators.82, 83, 84 Evidence showing that behavioral health parity laws do not increase insurance costs also played a major role in passing the MHPAEA at the federal level.9 Given that a systematic review of 12 studies concluded that C‐SBHPL does not increase insurance costs,4 this information should be a focus of dissemination materials about C‐SBHPL. The finding that only 16.1% of legislators believed that C‐SBHPL did not increase insurance premiums indicates that this evidence has failed to reach most legislators. Better dissemination of this information thus might help cultivate support for C‐SBHPL among Republican and ideologically conservative legislators, because research suggests that they value economic evaluation data more than their peers.60, 61, 85

Stigma toward people with mental illness was prevalent among legislators and inversely associated with support for C‐SBHPL, a finding consistent with public opinion research.18, 86 More broadly, this finding is also consistent with political science research, which suggests that legislators’ attitudes toward a policy are intertwined with their attitudes toward the populations that the policy affects.87, 88 Our findings underscore the importance of adapting stigma reduction interventions that have demonstrated effectiveness among the general public to legislators.39, 64, 89 Attitudes toward people with mental illness are a mutable individual characteristic that should be considered in behavioral health D&I research.

We found that state‐level contextual factors, such as a recent mass shooting and a large increase in the opioid overdose death rate, were associated in unadjusted analyses with support for C‐SBHPL. This suggests that temporal contextual factors like these might signal the opening of a “policy window” and an opportunity to disseminate evidence about C‐SBHPL to state legislators at a time when they are particularly supportive of adopting a policy.90 None of the state‐level contextual factors measured, however, were significantly associated with C‐SBHPL support in the fully adjusted models, indicating that legislators’ individual characteristics have more influence on C‐SBHPL support than do state‐level contextual factors.

Limitations

Our study has at least six main limitations. First, although the response rate of 16.4% is reasonable for legislators56 and is higher than the response rates of several recent surveys of legislators,46, 57, 58 it is low by public opinion research standards. Information about the gender, geographic region, and political party of nonrespondents, however, allowed us to develop and apply nonresponse weights. This in turn enabled us to account for these differences and improve our confidence that the results were not heavily influenced by nonresponse bias.59 Second, we measured stigma toward people with mental illness, not substance use disorders. Public opinion research suggests that US adults have more stigma toward people with substance‐use disorders than those with mental illnesses.64, 91 Thus, our findings do not account for the full range of stigma toward people with behavioral health conditions.

Third, the quality of overdose death reporting varies among states and some states’ estimates of opioid overdose death rates might not be precise.92 Fourth, we should emphasize that C‐SBHPL is only one type of policy, and the results of our study cannot necessarily be generalized to other types of behavioral health legislation, such as laws that increase taxes to fund behavioral health services. Fifth, our survey was limited to legislators and our results are not generalizable to administrative policymakers in the state government's executive branch who are responsible for enforcing C‐SBHPL.

Finally, although the evidence that C‐SBHPL increases access to behavioral health treatments and does not increase insurance premiums was produced by systematic reviews conducted by the US Community Preventive Services Task Force,3, 4 caveats related to this evidence should be noted, especially because beliefs about these C‐SBHPL impacts were most strongly associated with support for C‐SBHPL. While C‐SBHPL increases access to behavioral health treatments by expanding coverage, it does not necessarily increase the receipt of treatments, partly because of behavioral health workforce shortages. Although the review of economic evaluations studies published through 2011 found that C‐SBHPL was not associated with increases in insurance premiums,4 this could be the result of C‐SBHPL's prompting insurance companies to implement new managed‐care practices that actually restrict the utilization of behavioral health treatments. While we did not identify any studies published after 2011 that assessed the impacts of C‐SBHPL on insurance premiums, the economic impacts of these laws may have changed after the implementation of the ACA.

Conclusion

Beliefs about policy impact, behavioral health treatment effectiveness, and mental illness stigma were independently associated with legislators’ support for C‐SBHPL. These mutable characteristics were stronger predictors of C‐SBHPL support than were most of the fixed characteristics and all of the state‐level contextual factors. The dissemination of evidence about C‐SBHPL to state legislators should target mutable characteristics because they could be antecedent to increasing support for policy adoption.