The impact of the Affordable Care Act on health care access and self‐assessed health in the Trump Era (2017‐2018)

Abstract Objective To estimate the impact of the major components of the ACA (Medicaid expansion, subsidized Marketplace plans, and insurance market reforms) on health care access and self‐assessed health during the first 2 years of the Trump administration (2017 and 2018). Data Source The 2011‐2018 waves of the Behavioral Risk Factor Surveillance System (BRFSS), with the sample restricted to nonelderly adults. The BRFSS is a commonly used data source in the ACA literature due to its large number of questions related to access and self‐assessed health. In addition, it is large enough to precisely estimate the effects of state policy interventions, with over 300 000 observations per year. Design We estimate difference‐in‐difference‐in‐differences (DDD) models to separately identify the effects of the private and Medicaid expansion portions of the ACA using an identification strategy initially developed in Courtemanche et al (2017). The differences come from: (a) time, (b) state Medicaid expansion status, and (c) local area pre‐2014 uninsured rates. We examine ten outcome variables, including four measures of access and six measures of self‐assessed health. We also examine differences by income and race/ethnicity. Principal Findings Despite changes in ACA administration and the political debate surrounding the ACA during 2017 and 2018, including these fourth and fifth years of postreform data suggests continued gains in coverage. In addition, the improvements in reported excellent health that emerged with a lag after ACA implementation continued during 2017 and 2018. Conclusions While gains in access and self‐assessed health continued in the first 2 years of the Trump administration, the ongoing debate at both the federal and state level surrounding the future of the ACA suggests the need to continue monitoring how the law impacts these and many other important outcomes over time.


| INTRODUC TI ON
In 2014, the major components of the Affordable Care Act (ACA), including the individual mandate, subsidized Marketplace coverage, and state Medicaid expansions, were implemented. [1][2][3] A recently published review 2 summarizes the growing literature on the impact of the ACA on insurance coverage, [4][5][6][7][8][9][10][11][12][13][14][15][16] access to care, [17][18][19]  Some studies focusing on the ACA Medicaid expansions alone found mixed results, with some showing improvements in health, while others found no effect. 2,3 One study examining both the Medicaid expansions and the non-Medicaid expansion components of the ACA found that an increase in health (as measured by the probability of reporting excellent health) emerged with a lag in 2015. 28 In this paper, we estimate the causal effects of the ACA on access to care and self-assessed health during 2017 and 2018, the first 2 years of the Trump administration, using data from the Behavioral Risk Factor Surveillance System (BRFSS). Ours is among the first publications to include both 2017 and 2018 data. 25,[27][28][29][30][31] Our access outcomes are the likelihoods of having insurance coverage, costs being a barrier to seeking care, a primary care doctor, and a checkup in the past year. Our health outcomes include overall self-assessed health and days of the previous month not in good physical health, not in good mental health, and with health-related limitations.
There are multiple reasons why adding data from the first two years of the Trump administration to the prior analysis is important. Trump's first executive order encouraged the federal government to waive or delay the implementation of any components of the ACA that would impose a financial or regulatory burden. 33 In addition, funding for ACA outreach and education programs, including funding for navigators, was reduced for open enrollment periods associated with 2017 and 2018 coverage. 34 Potentially most consequential, in October 2017, the administration discontinued cost-sharing reduction (CSR) payments to insurers for silver Marketplace plans, at a time when insurers already submitted premium rates for the coming plan year with an expectation of receiving CSR payments in return for reducing the cost-sharing in plans for low income enrollees. 35 Political debate surrounding the ACA was prominently featured in the news, including the failed vote to repeal the ACA in July 2017 and the vote to pass the tax reform package that included a repeal of the ACA individual coverage mandate in December 2018. 36 Thus, the addition of 2017 and 2018 data allows us to examine the initial causal impact of these events. This is important as recent descriptive evidence suggests that the national coverage rate actually fell by 0.5 percentage points between 2017 and 2018. 37 Following a recently established literature, we estimate difference-in-difference-in-differences (DDD) models with the differences coming from time, state Medicaid expansion status, and local area pretreatment uninsured rate in order to estimate the impact of the full ACA. 11,[14][15][16]25,27,28,38 This approach stands in contrast to many studies that use a simpler difference-in-differences (DD) model comparing changes in expansion states to changes in nonexpansion states in order to identify the effect of the ACA Medicaid expansion alone. Identifying the impact of the national components of the ACA, such as the individual mandate and subsidized Marketplace coverage, requires a different approach because they were implemented in every state at the same time. The inclusion of a third difference in our model handles this issue because the national components of the ACA should provide the most intense "treatment" in local areas with the highest uninsured rates prior to the ACA.

| Data
We use data from the BRFSS, an annual telephone survey of health and health behaviors conducted by state health departments in collaboration with the CDC. The BRFSS is the largest continuous health survey in the United States, collecting information on more than 300 000 adults per year. Having a large sample size is critical to obtaining meaningful precision because the ACA affected insurance

What is already known on this topic
• The ACA led to significant improvements in coverage and access to care throughout 2014 to 2016, as well as a lagged emergence of improvements in self-assessed health.
• However, changes in ACA administration beginning in 2017 could have negatively affect these gains.
• This study evaluates the impact of the ACA on insurance coverage, access to care, and self-assessed health including 2017 and newly released 2018 data. coverage for only a fraction of the population, limiting plausible effect sizes. The BRFSS is therefore a commonly used data source in the ACA literature on access and self-assessed health. 23,25,27,28 Our sample period is 2011-2018. The sample starts in 2011 because this is the first year in which the BRFSS included cell phones in its sampling frame. The sample ends in 2018 because this is the last year currently available. This timeframe gives us three years of pretreatment data and five years of post-treatment data. We limit our sample to individuals 19-to 64 years old who were interviewed between 2011 and 2018. As is common in the literature, we drop observations with missing values for the variables used in our analysis. 23,39,40 Our outcome variables measure access to care and self-reported health status. Access outcomes include indicators for any health coverage, having a primary care doctor, having a regular physician checkup in the past 12 months, and having any care needed but foregone because of cost in the past 12 months. Self-reported health status is based on a rating of overall health as poor, fair, good, very good, or excellent. We use this to construct indictors for whether overall health is good or better (ie, good, very good, or excellent), very good or excellent, and excellent. Other health measures include number of days of the last 30 not in good mental health, not in good physical health, and with health-related functional limitations.

What this study adds
These sorts of subjective self-assessed health variables have been shown to be correlated with objective measures of health, such as mortality. [41][42][43] We construct a Medicaid expansion indicator that is based on information collected by the Kaiser Family Foundation. 44 45 We measure the intensity of the non-Medicaid components of the ACA using the uninsured rate in the respondent's "local area" in the pretreatment year of 2013. This measure captures the "dose" of ACA treatment the local area could have received. We compute each respondent's "local area" pretreatment uninsured rate within our BRFSS sample of nonelderly adults. The publicly available BRFSS does not include geographic identifiers narrower than the state, but does tell us whether the respondent resides in the center city of an MSA, outside the center city of an MSA but inside the county containing the center city, inside a suburban county of an MSA, or not in an MSA. We use this variable to construct four subgroups within each state: those living within a central city, suburbs, non-MSA, and within-state location unavailable (this is the case for respondents interviewed on their cell phone). Based on these four geographic categories, we calculate the pretreatment average uninsured rates by "location" (considering "cell phone" to be a location for the sake of convenience) within a state. To ensure that each area contains enough respondents from our sample to reliably compute pretreatment uninsured rates, we follow the    Table S1 reports the means and standard deviations for the controls.
We stratified our entire analytic sample into four groups based on whether the respondent's state expanded Medicaid and whether the local area's pretreatment uninsured rate was above or below the median within the sample. According to Table 1, 79 percent of the sample had some form of coverage prior to 2014. Individuals in expansion states (columns 2 and 3) were slightly more likely to have insurance prior to 2014 than those in nonexpansion states (columns 4 and 5). Residents who live in expansion states with prereform uninsured rates below the median (column 3) had, on average, better health care access and self-assessed health than the rest of the sample even before 2014. Our DDD model will account for these baseline differences. Our online Appendix describes trends in our outcome variables over time.

| Methods
Our goal is to estimate the effects of both the fully implemented • and is the error term, which is clustered by state and heteroscedasticity-robust.
Note that POST t is absorbed by the time-by-area-type fixed effects ( ) so it is not separately included in Equation (1) (1) Following prior literature, we consider 2 to represent unobserved confounders rather than capturing part of the Medicaid expansion's causal effect, though we test the sensitivity of our results to changes in this assumption. 11,[14][15][16]25,27,28,38,46 The effect of the "fully implemented" ACA, that is, in Medicaid expansion states, combines the impacts of the Medicaid expansion and the national non-Medicaid components of the ACA: 1 *UNINSURED + 3 * UNINSURED . We report the predicted or implied effect of the ACA at the sample mean pretreatment uninsured rate rather than the underlying regression coefficients.
These implied effects are given by 1 *−UNINSURED in nonexpansion states and 1 *−UNINSURED + 3 * −UNINSURED in expansion states.
While estimates based on Equation (1) Tables 2 and 3  In the bottom panel, we report the same implied effects over the combined 2014-2018 postperiod based on Equation (1). In Appendix Table S4 and S5, we discuss alternative specifications to Equation (1) and the robustness of our findings.

| Effects on access
The top panel of Table 2

| Effects on health
Similarly to previous work considering effects through 2016, our results reported in the top panel of Table 3 suggest that the emergence of an impact on the likelihood of having excellent self-assessed health appears particularly gradual. 28 The effect of the fully imple- The bottom panel of Table 3

| Testing identifying assumptions
Appendix Tables S2 and S3 present the event study results for the pre-ACA coefficients associated with our access to care and selfreported health regressions, respectively. In total, the event study regressions provide 40 coefficients in the pretreatment period (four coefficients in the pre-ACA period for each of the ten outcomes).
We observe only two statistically significant pre-ACA coefficients out of 40, or five percent, which is exactly what we would expect by chance with a 5 percent rejection rate. Both of the failures are for the checkup variable, suggesting that the results for that outcome should be interpreted with caution.

| Specification checks
The results of our many specification checks are described in detail in the Appendix. These include dropping those in the catch-all cell phone "area type," excluding 19-to 25-year-olds since they may TA B L E 2 Effects of ACA at mean pretreatment uninsured rate on health care access In addition, we explore the sensitivity of our results to treating the coefficient on the * t term as part of the causal effect of the Medicaid expansion. Appendix Tables S4 and S5 report the results of these specification checks. Taken as a whole, these estimates are broadly consistent with our baseline results.   The literature has documented multiple potential reasons why the national components of the ACA might be more effective than the Medicaid expansion at improving access and self-assessed health. 28  to examine the evolving impact of the ACA on access to care and self-assessed health, in addition to the other key outcomes that have been featured in the ACA literature. 2 While the changes to the ACA we examined did not lead to short-run reductions in access to care or changes in the trend in reporting increased gains in excellent self-assessed health in 2017 and 2018, that does not tell us whether the impact of these changes will differ in the long run or how the other potential changes described above will impact these outcomes.