The Effects of ACA Medicaid Expansions on Health after Five Years

Corresponding Author: George L. Wehby, Ph.D., Professor, University of Iowa, Departments of Health Management and Policy, Economics, and Preventive & Community Dentistry, and Public Policy Center, Research Associate, National Bureau of Economic Research, 145 N. Riverside Dr., 100 College of Public Health Bldg., Room N250, Iowa City, Iowa 52242-2007, Phone: 319-384-3814, Fax: 319-384-4371, ude.awoiu@ybhew-egroeg

Associated Data

Abstract

We examine the ACA Medicaid expansion effects on self-rated health status over five years. The study uses data from the Behavioral Risk Factor Surveillance System for 2011–2018 and a difference-in-differences design. There is improvement in health status on a 1–5 point scale from poor to excellent health among individuals below 100% of the federal poverty line by 0.031, 0.068, 0.031, 0.064, and 0.087 points in 2014, 2015, 2016, 2017, and 2018, respectively. Changes in 2015, 2017, and 2018 are statistically significant (p<0.05), and the 2014 change is marginally significant. The difference between 2014 and 2018 effects is statistically significant (p<0.05). In most years, we cannot distinguish changes in days not in good physical or mental health from no effect. Overall, there is only minimal evidence for effects intensifying over time, suggesting that health gains thus far have mostly occurred early on due to unmet needs among those previously uninsured.

Keywords: ACA, Medicaid expansion, difference-in-differences, low income individuals, poverty, health equity

The Patient Protection and Affordable Care Act (ACA) included a provision to expand Medicaid to adults below 138% of the federal poverty level (FPL). Nearly two thirds of the states have since adopted this expansion, and over half of these states have enacted their expansions over five years ago (Kaiser Family Foundation, 2018)

There is substantial evidence that the ACA Medicaid expansions have increased coverage among low-income adults, including across demographic subgroups (Berdahl & Moriya, 2018; Buchmueller et al., 2016; Stimpson et al., 2019; Stimpson & Wilson, 2018; Wehby & Lyu, 2018; Wherry & Miller, 2016). Overall, the Medicaid expansions increased coverage since 2014 by up to 7 percentage-points in 2017, with larger effects observed when focusing on low-income adults (Cawley et al., 2018; Courtemanche et al., 2019; Wehby & Lyu, 2018). There is evidence of improved access to care following the expansions, including increased routine visits to a health professional, dental visits, use of cancer screening services, and having a usual place of care (Choi et al., 2017; Courtemanche et al., 2017; Lyu & Wehby, 2019; Wehby et al., 2019; Lyu et al., 2020).

There is also growing evidence of the ACA Medicaid expansion effects on health although that evidence is much thinner than work on coverage and access to care. One study has reported a decline in the number of days not in good health combining data from 2014 and 2016 (Cawley et al., 2018). Another has reported an increase in reporting excellent health in 2015 and 2016 using data through 2016 (Sommers et al., 2017). Data from 2014 suggests improvement in mental health for adults with minor or moderate emotional conditions but not for those with severe emotional disorders, although other work shows increased access and health services utilization regardless of emotional health severity (Fry & Sommers, 2018; McMorrow et al., 2016). Finally, there is recent evidence of declines in disease-related mortality using data through 2017 (Miller et al., 2019).

Despite these initial findings on health, there is still a great need to understand the effects of the ACA Medicaid expansions on health over a longer period than included in the studies published to date, most of which have included data through 2016. Unlike coverage, access, and utilization, changes in health status from the Medicaid expansions may take longer to materialize.

NEW CONTRIBUTION

We examine the effects of the Medicaid expansions on self-rated health status of low-income individuals over five years from 2014, the first year when most of the expansions occurred, using one of the most recent national data sources. We estimate the expansion effects on health year by year from 2014 through 2018 to assess differences over time. We consider different income ranges to account for the possibility of churning into and out of Medicaid eligibility. Finally, we evaluate potential heterogeneity in effects across demographic subgroups, and consider changes in access to health services as a potential mechanism for changes in self-rated health.

METHODS

Conceptual Framework

There are two main reasons why the intent-to-treat effect of the Medicaid expansions on health could change over time. First, take-up of Medicaid gradually increased from 2015 to 2017 and was nearly double in 2015 compared to 2014, indicating that the proportion of low-income individuals affected by the expansion has increased over time especially in 2015 (Courtemanche et al., 2019; Wehby & Lyu, 2018). Second, while some individuals with unmet health needs can receive immediate benefits after gaining coverage (such as from diagnosis and treatment), the broader effects on health from increased preventive care (and any related changes in lifestyle) or from improved management of chronic conditions (such as due to greater access to prescription drugs) may not occur right away. Health benefits from these changes may accumulate over time for affected individuals, and more individuals can realize these benefits over time. If so, the effect of the expansions on health may increase over time. A third potential reason is that some of the health effects may occur from the added financial security and income resulting from coverage due to lower medical debt, increased access to credit, and disposable income, which can also build up over time (Bureau of Business and Economic Research 2018; Hu et al., 2018).

Another issue that can affect year by year estimates, particularly in cross-sectional samples, is the possibility of individuals churning into and out of the Medicaid eligibility income range over time based on income changes or errors in income reporting or measurement. If health improvements lead to greater income for some individuals (for example, due to more work hours), then it is possible that some may become income ineligible for Medicaid. Income reporting errors also prevent capturing health changes among all those affected by the Medicaid expansions. An implication is that income thresholds for eligibility and churning into and out of eligibility may limit long-run effects on health. Empirically, this requires considering the possibility that the expansion effects on health may also occur above 138% FPL not only below that threshold.

Data

We employ data from the Behavior Risk Factors Surveillance System (BRFSS), a nationally representative, cross-sectional survey, for years 2011 through 2018 (CDC 2018). Along with standard demographic and socioeconomic indicators, BRFSS respondents report their current health status on a five-category scale (excellent, very good, good, fair, or poor). We focus on self-rated health, a valid measure of health status (Idler & Angel, 1990; Miilunpalo et al., 1997), as the main outcome. In the main models, we use the five-category scale but also examine a series of binary indicators to understand if there are ranges in the scale that are changing more than others. Additionally, respondents indicate how many days in the past thirty they have lived not in good physical health and not in good mental health which we evaluate in additional models.

We also examine access to health services to further understand effects on health. We use the following three measures: 1- whether the individual had forgone medical care due to cost in past year; 2- having a personal doctor or health provider; and 3- whether the person has completed a routine medical checkup in past year. Together, these measures capture multiple dimensions of access to health services (cost, continuity of care, and preventive care use).

Sample

The Medicaid expansions increased income eligibility to 138% FPL. However, individuals above 100% FPL also received subsidies for the private insurance market exchanges in non-expanding states. Therefore, to focus on the group gaining coverage almost exclusively through Medicaid and to reduce the possibility of differential trends between expanding and non-expanding status, we limit the main analytical sample to individuals whose household income is below 100% FPL. At the same time, and as noted above, the possibility of churning into and out of Medicaid eligibility suggests that health effects may be observed at higher income thresholds. Therefore, we also examine the Medicaid expansion effects on self-rated health for samples below 138% FPL, 200% FPL, 300% FPL, and 400% FPL.

Another reason for considering alternative income ranges is that the BRFSS does not have an exact income measure. Instead, income is reported in the following ranges: Less than 10,000; 10,000–14,999; 15,000–19,000; 20,000–24,999; 25,000–34,000; 35,000–49,999; 50,000–74,999; and 75,000+. We use the midpoints of the reported income categories to calculate income as percent of FPL; for the last income category of 75,000+, we assign the midpoint as $87,500 (assuming a cap of $100,000). None of sample below 100% FPL and less than 1% of adults below 200% FPL are in the highest income group; 3.3% and 12.7% of the samples below 300% and 400% FPL, respectively, are in the highest income category. In transforming income into percent of FPL, we consider the number of adults and children in the household. For respondents in the cell-phone sample in 2011–2013 who were not asked about the total number of adults in the household (about 7% of the sample), we assign number of adults based on reported marital status (2 adults for married or cohabitating, 1 otherwise).

Because our goal is to examine health effects over five years from the 2014 expansions, we focus on individuals aged 25–64 years who would have been age eligible for the expansions from the first year. Therefore, we exclude from the main sample individuals younger than 25 because age is reported in ranges in BRFSS, and because (most of) those in the younger range interviewed in recent years would not have been age eligible in the first year. As a sensitivity check however, we include adults below 25. Supplementary Table 1 online shows weighted descriptive statistics for the main analytical sample (25–64 year olds below 100% FPL), which ranges between 225,674 and 231,415 depending on outcome. Supplementary Table 2 online shows weighted summary statistics for the health outcomes year by year and separately for expanding and non-expanding states.

State Assignment into Treatment/Control Groups

We focus on the 22 states that expanded in 2014 (20 of which expanded on January 1, 2014) and exclude states that expanded after December 31 st , 2014. In the main model, the control group includes states that did not expand under the ACA, and those that had full or near full expansions prior to 2014. We test sensitivity of the main estimates to excluding from the control group states that had full or near full expansions prior to 2014, or assigning those states to the treatment group. Supplementary Table 3 online shows the assignment of states into the treatment and control groups.

Study Design and Estimation

We use a difference-in-differences model to estimate the effects of the ACA Medicaid expansions on health, following many studies that have examined the expansion effects on a variety of outcomes (e.g., Ugwi et al., 2019; Wehby et al., 2019; Wehby & Lyu, 2018).

The model is specified as follows:

H e a l t h ismt = α + β 1 M e d i c a i d s ∗ Y 2014 + β 2 M e d i c a i d s ∗ Y 2015 + β 3 M e d i c a i d s ∗ Y 2016 + β 4 M e d i c a i d s ∗ Y 2017 + β 5 M e d i c a i d s ∗ Y 2018 + X i s t Γ + θ s + m m + ω t + ϵ i s t ( 2 ) .

Health is one of the health status outcomes described above for person i in state s interviewed in month m in year t , while Medicaid is a binary variable equal to 1 for expanding states and 0 for the control states. X i s m t is a set of sociodemographic variables including age, gender, race/ethnicity, marital status, educational attainment, and whether there are children in the household (Supplementary Table 1 online). θ s are state fixed effects accounting for time-invariant differences between states. m m is survey month fixed effects, and ω t includes year fixed effects which capture national changes in outcomes over time (shared between expanding and control states). Y2014-Y2018 is a series of binary variables equal to one for each year from 2014 through 2018 (2011–2013 is the reference period). The coefficients of their interactions with the Medicaid expansion status, β 1 , β 2 , β 3 , β 4 , and β 5 , are estimates of the Medicaid expansion effects in 2014, 2015, 2016, 2017, and 2018, respectively (compared to 2011–2013). To examine whether the expansion effects are significantly different over time, we test whether these five coefficients are significantly different from each other, and whether the effect in 2014 is different from 2018.

The difference-in-differences model assumes that without the Medicaid expansions, outcomes would have evolved similarly over time between the expanding and control states. We check this assumption by evaluating if outcome trends differed between expanding and treatment states before the expansion in years 2011 and 2012 relative to 2013. To do so, we re-estimate a model that adds indicators for 2011 and 2012 (with 2013 as the reference year) and test the joint significance of 2011 and 2012 interactions with Medicaid expansion status.

We employ weighted least squares with the BRFSS sampling weights and cluster the standard errors at the state level. As a sensitivity check, we adjust the weights for changes in each state sample size over years. As another check, we re-estimate the main model without sampling weights.

Prior studies have shown that the Medicaid expansions have increased coverage across demographic subgroups such as those defined by age, gender, and race/ethnicity, and with or without children in household, although coverage gains are larger among younger adults and childless adults. Because pre-expansion differences in health across some demographic subgroups might modify coverage effects on health, we also evaluate the heterogeneity in the expansion effects on self-rated health by demographic factors.

RESULTS

Medicaid Expansion Effects on Self-Rated Health

Figure 1 and Table 1 show estimates of Medicaid expansion effects on self-rated health status measured on the 1–5 point scale of from poor to excellent. Estimates are the year-by-year from 2014 through 2018 (relative to 2011–2013) for 25–64 year olds below 100% FPL. Supplementary Figure 1 shows the weighted-means of self-rated health status by year and Medicaid expansion status for adults under 100% FPL. There is an improvement in self-rated health by 0.031, 0.068, 0.031, 0.064, and 0.087 points in 2014, 2015, 2016, 2017, and 2018, respectively. Effects are statistically significant (p<0.05) except in 2014 (marginally significant) and 2016 (insignificant). These changes represent 1.1% to 3.1% of the pre-expansion mean. The difference in effects between 2014 and 2018 is statistically significant (p<0.05).

Difference-in-Differences Estimates of ACA Medicaid Expansion Effects on Self-Rated Health on a Five-Point Scale from Poor (1) to Excellent (5), 25–64 Year Olds below 100% FPL, Year-by-Year From 2014 to 2018 versus 2011–2013

Notes: The figure shows difference-in-differences estimates of Medicaid expansion effects on self-rated health among adults 25–64 years old below 100% FPL estimated using weighted least squares (and BRFSS sampling weights). Effects are estimated year by year from 2014 through 2018 with 2011–2013 as the reference year. The models adjust for age, race/ethnicity, sex, marital status, educational attainment, whether there were children living in the home, month of survey fixed effects, year of survey fixed effects, and state fixed effects. 95% confidence intervals are shown in bars.

Table 1.

Difference-in-Differences Estimates of ACA Medicaid Expansion Effects on Self-Rated Health Status of 25–64 Year Olds on a Five-Point Scale from Poor (1) to Excellent (5) by Income, Year-by-Year From 2014 to 2018 versus 2011–2013

Notes: The Table shows difference-in-differences estimates of Medicaid expansion effects on measures of self-rated health estimated using weighted least squares (and BRFSS sampling weights). Effects are estimated year by year from 2014 through 2018 with 2011–2013 as the reference year. The models adjust for age, race/ethnicity, sex, marital status, educational attainment, whether there were children living in the home, month of survey fixed effects, year of survey fixed effects, and state fixed effects. Standard errors (SE) are clustered by state and are reported in parentheses. Sample sizes range between 230,437 and 1,220,640.

As noted above, we also examine if there are categories on the health rating scale that are changing more than others by coding binary outcomes for better health in an increasing order. Specifically, we code the following four outcomes: fair, good, very good, or excellent health (versus poor health); good, very good, excellent health (versus poor or fair health); very good and excellent health (versus poor, fair, or good health); and excellent health (versus poor, fair, good, very good health). Then we re-estimate the model described above for these outcomes for individuals below 100% FPL (Supplementary Table 5 online). The only outcome with statistically significant changes in multiple years is reporting good or better health, as opposed to fair or poor. The likelihood of that outcome increases by 0.013, 0.027, 0.012, 0.042, and 0.042 points in 2014, 2015, 2016, 2017, and 2018 respectively; all are statistically significant (p<0.05) except for 2016. Differences in this outcome are statistically significant across years and between 2014 and 2018. Effects on the other binary outcomes are less pronounced and generally insignificant. There are no significant differential pre-trends in these binary outcomes between expanding and non-expanding states (Supplementary Table 6 online).

Medicaid Expansion Effects on Days not in Good Health

Supplementary Table 7 online shows the year-by-year estimates of the Medicaid expansion effects on two additional measures of health for 25–64 year olds below 100% FPL: days not in good physical health (in past 30 days), and days not in good mental health. Beginning with days not in good physical health (in past 30 days), there is an insignificant increase in 2014, a significant decline of 0.4 days in 2015, and smaller and insignificant declines in 2016–2018. For days not in good mental health, we find no strong evidence of improvement (declines) in any year, In relative terms, the magnitude of effects on days not in good mental health are close to those on self-rated health (1.4 to 3.6% relative to pre-expansion mean). However, there are significant differential pre-trends in this outcome (Supplementary Table 8 online).

Medicaid Expansion Effects on Self-Rated Health by Demographic Factors

In Supplementary Table 9 online, we report estimates of Medicaid expansion effects on the five-point self-rated health scale among subgroups defined by age, gender, race/ethnicity, and children at home or not, all below 100% FPL. All subgroups except non-Hispanic Whites show a significant or marginally significant improvement in self-rated health in at least one of the five years. Females, individuals of race/ethnicity other than non-Hispanic-White, non-Black Hispanic, or Hispanic, and those with children at home have larger improvement in 2018 compared to 2014 (p

Medicaid Expansion Effects on Access

Supplementary Table 10 online shows year-by-year estimates of Medicaid expansion effects on the three access measures mentioned above for 25–64 year olds below 100% FPL. There is a significant decline in the likelihood of forgoing necessary medical care due to cost in each year, with declines ranging from 0.028 points in 2014 to 0.058 points in 2016 and 2017, but differences across years are not statistically significant. There is also an increase in the likelihood of reporting a personal doctor/healthcare provider and completing a routine checkup that is significant in some years, but no evidence of significant differences across years. However, both of these outcomes show differential pre-trends between expanding and non-expanding states, so estimates should be viewed with caution (Supplementary Table 11 online).

Sensitivity Checks

We evaluate the sensitivity of the main model estimates, effects of the Medicaid expansions on the five-point scale of self-rated health for individuals below 100% FPL, to certain sample and estimation choices (Supplementary Table 12 online). Estimates are largely robust to excluding states with full or near full expansions prior to 2014. However, switching those states from control to treatment states results in smaller and insignificant estimates in 2014–2017, but the estimate in 2018 is similar to the main estimate and statistically different from 2014. Estimates are also largely robust to adding adults younger than 25; effect in 2014 is larger, while effect in 2018 is smaller than the main model (for 2018, the estimate is 0.069 versus 0.087 in the main model). Adjusting weights for differences in each state’s sample size across years reduces the estimate for 2018 slightly but it remains significantly different from 2014. The estimates however are noticeably smaller (for all years except 2016) and statistically insignificant without using sampling weights (only marginally significant in year 2016) and there are no significant differences across years. This difference between weighted and unweighted estimates might suggest treatment effect heterogeneity across states in which case the weighted estimate may be closer to the average partial effect of the expansion (Solon et al., 2015).

Finally, we check errors in assigning individuals to below 100% FPL based on their income range midpoints by using data from the American Community Survey (ACS) for the same years and age range as the BRFSS sample. In the ACS, individuals report their exact income, not in ranges as the BRFSS. We select two samples from the ACS based on income. The first sample includes individuals below 100% FPL based on their exact reported income. The second sample includes those below 100% FPL based on midpoints of the same income ranges used in BRFSS. Then we compare the two samples by year.

In the ACS sample below 100% FPL based on exact income, 6–11% (depending on year) are classified at/above 100% FPL based on BRFSS income range midpoints (Supplementary Table 13 online). Also, 1–3% of the ACS sample below 100% FPL based on BRFSS income range midpoints have exact income at/above 100% FPL. Next, we estimate the Medicaid expansion effects on Medicaid coverage in each of those two ACS samples using a similar difference-in-differences model to that described above. Despite the noted errors in classification based on income range midpoints, estimates of the Medicaid expansion effects on Medicaid coverage are identical between the ACS sample below 100% FPL based on exact income and the ACS sample below 100% FPL based on BRFSS income range midpoints (Supplementary Table 14 online). This similarity suggests that those close to the 100% FPL cutoff (on either side) experience a similar Medicaid take-up effect (from the expansion) as those with lower income and that there is little effect on estimates for the sample below 100% FPL in the BRFSS from using the income range midpoints.

DISCUSSION

We evaluate the effects of Medicaid expansion on self-rated health over a five-year post-expansion period from 2014 when most of the expansions first began till 2018. We find limited support based on our evidence for the notion of expansion effects on health increasing over the five post-expansion years. Using one of the largest and most recent nationally representative data sources, we do find evidence that the Medicaid expansion has improved self-rated health status among adults ages 25–64 below 100% FPL over this period. The effect on the five-point scale of self-rated health (from poor to excellent) in this sample is largest in 2018 (a nearly 3.1% improvement relative to the pre-expansion rate). The 2018 effect is more than twice as large as the 2014 effect, which is marginally significant, and the difference between the two years is statistically significant. However, differences in effects across the five years are not jointly significant. We observe the first statistically significant improvement in the five-point health rating scale soon after implementation in 2015, likely reflecting the larger increase in Medicaid take-up rates in that year compared to smaller relative gains in later years (Courtemanche et al., 2019; Wehby & Lyu, 2018). The effect in 2015 is almost 80% of that in 2018. The effect in 2016 is smaller and insignificant (although the estimate is imprecise and difference from 2018 is statistically insignificant). The effect in 2017 is close to that in 2015 and statistically significant. These results highlight the importance of continuing to watch trends in self-rated health across expansion versus non-expansion states as evidence continues to build.

Using a series of samples with increasing income ranges, we also show that improvements in the five-point self-rated health scale dissipate as income rises, and that effects are concentrated among individuals below 138% of FPL. There is a significant improvement in health status rating for almost all examined subgroups by demographic factors, suggesting widespread benefits in the affected population. There is little evidence for changes in days not in good physical or mental health; the only statistically significant change is decline in days not in good physical health in 2015 (5.2% relative to pre-expansion mean). It is worth noting, however, that the estimates for these outcomes are not precise, are within range of those observed for self-rated health in relative terms (i.e. compared to pre-expansion outcome means). Finally, there is improvement in health care access measures, but no clear evidence of effects intensifying over years.

Taken together, the results suggest that improvements in self-rated health over this period are likely to have been concentrated among those with unmet needs prior to the income-based expansion, a group also more likely to have sought coverage early on. Additional benefits from improved preventive care and greater access that would affect a wider proportion of low-income adults gaining coverage under the expansion may yet need more time to show up in a general health status measure. Additional work can also focus on disentangling the possible effects of financial security from other pathways to improved self-rated health.

It is worth noting that the ACA included other insurance provisions such as no exclusions from private coverage related to pre-existing conditions, insurance marketplaces, and the employer mandate. The research design assumes that these provisions had similar effects between Medicaid expanding and non-expanding states. However, estimates of the Medicaid expansion effects are conditional on the other ACA provisions that affect expanding and non-expanding states similarly. It is possible that effects of Medicaid expansions when enacted alone may differ from those enacted with other provisions such as the ACA did.