The effects of the American Rescue Plan Act on racial equity in health insurance coverage

Abstract

Objective

To evaluate the effects on racial disparities in health insurance coverage from the changes in the Premium Tax Credit (PTC) implemented in March 2021 as part of the American Rescue Plan Act (ARPA).

Data Sources and Study Setting

We use nationally representative individual‐level data from the Household Pulse Survey (HPS), which provides demographic, economic, and health insurance information for United States residents during the period April 2020–August 2022.

Study Design

While the PTC changes applied to all states, the 14 states that did not expand Medicaid received substantially more benefits than the expansion states since they had more uninsured individuals eligible for the PTC than the expansion states. In our analysis, the treatment (control) group includes all Medicaid nonexpansion (expansion) states. We use a difference‐in‐difference regression analysis to estimate the increase in the probability of insurance coverage after the expansion of the PTC. Furthermore, we conduct sensitivity and heterogeneity analyses.

Data Collection/Extraction Methods

We focus on survey respondents ages 18–64.

Principal Findings

The expanded PTC increased the probability of an individual having coverage through the Health Insurance Exchange (HIX) in a nonexpansion state by 0.95 (95% CI: 0.6136, 1.2900), 1.75 (95% CI: 1.1795, 2.3291), and 1.75 (95% CI: 1.1815, 2.3269) percentage points among White, Black, and Hispanic respondents, respectively. It also increased overall health insurance coverage among all groups.

Conclusions

The expanded PTC boosted HIX and overall health insurance coverage and reduced racial disparities.

What is known on this topic

• The American Rescue Plan Act (ARPA) expanded the Premium Tax Credits (PTCs) of the Affordable Care Act (ACA). The PTCs subsidize insurance purchased through the Health Insurance Exchange (HIX).

• The ACA reduced racial disparities by increasing health insurance coverage among Blacks and Hispanics. The ARPA introduced the first expansion of PTCs since 2010.

What this study adds

• This study evaluates how the recent expansion in PTC eligibility and amount affected enrollments through HIX by comparing enrollments in Medicaid expansion and nonexpansion states.

• Our results demonstrate that the PTC expansion increased HIX participation in the nonexpansion states at a higher rate than expansion states.

• The effects of the increased PTC were larger for Black and Hispanic respondents (compared to White respondents), indicating that the ARPA contributed to a reduction in racial disparities in health insurance coverage.

1. INTRODUCTION

Between September 2013 and February 2015, 16.9 million individuals gained insurance coverage due to the Affordable Care Act (ACA). ¹ Estimates suggest that about 40% of the increase was due to the Premium Tax Credit (PTC), a subsidy used by eligible individuals to buy health insurance from the Health Insurance Exchange (HIX). ² The ACA also mitigated racial disparities in health insurance coverage. After the main ACA provisions went into effect in 2014, the percentage of uninsured adults decreased by 3.0 percentage points among Whites, 5.1 percentage points among Blacks, and 7.1 percentage points among Hispanics. ³ A number of papers ³ , ⁴ , ⁵ found that the ACA reduced inequalities in health insurance coverage.

However, despite the ACA, an estimated 30 million Americans were uninsured in 2019 with huge racial disparities in access to health insurance. In 2019, the uninsurance rates among Whites, Blacks, and Hispanics were 7.8%, 11.4%, and 20.0% respectively. ⁶ The American Rescue Plan Act (ARPA) of 2021 expanded the eligibility and amount of the PTC that was established as part of the ACA. These expansions to the PTCs are the first since the ACA. At the time of the passage of the ARPA, researchers predicted that the ARPA may improve health equity. ⁷ This paper estimates the effect of the ARPA PTC expansion on racial disparities in insurance coverage. There are three major changes instituted by the ARPA that are relevant for our purposes. First, anyone who received unemployment compensation (UC) in 2021 is eligible for zero‐premium health insurance. This provision also applies to individuals in the nonexpansion states with incomes below 100% of the Federal Poverty Level (FPL) who were not eligible for PTCs (coverage gap) before the ARPA. Second, households below 150% of the FPL are now eligible for zero‐premium health insurance through HIX. Before the ARPA, these households were required to contribute a portion of their income to insurance premiums. Finally, households above 150% of the FPL received higher PTC amounts, including households above 400% of the FPL, who were previously not eligible for the PTC.

Although HIX is available in every state, it may be the only subsidized insurance option for low‐income nonelderly adults without Employer‐Sponsored Insurance (ESI) and who live in nonexpansion states. Although the ACA mandated Medicaid expansion in all states, a Supreme Court ruling made the expansion optional for states. A total of 36 states (and Washington DC) expanded Medicaid before January 1, 2021, leading to wide variability in Medicaid eligibility across states. In 2019, the uninsurance rates in nonexpansion states were almost double that of expansion states across all races and ethnicities. For example, in 2019, the uninsurance rate among Whites was 6% in expansion states compared with 11% in nonexpansion states. The corresponding numbers for Blacks and Hispanics were 8% and 15% in expansion states and 15% and 28% in nonexpansion states. ⁶

Therefore, it is plausible that nonexpansion states will see a larger increase in HIX participation than expansion states because of the ARPA PTC expansions. We compare HIX participation in expansion and nonexpansion states before and after the ARPA using nationally representative high‐frequency data from the Household Pulse Survey (HPS) and difference‐in‐difference regressions to explore how the expanded PTCs affected racial inequality in insurance coverage.

In addition, we expect the impact of these provisions may be higher among low‐income individuals. However, the effect may not be zero among higher income individuals because Medicaid eligibility is based on current income. Therefore, even individuals with a relatively high income in the previous year may become eligible for Medicaid if they live in an expansion state and their income falls below the cutoff level (for example, because of a job loss) since the ACA removed the asset test for Medicaid eligibility.

2. BACKGROUND: THE ARPA AND THE HIX

The basic tenets of the ARPA were part of President Biden's campaign. Three important events are most relevant to the implementation of the ARPA. First, all the major TV networks declared President Biden as the winner of the Presidential election on November 7, 2020 (during the 18th round of the HPS). Second, the Georgia Senate election, which allowed Democrats to control the Senate, was held on January 5, 2021 (between the 21st and 22nd rounds of the HPS). Third, the ARPA was signed into law on March 11, 2021 (during the 26th round of the HPS).

The open enrollment period for HIX enrollment for the calendar year 2021 was supposed to be from November 1, 2020, to December 15, 2020. Outside of the open enrollment period, individuals can purchase health insurance through HIX with a qualifying life event (such as a loss of a job, marriage or divorce, birth of a child, etc.). However, After President Biden took office, he issued an executive order that created a special enrollment period (SEP) from February 15, 2021, to August 15, 2021. During this SEP, individuals could purchase insurance through HIX without any qualifying life events. Individuals who already chose a plan could switch their plans during the SEP.

The PTC can be claimed at the end of the year when one files a tax return, or it can be claimed at the time of purchase through HIX. The latter is known as Advanced Premium Tax Credit (APTC). The APTC reduces the premium an individual has to pay by the amount of the APTC. However, the APTC is based on anticipated income. If the actual income differs from anticipated income (or the amount of PTC changes, which happened in the case of the ARPA), an individual will receive (pay) the difference between what they were supposed to receive and what they actually received as APTC.

3. METHODS

3.1. Data sources and study sample

We use data from the HPS, a nationally representative individual‐level repeated cross‐sectional survey, to estimate the effect of increased PTC on coverage. The HPS surveyed respondents on topics including demographics, socioeconomics, and source of health insurance.

We use the first 48 rounds from April 2020 to August 2022. Some individuals may have anticipated that PTCs will be increased/expanded after the presidential election and/or after the GA Senate election, even though the ARPA did not become a law until March 11, 2021. Therefore, we consider the period from November 7, 2020, to March 11, 2021 (rounds 18–26 of the HPS) as a transition period; since this period is technically neither a true “before” period, nor a true “after” period. In other words, in our baseline empirical implementation, we drop the data from these rounds. We consider rounds 1–17 as “before” and rounds 27–48 as “after” periods.

We focus on respondents ages 18–64 because adults over 64 qualify for Medicare in all states. This reduces the sample size from 3,720,633 observations to 2,754,608 observations.

From our sample of 2,754,608 respondents, we lose 595,078 observations because either the outcome or one of the control variables is missing. Finally, we eliminated rounds 18–26 of the HPS (except when testing for parallel trends) due to this being the policy transition period, leaving us with a final sample size of 1,802,922.

Following the HPS guidelines, we create five mutually exclusive racial‐ethnic groups: Non‐Hispanic White (NHW), Non‐Hispanic Black (NHB), Non‐Hispanic Asian (NHAsian), Non‐Hispanic Other Races (NHOther), and Hispanic (HISPAN). The sample sizes for each category are NHW (N = 1,343,001), NHB (N = 139,350), NHAsian (N = 95,541), NHOther (N = 74,020), and Hispanic respondents (N = 151,010).

The HPS asked individuals about their health insurance source(s). A respondent could pick up to eight sources: ESI, Marketplace (HIX), Medicare, Medicaid, Tricare, Veteran's Affairs (VA), Indian Health Service (IHS), and Other. In our sample, 21% of respondents reported having multiple sources of health insurance. This problem is not unique to our study or even to the HPS. Previous studies have reported difficulty in classifying the source of health insurance for respondents given the variety of choices and confusion among respondents, especially in the post‐ACA period. ⁸ , ⁹

We create mutually exclusive health insurance categories. ¹⁰ The ESI classification includes all respondents with ESI coverage. The HIX classification includes respondents with HIX coverage (but no ESI). The Medicaid classification includes respondents with Medicaid coverage (but no ESI, HIX, or other insurance). The Other Private Insurance classification includes respondents with any other type of private insurance (but no ESI, Medicaid, Medicare, HIX, Tricare, VA, or HIS). The Other Public Insurance classification includes respondents with Medicare, Tricare, VA, or HIS (but no ESI, HIX, Medicaid, or Other Private Insurance). Finally, respondents without any health insurance are categorized as uninsured.

There is some evidence that the HPS sample may not be representative due to the relatively low response rate. ¹⁰ For example, the estimates from the HPS overestimate the vaccination rate. ¹¹ Other papers found that HPS is representative in some dimensions but not others. ¹⁰ , ¹² There is also evidence that health insurance coverage rates in the HPS are different from some of the other surveys. ¹⁰

However, the biases in the levels of HPS health insurance coverage variables are similar across states and stable over time. ¹⁰ This is critically important for our purpose. Since we are relying on differences in differences, any bias in levels will be differenced out if the bias is similar in expansion and nonexpansion states. Furthermore, since our regression‐adjusted estimates include state and round fixed effects, the existence of bias in levels is not a problem as long as the bias is similar across states or over time.

3.2. Main outcome

Our primary outcome variable is whether an individual has health insurance through HIX. We also explore whether the ARPA PTCs changed health insurance coverage through Medicaid, ESI, other private sources, other public sources, and overall health insurance coverage. Please see Table A1 for a summary of respondent insurance sources across race‐ethnicity groups.

3.3. Primary explanatory variable

We treat the 14 states that did not expand their Medicaid programs before February 2021 as nonexpansion states. The nonexpansion states include Alabama, Florida, Georgia, Kansas, Mississippi, Missouri, North Carolina, Oklahoma, South Carolina, South Dakota, Tennessee, Texas, Wisconsin, and Wyoming. Two states, Oklahoma and Missouri, expanded Medicaid in 2021 (January 7 and 10, 2021, respectively). We treat them as nonexpansion states, but we show that the results are robust even if we omit these two states in the robustness section.

3.4. Covariates

To address potential confounders, we include controls for gender, marital status, age, education, income, and the number of children in the household all taken from the HPS survey data. We also control for the state‐level unemployment rate, taken from FRED Economic data. The mean values for all covariates for expansion states are in Table A2 and for nonexpansion states are in Table A3.

3.5. Parallel trend assumption

The difference‐in‐difference (DD) estimates represent the causal effect of the ARPA PTCs under the assumption that the nonexpansion and expansion states would have had similar trends in HIX participation in the absence of the ARPA. Testing this assumption is not possible since it requires observing the counterfactuals. Instead, we (and the rest of the literature) test whether the trends in the outcome variable (HIX participation rate in our case) were similar in nonexpansion and expansion states before the treatment (the ARPA in our case). To test this, we estimate the following regression equation

$$Y_{\mathit{it}}=\mathit{\alpha I}\left({\text{NONEXP}}_{\mathrm{it}}\right)+\sum _{j=2}^{48}{\beta }_{j}I\left(\text{WAVE}=j\right)+\sum _{j=2}^{48}{\gamma }_{j}I\left(\text{WAVE}=j\right)*I\left(\text{NONEX}{P}_{\mathrm{it}}\right)+\delta X_{\mathrm{it}}+S_i+{\epsilon }_{\mathrm{it}},$$ (1)

where $Y_{\mathrm{it}}$ is a binary outcome variable such as whether a respondent has health insurance through HIX or not. $X_{\mathrm{it}}$ includes individual‐level controls. $S_i$ includes state fixed effects. Since the parameters are identified from intra‐state variation, the parameter $\alpha$ is not identified. ${\beta }_j\mathrm{s}$ represent the time effects and ${\gamma }_j$s represent the difference between nonexpansion and expansion states during wave $j$. The parallel trend assumption requires that all the pre‐ARPA ${\gamma }_{j}s\left(j=2–17\right)$ are jointly insignificant. In addition to testing for the parallel trends assumption, this regression specification and the high‐frequency nature of the HPS (weekly from April 2020 to August 2020 and bi‐weekly after that) allow us to identify the timing of the response to this law precisely.

3.6. Identification strategy

After we establish the parallel trend assumption, we estimate the average effect of ARPA PTCs on outcomes. To that effect, we estimate

$$Y_{\mathit{it}}=\mathit{\alpha I}\left({\text{NONEXP}}_{\mathrm{it}}\right)+\mathit{\beta I}\left(\text{POSTARPA}\right)+\mathit{\gamma I}\left(\text{POSTARPA}\right)*I\left({\text{NONEXP}}_{\mathrm{it}}\right)+\delta X_{\mathrm{it}}+S_i+{\theta }_t+{\epsilon }_{\mathrm{it}},$$ (2)

where the coefficient of the interaction term ($\gamma$) is the primary coefficient of interest. Since we include state and time fixed effects, the parameters $\alpha$ and $\beta$ are not identified. All standard errors are clustered at the state level.

4. RESULTS

4.1. Parallel trend

As discussed above, the interpretation of DD estimates depends on the parallel trend assumption. Therefore, before discussing the main results, we check the parallel trend assumption. We estimate Equation (1) to test whether there are any differences in the pre‐ARPA trends in HIX participation across expansion and nonexpansion states. The estimated interaction terms (${\gamma }_{j}s;j=2–48)$ for the full sample are presented in Figure 1. A note below Figure 1 lists all the control variables included in the regressions. The bold line shows the estimated coefficients and the dotted lines show the estimated 95% confidence intervals. As shown in Figure 1, the estimated coefficients before the presidential election (rounds 2–17) are close to zero and never significantly different from zero. Furthermore, they do not show any trend. Next, to formally test the parallel trends assumption, we test that these 17 interaction term coefficients are jointly zero. The p‐values associated with the hypothesis is 0.44. Therefore, we conclude that the parallel trend assumption holds, and the DD estimates represent the causal effect of the ARPA PTCs.

FIGURE 1.

Estimated changes in the probability of Health Insurance Exchange (HIX) enrollment: By the HPS survey rounds. Controls include sex, marital status, age, age squared, number of children in the household, educational and income categories, state fixed effects, and survey round fixed effects. 95% CI is based on standard errors clustered at the state‐round level. Source: Author's Analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 1 also allows us to decipher the timing of the response. We can see that the interaction terms show an upward trend during the transition period (between November 2020 and March 2021), especially after the Georgia Senate runoffs. The coefficients after ARPA implementation in March 2021 are always positive and they are almost always statistically significant.

One concern is that the upward trend in the round effects seems to start around the 13th round. We are unaware of any particular event that would differentially increase HIX enrollment in the nonexpansion states after the 13th round. Individuals have 60 days after they lose their ESI to apply for health insurance through HIX. Therefore, it is possible that during the early part of the COVID‐19 emergency, individuals delayed their applications to get insurance through HIX (possibly because of widespread shutdowns and service disruptions). However, there is no way to confirm this in the data. We should note that none of the coefficient estimates between rounds 13 and 18 in Figure 1 are statistically significant. Furthermore, they are jointly insignificant. Please see the Sensitivity Analysis section for more on this issue.

From a policy evaluation perspective, it is crucial to understand how the ARPA affected the racial and ethnic inequality in health care access, given the pre‐existing differences in health insurance coverage across races and ethnicity. Moreover, there are differences in other socioeconomic characteristics across races and ethnicity, and therefore, we should expect different impacts of the ARPA across races and ethnicity. Therefore, we report the results for NHW, NHB, NHAsian, NHOther, and Hispanics separately. We estimated the parallel trend specification for each of the five racial‐ethnic groups. The results, presented in Figure A1, are similar to those in Figure 1.

4.2. Mean DD

We begin with a mean DD of the primary outcome variable. As discussed in the Backgrounds section, we take waves 2–17 as the pre‐ARPA period and waves 27–48 as the post‐ARPA period in our baseline analysis.

The summary mean DD results for HIX participation (our primary outcome variable) are shown on the left panel of Figure 2. The underlying data are given in Table A4. The percentage of people who purchased insurance through HIX in our sample is similar to previous studies (7.3% in our sample), which covers the period 2020–2022. If we restrict our sample to just 2020, then the corresponding number is 7.1%. For comparison, Bundorf et al. ¹⁰ report the percentage of people who purchased through HIX at 7.0% in the spring and summer and 7.1% in the fall and winter.

FIGURE 2.

Mean DD and regression‐adjusted DD estimates of the American Rescue Plan Act (ARPA) on Health Insurance Exchange enrollment. Controls include sex, marital status, age, age squared, number of children in the household, educational and income categories, state unemployment rate, state fixed effects, and survey round fixed effects. 95% CI is based on standard errors clustered at the state level. Stars indicate coefficient is significantly different from NHW group. Source: Author's Analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26. [Color figure can be viewed at wileyonlinelibrary.com]

For the full sample (all races combined; Table A4), coverage through HIX increased from 7.7% before the ARPA to 9.2% after the ARPA in the nonexpansion states, an increase of 1.5 percentage points (statistically significant at 1%). On the other hand, coverage through HIX in expansion states increased by only 0.3 percentage points (not significant at conventional levels). The mean DD estimate indicates that the ARPA increased HIX coverage by 0.9 percentage points (statistically significant at 1%) among NHW respondents in the nonexpansion states compared with the expansion states. This mean DD estimate is shown in the left panel of Figure 2. Similar analysis suggests that the mean DD estimates for NHW, NHB, NHAsian, NHOther, and Hispanic subsamples are 0.9, 1.7, 0.8, 0.7, and 1.7 percentage points, respectively (see left panel of Figure 2).

Tables A5 and A6 show the sample means for health insurance coverage through ESI and Medicaid, respectively. Table A7 shows the effects of the ARPA on uninsurance rates, which indicates that in the full sample, uninsurance rates in the nonexpansion states declined by 1.1 percentage points more than the expansion states. The corresponding estimates are 0.8, 1.8, 0.9, 1.3, and 1.4 percentage points among NHW, NHB, NHAsian, NHOther, and Hispanic respondents, respectively.

4.3. Regression‐adjusted estimates

Next, we estimate Equation (2) to estimate the average marginal effect. The estimated coefficients of the interaction terms for the full sample and for each racial‐ethnic group are presented in the right panel of Figure 2. The underlying estimates are given in Table A8. Estimates show that the additional PTCs increased the probability of an individual having HIX coverage by 1.13 percentage points (p < 0.001) in the full sample. We also estimated the effect separately for each of the racial‐ethnic groups. The effect among NHW respondents is 0.95 percentage points (p < 0.001). The effect among NHB, NHAsian, NHOther, and Hispanic respondents are 1.75 (p < 0.001), 1.03 (p‐value 0.084), 0.75 (p‐value = 0.044), and 1.75 (p < 0.001), respectively.

Next, to check if the differences across the racial‐ethnic groups are statistically significant, we estimate a regression where we include a triple interaction term (Post ARPA*Nonexpansion*Racial‐ethnic groups). The stars next to a racial‐ethnic group in Figure 2 signify that the group is significantly different than NHW respondents. The estimates for the triple interaction terms along with p‐values are in Table A9. The effect among NHB respondents is 0.8 percentage points more than the NHW respondents (p‐value = 0.003), and the effect among Hispanic respondents is also 0.8 percentage points more than the NHW respondents (p‐value = 0.045). This result is unsurprising given that more NHB and Hispanic respondents were uninsured to begin with. The estimated effects among the NHAsian and NHOther groups are not statistically different from that of NHW respondents.

We also estimated the regression‐adjusted effects of the ARPA on coverage through ESI, Medicaid, other public sources, and other private sources (please see Table A8). Estimates show that the additional PTCs reduced the probability of being uninsured by 0.96 percentage points (p < 0.001) in the full sample. The effect among NHW respondents is 0.75 percentage points (p < 0.001). The effect among NHB, NHAsian, NHOther, and Hispanic respondents are 1.41 (p‐value 0.004), 0.94 (p‐value 0.007), 0.96 (p‐value 0.11), and 1.26 (p < 0.001), respectively.

4.4. Sensitivity analyses

Next, we check the robustness of our results. The top row of Table 1 presents the baseline estimates for comparison purposes. When we use the original response to the source of the health insurance question (instead of creating mutually exclusive categories, which means some respondents are in multiple categories), the qualitative results remain the same (shown in Table 1).

TABLE 1.

Regression‐adjusted DD estimates of the ARPA on Health Insurance Exchange enrollment sensitivity analyses.

Sensitivity analysis	Non‐Hispanic White	Non‐Hispanic Black	Non‐Hispanic Asian	Non‐Hispanic Other	Hispanic	All
Baseline HIX	0.00952***	0.0175***	0.0103*	0.00746**	0.0175***	0.0113***
p‐value	(0.000)	(0.000)	(0.084)	(0.044)	(0.000)	(0.000)
Original HIX category	0.00633**	0.0250***	0.00632	0.00654	0.0171***	0.00909***
p‐value	(0.015)	(0.000)	(0.356)	(0.421)	(0.002)	(0.001)
Include rounds 18–26	0.00834***	0.0149***	0.00628	0.00665*	0.0150***	0.00969***
p‐value	(0.000)	(0.000)	(0.230)	(0.059)	(0.000)	(0.000)
Exclude OK and MO	0.00978***	0.0179***	0.00944	0.00503	0.0186***	0.0117***
p‐value	(0.000)	(0.000)	(0.114)	(0.237)	(0.000)	(0.000)

Note: Controls include sex, marital status, age, age squared, number of children in the household, educational and income categories, state unemployment rate, state fixed effects, and survey round fixed effects. Standard errors clustered at the state level. p‐values in parentheses. Stars indicate statistical significance at the 1%, 5%, and 10% levels.

Abbreviations: HIX, Health Insurance Exchange; MO, Missouri; OK, Oklahoma.

Source: Author's analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26.

Second, in our baseline specification, we dropped data from rounds 18–26 (the transition period). When we add the data from these rounds and classify them as the before period (since they are from before the formal passage of the ARPA), the qualitative results remain the same (shown in Table 1).

Third, we noted that two states (Missouri and Oklahoma) expanded their Medicaid in 2021. We exclude these two states from our sample and re‐estimate the treatment effects as a robustness check. The results (shown in Table 1) are similar to the baseline results.

Finally, we noted during the discussion on parallel trends that we also evaluate results in which rounds 1–12 are the “before” and rounds 27–48 are the “after” periods. We find qualitatively similar results, as demonstrated in Figure A2.

4.5. Heterogeneity analyses

In the Introduction section, we discussed that we expect the effect of the ARPA to be higher among low‐income people. We also expect a heterogeneous response by parental status. Since parents with children are more likely to be eligible for Medicaid than nonparents (even in nonexpansion states), we expect stronger effects among nonparents. We estimate how the effects of the ARPA differed by income, education, and parental status.

Table 2 presents the estimates. The HPS does not report the exact income or income‐to‐poverty ratio. Income is reported only in broad categories (Less than $25,000, $25,000–$34,999, $35,000–$49,999, $50,000–$74,999, $75,000–$99,999, $100,000–$149,999, $150,000–$199,999, and $200,000 and above). Therefore, we combine the bottom four income brackets (i.e., annual household income less than $75,000, which makes up 45% of the sample) and top four income brackets (i.e., annual household income of $75,000 or more). The HIX participation increased by 1.85 percentage points (p‐value <0.001) in the low‐income group and by 0.38 percentage points (p‐value = 0.005) in the high‐income group. Similarly, the HIX participation increased by 1.59 percentage points (p‐value <0.001) among those without a college degree and by 0.73 percentage points (p‐value <0.001) among those with a college degree (Table 2). Finally, HIX participation increased by 1.37 percentage points (p‐value <0.001) among adults without children and by 0.75 percentage points (p‐value <0.001) among adults with children.

TABLE 2.

Regression‐adjusted DD estimates of the ARPA on Health Insurance Exchange enrollment heterogeneity analyses.

Heterogeneity analysis	Non‐Hispanic White	Non‐Hispanic Black	Non‐Hispanic Asian	Non‐Hispanic Other	Hispanic	All
Education: Less than bachelor's	0.0146***	0.0168***	0.0147	0.0129**	0.0228***	0.0159***
p‐value	(0.000)	(0.001)	(0.364)	(0.012)	(0.000)	(0.000)
Education: bachelor's and up	0.00529***	0.0194***	0.0112***	0.000560	0.0117**	0.00730***
p‐value	(0.002)	(0.000)	(0.007)	(0.949)	(0.022)	(0.000)
Income: $75 k and under	0.0171***	0.0221***	0.0169	0.00954	0.0270***	0.0185***
p‐value	(0.000)	(0.000)	(0.112)	(0.112)	(0.000)	(0.000)
Income: Over $75 k	0.00330**	0.00733**	0.00764	0.00390	0.00387	0.00383***
p‐value	(0.041)	(0.013)	(0.152)	(0.517)	(0.238)	(0.005)
Households without children	0.0113***	0.0219***	0.0195***	0.0136**	0.0223***	0.0137***
p‐value	(0.000)	(0.000)	(0.008)	(0.019)	(0.000)	(0.000)
Households with children	0.00618***	0.0127**	0.00151	0.000622	0.0127***	0.00751***
p‐value	(0.001)	(0.012)	(0.825)	(0.927)	(0.002)	(0.000)

Note: Controls include sex, marital status, age, age squared, number of children in the household, educational and income categories, state unemployment rate, state fixed effects, and survey round fixed effects. Standard errors clustered at the state level. p‐values in parentheses. Stars indicate statistical significance at the 1%, 5%, and 10% levels.

Source: Author's Analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26.

5. DISCUSSION

Between 2013 and 2014, the uninsurance rate among nonelderly Whites, Blacks, and Hispanics went down by 3.1, 5.1, and 6.7 percentage points, respectively. The declines were 3.3 (2.3), 5.6 (4.0), and 5.4 (7.2) percentage points in the expansion (nonexpansion) states among Whites, Blacks, and Hispanics, respectively. ³ Therefore, the Medicaid expansion reduced the uninsurance rate by 1.0, 1.6, and 1.8 percentage points among Whites, Blacks, and Hispanics, respectively, during the first year of the ACA. Estimates also suggest that during the first years after the implementation of the ACA, the uninsurance rate declined by 7.7 and 11.5 percentage points among Whites and non‐Whites, respectively, and individually purchased insurance increased by 0.7 and 1.8 percentage points among Whites and non‐Whites, respectively. ¹³ In comparison, the ARPA subsidies increased individual purchase of insurance through HIX by 0.9, 1.75, and 1.75 percentage points among NHW, NHB, and Hispanic respondents, respectively. Therefore, the effect of the ARPA is comparable to previous major policy changes.

The ARPA may change the policy priorities in both expansion and nonexpansion states. The Inflation Reduction Act has already extended the ARPA PTCs until 2025. Blavin et al. ¹⁴ found that individuals below 150% of FPL in non‐expansion states spend $344 less in out‐of‐pocket expenditures than those in expansion states. Fiedler ¹⁵ found that the take‐up rate of HIX subsidy is about 50% for people below 400% of FPL. One of the concerns since the passage of the ACA has been that even with the ACA PTC and Cost Sharing Reduction (CSR), which is available to families earning up to 250% of FPL, leaves health insurance unaffordable for many families. The ARPA's increased PTC will make health insurance more affordable, and we find that it increased coverage through HIX. The increased PTC should also reduce the need for “wrap‐around” state subsidies offered by some nonexpansion states. Currently, California, Massachusetts, New Jersey, and Vermont offer such subsidies. States may spend that money to subsidize more targeted populations (such as those affected by “family glitch” or undocumented immigrants). ¹⁶ They may also focus on expanding CSR eligibility, which is not addressed by the ARPA.

On the other hand, the increased PTCs (especially the zero‐premium subsidies for people below 150% of FPL) and consequently increased health insurance coverage in nonexpansion states may reduce their incentive to expand Medicaid. Under the ACA, the federal government pays 90% of the cost of insuring adults who are covered under the Medicaid expansion. Under the ARPA, the federal government offered to pay 95% of the cost of expanded Medicaid coverage for states that chose to expand Medicaid (for the first 2 years). However, the federal government pays 100% of the PTC. This may reduce the incentive to expand Medicaid. If the nonexpansion states do not expand Medicaid, then people below 150% of FPL in those states will continue to pay higher out‐of‐pocket costs compared with their counterparts in expansion states. On the other hand, if ARPA subsidies are not extended beyond 2025, many people will lose health insurance, and the effects will be more pronounced in the nonexpansion states (assuming no additional Medicaid expansion). Our results also show that low‐income Black and Hispanic Americans will particularly be adversely affected.

APPENDIX A.

TABLE A1.

Insurance coverage by race.

Race categories	ESI	HIX	Medicaid	Other private	Other public	Uninsured
Non‐Hispanic White	1,020,300 (75.97%)	104,906 (7.81%)	81,678 (6.08%)	11,034 (0.82%)	41,344 (3.08%)	83,739 (6.24%)
Non‐Hispanic Black	94,263 (67.64%)	6713 (4.82%)	16,540 (11.87%)	1404 (1.01%)	4961 (3.56%)	15,469 (11.1%)
Non‐Hispanic Asian	78,027 (81.67%)	6364 (6.66%)	3642 (3.81%)	699 (0.73%)	1137 (1.19%)	5672 (5.94%)
Non‐Hispanic Other	48,190 (65.1%)	4425 (5.98%)	8874 (11.99%)	772 (1.04%)	5537 (7.48%)	6222 (8.41%)
Hispanic	100,412 (66.49%)	9827 (6.51%)	14,195 (9.4%)	2027 (1.34%)	4127 (2.73%)	20,422 (13.52%)
All	1,341,192 (74.39%)	132,235 (7.33%)	124,929 (6.93%)	15,936 (0.88%)	57,106 (3.17%)	131,524 (7.3%)

Note: Percent of total race‐ethnicity group in parentheses.

Abbreviations: ESI, Employer‐Sponsored Insurance; HIX, Health Insurance Exchange.

Source: U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26.

TABLE A2.

Covariate mean values for expansion states.

Covariate	Non‐Hispanic White	Non‐Hispanic Black	Non‐Hispanic Asian	Non‐Hispanic Other	Hispanic	All
Female	0.61	0.69	0.51	0.63	0.62	0.61
Marital status	0.60	0.38	0.64	0.47	0.52	0.58
Age	45.91	45.56	43.31	42.98	42.80	45.36
Total number of kids in household	0.73	0.85	0.78	0.87	0.95	0.77
Education: Less than high school	0.00	0.01	0.00	0.01	0.02	0.00
Education: Some high school	0.01	0.02	0.01	0.02	0.04	0.01
Education: High school graduate or equivalent	0.10	0.15	0.05	0.14	0.15	0.10
Education: Some college	0.19	0.26	0.11	0.27	0.25	0.20
Education: Associate's degree	0.10	0.11	0.06	0.12	0.11	0.10
Income: Less than $25,000	0.09	0.20	0.07	0.17	0.16	0.10
Income: $25,000–$34,999	0.07	0.13	0.06	0.10	0.12	0.07
Income: $35,000–$49,999	0.09	0.14	0.07	0.12	0.13	0.09
Income: $50,000–$74,999	0.16	0.18	0.13	0.17	0.17	0.16
Income: $75,000–$99,999	0.15	0.12	0.13	0.13	0.13	0.14
Income: $100,000–$149,999	0.21	0.13	0.20	0.15	0.15	0.20
Income: $150,000–$199,999	0.11	0.06	0.13	0.07	0.07	0.10
Income: $200,000 and above	0.14	0.05	0.22	0.08	0.07	0.13
Unemployment rate	7.49	8.16	8.83	8.23	8.40	7.71
N	994,054	81,645	79,165	55,261	104,580	1,314,705

Source: U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26.

TABLE A3.

Covariate mean values for nonexpansion states.

Covariate	Non‐Hispanic White	Non‐Hispanic Black	Non‐Hispanic Asian	Non‐Hispanic Other	Hispanic	All
Female	0.61	0.73	0.46	0.63	0.61	0.62
Marital status	0.62	0.39	0.69	0.50	0.56	0.59
Age	46.31	45.07	42.64	43.51	43.54	45.67
Total number of kids in household	0.74	0.92	0.88	0.89	0.93	0.79
Education: Less than high school	0.00	0.00	0.01	0.01	0.02	0.01
Education: Some high school	0.01	0.02	0.01	0.02	0.04	0.01
Education: High school graduate or equivalent	0.11	0.14	0.05	0.14	0.15	0.12
Education: Some college	0.21	0.25	0.10	0.27	0.23	0.22
Education: Associate's degree	0.11	0.13	0.06	0.13	0.12	0.12
Income: Less than $25,000	0.10	0.23	0.08	0.19	0.17	0.12
Income: $25,000–$34,999	0.08	0.15	0.06	0.12	0.13	0.09
Income: $35,000–$49,999	0.10	0.15	0.08	0.13	0.14	0.11
Income: $50,000–$74,999	0.18	0.19	0.14	0.18	0.18	0.18
Income: $75,000–$99,999	0.15	0.11	0.14	0.13	0.13	0.14
Income: $100,000–$149,999	0.20	0.10	0.21	0.14	0.13	0.18
Income: $150,000–$199,999	0.09	0.04	0.12	0.05	0.06	0.08
Income: $200,000 and above	0.11	0.03	0.17	0.05	0.06	0.09
Unemployment rate	6.52	6.92	6.82	6.38	7.30	6.64
N	348,947	57,705	16,376	18,759	46,430	488,217

Source: U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26.

TABLE A4.

Mean difference in difference estimates of the effects of the ARPA on Health Insurance Exchange coverage.

Race	Insurance	Expansion status	Before	After	Diff	Mean DD
NHW	HIX	EXPAN	0.073 (0.260)	0.077 (0.267)	0.004 [0.001]
NHW	HIX	NONEXPAN	0.081 (0.273)	0.095 (0.293)	0.014 [0.001]	0.009 [0.001]***
NHB	HIX	EXPAN	0.040 (0.195)	0.043 (0.204)	0.003 [0.001]
NHB	HIX	NONEXPAN	0.049 (0.215)	0.070 (0.255)	0.021 [0.002]	0.017 [0.002]***
NHAsian	HIX	EXPAN	0.064 (0.245)	0.062 (0.241)	−0.002 [0.002]
NHAsian	HIX	NONEXPAN	0.081 (0.274)	0.087 (0.282)	0.006 [0.004]	0.008 [0.004]*
NHother	HIX	EXPAN	0.056 (0.229)	0.055 (0.227)	−0.001 [0.002]
NHother	HIX	NONEXPAN	0.071 (0.256)	0.076 (0.266)	0.005 [0.004]	0.007 [0.004]*
HISPAN	HIX	EXPAN	0.052 (0.222)	0.054 (0.226)	0.002 [0.001]
HISPAN	HIX	NONEXPAN	0.083 (0.277)	0.103 (0.304)	0.020 [0.003]	0.017 [0.003]***
ALL	HIX	EXPAN	0.068 (0.252)	0.071 (0.257)	0.003 [0.000]
ALL	HIX	NONEXPAN	0.077 (0.267)	0.092 (0.289)	0.015 [0.001]	0.011 [0.001]***

Note: Standard deviations are in parentheses, and standard errors are in brackets.

Abbreviations: HISPAN, Hispanic; HIX, Health Insurance Exchange; MeanDD, Mean Difference in Difference; NHAsian, Non‐Hispanic Asian; NHB, Non‐Hispanic Black; NHother, Non‐Hispanic Other Race; NHW, Non‐Hispanic White.

Source: Author's Analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26.

^* p < 0.10;

^** p < 0.05;

^*** p < 0.01.

TABLE A5.

Mean difference in difference estimates of the effects of the ARPA on Employer‐Sponsored Insurance coverage.

Race	Insurance	Expansion status	Before	After	Diff	Mean DD
NHW	ESI	EXPAN	0.768 (0.422)	0.760 (0.427)	−0.008 [0.001]	‐
NHW	ESI	NONEXPAN	0.750 (0.433)	0.741 (0.438)	−0.009 [0.001]	−0.001 [0.002]
NHB	ESI	EXPAN	0.678 (0.467)	0.681 (0.466)	0.003 [0.003]	‐
NHB	ESI	NONEXPAN	0.669 (0.471)	0.677 (0.468)	0.009 [0.004]	0.006 [0.005]
NHAsian	ESI	EXPAN	0.813 (0.390)	0.828 (0.377)	0.015 [0.003]	‐
NHAsian	ESI	NONEXPAN	0.789 (0.408)	0.811 (0.392)	0.022 [0.006]	0.007 [0.007]
NHother	ESI	EXPAN	0.662 (0.473)	0.656 (0.475)	−0.007 [0.004]	‐
NHother	ESI	NONEXPAN	0.622 (0.485)	0.633 (0.482)	0.011 [0.007]	0.018 [0.008]**
HISPAN	ESI	EXPAN	0.676 (0.468)	0.679 (0.467)	0.003 [0.003]	‐
HISPAN	ESI	NONEXPAN	0.635 (0.482)	0.638 (0.481)	0.003 [0.004]	−0.000 [0.005]
ALL	ESI	EXPAN	0.753 (0.431)	0.749 (0.434)	−0.005 [0.001]	‐
ALL	ESI	NONEXPAN	0.726 (0.446)	0.722 (0.448)	−0.004 [0.001]	0.001 [0.001]

Note: Standard deviations are in parentheses, and standard errors are in brackets.

Abbreviations: ESI, Employer‐Sponsored Insurance; HISPAN, Hispanic; MeanDD, Mean Difference in Difference; NHAsian, Non‐Hispanic Asian; NHB, Non‐Hispanic Black; NHother, Non‐Hispanic Other Race; NHW, Non‐Hispanic White.

Source: Author's Analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26.

^* p < 0.10;

^** p < 0.05;

^*** p < 0.01.

TABLE A6.

Mean difference in difference estimates of the effects of the ARPA on Medicaid coverage.

Race	Insurance	Expansion status	Before	After	Diff	Mean DD
NHW	MEDICAID	EXPAN	0.065 (0.246)	0.076 (0.265)	0.011 [0.001]	‐
NHW	MEDICAID	NONEXPAN	0.031 (0.173)	0.040 (0.196)	0.009 [0.001]	−0.002 [0.001]***
NHB	MEDICAID	EXPAN	0.139 (0.346)	0.154 (0.361)	0.016 [0.002]	‐
NHB	MEDICAID	NONEXPAN	0.077 (0.267)	0.086 (0.280)	0.009 [0.002]	−0.007 [0.004]*
NHAsian	MEDICAID	EXPAN	0.042 (0.200)	0.045 (0.206)	0.003 [0.001]	‐
NHAsian	MEDICAID	NONEXPAN	0.014 (0.119)	0.013 (0.114)	−0.001 [0.002]	−0.004 [0.003]
NHother	MEDICAID	EXPAN	0.126 (0.332)	0.144 (0.352)	0.018 [0.003]	‐
NHother	MEDICAID	NONEXPAN	0.070 (0.255)	0.085 (0.279)	0.016 [0.004]	−0.003 [0.006]
HISPAN	MEDICAID	EXPAN	0.108 (0.310)	0.122 (0.327)	0.014 [0.002]	‐
HISPAN	MEDICAID	NONEXPAN	0.047 (0.211)	0.050 (0.218)	0.003 [0.002]	−0.011 [0.003]***
ALL	MEDICAID	EXPAN	0.074 (0.262)	0.085 (0.279)	0.011 [0.000]	‐
ALL	MEDICAID	NONEXPAN	0.039 (0.193)	0.047 (0.212)	0.008 [0.001]	−0.003[0.001]***

Note: Standard deviations are in parentheses, and standard errors are in brackets.

Abbreviations: HISPAN, Hispanic; MeanDD, Mean Difference in Difference; NHAsian, Non‐Hispanic Asian; NHB, Non‐Hispanic Black; NHother, Non‐Hispanic Other Race; NHW, Non‐Hispanic White.

Source: Author's Analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26.

^* p < 0.10;

^** p < 0.05;

^*** p < 0.01.

TABLE A7.

Mean difference in difference estimates of the effects of the ARPA on uninsurance.

Race	Insurance	Expansion status	Before	After	Diff	Mean DD
NHW	UNINSURED	EXPAN	0.062 (0.241)	0.045 (0.208)	−0.017 [0.000]	‐
NHW	UNINSURED	NONEXPAN	0.096 (0.295)	0.071 (0.257)	−0.025 [0.001]	−0.008 [0.001]***
NHB	UNINSURED	EXPAN	0.105 (0.306)	0.074 (0.262)	−0.030 [0.002]	‐
NHB	UNINSURED	NONEXPAN	0.160 (0.367)	0.111 (0.314)	−0.049 [0.003]	−0.018 [0.003]***
NHAsian	UNINSURED	EXPAN	0.063 (0.244)	0.046 (0.210)	−0.017 [0.002]	‐
NHAsian	UNINSURED	NONEXPAN	0.093 (0.290)	0.066 (0.249)	−0.027 [0.004]	−0.009 [0.004]**
NHother	UNINSURED	EXPAN	0.083 (0.275)	0.064 (0.244)	−0.019 [0.002]	‐
NHother	UNINSURED	NONEXPAN	0.127 (0.333)	0.096 (0.295)	−0.032 [0.005]	−0.013 [0.005]***
HISPAN	UNINSURED	EXPAN	0.126 (0.332)	0.102 (0.303)	−0.024 [0.002]	‐
HISPAN	UNINSURED	NONEXPAN	0.197 (0.398)	0.159 (0.366)	−0.038 [0.004]	−0.014 [0.004]***
ALL	UNINSURED	EXPAN	0.071 (0.256)	0.052 (0.223)	−0.018 [0.000]	‐
ALL	UNINSURED	NONEXPAN	0.115 (0.319)	0.085 (0.279)	−0.029 [0.001]	−0.011 [0.001]***

Note: Standard deviations are in parentheses, and standard errors are in brackets.

Abbreviations: HISPAN, Hispanic; MeanDD, Mean Difference in Difference; NHAsian, Non‐Hispanic Asian; NHB, Non‐Hispanic Black; NHother, Non‐Hispanic Other Race; NHW, Non‐Hispanic White.

Source: Author's Analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26.

^* p < 0.10;

^** p < 0.05;

^*** p < 0.01.

TABLE A8.

Regression‐adjusted estimates of the effect of the ARPA on insurance coverage.

Insurance	Non‐Hispanic White	Non‐Hispanic Black	Non‐Hispanic Asian	Non‐Hispanic Other	Hispanic	All
ESI	−0.00150 (0.672)	−0.00125 (0.811)	0.00247 (0.658)	0.00913 (0.258)	−0.00465 (0.336)	−0.00168 (0.622)
HIX	0.00952 (0.000)***	0.0175 (0.000)***	0.0103 (0.084)*	0.00746 (0.044)**	0.0175 (0.000)***	0.0113 (0.000)***
Medicaid	−0.00304 (0.121)	−0.00450 (0.319)	−0.00210 (0.310)	0.00112 (0.866)	−0.00804 (0.016)**	−0.00239 (0.234)
Uninsured	−0.00752 (0.002)***	−0.0141 (0.004)***	−0.00944 (0.007)***	−0.00959 (0.110)	−0.0126 (0.000)***	−0.00964 (0.000)***
Other Private	−0.000116 (0.761)	0.000130 (0.908)	−0.00145 (0.202)	−0.000450 (0.761)	0.00221 (0.034)**	0.000130 (0.731)
Other Public	0.00266 (0.010)**	0.00221 (0.258)	0.000178 (0.933)	−0.00767 (0.351)	0.00556 (0.061)*	0.00233 (0.068)*

Note: Controls include sex, marital status, age, age squared, number of children in the household, educational and income categories, state fixed effects, and survey round fixed effects. Standard errors clustered at the state level. p‐values in parentheses.

Abbreviations: ESI, Employer‐Sponsored Insurance; HIX, Health Insurance Exchange.

Source: Author's Analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26.

^* p < 0.10;

^** p < 0.05;

^*** p < 0.01.

TABLE A9.

Regression‐adjusted estimates of the effect of the ARPA on insurance coverage with race interaction.

Race category interaction	ESI	HIX	Other private	Medicaid	Other public	Uninsured
NHB	0.00229 (0.698)	0.00830 (0.003)***	0.000292 (0.803)	−0.00161 (0.706)	−0.000197 (0.929)	−0.00907 (0.034)**
NHAsian	0.00439 (0.378)	−0.00143 (0.811)	−0.00138 (0.208)	0.00150 (0.520)	−0.00263 (0.230)	−0.000443 (0.906)
NHOther	0.0114 (0.103)	−0.00160 (0.638)	−0.000680 (0.632)	0.00387 (0.493)	−0.0106 (0.180)	−0.00236 (0.652)
Hispanic	−0.00240 (0.630)	0.00800 (0.045)**	0.00223 (0.042)**	−0.00502 (0.104)	0.00315 (0.216)	−0.00596 (0.071)*

Note: Controls include sex, marital status, age, age squared, number of children in the household, educational and income categories, state fixed effects, and survey round fixed effects. Standard errors clustered at the state level. p‐values in parentheses.

Abbreviations: ESI, Employer‐Sponsored Insurance; HIX, Health Insurance Exchange; NHAsian, Non‐Hispanic Asian; NHB, Non‐Hispanic Black; NHother, Non‐Hispanic Other Race.

Source: Author's Analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 18–26.

^* p < 0.10;

^** p < 0.05;

^*** p < 0.01.

FIGURE A1.

Estimated changes in the probability of Health Insurance Exchange (HIX) Enrollment: by the HPS survey rounds and race/ethnicity. Controls include sex, marital status, age, age squared, number of children in the household, educational and income categories, state fixed effects, and survey round fixed effects. 95% CI is based on standard errors clustered at the state‐round level. Source: Author's Analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48. [Color figure can be viewed at wileyonlinelibrary.com]

FIGURE A2.

Mean DD and regression‐adjusted DD estimates of the ARPA on Health Insurance Exchange enrollment. Controls include sex, marital status, age, age squared, number of children in the household, educational and income categories, state unemployment rate, state fixed effects, and survey round fixed effects. 95% CI is based on standard errors clustered at the state level. Stars indicate coefficient is significantly different from NHW group. Source: Author's Analysis of U.S. Census Bureau Household Pulse Survey Rounds 1–48, excluding rounds 13–26. [Color figure can be viewed at wileyonlinelibrary.com]

Katsikas A, Mukhopadhyay S. The effects of the American Rescue Plan Act on racial equity in health insurance coverage. Health Serv Res. 2023;58(4):924‐937. doi: 10.1111/1475-6773.14166