State earned income tax credits and general health indicators: A quasi‐experimental national study 1993‐2016

Abstract Objective To assess the relationship between the presence and generosity of state‐level Earned Income Tax Credits (EITC) and multiple self‐reported measures of general health. Data Sources Data on state‐level tax credits and covariates were obtained from the National Bureau of Economic Research and University of Kentucky Center for Poverty Research, respectively. These data were merged with Behavioral Risk Factor Surveillance System survey records from 1993‐2016. Study Design Using difference‐in‐differences approaches and survey‐weighted Poisson regression that accounted for clustering of observations and included state and year fixed‐effects, we assessed relationships between EITC and self‐reported overall health, frequent mental distress, and frequent poor physical health in the prior 30 days. Covariates included state minimum wage, state GDP, and adoption of Medicaid expansion. Sensitivity analyses revealed that parallel trends were plausible; there were no significant lead and lag effects. Data Extraction Methods Analyses were restricted to respondents with no more than a high school diploma or equivalent because less‐educated adults are more likely to be low‐wage earners and therefore qualify for EITC. Principal Findings Among adults with no education beyond high school (n = 2 884 790), each additional 10‐percentage‐point increase in the generosity of state EITC—relative to the federal credit—was associated with fewer reports of frequent mental distress (−97.3 per 100 000; 95% CI: −237.2, 42.6) and frequent poor physical health (−149.6 per 100 000; 95% CI: −284.4, −14.9). When restricted to individuals interviewed during the three months when tax rebates are commonly disbursed, the magnitude of the association between EITC and prevalence of reported frequent mental distress was greater (−329.7 per 100 000; 95% CI: −636.0, −23.5). Conclusions The generosity of state EITC policies is positively associated with significant reductions in frequent mental distress and poor physical health, especially during months when the credit is received. Interventions to reduce poverty may positively impact health by reducing material hardship and stress.


| INTRODUC TI ON
Many risk factors for suboptimal health share poverty as an upstream determinant. 1,2 Financial precarity is a contributing factor for mental and physical illness. 1,[3][4][5] Income and employment strongly predict health outcomes, especially among those living at or below the poverty line. 1,6,7 In 2000, there were an estimated 291 000 deaths attributable to poverty and income inequality in the United States (US). 8 Some of these impacts are so substantial that they contribute to intergenerational health disparities in mental and physical health. 1,9 Poverty can compromise health via different pathways. 4 Insufficient income is associated with food insecurity, lack of health insurance, and exposure to environmental hazards (eg, lead paint and air pollution). [10][11][12] Poverty has potentially long-lasting psychological and physiological affects through stress processes. Low-income families are more likely to live in under-resourced neighborhoods with insufficient infrastructure and high crime rates; they are also more likely to depend on unpredictable employment. 13 These stressors not only contribute to higher rates of anxiety and depression among low-income adults 4,14 but also to overloading and fundamentally altering the body's stress response system causing long-term damage to one's health. 15,16 The co-occurrence of mental and physical health concerns may be caused by shared upstream determinants like poverty. This may be partially explained by the physiological manifestation of stress, or allostatic load, causing long-term damage to one's health. 16 In addition, poor physical and mental health have a bidirectional relationship with poverty if they disrupt work, which can create feedback loops perpetuating these problems. 17,18 Given the important and lasting impact of poverty on health, policies that reduce poverty may produce large improvements to population health across generations. 19 The federal Earned Income Tax Credit (EITC) was introduced in 1975 as a federal tax credit to low-earning working families. 20 Currently, the EITC is the largest anti-poverty program for working-aged adults in the US 21 The credit received by EITC beneficiaries as a tax refund is based on pretax earnings, marital status, and the number of children in the household. 22 It is "refundable," meaning that tax filers can receive the credit even if they do not have tax liability. In 2018, approximately 25 million beneficiaries received about $63 billion from the federal program. 23 Since 1975, many states have enacted their own EITC policies as a supplement to the federal credit and, as of 2017, 29 states and the District of Columbia had their own EITC. 22 State EITCs were implemented in different years and vary in their generosity between states and over time. This variation creates the opportunity to examine the impacts of EITC policies on health at the state population level using a quasi-experimental design.
Much of prior literature on the intersection of EITC and health has focused on maternal and child health, reflecting the primary target population of the policy. [24][25][26] EITC may reduce complications in pregnancy, adverse birth outcomes, and unfavorable child development at individual population levels. [24][25][26][27][28][29] The consistency of findings across study designs and outcomes supports a causal impact of EITC on maternal and child health. However, recent studies of EITC and multiple measures of short-term health outcomes for children found few associations. 27 The relationship between EITC and adult health more broadly has not been well studied. 30 Authors of a 2013 systematic review reported that the evidence for the impact of EITC policy on measures of overall adult health and well-being was limited for some outcomes (eg, self-reported health, mental health/psychological distress, and during months when the credit is received. Interventions to reduce poverty may positively impact health by reducing material hardship and stress. frequency of poor physical health days) and missing entirely for others (eg, diagnosed mental illness and alcohol consumption). A study of EITC and general health focused on federal policy expansions occurring in the 1990s and spanned 1993-2001. 31 Though this study found a relationship between EITC and reduced blood pressure, fewer poor mental health days, and fewer poor physical health days, the study population consisted only of women with children and a high school education or less-which is not uncommon in the extant literature on health and EITC. While women with children in the home are more likely to receive EITC benefits, this inclusion criterion does not capture fathers providing care for their children or individuals without children who may benefit from the tax credits and also substantially limits the size of the study population.
Since 1993, 19 states and the District of Columbia have created refundable EITCs or made existing EITCs refundable. 5,32,33 Increasing the availability and generosity of the EITC (eg, via state supplements) could likely provide several health benefits for low-income single mothers. 34 We sought to provide a contemporary assessment of the effect of state-level EITC policies on general indicators of physical and mental health which incorporated these recent changes in policy. Identifying policies that improve well-being could substantially contribute to public health efforts.
We assessed the prevalence of reported poor overall health, frequent mental distress, and frequent poor physical health in relation to state EITC among US adults with no postsecondary education.
The knowledge generated by this investigation can inform policymakers of the health benefits of EITC implementation and generosity, which go above and beyond the policy's direct outcomes of employment, income, and reduced poverty.

| Study design and population
We utilized cross-state variation in the existence and generos- We restricted our analysis to adults who had no more than a high school education or GED equivalent, a population of workers who are most likely to be affected by state EITCs. Annual earnings would be the most direct measure of eligibility for the EITC, but it is problematic to restrict the sample by an outcome directly affected by the policy and plausibly on the path between the policy and health.
Education is a reasonable proxy for EITC eligibility, while also not being affected directly by the policy. 35 Adults without postsecondary education frequently have lower earnings and are therefore most likely to be eligible for and benefit from EITC policy. 36,37 After excluding respondents with postsecondary education, the study period included 1272 state-years of data (DC being treated as a state) with 2 884 790 respondents with high school education or less (characteristics of individual respondents can be found in Appendix S1); a subset were the 743 174 observations included in the analysis of outcomes collected during tax season in which rebates are usually sent (February-April of each year). 27,38

| Outcomes
The outcomes of interest included the following: Respondents ranking health as "fair" or "poor" were considered to have suboptimal overall health. FMD and FPPH were measured via the "Healthy Days" component of the survey. FMD was assessed through the question, "Now thinking about your mental health, which includes stress, depression, and problems with emotions, for how many days during the past 30 days was your mental health not good?" As has been suggested by prior work by CDC, individuals reporting 14 or more days of poor mental health were considered to have FMD. 39 Similarly, participants were asked, "Now thinking about your physical health, which includes physical illness and injury, for how many days during the past 30 days was your physical health not good?" A cut-off of 14 or more days of health being not good was also used for physical health.
Though they are nonspecific, broad indicators of mental and physical health are readily available and informative in surveillance and research. 40 Public Health Practice agencies ranging from the local to national levels frequently use these indicators in assessing the overall health of the population and in planning. [41][42][43] The measures have been used by health plans and researchers as a measure of overall physical and mental health and been found to appropriately map on to chronic conditions. 44,45

| Policy exposure
The primary exposure was the existence and generosity of a state EITC. State EITC status is presented in Figure 1. These data were obtained from the National Bureau of Economic Research. 33 In most states, the benefits are set as a percentage of the federal benefit.
In our analysis, states with nonrefundable EITCs-where the value of the credit cannot exceed the recipient's tax liability-were considered to not have an EITC because individuals who qualify for the EITC benefits tend to have limited tax liability and thus would not receive any substantial rebate.
Three states are exceptions to the usual pattern of state EITC.
The Wisconsin EITC follows the structure of the federal EITC, but its value varies with the number of dependents. In these analyses, the EITC value assigned to Wisconsin reflects the amount received by households with two dependents; this was 25 percent of the federal credit in 1993, 16  differs from the federal benefit in that the value of the credit increases more sharply at a lower income and begins to phase out at a lower income-level than the federal. California did not introduce the state EITC until 2015, and benefits were not received until 2016 meaning the difference in structure had minimal impact on the analysis; consequently, we did not account for differences in the EITC structure.

| Covariates
We controlled for a set of time-varying state-level covariates that could have changed over the study period differentially by state and that are plausibly related to the outcomes of interest.
Time-varying covariates included in the model were state minimum wage (in 2010 inflation-adjusted dollars) and state GDP (in 2010 inflation-adjusted dollars). We elected to include minimum wage as a measure of a low-wage worker's earning potential. The state GDP served as an indicator of the overall economic productivity in a state; after including state and year fixed-effects, any large changes in the state GDP represent changes in the fiscal landscape that are unique to that state. These financial covariates were obtained from the University of Kentucky Center for Poverty Research (UKCPR) database. 5 We additionally included an indicator for adoption of Medicaid expansion. After expansion of Medicaid, many low-wage workers who may have been previously uninsured or under-insured would have increased access to health care, potentially impacting physical and mental health.
Indicators of Medicaid expansion were adapted from data provided by the Henry J. Kaiser Foundation. 46 These covariates were selected a priori based on existent literature and our conceptual model ( Figure 2). Covariates were available for all state-years of the study period.
While they are important factors in health, we intentionally excluded direct measures of employment and insurance from our models. The EITC can operate through employment which would then potentially impact insurance status. Due to the possible endogeneity of these values and the potential for them to act as mediators on the causal pathway, we did not include them in the models.

| Statistical analysis
This analysis used a difference-in-differences approach, comparing the change in prevalence of each health outcome across states with more and less generous EITC benefits as the benefits changed. In fully adjusted models, we included state and year fixed-effects and time-varying state-level covariates hypothesized to be related to both EITC generosity and health. We used survey-weighted Poisson regression and the margins command to calculate prevalence differences (PDs) and 95% confidence intervals (CIs). These models took the following form: where i indexed individuals, t years, and j states. Y itj was one of three general health outcome measures. EITC tj denoted a continuous variable, for which each one-unit increase represents a 10-percentage-point increase in the state refundable EITC (as a percentage of the Federal EITC). X tj was a vector of state-year covariates and j and t were state and year fixed-effects, respectively. 47 Error terms in analyses were clustered by state. The complex survey weighting method used in the BRFSS took clustering into consideration during calculations.
We additionally conducted sensitivity analyses using weighted linear regression.
For difference-in-differences models to produce unbiased estimates of the causal effect of the policy, the parallel trends assumption must be plausible. The assumption is that the trends in the outcomes being studied would be parallel in state with and without an EITC before the implementation of the policy and continue the same trend if not for the implementation of the policy.
It is not possible to observe the counterfactual after implementation, but it is conventional to examine parallel trends prior to the intervention as evidence of the plausibility of the assumption.
We assessed parallel trend assumptions visually (Appendix S3) and quantitatively via two different methods. First, the prevalence of the outcome in each state over time until the point at which EITC was introduced was visually compared between states that did and did not introduce a refundable EITC during the study period. To We anticipate that effects of EITC on mental distress and physical health would operate through stress reduction and thus would take effect shortly after receipt of cash. It follows that if the state EITC impacts mental distress, this effect should be stronger in the months immediately following receipt. To consider the shorter-term effects of EITC, additional analyses restricted BRFSS responses to interviews conducted during the tax season months of February-April. 20,38 This period has been used in prior analyses of the EITC to assess the policy's immediate impact. 27  When we further focused on the period in early spring (  Mean prevalence of frequent poor physical health at baseline per 100 000 (SD) health, which is congruent with prior research that focused exclusively on women with children. 31 It is reassuring that our study which covered a longer time period-thus including additional changes in EITC-and used a broader study population, results

| D ISCUSS I ON
were consistent with prior studies.
While our work was focused on a broader population than much of the prior literature, the research assessing maternal and child health in relation to EITC is helpful in contextualizing these findings.
The positive impact of EITC on health becomes more pronounced during the February-April period when eligible individuals tend to receive their EITC benefits. 27,48 Our findings add to the growing body of literature indicating health benefits immediately following the dispersal of state EITC funds 27 and are consistent with the notion that the induction period, or time between exposure and observable manifestation of the outcome, is shorter for financial stress and mental health than for the physical health impacts of poverty.
Based on the evidence of relationships between poverty and mental illness, prior work has also investigated EITC in relation to suicide. These analyses found that increased values of refundable state EITCs were associated with decreased suicide rates. 49,50 These findings were particularly pronounced among women-a group who is more likely to benefit from EITC. 50

| Limitations
The "Healthy Days" metric, as a self-reported measure, is intrinsically subjective. It is important to note that the results indicate shifts in perceived health; while perceived well-being is arguably more important than diagnosed poor health, the two are not interchangeable. Healthy Days Measures appropriately map onto detailed health inventories, indicating that misreporting and discrepancies between perceived and measurable health status are likely to be minimal. 53 Individuals may feel social pressures to report outcomes that differ from the true perception of their health-this is especially of concern when considering stigma around mental illness. We also have no reason to suspect that the accuracy of reporting would vary over time, which suggests our findings are unlikely to be an artifact of social pressures around

| CON CLUS IONS
During 1993-2017, 20 states implemented refundable state Earned Income Tax Credit policies designed to increase employment and household income among lower-earning workers. We found that these policies significantly reduced frequent mental distress and poor physical health among less-educated adults-especially during months when the credit was most likely to be received. Interventions to reduce poverty may improve health by reducing material hardship and stress.

CO N FLI C T O F I NTE R E S T
No disclosures were reported.

AUTH O R CO NTR I B UTI O N S
Ms Morgan had full access to all the data in the study and takes full responsibility for the integrity of the data and the accuracy of the data analysis.

RO LE O F TH E FU N D E R /S P O N S O R
The sponsor had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.