Short-term effects of the earned income tax credit on children’s physical and mental health

Abstract

Purpose:

Child poverty is associated with worsened health, although there is limited research on whether U.S. poverty alleviation policies improve children’s health. We examined the short-term effects of the earned income tax credit (EITC), among the largest U.S. poverty alleviation programs, on children’s food insecurity, weight status, and mental health.

Methods:

Using data from the National Health Interview Survey (NHIS, 1998–2016), we examined the effects of the EITC using a quasi-experimental difference-in-differences methodology. About 90% of EITC-eligible individuals receive tax refunds in February-April, while NHIS interviews occur throughout the year. We took advantage of this timing of refund receipt to compare EITC-eligible families interviewed in February-April with those interviewed in the other months, “differencing out” seasonal trends in outcomes among non-eligible families. Analyses involved multivariable linear regressions.

Results:

We found that food insecurity decreased in the months following EITC refund receipt, with no effects for weight status or mental health. Results were robust to alternative specifications.

Conclusion:

While these findings suggest that food insecurity among vulnerable children was reduced immediately after EITC refund receipt, this also means that the EITC may contribute to cyclical food insecurity. Policies to enhance income stability may be one solution to address these findings.

Background

Poverty in early life is linked to poor physical and cognitive development as children living in poverty have limited access to healthcare, healthy food, and safe housing.(1) This in turn may adversely influence children’s social and health outcomes in later life, including educational and economic opportunities, chronic disease, and mortality (see conceptual framework, Appendix Figure A1).(2–4) There is therefore increasing interest in policies to address childhood poverty, both as a way to reduce health disparities in childhood, and to maximize opportunities for vulnerable children to succeed throughout life.(2) The earned income tax credit (EITC) is among the largest poverty alleviation programs in the U.S. (5) It provides tax refunds for low-income working families as a lump-sum disbursement during tax season. Individuals with children receive substantially higher benefits than those without, and about half are single mothers. More than 25 million tax-filers received almost $63 billion in EITC refunds in 2018, with an average refund size of $3,000.(5) The EITC has brought millions of families out of poverty, including 5.6 million in 2018 alone.(6–8)

Most studies in the past two decades have shown that the EITC improves health outcomes and behaviors, although some studies have mixed findings. Research has suggested improvements in food insecurity and mental health among adults, and reductions in smoking particularly for pregnant women.(9–15) Conversely, studies also suggest that the EITC may increase adult obesity and worsen metabolic markers like cholesterol, with no effect on short-term healthcare utilization.(11, 16, 17) For child health, most studies examined infant birthweight and other perinatal outcomes, with the majority finding beneficial effects.(18–23) Fewer have examined outcomes later in childhood, with two studies finding improved child development and test scores, and another finding improved overall health but possibly mixed results for short-term food insecurity.(24–27) Importantly, since the EITC is received as a lump-sum payment each spring, recipients may spend it differently than income disbursed more frequently, like wages.(28, 29) Thus, the EITC may affect short-term health differently than it affects long-term outcomes that depend on the accumulation of material resources; both pathways are important to understand. For example, although the Supplemental Nutrition Assistance Program (SNAP, i.e., food stamps) is recognized as an important policy for improving nutritional health among low-income families, the fact that it is received as a monthly benefit often leaves families with food shortages by the end of the month, with dangerous health consequences.(30, 31)

In this study, we examined the effects of the EITC on several measures of child health that have received limited attention previously. Our goal was to estimate the short-term effects of the EITC in the weeks and months after the refund is received in the spring. This study adds to evidence that federal and state policymakers can use in deciding on the scope and design of one of the largest U.S. poverty alleviation policies.

Study Data and Methods

Sample

The sample was drawn from the 1998–2016 waves of the National Health Interview Survey (NHIS), a large nationally representative serial cross-sectional survey (see sample flowchart, Appendix Figure A2). While NHIS collects demographic information for all household members (N=1,807,540), we restricted the analysis to children (0–17 years old) for whom data were collected on the outcomes of interest, since NHIS only asks about the health of one child in each family (N=229,495). We then restricted the sample to 172,281 children with non-missing values for all variables necessary to calculate EITC refund size. Next, we restricted the sample to children with family income greater than $0 and less than $100,000, since those who are unemployed or with higher incomes are unlikely to be an appropriate control group for EITC recipients. Since not all outcomes were collected for all children in all years (Appendix Table A1), the number of individuals included in the analysis varied by outcome.

Exposure

The primary exposure was whether an EITC-eligible child was interviewed in the months immediately after EITC receipt. In practice, this variable was an interaction term between two variables: (1) the size of the EITC refund for which a child’s family was eligible and (2) whether they were interviewed immediately after EITC receipt.

As with most surveys, NHIS does not query participants about taxes or EITC receipt. To determine EITC eligibility, EITC refund size was instead calculated using Internal Revenue Service (IRS) formulas, implemented using the Taxsim27 package for Stata.(32) Inputs included parents’ age, marital status, number of children, pre-tax household income, state of residence, and tax year. Variation therefore results from year-to-year changes in IRS formulas and state EITC policies, since about half of states have implemented supplements to the federal EITC.(33) We assumed that all individuals eligible for the EITC received their refunds, similar to other studies of the EITC.(11, 25, 26) About 80% of eligible families actually receive their refunds, making this approach analogous to an intent-to-treat design.(34, 35) While subject to measurement error, this approach is an alternative—and perhaps an improvement—on prior literature that used simpler proxies to impute EITC eligibility (e.g., educational attainment or Medicaid eligibility).(15, 36, 37) We rescaled this variable to thousands of U.S. dollars for interpretability.

As noted above, to create the primary exposure, this variable for EITC refund size was interacted with another variable representing when the child’s family was interviewed. Notably, 90% of the EITC-eligible population receives their refunds in February- April, soon after filing taxes.(38) NHIS interviews occur throughout the year and are not associated with individual characteristics. We therefore dichotomized this modifier as 1 if the child’s family was interviewed in February-April and 0 for those interviewed in May-January.

Outcomes

We examined child health outcomes that were likely to be affected in the short term after receiving EITC refunds. These outcomes also captured hypothesized mechanistic pathways linking income with child health, like nutrition and stress (Appendix Figure A1).

First, we included food insecurity, measured using the U.S. Department of Agriculture’s 10-item Food Security Scale,(39) a validated measure including questions about the past 30 days. Higher scores indicate greater food insecurity (range 0–10).

Second, since prior work found that the EITC affects adult body mass index (BMI) in the short term, we used parent-reported height and weight to calculate each child’s BMI adjusted for age and gender. We then created binary variables representing overweight or underweight using standard cutoffs that vary by age, using the zanthro package in Stata.(40–44) Third, we included two measures of mental health. These may be influenced in the short run because of the hypothesized short-term effects of the EITC on parent stress and child nutrition, both of which may affect child mental health and behavior.(45, 46) The Mental Health Indicator (MHI), adapted from the Child Behavior Checklist, was asked for children aged 2–3 years.(47) Questions include whether the child is uncooperative, unhappy, or having sleep problems. Questions are slightly different for boys and girls, but since they are intended to capture the same construct and are scored equivalently, we combined them into a single outcome for both genders. This approach has been used in prior work,(48) and it is accepted practice to combine scores for both genders on the full Child Behavior Checklist.(49–52) Higher scores indicate greater risk of mental health problems (range 0–8). For children aged 4–17 years, NHIS administers the Strengths and Difficulties Questionnaire (SDQ), a screening tool for child behavioral and emotional problems.(53, 54) The score includes questions on whether the child is obedient, worried, or unhappy. Higher scores indicate greater risk (range 0–12).

Covariates

Continuous covariates included the child’s age and age-squared, inflation-adjusted income and income-squared, and number of children. We included income-squared and age-squared to account for possible non-linear relationships between these covariates and the outcomes. Categorical covariates included child’s gender and race, and parent marital status and education. Missing covariates were imputed (see Appendix). We also included state fixed effects (i.e., indicator variables) to adjust for time-invariant characteristics of states that might influence both EITC refund size and outcomes, and fixed effects for year to account for secular trends.

Analysis

Our main objective was to estimate the EITC’s short-term effects on health in the weeks and months immediately after refund receipt. As described above, the fact that NHIS interviews participants year-round creates a natural experiment, in which we compared outcomes among EITC-eligible children interviewed in February-April with those interviewed in May-January, “differencing out” seasonal differences in outcomes among non-eligible children. Recent studies have used this approach to evaluate the short-term effects of the EITC.(11, 17, 27) Thus, we first tabulated sample characteristics by EITC eligibility and interview season.

Next, we estimated the short-term effect of the EITC using a difference-in-differences (DiD) approach.(55) This quasi-experimental method allows us to account for time-invariant differences and secular trends among EITC-ineligible families, if DiD assumptions are met. See Appendix for additional details, including the equation. We present standardized effect sizes for continuous outcomes, and we scaled all estimates per $1,000 for interpretability.

Secondary Analyses

The EITC’s effects may differ for single versus married recipients, in particular because the income boost may be particularly salient for single mothers, a vulnerable subgroup.(14, 56) We therefore carried out stratified models separately for single and married households.

To test the robustness of the results, we also conducted several sensitivity analyses. First, we implemented a model in which the primary exposure was the federal EITC refund for which a family was eligible (rather than federal plus state). About half of states have implemented their own supplements to the federal EITC;(33) since there may be state characteristics that influence both state EITC generosity and health, this analysis reduces confounding from unobserved state-level characteristics. This analysis did not include fixed effects for state.

Second, we carried out an analysis in which we did not adjust the models for covariates that are also determinants of EITC refund size (i.e., number of children, income, and marital status). Adjusting for these variables may reduce the variability in our exposure and contribute to null findings.

Ethics Approval

Ethics approval was obtained from the institutional review board of the senior author’s university (protocol #17-23255).

Results

Sample Characteristics

EITC-eligible parents were more likely to be less educated, unmarried, Black, or Hispanic, with lower income (Table 1). Demographic characteristics e.g., child age and gender, among those interviewed during February-April were similar to those interviewed during May-January for both EITC-eligible and non-eligible individuals. Meanwhile, health outcomes were worse on average among EITC-eligible children compared with non-eligible children.

Table 1.

Sample characteristics by EITC eligibility status and interview season

	EITC-Eligible, Feb-Apr	EITC-Eligible, May-Jan	Non-Eligible, Feb-Apr	Non-Eligible, May-Jan
	Mean (SD) or %	Mean (SD) or %	Mean (SD) or %	Mean (SD) or %
Demographics
Female (%)	48	49	48	49
Age (years)	8.3 (5.2)	8.3 (5.1)	8.3 (5.5)	8.3 (5.4)
Race (%)
White	32	30	56	55
Black	22	22	13	13
Hispanic	35	36	20	21
Other	11	12	11	11
Parent married (%)	50	51	75	74
Parent education (%)
Less than high school	33	33	12	12
High school	31	31	28	27
College	29	29	37	38
More than college	7	7	23	23
Family income (US$)	24300 (11797)	24217 (11695)	66833 (18452)	66774 (18687)
EITC refund size (US$)	2981 (1779)	3028 (1787)	0 (0)	0 (0)
Outcomes
Food insecurity	1.6 (2.4)	1.7 (2.5)	0.5 (1.5)	0.6 (1.5)
Overweight (%)	48.0	47.3	40.8	40.8
Underweight (%)	7.2	7.5	7.9	8.2
MHI score	1.4 (1.6)	1.4 (1.6)	1.2 (1.4)	1.3 (1.5)
SDQ score	2.3 (2.4)	2.3 (2.5)	2.3 (2.6)	2.3 (2.6)
No. observations	15500	43632	16728	47061

The study sample was drawn from the National Health Interview Survey for survey years 1998–2016 and includes those making more than $0 and less than $100,000 in annual household income. Sample characteristics were calculated using unimputed data. EITC and income are presented in inflation-adjusted U.S. dollars.

EITC: earned income tax credit; MHI: Mental Health Indicator; SDQ: Strengths and Difficulties Questionnaire

DiD Assumptions

Note that DiD assumes that the trends (not the levels) in the outcomes are similar among EITC-eligible and non-eligible families during the non-exposure window, an assumption that we tested empirically (see Appendix). Descriptions of other tests of DiD assumptions are available in the Appendix, Appendix Figure A2, and Appendix Tables A3–A4.

Short-term Effects of the EITC

EITC refund size was associated with improved food insecurity in the short term (β −0.0094 per $1,000; 95%CI: −0.019, 0.00010; p-value 0.052) (Figure 1). We were unable to rule out the null hypothesis that there was no association for measures of child mental health or weight status.

Figure 1. Short-term Effects of the EITC on Child Health, Main Analysis.

The study sample was drawn from the National Health Interview Survey for survey years 1998–2016 and includes those making more than $0 and less than $100,000 in annual household income. Coefficients represent the interaction of a continuous variable for total EITC refund amount (in thousands of inflation-adjusted U.S. dollars) and a binary variable for being interviewed in February-April. These regression models controlled for child’s gender, age, age-squared, and race; parent’s education, marital status, family income and income-squared, and number of children in the family; and state and year fixed effects. Robust standard errors were clustered at the state level.

*p <0.10; ** p <0.05

EITC: earned income tax credit; MHI: Mental Health Indicator; SDQ: Strengths and Difficulties Questionnaire

Subgroup Analysis

When examining the short-term effects of the EITC separately for single and married parents, we found that there were no differences in the association of the EITC with any outcome by parents’ marital status (Appendix Table A2).

Secondary Analyses

When using only the federal EITC as the main exposure to reduce confounding by unobserved state characteristics (Figure 2), EITC refund size was again associated with improved food insecurity (β −0.011 per $1,000; 95%CI: −0.022, −0.00079; p-value 0.036). This association was similar in models unadjusted for EITC determinants (β −0.0089 per $1,000; 95%CI: −0.019, 0.0011; p-value 0.082) (Figure 3). As in the main models, we did not observe an association between the EITC and other outcomes in secondary analyses.

Figure 2. Short-term Effects of the EITC on Child Health, Federal EITC Only.

The study sample was drawn from the National Health Interview Survey for survey years 1998–2016 and includes those making more than $0 and less than $100,000 in annual household income. Coefficients represent the interaction of a continuous variable for federal EITC refund amount (in thousands of inflation-adjusted U.S. dollars) and a binary variable for being interviewed in February-April. These regression models controlled for child’s gender, age, age-squared, and race; parent’s education, marital status, family income and income-squared, and number of children in the family; and year fixed effects. Robust standard errors were clustered at the state level.

*p <0.10; ** p <0.05

EITC: earned income tax credit; MHI: Mental Health Indicator; SDQ: Strengths and Difficulties Questionnaire

Figure 3: Short-term Effects of the EITC on Child Health, Unadjusted for EITC Inputs.

The study sample was drawn from the National Health Interview Survey for survey years 1998–2016 and includes those making more than $0 and less than $100,000 in annual household income. Coefficients represent the interaction of a continuous variable for EITC refund amount (in thousands of inflation-adjusted U.S. dollars) and a binary variable for being interviewed in February-April. These regression models controlled for child’s gender, age, age-squared, and race; parent’s education; and state and year fixed effects. Robust standard errors were clustered at the state level.

*p <0.10; ** p <0.05

EITC: earned income tax credit; MHI: Mental Health Indicator; SDQ: Strengths and Difficulties Questionnaire

Discussion

Using a large diverse national sample, this study provides evidence on the short-term effects of the EITC on the health of vulnerable children. Our use of a quasi-experimental design adds to the existing observational literature linking income with child health.

We found that the EITC resulted in reduced food insecurity in the weeks and months after refund receipt, equivalent to about 1% of a standard deviation per $1,000. Note that average refund size is about $3,000, and up to $10,000 in states with more generous EITC programs. This effect size is small at the individual level and represents a modest effect at the population level. The findings are consistent with one prior study using data from the 1980s-1990s that found short-term improvements in food insecurity among adult EITC recipients using a similar analytic strategy,(11) although a similar study from that historical period also found null or possibly negative effects for children.(27) Our study used a larger sample and more recent data, and findings are robust to numerous model specifications, suggesting that added income from the EITC contributes to a modest increase in household food purchases for families with children. Conversely, this also implies that food insecurity is worse during other months of the year more distant from receipt of the EITC tax refund, similar to prior work that found that monthly receipt of SNAP benefits results in adverse health effects at the end of each month as food run out.(31) To reduce the cyclical nature of food insecurity among EITC recipients over the course of the year, policymakers may want to consider increasing the scope of the “Advance EITC,” a program that allows recipients to receive their EITC refunds in installments throughout the year, instead of as an annual lump sum. Recent studies investigated the advanced periodic payment of EITC and found lower perceived financial stress and lower levels of food insecurity.(57, 58) Historically, the Advance EITC has had low take-up (below 3%), although it may be that this is due to the administrative burden of enrolling in the program.(59) Scaling up this option would require careful policy design to ensure that the beneficiaries’ needs are met while minimizing logistical hurdles.

Meanwhile, we did not observe a short-term effect of the EITC on child weight status, mental health, or behavioral problems. This contrasts with prior studies of the long-term effects of the EITC, including studies finding lower BMI and reduced behavioral problems.(25, 60, 61) Of note, these prior studies were examining year-to-year changes in these outcomes, while our study examines the effects on these health measures in the weeks and months immediately after refund receipt. There are several explanations for these contrasting findings. It may be that these outcomes are unlikely to change in the short-term in response to this type of income boost. Prior studies have found that children’s BMI can change in the short term even in the absence of any targeted intervention, particularly for obese children,(62) and that the EITC resulted in short-term changes to adult BMI using an identical analytic strategy to the one used here.(11) Yet it may be that changes in food availability influence the household’s perceived food insecurity, but that shifts in the nutritional content of food are not substantial enough to affect children’s biometric outcomes like weight status. Alternately, it may be that children’s nutrition is prioritized over that of adults throughout the year, such that their nutritional intake is relatively more protected from income shifts. For children’s mental health and behavioral problems, it may be that the short-term hypothesized improvements to nutrition and parents’ mental health do not translate into improvements in similar outcomes among children in the short run. Alternately, it may be that these outcomes are determined by changes in other resources (like healthcare, schooling, or parenting style) whose variation may be more pronounced in the long run as found in prior studies(63, 64). It may also be that there are no short-term changes in child mental health and behavioral problems because households tend to “smooth” consumption of relevant tangible resources over the year, e.g., by taking on debt in other parts of the year in anticipation of the receipt of an EITC refund in the spring. Indeed, other studies have found that the EITC is often spent on paying down debt(65, 66) and one found no short-term effects on healthcare utilization.(67)

The study has several limitations. First, we cannot determine the exact timing of EITC refund receipt, assuming instead that refunds are predominantly received during February-April.(38) For example, some individuals may be observed two months after refund receipt, which may bias estimates downward. For others, they may not have received the refund at all, in which case, they are actually in the control group and are misclassified. Future studies should attempt to link health data with administrative data on tax refund receipt. Second, we estimated EITC refund size based on self-reported demographic characteristics, which may lead to measurement error and contribute to null or biased results, and there may be residual confounding by parental health status or other characteristics that may influence both EITC amount and child health; however, this technique leverages the granular individual information available in NHIS rather than using coarse proxies like educational attainment or state-level EITC availability, and represents an alternative approach with different strengths and limitations. A related problem is that we adjusted for covariates that were used in the creation of the EITC exposure variable—e.g., number of children, marital status—since they also represent potential confounders of the relationship between the exposure and outcome. This complicates the interpretation of model estimates, although we were not able to carry out additional sensitivity analyses (e.g., dichotimization of the exposure variable) due to restrictions in data access during the COVID pandemic. Third, given the large size of our dataset (over 1.8 million individuals) we restricted imputation to the sample for whom we had data on EITC inputs. This may result in biased results or loss of power due to the loss of 25% of the sample. Finally, to test the pre-treatment parallel trend assumption, we compared trends between EITC-eligible and ineligible families during May-January; yet this assumes that there is no long-term effect of being “treated” with the EITC in prior years, and some of these families may receive the EITC repeatedly. Nevertheless, we may expect those with more recent income boosts to have a short-term change in outcomes over and above this long-term trend.

Conclusion

This study found short-term effects of the EITC for food insecurity but not for child mental health or weight status. The current design of the EITC involves a lump-sum tax refund that may result in modest seasonal variation in food insecurity among households with vulnerable children. The results are particularly important in light of increased interest in local, state, and national policymaking around safety net programs in the wake of the COVID-19 pandemic. Future work should focus on evaluating policies that provide more frequent payments and more income stability throughout the year (e.g., the minimum wage).