Measuring immigrant public benefit participation: A methodological guide

Introduction

There is no clear consensus among researchers on how best to measure immigrant public benefit participation, despite these estimates’ relevance to public opinion and ongoing policy debates.

Widely circulated figures—such as those posted by President Trump in January—seem only to reinforce preexisting ideological views on immigration, including claims that immigrants overuse public benefits or constitute a fiscal burden on U.S.-born citizens. In accordance with these views, the Trump administration has pursued a number of policy changes related to the use of public benefits by immigrants, including reinterpretations of what constitutes a “federal public benefit” under the Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) of 1996¹ and what receipt of benefits should be a basis for denying a green card because an applicant is “likely to become a public charge.”

In the context of these proposed policy changes, rates of immigrant benefit participation take on heightened importance. Yet, estimates of these rates vary widely. Frequently cited analyses by the Cato Institute and the Center for Immigration Studies (CIS) use similar population survey data as input but arrive at distinct portrayals of benefit receipt by immigrants; Cato estimates that immigrants used 21% less welfare and entitlement benefits than native-born Americans on a per capita basis in 2022, while CIS finds that 54% of households headed by immigrants used at least one major welfare program in 2022, compared to 39% for U.S.-born households.²

How can two analyses using similar data produce such divergent portrayals of immigrant benefit participation? This brief will answer this question by examining the choices that go into measuring benefit participation among immigrants and demonstrating that even small differences in methodological approaches can produce substantial differences in the results. It is intended to serve both as a guide for interpreting existing estimates and as a set of suggestions for researchers on best practices in their construction.

We use the 2023 American Community Survey (ACS) to illustrate the wide range of numbers that can be reported and to assess the implications of different approaches. We highlight four key methodological decisions that shape headline findings: what is the relevant economic unit (individual versus household), how is participation adjusted (given the over-imputation and under-reporting present in survey data), what is the immigrant population of interest (foreign-born or non-citizen), and what programs are counted as public benefits. We show how each choice affects estimated participation rates and conclude by presenting our preferred estimates, which implement our recommended methodological approach.

For additional estimates and analysis, see our companion piece.

Background: Measuring benefit use

Most analyses of immigrant benefit use rely on nationally representative survey data. We use the 2023 American Community Survey (ACS) 1-year estimates, produced annually by the U.S. Census Bureau, to construct statistics on the participation rates of certain benefits programs across groups. We use the 2023 survey due to data availability constraints in more recent datasets.³ The 2023 ACS includes a representative sample of approximately 3,400,000 individuals in approximately 1,500,000 households, corresponding to about 1% of the population and including unauthorized immigrants. Figure 1 shows the weighted national totals from the 2023 ACS sample, broken down by citizenship and nativity at both the individual and household levels. Common alternative data sources include the Survey of Income and Program Participation (SIPP) and the Current Population Survey (CPS). Program participation studies, like those conducted by CIS and Cato, often rely on the SIPP due to its longitudinal design and detailed monthly benefit reporting, though it is a smaller sample than the ACS.

The U.S. Census Bureau defines “social welfare programs” as benefits provided by a governmental entity and targeted to individuals or households based on means-tested eligibility criteria. These include forms of cash assistance such as Supplemental Security Income (SSI) and Temporary Assistance for Needy Families (TANF), forms of food assistance such as the Supplemental Nutrition Assistance Program (SNAP, formerly known as the Food Stamp Program) and the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC), and healthcare coverage through Medicaid or the Children’s Health Insurance Program (CHIP). In analyses of benefit use, some scholars consider non-means-tested programs such as Social Security and Medicare, and others incorporate refundable tax credits.

In the ACS, some benefit programs are measured through direct self-reports. For SSI, TANF, and Social Security, respondents are asked whether they personally received benefits during the prior 12 months, and for Medicaid and Medicare, respondents are asked whether they are currently covered by that type of health insurance. However, for SSI and TANF, benefits received on behalf of a child may sometimes be reported by an adult household member, complicating efforts to attribute receipt to a specific individual. For SNAP, the question is asked at the household level (“Did you or any member of this household receive benefits from the Food Stamp Program or SNAP?”), meaning participation reflects household receipt rather than individual enrollment.

Other programs, including the Earned Income Tax Credit (EITC), housing subsidies, WIC, and school lunch are not directly reported in the ACS but are imputed through the Supplemental Poverty Measure (SPM). The ACS does not directly ask about housing assistance; instead, the Census Bureau imputes a housing subsidy value for the SPM based on data from the Current Population Survey. EITC benefits are simulated using TAXSIM, a tax model developed by the National Bureau of Economic Research, which calculates expected credits based on reported income and household characteristics—but without accounting for immigration-status restrictions on eligibility. Both WIC and free or reduced-price school meal participation are imputed using a logistic regression model based on CPS data, with benefit amounts then assigned based on programmatic data on average benefit levels.

Because these SPM-based figures are modeled rather than self-reported, they introduce additional measurement uncertainty and may not accurately reflect benefits received. These variables are also reported at the family unit level rather than for individuals. For school lunch, this makes interpretation particularly challenging, because citizen children in mixed-nativity families may count towards reports of immigrant benefit receipt. We propose a way to address this issue in our “Adjusting usage” section below.

ACS-based calculations for the EITC may similarly overstate participation among immigrants, because TAXSIM does not account for the immigration-status restrictions on eligibility. We address this over-imputation, along with incomplete take-up, in the “Adjusting usage” section. (The ACS does not include child tax credit (CTC) imputations so we do not include that program in our analyses.)

For most other programs and groups, under-reporting of benefit participation in household surveys is well documented in the research literature, particularly for means-tested programs such as SNAP and TANF. Survey respondents may misreport participation due to recall error, stigma, confusion about program names, or misunderstanding of eligibility. This concern is also discussed below.

Methodological considerations

1. Choosing a unit

In reports of benefit participation rates by immigrants, the choice of whether to measure benefit use at the individual, household, or any other unit level is highly influential for resulting estimates and a point of dispute among researchers. CIS reports on the household level and cites studies by The National Research Council, The Heritage Foundation, and two by the Census Bureau that do the same. CIS defends this decision by quoting the National Research Council: “the household is the primary unit through which public services are consumed.” By contrast, Cato uses the individual as the unit, primarily to account for the fact that immigrants often live in mixed-nativity households with native-born Americans. They argue that “[c]ounting native-born welfare consumption as immigrant consumption improperly inflates estimates of immigrant welfare use and deflates native-born consumption.” 

One natural research approach would be to align the unit of measurement with the way each program is administered. Benefits administered for a household, tax unit, or assistance unit are treated as household benefits here. Benefits with designated individual recipients are treated as individual programs. We outline guidelines for determining the proper unit of measurement for each program below.

The decision of the unit can have a significant impact on estimates; Figure 2 shows how using the household rather than the individual as the measurement unit affects rates for programs administered to individuals. The effect is largest for Medicaid: at the individual level, non-citizens use it at about the same rate as U.S.-born citizens, but counting a household as a beneficiary when one member participates opens a roughly 20-point gap, making non-citizens appear to use the program far more. The gap is mostly generated by the fact that U.S.-born citizen children enroll in Medicaid while their non-citizen parents often do not. The household measure overstates how much immigrants themselves use the program.

Household, family, or assistance unit programs: SNAP, TANF, EITC, and housing vouchers

SNAP is administered and recorded at the household level, with eligibility and benefit amounts determined by household income and composition. EITC is recorded at the family level in the ACS and is distributed to tax filing units (which typically include the set of eligible household members). TANF benefits are provided to a family assistance unit (typically an adult and any children), with eligibility determined based on unit income, the presence of dependent children, and immigration-related restrictions.⁴ Housing vouchers are administered at the household level, with the subsidy attached to the dwelling unit and shared by all residents. For each of these programs, participation is aptly measured at the level of the relevant assistance unit rather than the individual.

When interpreting these estimates, however, it is crucial to consider that an immigrant may reside in a SNAP or TANF-receiving household (or unit) but be excluded from the benefit calculation due to their immigration status. In fact, income of ineligible unit members may be considered as a resource even though they are not considered in the need calculation, thereby further lowering the benefits for other members. Program rules effectively penalize the mixed-citizenship unit for the immigrant’s presence by reducing or eliminating the total benefit available to citizens in the unit. 

For programs that are not analyzed on the individual level, measuring “immigrant use” thus requires special attention. A conceptually grounded approach to measuring immigrant receipt of benefits such as SNAP or TANF should count benefits as accruing to immigrants only when their presence increases unit eligibility or benefit size. Figure 3 illustrates this distinction for programs in this category, showing both “any receipt” rates—which include benefits flowing through citizen or immigrant members of mixed-citizenship or mixed-nativity households—and “immigrant use” rates that reflect only benefit use attributable to the immigrant or non-citizen member.

In the case of SNAP, ineligible immigrant members are excluded from the SNAP unit when determining household size, but their income is included—often on a prorated basis—in the household income test. According to USDA rules, the ineligible immigrant’s income must be considered, but the household’s benefit is not adjusted upward based on their presence. As a result, the presence of an ineligible immigrant can only reduce a household’s SNAP benefit or threaten eligibility relative to a scenario in which that person is absent. Researchers should account for this population when measuring SNAP participation; immigrant program participation should only be counted if the immigrant household member has an immigration status making them eligible to receive program support. 

TANF rules operate similarly to SNAP, in that an ineligible immigrant’s income is often counted as available to other eligible members of the household, but the ineligible person is not counted as a member of the “assistance unit,” thereby reducing the benefit available to the family. In these cases, we recommend that researchers avoid counting TANF receipt by the assistance unit as “immigrant use,” because the immigrant’s presence in the household is subtracting from rather than adding to the size of the benefit. 

Although ineligible immigrants in mixed-citizenship or mixed-nativity families may benefit indirectly from household SNAP or TANF resources going to citizens or eligible non-citizens, it is inappropriate to characterize their residence in a recipient household as constituting receipt of a public benefit. Other members of the household would be receiving a benefit at least as large if the ineligible immigrant were absent. 

The presence of an immigrant in a household may also disqualify other household members from receiving the benefit. In the case of EITC, if any member of the filing unit does not have a valid Social Security Number (SSN), the entire unit is ineligible for the credit. In the case of housing subsidies, the Department of Housing and Urban Development (HUD) recently issued a proposed rule that would require proof of U.S. citizenship or eligible status for every resident in HUD-funded housing. 

Individual programs: Medicaid, SSI, school lunch, WIC, Social Security, and Medicare

Other programs—Medicaid, Supplemental Security Income (SSI), social insurance programs like Social Security and Medicare, WIC and school lunch—are more appropriately analyzed at the individual level. Medicaid and Medicare are administered strictly at the individual level, with eligibility and enrollment determined separately for each person. Unlike cash transfers, health services cannot be shared across members of a household. School meals are similar: they are provided directly to individual children at school and cannot be redistributed to other household members. Household-level measures of these programs are therefore conceptually inappropriate, as they obscure individual participation and can mischaracterize access among immigrants and their U.S.-born family members.

Because cash assistance may be shared within the household, the lines of individual support are more blurred here than for health programs. Even so, Social Security and SSI are administered and best analyzed at the individual level. The same logic applies to WIC, which is also awarded on an individual basis—to pregnant women and children under age five specifically, regardless of immigration status. Counting non-recipient household members as “users” in these cases risks overstating immigrant benefit use. Indeed, similar to SNAP and TANF, SSI’s income deeming rules mean that a portion of an ineligible immigrant’s income or resources can be counted against the eligible recipient’s benefit calculation, so the non-immigrant receiving SSI may well have received more public support in the absence of the immigrant household member.

2. Adjusting usage: Over-imputation and under-reporting

The ACS’s use of imputed participation rates for programs such as the EITC and free or reduced-price school lunch can mischaracterize the extent of immigrant participation if not carefully specified. 

The ACS school lunch variable is reported at the household level, which can be misleading because many mixed-citizenship or mixed-nativity households include citizen children, who may be the only household members of school age. Counting these cases as non-citizen or immigrant use therefore overstates immigrant participation. We recommend that researchers avoid counting lunch receipt by citizen children as benefit participation by non-citizens. Moreover, we recommend characterizing school lunch participation at the individual level, since the benefit is received by individual children rather than shared across a household. Individual receipt must be imputed in the ACS: researchers should assume that if a child is in a household flagged as receiving the benefit, that child is a recipient and the child’s own immigration status should be used to classify use. 

Accurate EITC imputation similarly requires incorporating immigration-related eligibility constraints. The NBER TAXSIM model uses household characteristics and income to simulate eligibility for tax credits, but it does not account for SSN requirements. Because claiming the EITC requires valid SSNs for all filers in the unit, some low-income immigrants who appear eligible based on income can not actually receive the credit. If one spouse does not have a valid SSN, no one can be eligible for the EITC. We recommend adjusting imputed eligibility to reflect this restriction. Because the ACS does not ask about SSNs, we proxy for SSN possession using a probabilistic model of likely unauthorized status, described in “Our preferred numbers.”

Researchers should also adjust for under-reporting of means-tested benefits in self-reported survey data such as the ACS. A substantial literature comparing survey responses to administrative records finds that program participation in benefits including SNAP and Medicaid is frequently under-reported, with roughly one-third of true SNAP recipients and one in five Medicaid participants in the ACS failing to report participation. We recommend that researchers incorporate a correction procedure based on program and survey-specific reporting rates as documented in recent literature. 

For programs measured using Census Bureau-modeled SPM estimates rather than self-reported survey responses, under-reporting corrections are unnecessary. Instead, because receipt is imputed from CPS data rather than observed, researchers should adjust estimates using program–specific take-up rates to reflect that imputed receipt captures modeled participation rather than actual take-up.  

3. Defining an “immigrant” and an “immigrant household”

Another major source of variation in estimates of immigrant benefit use arises from how researchers define an “immigrant household.” Common approaches include classifying a household as “immigrant” if the head of household is foreign-born or if any household member is foreign-born. These choices can have significant effects on rates and can blur other important legal distinctions between the documented and undocumented population and foreign-born naturalized citizens and non-citizens. Figure 3 shows the effect different “household” definitions have on rates of program use. 

Further variation stems from how “immigrant” itself is defined. Some analyses group all foreign-born individuals together, including naturalized citizens, while others restrict the population to non-citizens. As shown in Figure 2, naturalized citizens tend to exhibit benefit use patterns distinct from those of non-citizens, reflecting differences in eligibility, duration of residence, and labor market attachment. 

Finally, researchers often impose additional sample restrictions, such as limiting analyses to groups of immigrants who have resided in the U.S. for a certain number of years, households below the poverty line, or to families with children. These restrictions are appropriate for specific research questions but significantly affect reported levels of benefit use. Clear justification and transparent reporting of these choices are essential for meaningful comparisons across reports and immigrant groups.  

Figure 4 shows how the gap between non-citizen and citizen participation in Medicaid can flip directions depending on how and which part of the population is examined. This variability means that statistics about immigrant benefit participation can be cherry-picked to tell almost any story, which makes transparency about methodological choices especially important. For a more detailed picture of how sample restrictions can affect rates, see our companion analysis, which includes reports of benefit participation rates for a variety of demographic groups.

4. Deciding which benefits should count

Variation in research output also arises from decisions about whether to include programs such as the EITC, subsidized housing, and school lunch. While the EITC is distinct from forms of traditional cash assistance programs like SSI or TANF in that it is a tax credit rewarding work, it can be seen as a part of the broader safety net that supports low-income populations.⁵ As such, prior analyses such as those by CIS and Cato have included the EITC as a benefit program when measuring immigrant benefit use. 

A related source of divergence in the existing literature concerns the inclusion of social insurance programs such as Medicare and Social Security. Unlike means-tested programs, eligibility for social insurance is primarily determined by age and work history rather than income. Including these programs tends to skew comparisons between non-citizen immigrant and citizen populations, particularly given non-citizens’ younger age profiles and shorter average contribution histories.

While the Census distinguishes means-tested programs from social insurance, some analyses (such as Cato’s inclusion of these programs as entitlements) combine the two. Classifying Medicare and Social Security as benefits in these calculations also justifies the inclusion of unemployment insurance, workers’ compensation, and paid family leave, which fall under the same social insurance umbrella. While Cato includes Medicare and Social Security, it does not account for these latter programs, and neither do comparable reports by MPI or CIS, which exclude all forms of social insurance. Researchers should consider reporting social insurance separately to avoid conflating different forms of public support.

Our preferred numbers

In Table 1, we present our estimates of non-citizen versus citizen usage of safety net programs we consider reasonable to include under the “public benefit” umbrella. We find that while households with non-citizens appear to have higher participation in most public benefit programs, this pattern is largely explained by the eligible citizen members living in those households. At the individual program level, non-citizens use nearly every program at lower rates than citizens—free or reduced-price school lunch being the only exception. For further analysis and sample restrictions, see our companion piece.

We focus on the citizen/non-citizen distinction, and we analyze programs at the individual or household level depending on how each program is administered; for example, SNAP rates are calculated over households while Medicaid rates are calculated over individuals, so the rates for these programs are not directly comparable. 

For household-level programs, we report three rates. The first is the participation rate for all-citizen households. The remaining two columns report rates for households with at least one non-citizen member (these households may also include citizen members). The second column reports the share of these households that receive a benefit through any member, which includes benefits received by citizen members and non-citizen members. The third column reports the share of these households whose benefit receipt we determine is specifically attributable to the non-citizen member; this third column is effectively a subset of the second. More detail on how we determine non-citizen attribution is reported in Table 2 below.

To identify which non-citizens are likely unauthorized—and therefore ineligible for status-restricted programs—we use a probabilistic model that scores each adult non-citizen based on characteristics observable in the ACS. Non-citizen children are classified as eligible if at least one adult non-citizen in their household is classified as likely lawful permanent resident (LPR) or likely refugee, and as likely unauthorized otherwise. For programs where the benefit accrues to a defined sub-unit rather than the whole household, we additionally apply program-specific rules to determine whether the non-citizen specifically is the relevant recipient. We finally adjust for known under-reporting rates in the ACS for programs that are self-reported and for take-up rates for certain programs that are imputed in the SPM module of the ACS.  

Table 3 summarizes all adjustments by program. 

Conclusion

Estimates of immigrant public benefit participation vary widely because the methodological choices researchers make—how they define the unit of analysis, who counts as an immigrant, and which programs are included—exert enormous influence over results. As this brief demonstrates using estimates from the 2023 ACS, these decisions can shift reported participation rates dramatically.  

Using our recommended method, we find that households with non-citizens participate in most household-level programs more than those without, but this is largely due to benefit receipt by citizens in those households. When non-citizen participation is isolated within these households, the relationship inverts—non-citizens use these household programs at lower rates than citizens. For most individual programs with the exception of free or reduced-price school lunch, non-citizens have lower participation rates. For additional estimates and a more detailed discussion of these findings, see our companion analysis.

Statistics on immigrant benefit participation inform current policy debates regulating immigrant access to public benefits, including proposed changes to interpretations of PRWORA and new guidance on the public charge doctrine. They also bear on broader discussions about the direction of immigration policy. When headline figures fail to account for program-specific eligibility rules and the co-residence of citizens and non-citizens within participating households, the resulting estimates do not accurately characterize immigrant participation in the safety net. Such figures risk both overstating the fiscal impact of immigration and misidentifying which populations actually receive public support.