Medicaid spent $355 billion in 2009 on medical and long-term care and covered approximately 62 million low-income people, 48% of whom were children, 23% adults, 7% aged, 14% disabled, and 8% “others.”1 Currently, Medicaid eligibility is limited to citizens and documented residents of limited incomes within targeted groups—families with dependent children; pregnant women; and persons who are aged or disabled. A nondisabled single adult with no children is generally not eligible but, beginning in 2014, the Affordable Care Act extended eligibility to all adults less than 65 years with incomes up to 138% of the Federal Poverty Level. In 2009, 15.7% of the total US population was covered by Medicaid, including 26.5% of the Hispanic population.2
In 2010, there were an estimated 11.2 million undocumented immigrants accounting for 3.2% of the US population.3,4 Many Americans believe that Medicaid spending on undocumented immigrants is excessive, that it is partially responsible for soaring medical costs, and that it should be prohibited.5–7 But other Americans believe that spending for Medicaid for undocumented immigrants should be expanded.8 Little evidence exists on the dollar amount of Medicaid use by undocumented immigrants. A Congressional Budget Office report cites an Oklahoma Health Care Authority study, showing that less than 1% of Medicaid spending went to undocumented immigrants in that state.9
Only a few studies estimate the magnitude of the disparity in health insurance coverage between undocumented immigrants and natives, and even fewer consider disparities for Medicaid.10,11 The dearth of studies is due, in part, to the fact that immigration status is generally not identifiable in publicly available surveys. We are not aware of any studies that use nationally representative data on undocumented workers outside of agriculture.
This study addressed the following three questions: Were undocumented or documented households more or less likely to participate in Medicaid? How large were the differences? What accounted for those differences? We hypothesized that the differences comported with legal restrictions. Undocumented worker households are not eligible for Medicaid, except for (1) emergencies, (2) pregnancy and delivery, and (3) children born in the United States. Whereas we had no measure for emergencies, we hypothesized that our variables for sex and children in the household would be important predictors of Medicaid use.
We used data from the National Agricultural Workers Survey (NAWS). The great advantage of the NAWS is that it contains information on the legal status of farm workers; we did not have to estimate that status with simulation models as other investigators have been required to do given the limitations of their data.4,10 Despite this significant advantage, the NAWS is relatively underutilized. For example, whereas government reports provide some descriptive statistics on Medicaid use in the NAWS, we are not aware of any study with multivariable analysis of Medicaid.12
This study adds to the growing literature on related subjects by occupational health researchers. With rising national health care costs garnering increasing public attention, greater numbers of studies have addressed the use and costs of health services by employees. Berger et al,13 Tang et al,14 and Johnston et al15 analyzed medical costs and claims data for employees with osteoarthritis, cancer, and acute coronary syndrome, respectively. Medicaid, specifically, has been studied for two reasons—some of the true costs of occupational injuries and illnesses are shifted to Medicaid, and some evidence suggests that employers that do not offer employer-sponsored health plans shift the cost of care of nonoccupational illnesses onto Medicaid.16,17 In addition, whereas there have been few studies on Medicaid use by undocumented workers, many more have addressed insurance and access to care issues by all immigrants, including those in agriculture.18,19
The NAWS is a repeated annual cross-sectional survey of nationally representative samples of hired farm workers that began in 1989.20 Interviews were conducted three times per year, collecting detailed information on individual and household demographics, employment, health history, and income. From 1989 through 2009, 52,479 in-person interviews were completed in 467 counties across 40 states.20 Our sample consisted of all hired farm workers from 1993 to 2009 with nonmissing data for all relevant variables, except income for which we allowed some missing values (explained below). Our variables included Medicaid receipt, current legal status, ethnicity, race, age, marital status, the number of children, education, and income. The first 4 years were excluded from analysis because the Medicaid variable was only available from 1993 onwards. The final sample consisted of 41,324 farm workers, aged 14 to 90, including 21,273 documented and 20,051 (48.5%) undocumented workers. The NAWS composite weights were used to obtain population estimates.
The NAWS data are de-identified secondary data and qualify under exemption #4 of the National Institutes for Health categories, indicating that a review by a Human Subjects Ethics committee is not required. These data are available to the public.20
The outcome variable, Medicaid, was binary and equaled 1 if the respondent or anyone else in his or her household used Medicaid in the preceding 2 years. The independent variable of interest, undocumented, was also binary and equaled 1 if the farm worker was undocumented, that is, a noncitizen who was neither a legal permanent resident nor a holder of other work authorization. An undocumented or documented household was defined as the household to which the worker-respondent belonged.
Covariates included indicator (dummy) variables for year, sex, age intervals, race and ethnicity, region, education categories, marital status, the number of children, and individual and family income intervals. Age was categorized in 5-year bins with the last category being older than 65 years of age.
The sample contained 82% with Hispanic ethnicity; 44% identified as white, 49% as “other,” and the remainder as either black, American Indian/Alaskan Native/Indigenous, Asian, or Pacific Islander. We classified the sample into the following four groups: white Hispanic, white non-Hispanic, nonwhite Hispanic, and nonwhite non-Hispanic.
The NAWS defined six geographic regions. California was its own region and comprised one third of the sample. The five additional regions included East (16%), Southeast (13%), Midwest (19%), Southwest (7%), and Northwest (12%). Education was divided into the following four categories: primary school graduate, some high school, high school graduate, and some college or more. Marital status equaled 1 for “currently married or cohabiting” and 0 otherwise. The number of children referred to those residing in the United States; undocumented workers often leave families behind. In our data, more than half of undocumented married workers did not live with their spouse in the United States.
We used the NAWS-created family income variable with the same 15 income categories every year. The NAWS coded incomes as the average within each income category, except the bottom and top codes. The bottom income code “$250” corresponded to “less than $500,” whereas the top two income codes “$37,500” and “$50,000” corresponded to “$35,000 to $49,999” and “$50,000 or more,” respectively. We converted the NAWS average incomes into 2009 dollars using the Consumer Price Index for All Urban Consumers and combined into our own four income categories (less than $11,250, $11,250 to $22,500, $22,501 to $32,499, and $32,500 or more). More than 20% of the sample had missing income. Rather than discarding such a large percentage of the sample, we created a dummy variable for missing income and recorded the 20% missing values with the sample modal value for each year. Finally, the number of children in the household was categorized into the following four groups: no child, one child, two children, and three or more children.
We examined Medicaid receipt and undocumented work status with logistic regressions that accounted for complex survey sample design with stratification, clustering, and weighting using Stata12 (StataCorp LP, College Station, TX). We first ran 12 logistic regressions with Medicaid as the dependent variable, undocumented as the key independent variable, and various combinations of covariates. These 12 combinations were constructed on the basis of hypotheses from the legal requirements for Medicaid and the study by Dubard and Massing,21 suggesting special attention to variables for children, sex, martial status, income, and region. The first regression contained only undocumented. The next five regressions contained undocumented and only one of the other “special attention” categories. The final six regressions included all basic demographic variables (sex, race/ethnicity, age, and marital status), geographic variables (regions), 16 of 17-year indicator variables, and systematically added other covariates. Our Table 2 presents results on four of the most informative of the 12 regressions. (Other regression results are available on request.)
Table 1 reports descriptive statistics on selected variables within documented and undocumented households separately. (A separate Table 1 with all variables is available from the corresponding author). Medicaid receipt rate for undocumented households was 10.4 percentage points less than that of documented households (P = 0.007). The groups differed in most characteristics. Undocumented farm workers were more likely to be male, Hispanic, and younger; to lack high school or higher degrees; to be single; to have no kids, and to have lower income. All of the above differences were statistically significant.
Two time-trend plots are available from the authors. The first showed that Medicaid receipt increased for both groups from 1993 to 2009. The plot also revealed a slight narrowing of the gap in Medicaid receipt between the two groups. A second plot showed that this narrowing coincided with an increase in the percentage of undocumented households with children in the 2000s.
Table 2 presents logistic regression results. The first regression (first column of numbers) included only undocumented and indicator (dummy) variables for years; the odds ratio was 0.48 (95% confidence interval [CI], 0.40 to 0.56). This “raw” odds ratio indicated that undocumented workers were 52% less likely than documented workers to report Medicaid benefits. In column 2, which included demographic, regional, and annual covariates but not variables for children, the odds ratio fell to 0.34 (95% CI, 0.29 to 0.40). But in regression #3, which included only children and no other covariates, the odds ratio climbed to 0.86 (95% CI, 0.73 to 1.02). In addition, the upper bound CI crossed 1.0 in regression #3, and the confidence intervals did not overlap between regressions #2 and #3. Regression #4 included all covariates and resulted in an estimated odds ratio of 0.56 (95% CI, 0.47 to 0.66). The children covariates were highly statistically significant in both regressions #3 and #4 and in every regression that included them. Moreover, odds ratios were large. In our preferred regression #4, households with one child and households with three or more children were 15 and 29 times more likely to receive Medicaid than childless households.
These findings—very large odds ratios, statistical significance on the children covariates, and findings that by merely adding or deleting covariates for children had such a striking effect on the odds ratio for the undocumented—were noteworthy. We, therefore, ran additional regressions that allowed for interactions between undocumented and the children covariates. Results for our preferred regression—that included all the covariates in Table 2—appear in Table 3. The regression in Table 3 estimated that the odds of childless undocumented households receiving Medicaid benefits were 12% that of childless documented households. On the contrary, having a child dramatically increased the odds of Medicaid receipt—the odds of receipt for one child documented households were 6.57 times greater than that of the childless documented households. The effect was greater for undocumented households—the odds of receipt for one child undocumented households were (0.12*6.57*10.63 = 8.38) or 8.4 times greater than that of the childless documented households. All of these results on undocumented, and the interaction terms were strongly statistically significant.
We now consider control variables. Women were more likely to receive Medicaid than men (P < 0.01), independent of the presence of children. All the minority groups were more likely to receive Medicaid than the white non-Hispanic group. Being married was positively associated with Medicaid receipt. Regression #4 in Table 2 indicated that the odds of Medicaid receipt for those with some high school education were 1.17 times greater than the odds for those with only primary school education, whereas those with a high school diploma or some college were too imprecise to be conclusive (the upper CIs were greater than 1.0). Higher income, either for individuals or families, was associated with lower odds of Medicaid receipt. Medicaid receipt also varied by region—California had the highest odds.
Results from the NAWS indicated that without adjustments for covariates, approximately 22.6% (n = 4805) of documented households and 12.2% (n = 2447) of undocumented households received Medicaid benefits over 2-year intervals from 1993 to 2009. The difference between these two “raw” percentages was 10.4 percentage points (n = 2358) and corresponded to an odds ratio of 0.48 for undocumented households (P = 0.007). In regressions that alternately excluded and included variables measuring the number of children in the family, the odds for undocumented ranged from 0.33 (excluded) to 0.86 (included). In our preferred regression that included all control variables, the odds ratio was 0.56 on undocumented. All odds ratios were strongly statistically significant. Additional analyses were conducted on interactions between the children variables and undocumented. The presence of children had a disproportionately larger effect for undocumented than documented worker households. We identified the variables measuring children as those most responsible for changes in estimates of differences in Medicaid receipt by documented and undocumented households. We also found that female respondents and married persons were more likely to report Medicaid use, and respondents with relatively high income ($35,000 or more) were less likely to report Medicaid use.
Our results on children, females, marriage, and high income were consistent with Medicaid eligibility laws. The additional result that there was a disproportionate effect of children on undocumented versus documented households may be explained by the relative difference in access to employer-sponsored health insurance. Studies, including one using the NAWS, show that documented households were much more likely than undocumented ones to have employer-sponsored health insurance (EPHI).10,11,22 Given that EPHI tends to be more generous than Medicaid and the physician pool is larger for EPHI than Medicaid, low-income families with EPHI are likely to choose EPHI over Medicaid. But undocumented households, even with citizen-children, rarely have that choice.
Our findings are consistent with the literature. Ortega et al23 estimated that undocumented immigrants were roughly half as likely as US-born Hispanics to visit emergency departments in Los Angeles. DuBard and Massing21 analyzed “Emergency Medicaid” spending in North Carolina between 2001 and 2004 using state Medicaid administrative data. They found that 99% of recipients were undocumented, 95% were female, and 82% to 86% of the spending was for pregnancy and neonatal care.
This study has strengths and limitations. The major strength is that the NAWS contains (we believe the only) nationally representative sample of undocumented workers and it is large. Another strength is that the NAWS data pertain to households. Given that almost 50% of Medicaid recipients were children, households rather than individual adults are the relevant category.1 Finally, the NAWS has detailed information on demographics and income, a critical parameter used in determining eligibility for Medicaid. One limitation is that farm workers are different from other immigrants, detracting from our ability to generalize these findings to all undocumented households. For example, farm worker immigrants are more likely to be seasonal/migrant, nonsalaried workers, and work in rural areas than nonfarm worker immigrants; thus, access to health care and information, lifestyle, and other unobservable factors may differ significantly across these two groups. Another limitation is that the composition of the farm worker populations has likely changed from 1993 to 2009. Although we have controlled for race, year, and age factors, cohort effects have not been addressed, which would require panel data. Using our findings to forecast Medicaid use into the future would be problematic.
Political controversy swirls around whether and how much undocumented immigrants use government programs. Evidence in the scientific literature is lacking, in part, because few data are available on undocumented immigrants. We analyzed a nationally representative sample of documented and undocumented farm workers (N = 41,342) from 1993 to 2009. Our analysis revealed that undocumented farm worker households were roughly half as likely as documented households to use Medicaid. We also found that undocumented use was strongly influenced by the presence of children, as expected given legal allowances.
1. Centers for Medicare & Medicaid Services. Annual Medicare and Medicaid Statistical Supplement Report 2011 Edition, CMS Office of Information Products and Data Analysis. Available at: http://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/MedicareMedicaidStatSupp/2011.html
. Published 2011. Accessed April 10, 2013.
2. US Census Bureau, Statistical Abstract of the United States: 2012. 131st ed. Washington, DC. Available at: http://www.census.gov/compendia/statab/2012/tables/12s0155.pdf
. Published 2011. Accessed August 27, 2013.
3. U.S. Census Bureau, Statistical Abstract of the United States. 2012. 131st ed. Washington, DC. Available at: http://www.census.gov/compendia/statab/2012/tables/12s0007.pdf
. Published 2011. Accessed February 13, 2013.
4. Passel J, Cohn D'V. Unauthorized immigrant population: national and state trends, 2010. Pew Research Hispanic Center. Available at: http://www.pewhispanic.org/2011/02/01/unauthorized-immigrant-population-brnational-and-state-trends-2010/
. February 1, 2011. Accessed February 11, 2013.
5. Wolf R. Rising health care costs put focus on illegal immigrants. USA TODAY. January 22, 2008. Available at: http://usatoday30.usatoday.com/news/washington/2008-01-21-immigrant-healthcare-N.htm
. Accessed November 7, 2014.
6. Cover M. Illegal immigrants account for $10.7 billion of nation's health care costs, data show. CNS News. July 29, 2009, 5:57 PM. Available at: http://cnsnews.com/news/article/illegal-immigrants-account-107-billion-nation-s-health-care-costs-data-show
. Accessed August 23, 2013.
7. Rector R, Richwine J. The Fiscal Cost of Unlawful Immigrants and Amnesty to the U.S. Taxpayer. Heritage Foundation Special Report #133 on Immigration. Washington, DC. Available at: http://www.heritage.org/research/reports/2013/05/the-fiscal-cost-of-unlawful-immigrants-and-amnesty%20to-the-us-taxpayer
. May 6, 2013. Accessed August 23, 2013.
8. Chesler ER. Notes: Denying undocumented immigrants access to Medicaid: a denial of their equal protection rights. The Boston University Public Interest Law Journal. 2008;17:255–287.
9. Congressional Budget Office. The Impact of Unauthorized Immigrants on the Budgets of State and Local Governments. The Congress of the United States—Congressional Budget Office. Available at: http://www.cbo/gov/sites/default/files/12-6-immigration.pdf
. Accessed November 7, 2014.
10. Zuckerman S, Waidmann TA, Lawton E. Undocumented immigrants, left out of health reform, likely to continue to grow as share of the uninsured. Health Aff. 2011;30:1997–2004.
11. Goldman DP, Smith JP, Sood N. Legal status and health insurance among immigrants. Health Aff. 2005;24:1640–1653.
12. Carroll DJ, Samardick R, Bernard S, Gabbard S, Hernandez T. Findings from the National Agricultural Workers Survey (NAWS) 2001–2002: A Demographic and Employment Profile of United States Farm Workers. Washington, DC: US Department of Labor, Office of the Assistant Secretary of Policy; 2005. Research Report No. 9. Available at: http://www.doleta.gov/agworker/report9/naws_rpt9.pdf
. Accessed February 13, 2013.
13. Berger A, Hartrick C, Edelsburg J, et al. Direct and indirect economic costs among private-sector employees with osteoarthritis. J Occup Environ Med. 2011;53:1228–1235.
14. Tang DH, Alberts DS, Nevins R, et al. Health care expenditures, hospitalizations, and productivity associated with cancer in US employer settings. J Occup Environ Med. 2012;54:1453–1460.
15. Johnston SS, Curkendall S, Makenbaeva D, et al. The direct and indirect cost burden of acute coronary syndrome. J Occup Environ Med. 2011;53:2–7.
16. Leigh JP, Marcin JP. Workers' compensation benefits and shifting costs for occupational injury and illness. J Occup Environ Med. 2012;54:445–450.
17. Waddoups CJ. Public subsidies of low-wage employment: the case of uncompensated health care. J Econ Issues. 2006;40:813–824.
18. Frank AL, Liebman AK, Ryder B, et al. Health care access and health care workforce for immigrant workers in the agriculture, forestry, and fisheries sector in the Southeastern US. Am J Ind Med. 2013;56:960–974.
19. Panikkar B, Woodin MA, Brugge D, et al. Characterizing the low wage immigrant workforce: a comparative analysis of the health disparities among selected occupations in Somerville, Massachusetts. Am J Ind Med. 2014;57:516–526.
20. US Department of Labor. National agricultural worker survey. Available at: http://www.doleta.gov/agworker/naws.cfm
. Accessed February 14, 2013.
21. Dubard C, Massing MW. Trends in emergency Medicaid expenditures for recent and undocumented immigrants. JAMA. 2007;297:1085–1092.
22. Kandilov AMG, Kandilov IT. The effect of legalization on wages and health insurance: evidence from the National Agricultural Workers Survey. Appl Econ Perspect Policy. 2010;32:604–623.
23. Ortega AN, Fang H, Perez VH, et al. Health care access, use of services, and experiences among undocumented Mexicans and other Latinos. Arch Intern Med. 2007;167:2354–2360.