Word to the mother(tongue): language access and Medicaid for Limited English Proficient migrants


 Limited language proficiency among migrants has been shown to limit migrants from various aspects of everyday life, from the labor market to government services. In the USA, language access laws have been enacted to help provide Limited English Proficient (LEP) individuals and households access to public benefits. The most extensive efforts in language access are in health care, with numerous states requiring Medicaid agencies to provide interpreters or translated documents. In this paper, I take advantage of heterogeneity in state-level language access laws to examine whether these efforts do indeed increase Medicaid take-up rates. I find that language access improves Medicaid take-up rates among LEP migrants without crowding out private health insurance; in fact, private health insurance coverage improves with the enactment of language access laws. There is some variation in efficacy across states, with some evidence that California and New York are the main drivers of the increased take-up rates. Lastly, I find that even though many of the language access laws primarily target Spanish speakers, the laws might not be as helpful to Spanish-speaking migrants.


Introduction
In a typical Economics 100 class, the concept of adverse selection is usually introduced, with health insurance as a primary example. The example is particularly relevant in the United States, as the relationship between health insurance coverage (or lack thereof) and the high costs of medical care in the U.S. is scrutinized. To address this problem of adverse selection with regards to health insurance, an emphasis has been placed on insuring everyone; to help poor individuals obtain health insurance, the U.S. has made an effort to expand Medicaid. However, expanding Medicaid coverage via increased eligibility levels could still leave a segment of the poor population without Medicaid -poor Limited English Proficient (LEP) individuals and households will still not have insurance if they are unable to understand the Medicaid application procedure. This is not a trivial portion of the population; LEP migrants made up 9% of the entire U.S. population (not just migrant population) as recently as 2011 (Whatley and Batalova, 2013).
In theory, an inability to speak or read English should not prevent access to government services. Title VI of the 1964 Civil Rights Act states that "No person in the United States shall, on the ground of race, color, or national origin, be excluded from participation in, be denied benefits of, or be subject to discrimination under any program or activity receiving Federal financial assistance" (Pub. L. 88-352, title VI, Sec. 601, July 2, 1964, 78 Stat. 252).
Executive Order 13166, issued in 2000, clarified Title VI's national origins aspect, stating "each Federal agency shall examine the services it provides and develop and implement a system by which LEP persons can meaningfully access those services consistent with, and without unduly burdening, the fundamental mission of the agency" (Executive Order 13166).
According to these two laws, language access, or translation services, should be provided in the case of Medicaid enrollment. Title VI applies broadly, regardless of whether federal funding is direct or indirect and the amount of funds received, and "in the health care context, this includes virtually all: hospitals; doctor's offices; nursing homes; managed care organizations; state Medicaid agencies; home health agencies; health insurance providers; and social services organizations" (Youdelman, 2009, p.2). In reality, Perkins and Youdelman (2008) argue that " [D]espite such federal requirements, lack of knowledge and enforcement leave millions of LEP individuals without meaningful access to health care" [p.4].
To be fair, all states have passed some degree of language access laws in the health care industry. The amount of language access varies greatly across states, however. Some states have enacted comprehensive Title VI-like laws for state agencies, while some states merely require information pamphlets on a variety of diseases be translated. Provisions also include rules on interpreter certification and facility licensure, access plans, and rights at hearings. To see if language access plays a role in the goal of extending health insurance coverage, I take advantage of the heterogeneity in state-level language access laws to analyze whether language access increases Medicaid take-up rates for LEP migrants. Due to the similarities between the State Children's Health Insurance Program (SCHIP) and Medicaid 1 , "Medicaid" from now on means non-Medicare, non-military public health insurance. To my knowledge, this is the first paper to look into the effects of language access on access to government services. This is also the first paper to examine individual outcomes from language access/barriers in health care using public use microdata. My main result is that language access improves Medicaid take-up rates among likely LEP migrants. What is more, there is no evidence of government crowd-out; private health insurance coverage even increases after the passage of language access laws. The effect of the different state laws varies across states, and the main results could be driven by New York and California. Furthermore, even though many of these translation services primarily target Spanish-speakers, the results suggest that the laws are not more helpful to Spanish-speaking migrants.
This fits into the scarce literature on the importance of language in accessing health care.
There is some some literature on adverse health outcomes from being unable to communicate with doctors and nurses effectively, primarily published in the biomedical and life science fields. These articles usually point to how language barriers contribute to a low level of understanding of diagnoses and treatment options (Hadler et al., 2013;Khan et al., 2013).
In a technical report, Quan and Lynch (2010) examines how costly language barriers are when miscommunication leads to medical malpractice. Analyzing the medical malpractice claims of a malpractice carrier that insures in four states, Quan and Lynch (2010) identified 35 closed claims that involved language barriers from January 2005 through May 2009, 2.5% of total claims reviewed by the malpractice carrier. These claims resulted in over $5 million in damages, settlements, and legal fees. The lack of malpractice carriers providing interpreters to hospitals to reduce these types of claims could, ironically, be a case of adverse selection.
Providing interpretors would likely increase the costs of using the malpractice carrier, and hospitals that deem themselves to have little need for interpreters would not pay higher costs for a malpractice carrier that offers interpretors the hospitals considers unnecessary. This affects the malpractice carrier's risk pool, as it insures hospitals that are increasingly more likely to need interpretors. In this analogy, hospitals with little need for interpretors are healthy people, and malpractice carriers with interpretors are health insurance agencies.
There is also some literature regarding immigrants' lower rates of obtaining health insurance. A Migration Policy Institute report by Ku and Jewers (2013) illustrates the disparity of health insurance and health care access between immigrants and native-borns. In 2011, 62% of non-citizen low-income adults (ages 19 to 64) were uninsured, compared to 42% of naturalized citizens and 35% of native borns (Ku and Jewers, 2013, fig.2). Foreign-borns made approximately 20% fewer office-based medical visits than native-borns (Ku and Jewers, 2013, fig.3,4). Ku and Jewers (2013) point out that while these differences could be due to income differences and eligibility restrictions placed on non-citizens, they also stress the importance of language barriers, particularly with regards to communication between patients and clinicians. Feinberg et al. (2002) more carefully examine the language access issue in obtaining Medicaid, examining the effects of language barrier on education and outreach, and initial enrollment. They find that the major barriers to Medicaid enrollment related to "'know-how' -that is, knowing about the Medicaid program, if their child was eligible, and how to enroll" (Feinberg et al., 2002, p.5).

Data and Methodology
There are three main pieces to the puzzle of whether language access improves Medicaid take-up rates for LEP migrants: if the individual is enrolled in Medicaid, if the individual is an LEP migrant, and whether the state the individual resides is treated by having language access laws that would affect Medicaid accessibility. The data come from the Current Population Survey (CPS), which has rich data about health insurance coverage. The sample period covers 1994-2008. Unfortunately, the CPS has no English proficiency question, so English proficiency has to be inferred from other survey questions. First, I look at where an individual was born. Individuals born in countries that do not have English as an official, primary, or widely spoken secondary language clearly are more likely to be LEP. This standard for identifying LEP individuals is obviously limited; some countries, such as India, do not have a large English-speaking population in general, but educated individuals almost all speak English 2 . Further, Chiswick and Miller (2001) find evidence that education increases destination-language skills. Thus, I do not consider respondents with a high school degree or higher as LEP. Lastly, considering the "critical period" of second language acquisition as in Bleakly and Chin (2010;2004) 3 , I use an individual's year of arrival, age, and year of survey to determine the individual's age of arrival. If the individual arrived in the United States prior to nine years of age, I do not consider them as LEP. Unfortunately, this identification strategy probably misidentifies some LEP individuals in both directions; some individuals identified as LEP by these standards might be fluent in English, for example by moving to an English-speaking, non-U.S. country prior to age nine. Conversely, some individuals identified as being proficient English speakers by this strategy might actually be LEP, as might happen in ethnic enclaves (Chiswick and Miller, 2005).
At its most basic, this paper is analyzing whether legally mandated translation services assist LEP adults in accessing health services for themselves and their children. Translation assistance does not always come from policy-mandated translators or translated documents, however. Someone who is not fluent in English could live in a household that has a proficient English-speaker who can provide translation services for them; language access laws should not affect these types of LEP individuals. Alternatively, children born in an English-speaking country might be proficient in English, but have parents who are not English proficient. In this scenario, the child is proficient in English, is too young to understand the application process, has LEP parents, and therefore would likely be affected by language access laws.
To address these two cases, I look at whether all adults, defined as being older than 16 years of age, in a household are LEP. If at least one adult in the household is not LEP, than all members in the household are recoded as not being LEP. If all adults in the household are defined as LEP, the children are all recoded as being LEP.
To determine whether a state is considered treated by a language access state in a particular year, I go through the list of state laws provided in Perkins and Youdelman (2008) and find state laws that should, in theory, increase access to Medicaid. I look for laws that require Medicaid agencies to provide translated documents, translators throughout the application process, or non-English advertisements and notifications of eligibility. There are 28 "states" 2 I get information about a country's languages spoken from the CIA Factbook. 3 According to the "critical period hypothesis", individuals with exposure to a language during the critical period can acquire the language up to native ability. Exposure after the critical period, however, and an individual is less likely to be proficient. The authors consider ages 0-9 years as the critical period.
(27 states and the District of Columbia) that, as of 2008, have some degree of improved language access to Medicaid 4 ; of these 28, I was able to find the date of enactment for 22 states. See appendix A for the list of states with improved language access to Medicaid, what the content of these laws are, and when the laws were enacted.
Careful attention is paid to two particular types of language access laws that might have minimal effects on Medicaid take-up rates. The first type of laws are laws that require Medicaid providers and/or affiliates to have an access plan that addresses language access issues. The language of these laws generally do not explicitly require that the organization improve access, they merely require a plan that includes the organization's efforts to address access needs. In other words, it seems that a plan stating that limited to no efforts are being made to address the needs of LEP migrants seeking coverage would still be in compliance with the law. Any minimal effort to actually provide translation services would have minimal effects on Medicaid take-up rates. The second type of laws are laws that affect certain segments of Medicaid seekers or enrollees. For example, in Indiana and Mississippi, language access laws were passed for health maintenance organizations (HMOs), but not for Medicaid carriers as a whole. If some, but not all, Medicaid benefits are distributed via HMOs, the impact of these language access requirements might be limited on Medicaid take-up rates. In Tennessee, only those who are already enrolled in Medicaid can obtain translation assistance; would this increase Medicaid take-up rates, since LEP migrants seeking Medicaid coverage do not receive government mandated language access? I consider the states as treated for the purpose of this analysis, to estimate a lower bound of the treatment effect. Appendix A makes note of the states with laws that possibly have a limited effect on Medicaid take-up rates.
Pre-treatment characteristics are compared between states that have passed Medicaid access-improving language access laws and states that have not to see if there are any systemic differences between the two. I use 1970 as the year to compare pre-treatment characteristics because the immigration literature uses the stock of migrants in this year as an instrument to control for the endogeneity of immigrants' migration decisions (Altonji and Card, 1991).
The variables chosen reflect the size of the migrant population, particularly with respect to migrants who are less likely to speak English, and economic conditions that migrants might consider in their migration decision. The party affiliation of the governor is used to capture the political atmosphere that might encourage or discourage the passage of migrant-friendly laws. Notes: Democrat = 1, Republican = 2 for "Governor party". * Includes four states whose date of passage could not be found. There is no statistically significant difference between the means for all the variables.
An important concern for this paper is the exogeneity of the language access laws. To analyze the effects of language access on Medicaid take-up rates for LEP migrants, I use a difference-in-differences strategy, comparing Medicaid enrollment between LEP and non-LEP migrants before and after language access laws are enacted. I run a linear probability model on the following regression to evaluate the effects of language access laws on the Medicaid take-up rates of LEP migrants: where M ED i,t is an indicator variable that is one when an individual i has Medicaid in time period t and a zero otherwise, LA i,t is the treatment variable that is one when the individual lives in a state that has a Medicaid-access-improvement law as described below at time t and zero otherwise, and LEP i is a dummy variable that is one when an individual is  does not necessarily need to be an American citizen to obtain Medicaid benefits -in all states, immigrants who have been in "qualified" 5 immigrants status for 5 years or more are eligible, while some states cover benefits even if an immigrant is "not-qualified." Obtaining an immigrant's status is all but impossible from the CPS, so I do not attempt to limit the sample by "qualified" immigrant status. Medicaid take-up rates for these individuals do not follow a perfectly parallel trend between 1994-2008, particularly in states with small migrant populations. The variation for these states with small migrant populations is likely due sampling, though; going from 0 out of 1 migrants enrolled in Medicaid in one year to 2 out of 2 the next year is as dramatic of a change in take-up rates as possible, but is probably

Results
Initial results (table 3, column 3) show that language access laws increased Medicaid take-up rates among LEP individuals by 1.8 percentage points. Approximately 37% of the sample has Medicaid, so an increase of 1.8 percentage points is equal to an increase of almost 5%.   One source of the increase in Medicaid take-up rates could be from private insurance users switching to Medicaid. The language access laws are doing nothing to expand health insurance coverage if these laws result in the government crowding out private insurance, by encouraging private insurance users to use public insurance (Brown et al., 2007;Brown and Finkelstein, 2008  This is supported by the lower interaction coefficient for the lowest income cutoff -these individuals probably would not be able to afford private insurance, even if they preferred it to public insurance.   with accessing Medicaid. A major concern is that even though non-Hispanic LEP migrants have little trouble with Medicaid take-up, language access is further improving their access to Medicaid, while Spanish-speakers who are not defined as LEP do worse after language access laws have been passed, according to the "all Spanish-speaking countries" and "Puerto Rico and Mexico" definition of Spanish-speakers. How much do translators help with communicating diagnoses for LEP migrants? Migrants will continue to arrive in large numbers and not all these immigrants will be healthy and proficient in English; any concerns with public health will need to consider how to approach the language barriers that exist.  (column 2). Focusing in on states that have passed a Medicaid access-improving language access law, the employed-to-unemployed ratio is statistically significant at the 10% level (columns 3 and 4). The direction is negative, perhaps a little counterintuitive -the more people that are employed for each unemployed individual, the fewer health care language access laws there are. One might expect that a higher employed-to-unemployed ratio would encourage migrants to move to a state, thereby encouraging the passage of more laws. The governor's party stays positive and statistically significant; having a Republican governor in 1970 increases the number of health care language access laws in 2008. Aside from the governor's party, state characteristics seem to do little to predict migrant friendliness with respect to the number of health care language access laws.
The second test's results are in table 11. Results from regressing the date of passage on the covariates are in columns (1)-(3). Column (1) contains all states with known dates; column (2) takes out California, due to its early adoption of Medicaid access-improving language access laws; and column (3) contains only states that have passed Medicaid access-improving language access laws in the 1994-2008 sample. Governor's party is statistically significant again, but this time negative. Having a Democratic governor will lower the number of health care language access laws, but result in earlier adoption of Medicaid access-improving language access laws. When limiting the sample to states that passed laws in the 1994-2008 sample, and the percent of the population born in a non-English speaking birthplace, the percent of the population with English as a mother tongue, and total family income become statistically significant at the 10% level, while governor's party loses significance.
The direction of some of these coefficients are again perhaps counterintuitive -having more a higher percent of the population with a non-English speaking birthplace is more likely to encourage migrants, thereby putting more pressure to pass a Medicaid access-improving language access law, but the coefficient indicates a later date of passage with a higher percent of the population born in a non-English speaking country.
Since there are only 22 states with known dates for the passage of Medicaid accessimproving language access laws, a regression with all the covariates might not be meaningful. To address this, separate regressions are run where only one covariate is used as the explanatory variable; results are in table 11 columns (4)-(6). Total family income is barely statistically significant at the 10% level for all 22 states, but loses significance as soon as California is dropped from the sample. R 2 values are reported to show the explanatory power of each covariate. Only governor's party can explain more than 20% of the variation. Since the governor's party is the most consistent explanatory variable for predicting the date of passage of Medicaid access-improving language access laws, but predicts something almost opposite to the first test for the number of health care language access laws, I conclude that the passage of the Medicaid access-improving language access laws is unpredictable.  Notes: Dependent variable is date of passage of Medicaid access-improving language access law. Columns (4)-(6) have covariate as only explanatory variable; R 2 in brackets. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Democrat = 1 and Republican = 2 for "Governor's party".