U.S. Immigration Policy Regimes and Physical Disability Trajectories Among Mexico–U.S. Immigrants

Abstract

Objectives

Although immigration policies have shifted dramatically over the course of U.S. history, few have examined their role as a source of health heterogeneity among immigrants. We address this gap by evaluating whether exposure to U.S. Immigration Policy Regimes (IPRs) corresponds with later-life disability disparities among Mexico–U.S. migrant women and men, and assess the degree to which observed differences may also be associated with immigration policies and occupational composition.

Method

We analyze 8 waves of data from the Hispanic Established Populations for the Epidemiologic Study of the Elderly (3,044 individuals and 14,474 observations from 1993/1994–2013). Using hierarchical linear models, we estimate trajectories of physical disability associated with gender, occupation, and IPR.

Results

We find differences in disability trajectories by IPR. Associations are not different between men and women, and are not mediated by occupational composition. We also observe an additive effect for certain occupations among women, but not among men.

Discussion

Findings demonstrate that exposure to different IPRs is associated with disparate disability trajectories among Mexico–U.S. migrants. Future research is needed to contextualize the role of IPRs amid other mechanisms of gendered racialization that powerfully contribute to cumulative health differences across the life course.

How does U.S. immigration policy shape the accumulation of disabilities among aging Mexican Americans in the United States? Scholars have documented the importance of relative social position for the functional health of Mexico–U.S. migrants, both in terms of migration selection mechanisms and the exposure to health-averse conditions (Salinas, Su, & Al Snih, 2013). Health researchers have generally treated selection and exposure mechanisms as invariant across periods marked by different immigration policies, although some have recently called for closer attention to the role of changing structural factors in shaping health outcomes (Viruell-Fuentes, Miranda, & Abdulrahim, 2012). Recent research documents the importance of naturalization and life course timing (Gubernskaya, Bean, & Van Hook, 2013), but none to date have examined the role that changing immigration policies have in shaping physical disability among Mexico–U.S. migrants. The present study draws on insights from research linking changes in immigration policy to varying labor market conditions and the composition of migrant cohorts (Donato, Wakabayashi, Hakimzadeh, & Armenta, 2008; Durand, Massey, & Charvet, 2000; Garip, 2017) to guide an investigation of the relationship between immigration policy and disability.

Migration scholars have theorized that both the composition of migrant cohorts and their exposure to conditions in receiving contexts are shaped by Immigration Policy Regimes (IPRs), or broad sets of social structures organizing the racialization and incorporation experiences of immigrants in terms of their possibilities for citizenship, work, and participation in everyday economic, cultural, and political life (Faist, 1995; Sainsbury, 2006). IPRs correspond with the kinds of selection and exposure mechanisms theorized to drive functional health differences among aging Mexican American immigrants. Investigating the relationship between IPRs and physical disability trajectories provides an avenue to examine health consequences of politically contingent structural conditions in historical time (Gentsch & Massey, 2011; Viruell-Fuentes et al., 2012). Assessing the impact of IPRs on disability is also important because disparities may contribute to the reproduction of wider inequalities by exposing immigrants and their families to additional hardships, and understanding these linkages is especially important given current public policy debates regarding health care and immigration reform (Castañeda et al., 2015; Hummer & Hayward, 2015).

In the present study, we first review prior research on social sources of health disparity among Mexican Americans, focusing on the links between selection and exposure mechanisms and health in later life. Then, we assess the role that exposure to different IPRs may play in disability decades later. Using eight waves (1993/1994–2013) of data from the Hispanic Established Populations for the Epidemiologic Study of the Elderly (H-EPESE), we examine the associations of four distinct IPRs with disability among aging Mexican Americans. Findings suggest that IPRs are indeed associated with heterogeneous disability trajectories. The connection between IPRs and occupation, however, is less straightforward and more complex than prior research on migration selection suggests. Occupational differences at the individual level are associated with different trajectories among women, but not among men, and occupation does not mediate the observed association between IPRs and disability.

Heterogeneity in Mexican American Immigrant Health

Health research on U.S. immigrants born in Mexico has flourished in recent years. Much of this research concerns the “epidemiological paradox,” an empirical puzzle where Mexico–U.S. migrants experience lower mortality rates than U.S.-born non-Hispanic whites and African Americans, despite lower income, education, and access to health-promoting resources (Turra & Goldman, 2007). Beyond mortality, however, research has also consistently documented that Mexican American migrants experience prolonged periods of illness and disability compared with U.S.-born groups (Angel, Angel, & Hill, 2015; Eschbach, Al Snih, Markides, & Goodwin, 2007; Hayward, Hummer, Chiu, González-González, & Wong, 2014).

Some observed health differences among Mexican American immigrants may be attributed to health selection, as relatively healthier Mexicans are more likely to migrate to the United States, and those with poorer health are more likely to return to Mexico (Palloni & Arias, 2004). Others note a dose–response relationship between exposure to adverse social and economic conditions in the United States and declining functional health (Abraído-Lanza, Echeverría, & Flórez, 2016; Gorman, Read, & Krueger, 2010). Among migrants, those with legal documentation status are relatively protected from mechanisms contributing to physical disability, whereas undocumented migrants face elevated exposure to occupational health risks (Hall & Greenman, 2015) and lower levels of health care utilization (Torres & Waldinger, 2015). The effects of exposure to the U.S. context on health have been explored by estimating heterogeneity due to age at immigration, and theorized mechanisms buffering against the negative effects of exposure to the United States include protective individual-level health behaviors and geographical factors (Abraído-Lanza et al., 2016; Eschbach, Ostir, Patel, Markides, & Goodwin, 2004; Gubernskaya, 2015).

Recent research has explored the role of temporal changes in the U.S. context as a source of health disparity (Martinez et al., 2015; Viruell-Fuentes et al., 2012). Policy changes not only structure the composition of migrant cohorts (Garip, 2017), but also shape access to Medicare, Medicaid, Supplemental Security Income (SSI), and health services that mitigate the accumulation of health problems (Martinez et al., 2015; O’Neil & Tienda, 2015). These insights raise the question of whether exposure to policy structures in the past is also associated with health decades later.

Health Effects of Changing U.S. IPRs

Migrant selectivity and the sets of social conditions faced by immigrants upon arrival to the United States change over time as a result of shifting policies, social, and economic conditions (Donato, Wakabayashi, et al., 2008). However, despite a general acknowledgement that treating the health consequences of distinct historical conditions in a life course perspective is important and the burgeoning literature on the importance of cohort membership to health outcomes (Masters, Hummer, & Powers, 2012; Yang & Lee, 2009), little work attempts to identify whether this continual changing exposure to U.S. immigration policies is associated with different health profiles in the future.

One way to assess the relationship between immigration policy and health is to measure the association of health with particular IPRs. While researchers often conduct comparisons of IPRs cross-nationally, differences may also occur within countries over time (Ford, Jennings, & Somerville, 2015). Applying insights from migration research (Bean, 1997; Durand et al., 2000; Garip, 2017), we identify four broad historical periods characterizing the flow of migrants from Mexico to the United States: the Railroad, Bracero, Limiting, and Post-IRCA IPRs.

The Railroad IPR, spanning 1910–1941, was broadly characterized by the recruitment of male Mexican workers to meet rapidly expanding labor demands by U.S. railroads, farms, and mining companies amidst the backdrop of exclusionary policies targeting Asian American immigrant labor sources. At first mostly characterized by migration to Texas, this period saw the rise of California as the center of Mexican American settlement by 1920, before the onset of the Great Depression accompanying a period of deportations and overall declining Mexican American population during the 1930s (Durand et al., 2000).

The Bracero IPR began in 1942, as wartime labor demands led to formal policies governing the annual migration of Mexican farmworkers. Although ostensibly welcomed as a source of low-wage labor in some legislation, such as a bilateral 1949 agreement legalizing many undocumented farmworkers (Bean, 1997), migrants faced everyday hostility, civil rights violations, and threats of mass deportation such as during “Operation Wetback” in 1954 (Astor, 2009).

The Limiting IPR replaced the Bracero Program, which ended in 1964. This period was characterized by consistent demand for unskilled labor and the flow of undocumented migrants (Durand et al., 2000; Garip, 2017). While male unskilled laborers composed the majority of migrants in earlier IPRs, the Limiting IPR saw the rise of gendered network migration, as female migrants reunited with their male spouses and family members who had previously come to the United States for work (Donato, Wagner, & Patterson, 2008; Garip, 2017).

The Post-IRCA IPR began with the passage of the Immigration Reform and Control Act of 1986 (IRCA), which provided amnesty to nearly 3 million undocumented migrants while imposing a new set of sanctions on employers who sought to hire undocumented workers (Baker, 1997). Employers shifted to subcontractors, undocumented immigrants faced fewer workplace legal protections and declining wages, and newly legalized immigrants faced greater wage competition and rising levels of hostility in established immigrant destinations such as California (Durand et al., 2000; Durand, Massey, & Pren, 2016).

Changes in IPRs over time raise a number of questions regarding the immigrant selection and incorporation mechanisms presumed to be constant by much of the prior research on immigrant health. If exposure to IPRs are associated with latent health resources, disparities in physical disability should be observable even decades after exposure to particular IPRs. As outlined above, many of the barriers and rules structuring IPRs operate through gender and occupation, making these important potential mechanisms through which IPRs affect observed differences in disability trajectories.

Hypotheses

Mexico–U.S. migrants may experience relatively advantaged social positions in Mexico and relatively disadvantaged social positions after arrival to the United States (Garip, 2017). Because IPRs correspond with distinct sets of conditions that vary in the degree to which they expose immigrants to relative disadvantage, varying exposure to IPRs should be correlated with health differences. We are unable to make specific predictions regarding the impact of particular IPRs for two reasons, however. First, we cannot definitively say that any IPR provides a welcoming and beneficial migration context. Second, because IPRs occur sequentially, exposure to any IPR is highly correlated with birth cohort, immigration age, and exposure to other IPRs. For example, a respondent in the H-EPESE who has any exposure to the Bracero IPR will also be exposed to the Limiting and Post-IRCA IPRs; otherwise they would fall out of the sampling frame through mortality or return migration. We therefore expect:

Hypothesis 1: The life course timing and duration of exposure to IPRs are associated with different physical disability trajectories.

Formal and informal means of immigrant incorporation under IPRs are stratified by class and gender. Based on research documenting gender differences in disability among older Mexican Americans (Hayward et al., 2014) and in line with intersectional approaches (Brown & Hargrove, 2013; Viruell-Fuentes et al., 2012), we expect that health-promoting resources are unequally distributed to the disadvantage of women. We therefore expect:

Hypothesis 2: Exposure to some IPRs will disadvantage women relative to men.

Occupational composition of migrant cohorts varies by gender across IPRs (Durand et al., 2000; Garip, 2017; Gentsch & Massey, 2011). If changes in preferred occupation across IPRs also change migrant selectivity, occupation measures may correlate with health outcomes. Occupation may also affect exposure to physical health risks such as pesticides among farmworkers or physical injury among construction laborers. Because IPRs change occupational composition, and prior work suggests a relation between occupation and disability through some combination of migration selection and post-migration risk exposure, we expect:

Hypothesis 3a: Different occupations will be associated with different disability trajectories.

Hypothesis 3b: Occupation will partially mediate differences in disability trajectories across IPR exposure.

Method

Data

We use the H-EPESE, a panel of aging Mexican Americans in five states: Texas, California, New Mexico, Arizona, and Colorado. The H-EPESE began with 3,050 Mexican Americans aged 65 or older in 1993/1994 (Markides, 2009). Respondents were followed in 1995, 1998, 2000, 2004, 2006, 2010–2011, and 2013 (Markides, 1999, 2005, 2009; Markides, Ray, Angel, & Espino, 2009). Beginning with Wave 5 (2004–2005), the study added 902 new respondents (Markides, 2009). This refresher sample was 75 years or older at baseline. For all time-invariant variables, we use the first observation as baseline (Wave 1 for original respondents and Wave 5 for newly added respondents). The H-EPESE reports vital statistics data for the first five waves, but has not published subsequent mortality data. Attrition across waves ranges between 20% and 30%, much due to mortality.

Measures

Disability

We sum activities of daily living (ADL) and instrumental activities of daily living (IADL) to measure disability. This includes difficulty with the following: walking across a room, bathing, personal grooming, dressing, eating, getting into bed from a chair, using the toilet, using the telephone, traveling alone, shopping, preparing meals, doing light housework, taking medicine, handling money, doing heavy house work, walking up and down stairs, and walking half a mile. Summing disability indexes is a common way to measure disability (Brown, O’Rand, & Adkins, 2012), and a summed index of ADL/IADL items captures a wide range of disabilities (Taylor, 2010).

Exposure to IPR

We calculate IPR exposure as the number of years a respondent lived in the IPR: Railroad (from 1886 through 1941), Bracero (1942–1964), Limiting (1965–1985), and Post-IRCA (1986 forward). The H-EPESE does not include detailed migration histories, but does ask at what age the individual came to the United States “to stay.” We calculate exposure to IPR based on immigration age, presuming continuous residence in the United States from reported age at immigration. We measure in decades (dividing by 10) to scale effects. In our regression models, IPR exposure coefficients are relative to spending a decade in Mexico during the same period. In other words, the estimates are relative to postponing immigration for a decade, and avoiding additional exposure to the relevant IPR.

Occupation

The H-EPESE reports 1990 Occupational codes to identify the occupation respondent “ha[s] done most of [his/her] life.” We aggregate these codes into a dummy variable series as follows: (a) Homemaker; (b) Unclassified/Unstated; (c) Unemployed (includes Disabled); (d) Farm (includes Forestry and Fishing) (codes 473–499); (e) Labor (includes Craft and Military) (codes 503–905); (f) Office (includes Managerial and Technical) (codes 3–391); and (g) Service (codes 403–469) (Scopp, 2003; Supplementary Appendix).

Birth cohort and age

Age is measured continuously, and we use dummy variable indicators for 5-year birth cohort spans: prior to, and including 1904 (reference); 1905–1909, 1910–1914, 1915–1919, 1920–1924, and 1925 or over.

Covariates

We control for other measures which relate to disability status. These include time-invariant sociodemographic variables: female (dummy) and education (years). Marital status is a time-varying dummy variable series comprised of married (reference), widowed, and unmarried.

We include health characteristics and access to care measures which have been associated with disability. Time-varying measures include BMI (continuous), whether the respondent visited a physician within the last year (dummy, representing health care utilization), depressive symptoms (continuous, Center for Epidemiological Studies-Depression [CESD] scale 0–54) (following Radloff, 1977), and cognition (continuous, Mini-Mental State Examination [MMSE] scale 13–30) (following Hill, Angel, Balistreri, & Herrera, 2012). Time-invariant health characteristics include whether the respondent has ever smoked (dummy identifying whether the respondent has ever smoked 100 cigarettes or more in his/her life time), whether the respondent had been told by baseline that they have diabetes, hypertension, any type of cancer, or cardiovascular disease. We use baseline measures because the H-EPESE questions on conditions are inconsistent across waves. We also include a baseline measure of self-rated health, ranging from excellent (1) to poor (4). Self-rated health is closely correlated with health status, sometimes performing better than physician analysis of biomarkers (Todd & Goldman, 2013).

Finally, we include a dummy indicator for attrition (attrition), mortality (died), and the number of waves observed to correct for bias related to mortality and attrition (Brown et al., 2012; Warner & Brown, 2011). We also control for whether the respondent answered by proxy.

The H-EPESE sampling frame is limited to individuals self-identifying as Hispanic origin, and race was not measured. For our analytic sample, we drop 11 individuals with missing data on nativity status and/or marital status. While missing data proportions are relatively small, we follow standard practice by using five multiple imputation data sets, and estimate in “wide” format before transforming the data to long format for analysis (Allison, 2001).

Analytic Strategy

Following prior studies, we use hierarchical growth curves with individuals nested in time (Brown et al., 2012). We determined a quadratic effect was the appropriate functional form by comparing Akaike information criterion (AIC) and Bayesian information criterion (BIC) values and performing likelihood ratio tests (Raudenbush, Bryk, & Congdon, 2002, p. 173). All models have the following form:

Level 1:

$$\begin{array}{ll}{y}_{it}\hfill & ={\beta }_0+{\beta }_1(a_{it}-{\overline{a}}_{b=b_i})+{\beta }_2{(a_{it}-{\overline{a}}_{b=b_i})}^2 +{\beta }_3{x}_{it}+{\beta }_4{p}_{it}\hfill \\ \hfill & +{\beta }_5[p_{it}(a_{it}-{\overline{a}}_{b=b_i})]+{\beta }_6[p_{it}{(a_{it}-{\overline{a}}_{b=b_i})}^2]+{ϵ}_{it}\hfill \end{array}$$

Level 2:

$$\begin{array}{l}{\beta }_0={\beta }_{00}+{\beta }_{01}{b}_i+{\beta }_{02}{r}_i+{\beta }_{03}{o}_i+{\beta }_{04}{c}_i+ {\omega }_0\\ {\beta }_1={\beta }_{10}+{\beta }_{11}{b}_i+{\beta }_{12}{r}_i+{\beta }_{13}{o}_i+{\omega }_1\\ {\beta }_2={\beta }_{20}+{\beta }_{21}{b}_i+{\beta }_{22}{r}_i\end{array}$$

To conserve space, many of the effects are represented in vector or matrix format. y represents the outcome variable, i indexes the individual, and t indexes the wave. The level 1 model includes intercept (β₀), linear (β₁), and quadratic (β₂) effects. Age (a) is centered on group mean age for cohort $({\overline{a}}_{b=b_i})$, where b is a matrix of the dummy variable series for birth cohort. This design reduces bias in the coefficients (Raudenbush et al., 2002, pp. 183–185). Level 1 also includes effects for time-varying controls (β₃), and for proxy answers (p) across the intercept, linear, quadratic age effects (β₄, β₅, β₆). ∈_it is the stochastic component of observation level, normally distributed with mean 0.

Level 2 defines random effects for the intercept and linear terms, and a fixed effect for the quadratic term. All estimates include a matrix of effects for birth cohort (β_·1), IPR exposure (β_·2), and occupation (β_·3). b and o are matrices representing dummy variable series with the reference column removed. In all models, we adjust for time-invariant controls (c_i). We use an unstructured design to estimate level 2 errors (ω) (Singer, 1998).

To test our three hypotheses, we estimate seven models. Model 1 is the baseline model. Model 2 adds IPR exposure to test associations between IPR and disability (Hypothesis 1). Model 3 interacts gender with IPR exposure to test whether disability trajectories across IPR exposure differ by gender (Hypothesis 2). Because occupation is highly stratified by gender, we estimate separate occupation models for males and females. Baseline exposure models for men (Model 4) and women (Model 6) include the same covariates as Model 2. To test whether different occupations are associated with different disability trajectories (Hypothesis 3a) and whether occupation mediates observed differences in disability trajectories across IPRs (Hypothesis 3b), we include the dummy variable series for occupation (Models 5 and 7) to see if the IPR associations in the baseline model are diminished.

Results

Descriptive Statistics

Descriptive statistics lend preliminary evidence to support Hypotheses 1, 2, and 3b, revealing (a) correlations between IPR exposure and levels of disability, (b) differences in disability and variation in IPR exposure by gender, and (c) different distributions of occupations by gender.

Pairwise correlations reveal positive associations between disability and Post-IRCA IPR exposure, a smaller positive association with Railroad IPR exposure, and relatively small negative associations with Bracero and Limiting IPR exposure. Supplementary Figure 1 depicts these pairwise correlations. Supplementary Table 1 displays descriptive statistics across gender, relevant to evaluating Hypothesis 2. First, Supplementary Table 1 shows that women have, on average, 1.4 disabilities more than men (p < .05).

Second, men and women experience slightly different patterns of exposure to IPRs. Although average age at migration is in the early 30s for both men and women, women migrate when they are 1.1 years older than men (p < .05). Additionally, the variance of migration ages for women is larger than for men, suggesting more women migrating at older ages, consistent with the “network migration” effects noted in prior research (Cerrutti & Massey, 2001; Garip, 2017). These differences in migration patterns correspond to a few statistically significant gender differences in exposure to IPRs. Compared with men, women had more exposure to the Railroad and Limiting IPRs (p < .05), but less exposure to the Post-IRCA IPR (p < .05).

As expected, the distribution of occupations is significantly different by gender and nativity (p < .001). Fewer than 10% of men report their occupation as homemakers, unemployed, service, or unstated/unclassified occupations, as illustrated in Supplementary Figure 2. In contrast, approximately 63% of women report one of these occupational categories. Occupational distributions are also stratified by nativity: higher proportions of foreign-born women worked in service occupations, but higher proportions of U.S.-born women were homemakers.

Growth Curves

The growth curves confirm the expectations of Hypotheses 1 and 3a, but do not confirm Hypotheses 2 or 3b. Specifically, we find (a) different patterns of exposure to IPRs are associated with different disability trajectories and (b) occupation is associated with disability trajectories for women only. We find no evidence of different disability trajectories from IPR exposure by gender, nor do we find that occupation mediates the association between IPR exposure and disability.

Model 2, reported in Table 1, supports Hypothesis 1: differential exposure to IPRs has different associations with disability accumulation. Disadvantages are associated with exposure to the Post-IRCA IPR relative to spending time in Mexico at baseline (β₀) and in linear growth (β₁), although the quadratic effect (β₂) indicates a slowing accumulation of disabilities. Notably, these effects are also statistically different from all other estimated IPR coefficients (p < .05). In contrast, exposure to the Limiting IPR is associated with lower linear growth (β₁) relative to spending time in Mexico. This coefficient is lower than estimates of the Post-IRCA and Railroad IPRs at significant levels (p < .05), but the difference between Limiting exposure and Bracero exposure is only marginally significant (p < .10).

Table 1.

Results of Linear Growth Curves for ADL/IADL Disability

	Model 1	Model 2	Model 3
Intercept (β0)
Intercept	7.895 (0.537)***	7.607 (0.627)***	7.063 (0.768)***
Female	0.880 (0.111)***	0.871 (0.111)***	1.698 (0.691)*
IPR exposure
Bracero (1942–1964)		0.044 (0.118)	0.053 (0.187)
Limiting (1965–1985)		−0.219 (0.155)	−0.074 (0.248)
Post-IRCA (1986–2014)		0.742 (0.261)**	0.944 (0.367)*
Railroad (1886–1941)		−0.023 (0.089)	−0.074 (0.137)
Interaction with gender
Bracero × Female			−0.019 (0.249)
Limiting × Female			−0.220 (0.311)
Post-IRCA × Female			−0.361 (0.411)
Railroad × Female			0.092 (0.176)
Linear effect (β1)
Intercept	0.164 (0.090)+	0.046 (0.103)	0.100 (0.126)
Female	0.085 (0.013)***	0.083 (0.013)***	−0.009 (0.117)
IPR exposure
Bracero (1942–1964)		−0.001 (0.014)	0.008 (0.022)
Limiting (1965–1985)		−0.053 (0.019)**	−0.055 (0.031)
Post-IRCA (1986–2014)		0.179 (0.041)***	0.117 (0.057)*
Railroad (1886–1941)		0.006 (0.012)	−0.002 (0.018)
Interaction with gender
Bracero × Female			−0.015 (0.028)
Limiting × Female			0.001 (0.038)
Post-IRCA × Female			0.103 (0.067)
Railroad × Female			0.013 (0.023)
Quadratic effect (β2)
Intercept	−0.005 (0.010)	−0.015 (0.011)	−0.008 (0.013)
Female	0.000 (0.002)	0.000 (0.002)	−0.007 (0.010)
IPR exposure
Bracero (1942–1964)		0.000 (0.002)	0.001 (0.003)
Limiting (1965–1985)		0.001 (0.002)	−0.002 (0.004)
Post-IRCA (1986–2014)		−0.004 (0.001)**	−0.002 (0.002)
Railroad (1886–1941)		−0.000 (0.002)	−0.000 (0.002)
Interaction with gender
Bracero × Female			−0.001 (0.004)
Limiting × Female			0.004 (0.005)
Post-IRCA × Female			−0.004 (0.003)
Railroad × Female			0.000 (0.003)

Note: N = 14,474 (multiply imputed, M = 5); standard errors in parentheses; all models control for birth cohort, depression, cognitive ability, marital status, BMI, physician visits, proxy answers, attrition, number of waves, and baseline measures of self-reported health, cancer, hypertension, diabetes, and cardiovascular disease; baseline measures are from 1993/1994 for original sample and 2004 for refresher (new) sample. ADL = activities of daily living; BMI = body mass index; IADL = instrumental activities of daily living; IPR = Immigration Policy Regime; IRCA = Immigration Reform and Control Act.

⁺ p < .1. *p < .05. **p < .01. ***p < .001.

The quadratic curves, together with the close correlations of the IPRs pictured in Supplementary Figure 1, make it difficult to visualize these effects. To simplify the interpretation of Model 2, we generate a series of predicted growth curves based on a synthetic data set of 18 individuals. Each is born at the beginning of the six different birth cohorts, and is presumed to have either been born in the United States, migrate at age 30, or migrate at age 60; baseline disability is assumed at 1998. All other variables are estimated based on birth-cohort mean levels. The synthetic data set assumes varying exposure to IPRs as summarized in Supplementary Figure 3. This figure shows that older cohorts have the greatest levels of exposure to all four IPRs. Levels of exposure within birth cohort are reduced as age at immigration increases. Using the estimates of Model 2 in Table 1, we age these synthetic individuals 10 years from 1998 to 2008, plotting points every 2 years in Figure 1.

Figure 1.

Predicted disabilities across 16 immigration/birth cohort profiles. Note: Predicted curves use estimates from Model 2 of Table 2. All variables held at birth cohort group-level means apart from birth cohort dummy variable series and Immigration Policy Regime (IPR) exposure variables. Summary of IPR exposure levels across each birth cohort in synthetic data appears in Supplementary Figure 3.

The predicted disabilities in Figure 1 identify exposure by age at immigration, from U.S.-born (darkest) to immigrants who migrated at 60 (lightest). Importantly, this figure shows a disadvantage to migration at older ages for younger cohorts, which disappears for older cohorts. These predicted disabilities are driven by the IPR exposures pictured in Supplementary Figure 3. In particular, the variations in exposure to the Limiting IPR (third panel) lead to increased relative linear growth terms as described above. Ultimately, Figure 1 shows that age at immigration may be associated with slower or faster disability accumulation, depending on IPR exposure, confirming Hypothesis 1.

In contrast, Model 3 of Table 1 lends no evidence that IPR exposure varies between men and women, so Hypothesis 2 is not supported. We test for different effects across gender by interacting the dummy variable indicator for female with all IPR exposure terms across intercept, linear, and quadratic effects. None of these interactions are statistically significant, however.

Table 2 shows evidence consistent with Hypothesis 3a, that different occupations are associated with different disability growth profiles, but we find no evidence that occupation mediates IPR exposure (Hypothesis 3b). Models 4 and 6 provide baseline effects of exposure by gender. Models 5 and 7 include the dummy variable series of occupation to these baseline models. With respect to men (Models 4 and 5 in Supplementary Table 2), occupation effects are not statistically significant and none of the IPR exposure effects are diminished.

Table 2.

Results of Linear Growth Curves for ADL/IADL Disability for Women

	Model 6	Model 7
Intercept (β0)
Intercept	8.570 (0.851)***	7.741 (0.866)***
IPR exposure
Bracero	0.066 (0.166)	0.037 (0.165)
Limiting	−0.308 (0.203)	−0.216 (0.203)
Post-IRCA	0.716 (0.345)*	0.734 (0.343)*
Railroad	−0.014 (0.125)	0.020 (0.124)
Occupation
Unemployed		0.583 (0.372)
Homemaker		0.658 (0.204)**
Office worker		−0.216 (0.231)
Service		Ref
Farmworker		0.672 (0.243)**
Laborer		−0.096 (0.208)
Unclassified/unstated		0.673 (0.193)***
Linear effect (β1)
Intercept	0.155 (0.149)	0.123 (0.150)
IPR exposure
Bracero	−0.006 (0.019)	−0.009 (0.019)
Limiting	−0.054 (0.024)*	−0.048 (0.024)*
Post-IRCA	0.221 (0.053)***	0.229 (0.053)***
Railroad	0.010 (0.017)	0.013 (0.016)
Occupational categories
Unemployed		0.013 (0.053)
Homemaker		−0.003 (0.027)
Office worker		−0.065 (0.030)*
Service		Ref
Farmworker		0.055 (0.032)+
Laborer		−0.045 (0.027)+
Unclassified/unstated		0.037 (0.026)
Quadratic effect (β2)
Intercept	−0.010 (0.016)	−0.014 (0.016)
IPR exposure
Bracero	−0.001 (0.003)	−0.001 (0.003)
Limiting	0.002 (0.003)	0.003 (0.003)
Post-IRCA	−0.006 (0.002)**	−0.006 (0.002)**
Railroad	0.001 (0.002)	0.001 (0.002)

Note: N = 8,379 (multiply imputed, M = 5); standard errors in parentheses; all models control for birth cohort, depression, cognitive ability, marital status, BMI, physician visits, proxy answers, attrition, number of waves, and baseline measures of self-reported health, cancer, hypertension, diabetes, and cardiovascular disease; baseline measures from 1993/1994 for original sample and 2004 for refresher (new) sample. Male-only models reported in Supplementary Table 2. ADL = activities of daily living; BMI = body mass index; IADL = instrumental activities of daily living; IPR = Immigration Policy Regime; IRCA = Immigration Reform and Control Act.

⁺ p < .1. *p < .05. **p < .01. ***p < .001.

With respect to women (Models 6 and 7 in Table 2), we find a benefitted and a disadvantaged class of occupations. Women who were homemakers, farmworkers, or unclassified had significantly higher levels of baseline (β₀) disability than service workers and office workers. In addition, office occupations are associated with a statistically significant slower rate of linear growth in disability (β₁). As with men, inclusion of the occupation dummy variable series does not reduce the statistical association between IPR exposure and disability trajectories, inconsistent with classical expectations of mediator variables.

Discussion

This study advances efforts to understand the link between structural conditions and health among Mexican American immigrants by exploring the effects of IPRs on physical disability in later life. We find that exposure to distinct IPRs corresponds with divergent disability trajectories, supporting Hypothesis 1. In particular, exposure to the Limiting IPR is associated with slower disability growth. This may be a result of relatively broader social inclusion of immigrants after large-scale deportation programs such as Operation Wetback ended. In contrast, the most health-averse period of exposure to the United States relative to Mexico is the Post-IRCA period. The Post-IRCA period, characterized by increased levels of border enforcement, includes increasingly precarious labor market conditions and threat of deportation (Durand et al., 2016; Martinez et al., 2015). Neoliberal welfare and health reforms placed access restrictions on health-promoting resources upon which aging immigrants may have otherwise utilized to mitigate functional health declines (Hagan, Rodriguez, Capps, & Kabiri, 2003; Marrow & Joseph, 2015).

No statistically significant differences in the effects of IPR exposure on physical disability were found between men and women (in contradiction to Hypothesis 2), and the effects of IPRs were not mediated by occupation (in contradiction to Hypothesis 3b). Importantly, as illustrated graphically in Figure 1, different associations in disability across IPR exposure have important implications for the health impact of immigration age and nativity status. Specifically, we find that immigration age may be associated with either better or worse disability trajectories. The difference depends on IPR exposure, which is determined by year of birth and age at immigration together. This suggests that age-at-immigration effects are susceptible to confounding by IPR exposure.

We find differences in disability trajectories across occupation, but for women only, partially supporting Hypothesis 3a. Substantively, we find two classes: health-protective occupations (laborer, office worker, and service worker), and health-averse occupations (unclassified/unstated, farm, and homemaker). The distribution of these occupations varies across both nativity status and gender as illustrated by Supplementary Figure 2, and these differences lead to significantly negative health impacts for women. This finding suggests that occupation is an important site for researchers to continue to explore as a cause of disparate disability trajectories between men and women.

We note a number of limitations in this study that may be addressed in future research. First, our data do not include respondents’ documentation status. This limitation is partially mitigated by the fact that strong associations exist between documentation status and other variables included in our models such as health care utilization and occupation (Hall & Greenman, 2015; Torres & Waldinger, 2015). Nevertheless, our findings suggest that an IPR’s treatment of documentation status may be a critical feature.

Second, our data do not include detailed migration history, requiring us to assume no exposure to the U.S. IPR(s) before an individual’s age at immigration, and to assume continuous exposure to the U.S. IPR(s) effective from reported age at immigration forward. To the extent this assumption is inaccurate, and respondents engaged in return-migration spells either before or after reported age at immigration, our measures may under- or overestimate exposure to U.S. IPRs.

Third, while the sample benefits from long-term follow-up, there is significant unmeasured mortality. To the extent that there is differential mortality caused by IPRs, our reported findings may be altered, particularly with regard to the Post-IRCA IPR, which includes the oldest migrants. In addition, the force of mortality limits statistical leverage for our male subsample (women are observed 1.5 times more than men in the sample). This reduces our confidence in the non-findings of occupation for men relative to women, but differential mortality conditional on IPR exposure would prove the broader point: that differential exposure to IPRs is associated with significant health effects.

Fourth, the sampling frame requires us to rely on retrospective migration data, so we are unable to isolate the effects of migration selectivity for IPR exposure. Finally, our study was limited in its scope by focusing only on Mexican Americans who resided in the Southwest United States in the mid-1990s; future research should assess the extent to which geographical differences mitigate the effects of IPR exposure on disability, for example, by comparing disability trajectories in established and new immigrant destinations. While the present study underscores theories linking migration policy and health, future research should assess changing policies structuring immigrant incorporation in relationship to other contemporaneous domestic U.S. racial policy regimes.

Despite these limitations, this study suggests that there are important associations between health of aging populations and the policy regimes they experience. Investigating changing IPRs in relation to other mechanisms of racialization in the United States may greatly advance efforts to understand the effects of public policy on the reproduction of health disparities and further contribute to a growing body of research linking social policy and the reproduction of racial/ethnic inequality.

Funding

This work was supported by a training grant through the National Institute of Aging at the National Institutes of Health (5T32AG000139-27 to B. J. Bartlett).

Conflicts of Interest

The authors declare no conflicts of interest.