The Comparability of Perceived Physical and Mental Health Measures Across Immigrants and Natives in the United States

Abstract

The comparative study of perceived physical and mental health in general—and the comparative study of health between the native-born and immigrants, in particular—requires that the groups understand survey questions inquiring about their health in the same way and display similar response patterns. After all, observed differences in perceived health may not reflect true differences but rather cultural bias in the health measures. Research on cross-country measurement equivalence between immigrants and natives on self-reported health measures has received very limited attention to date, resulting in a growing demand for the validation of existing perceived health measures using samples of natives and immigrants and establishing measurement equivalence of health-related assessment tools. This study, therefore, aims to examine measurement equivalence of self-reported physical and mental health indicators between immigrants and natives in the United States. Using pooled data from the 2015–2017 IPUMS Health Surveys, we examine the cross-group measurement equivalence properties of five concepts that are measured by multiple indicators: (1) perceived limitations in activities of daily life; (2) self-reported disability; (3) perceived functional limitations; (4) perceived financial stress; and (5) nonspecific psychological distress. Furthermore, we examine the comparability of these data among respondents of different ethnoracial origins and from different regions of birth, who report few versus many years since migration, their age, gender, and the language used to respond to the interview (e.g., English vs. Spanish). We test for measurement equivalence using multigroup confirmatory factor analysis. The results reveal that health scales are comparable across the examined groups. This finding allows drawing meaningful conclusions about similarities and differences among natives and immigrants on measures of perceived health in these data.

Introduction

Differences in the prevalence of physical and mental health between immigrants and native-born populations are well established (Singh and Siahpush 2002). The literature suggests that upon arrival in a host country, immigrants are likely to be healthier than the native-born population but that this health advantage dissipates with time (Boulogne et al. 2012). This phenomenon has been termed the healthy immigrant effect. Empirical evidence of the healthy immigrant effect has been documented for a wide variety of health indicators (e.g., obesity, hypertension, functional disabilities, chronic diseases, and mortality from almost all causes of death) (Akresh 2007) and in major immigrant-receiving countries (see Cunningham et al. [2008] for the United States, McDonald and Kennedy [2004] for Canada, and Martini et al. [2010] for Europe).

A review of the literature on the subject reveals that the use of the term healthy immigrant effect is quite vague. On the one hand, some studies refer to immigrants as a homogeneous population, arguing that upon arrival at their destination country, immigrants are in relatively better health, on average, than the members of the native population. On the other hand, empirical evidence has documented this advantage for only certain ethnoracial subgroups of immigrants (Huh et al. 2008). For example, findings on the initial health advantage of immigrants compared with their native-born counterparts (of the same ethnic origin) are less consistent for Asian immigrants than for Hispanic immigrants (Huh et al. 2008). Some studies found that Asian immigrants are likely to report poorer health status relative to their U.S.-born counterparts or the general U.S.-born population (e.g., Huh et al. 2008). Conversely, other studies found that Asians and Pacific Islanders are, on average, healthier than the general U.S. population (e.g., Frisbie et al. 2001).

Furthermore, it is unclear whether these observed differences reflect true differences or merely reflect cultural bias in the measures (Snowden 2003). Indeed, most health indicators were initially developed and tested on culturally homogeneous groups (e.g., White native English-speaking populations) (Snowden 2003). Consequently, concepts and constructs may not be understood in the same way when applied to other cultures (Teresi 2006). Therefore, measurement nonequivalence might result in biased conclusions about differences and similarities in health indicators across different population groups (Hardy et al. 2014). For example, Burgard and Chen (2014) showed how nonequivalence can lead to an underestimation or overestimation of health inequality, and they argued that researchers must take into account the varying levels of presence, quality, comparability, validity, and reliability of health indicators. However, to the best of our knowledge, only a few studies have examined the comparability of specific self-reported health measures between immigrants and natives (see Perreira et al. [2005] and Klokgieters et al. [2021] for a depression scale; see Buchcik et al. [2017] for health-related quality of life). Moreover, no one has systematically examined the comparability of several multidimensional self-reported health measures between immigrants and natives in a single study.

Using data from the 2015–2017 U.S. IPUMS Health Surveys (Lynn et al. 2019), this study addresses this gap by analyzing the comparability of some commonly used physical and mental health indicators between immigrants and U.S.-born natives. Specifically, we aim to evaluate the cross-group comparability of measures of nonspecific psychological distress, perceived disability, perceived limitations in activities of daily living (ADLs), and perceived functional limitations between immigrants and U.S. natives among the adult population aged 18 to 85 years. Furthermore, we examine the comparability of these data among respondents (1) deriving from different ethnoracial origins and (2) different regions of birth, who report (3) different rates of elapsed time (years) since migration, as well as their (4) age, (5) gender, and (6) language of interview (English vs. Spanish). By addressing measurement issues in the assessment of health status, this study contributes to a better and more reliable understanding of health disparities in a comparative framework.

Theoretical Background

The Problem of Nonequivalent Measures in Health Inequality Research

Questions relating to the healthy immigrant effect are centrally positioned in the migration literature. However, the reasons underlying this phenomenon remain a topic for vigorous research and debate. The most common explanations for the immigrant health advantage are positive health selection (immigrant self-selection as well as selective immigration policies; Dean and Wilson 2010) and negative health selection of return migrants (the so-called salmon bias effect; Goldman et al. 2014). The prevailing explanations for changes in immigrant health status over time are related to acculturation (i.e., immigrants adopt the health-related behaviors of the native-born population) (Abraído-Lanza et al. 2005), inadequate access to healthcare services (Bustamante et al., 2012), and socioeconomic inequality (Viruell-Fuentes et al. 2012). Although researchers have repeatedly used these explanations to account for health disparities between natives and immigrants, measurement nonequivalence between immigrants and natives has remained largely understudied (Morris 2018). Indeed, individuals from different countries may perceive their health in different ways and may understand questions inquiring about their health differently. As a result, questionnaire items measuring their perceived health might not have the same meaning for immigrants and natives. Individuals from different countries may also differ in how they use subjective response categories because of their different cultural backgrounds (e.g., Grol-Prokopczyk et al. 2015). Thus, their self-reported health responses to survey items may not be comparable. Therefore, cross-group equivalence of measurements of a latent construct that measures physical or mental health is necessary to draw meaningful conclusions in research comparing immigrants and natives. Lack of measurement equivalence hampers comparisons because differences in mean scores or in regression coefficients across groups may be a result of systematic biases in responses across groups or a different understanding of the questions rather than reflect real differences. At the same time, real differences may exist even when findings show no differences in mean scores or regression coefficients across groups (Davidov, Cieciuch et al. 2015; Davidov, Meuleman et al. 2014).

To date, only a few studies have attempted to examine measurement equivalence in health inequality research. These studies systematically tested for comparability and reported errors stemming from nonequivalence (Burgard and Chen 2014). For example, using data on 11 European countries from the Survey of Health, Ageing and Retirement in Europe, Hardy and colleagues (2014) examined how respondents translate morbidity and disability into self-rated health (SRH), how national populations differ in SRH, and how normative and person-specific reporting styles shape SRH. They found that observed country differences in SRH reflect compositional differences, cultural differences in reporting styles, and differing perceptions of how health restricts typical activities. Grol-Prokopczyk et al. (2015) examined vignette equivalence and response consistency in several widely used vignettes: the health vignettes that are implemented in the World Health Organization (WHO) Study on Global Ageing and Adult Health (SAGE) and World Health Survey, as well as similar vignettes in the Health and Retirement Study (HRS). Their results revealed substantial violations of vignette equivalence both cross-nationally and across socioeconomic groups: members of different sociocultural groups appeared to interpret vignettes as depicting fundamentally different levels of health. The authors concluded that the evaluated anchoring vignettes do not fulfill their promise of providing interpersonally comparable measures of health.

Notably, most of these health studies that addressed issues of measurement cross-group equivalence focused on cross-country comparisons of health measures. Single-country studies presumed social or cultural homogeneity in terms of response patterns and, therefore, neglected the question of whether health measures used in these studies were equivalent across groups within the country (e.g., Morris 2018). However, bias related to nonequivalence may “apply to all forms of comparative research, including comparisons of groups within the same single-country data set” (Morris 2018:559). Measurement nonequivalence within countries might lead to biased estimates of the incidence of perceived health problems and, therefore, affect the development of healthcare policies and delivery of healthcare to particular groups (Skinner et al. 2001). Thus, the comparability of health measures must be considered not only in cross-country comparative studies but also within single countries (Morris 2018).

Comparability of Health Measures Between Immigrants and Natives

Studies have repeatedly shown that, on average, recent immigrant population groups (compared with native-born population groups) manifest substantially lower rates of smoking, drinking, obesity, and hypertension; have a lower proportion of disabled persons; report lower levels of chronic diseases; and have a lower risk of mortality from almost all causes of death (Akresh 2007). Furthermore, both aggregate (Singh and Siahpush 2002) and within-group (Finch and Vega 2003; Muennig and Fahs 2002) studies have found that upon arrival, immigrants enjoy better health than their native-born counterparts in the host country. Despite such consistent findings, several converging lines of evidence demonstrate that the experience and conceptualization of health as reported in perceived health measures may differ across immigrants and natives.

Previous studies have suggested that mental health, for example, is contextually based and culturally embedded (e.g., Kleinman 1986). Therefore, it is plausible that the scales used for identifying depression, for instance, may tap into a different construct across the groups being compared. Immigrant groups may report higher/lower depression levels not only because of the actual higher/lower incidence of depression but also because the group expresses psychopathology in a way not captured by measures developed from a conceptualization of depression applicable to majority group members of the host country (e.g., White native-born) (Vega and Rumbaut 1991). Specifically, Hispanic and Black people are more likely to somatize mental health problems, reporting more physical symptoms of distress than European Americans (Guarnaccia et al. 1989). Asians do not differentiate between mental and physical well-being and identify a strong interpersonal element to depression (Kim 2002). Fillenbaum and colleagues (1990) examined seven cognitive screening or neuropsychological tests and compared them with clinical diagnoses. The authors reported that most measures, when adjusted for race and education, had lower specificities for Blacks than for Whites. They suggested that most measures were culturally or educationally biased. Similarly, Teresi et al. (2001) reviewed studies using differential item functioning and examining bias in direct cognitive assessment measures with respect to race/ethnicity and education. They found that item performance varied across groups that differed in terms of education, ethnicity, and race.

Despite the potential prevalence of measurement nonequivalence across groups within countries—in particular, across natives and immigrants—relatively little research has addressed the question of whether immigrants and natives perceive and respond to health measures equivalently (Buchcik et al. 2017; Klokgieters et al. 2021). For example, Perreira et al. (2005) tested the Center for Epidemiologic Studies Depression Scale (CES-D) for use in the multiethnic and foreign-born populations living in the United States. Their results demonstrated that the CES-D is not psychometrically equivalent across either race/ethnicity or immigrant generations. Perreira et al. argued that research using the CES-D might lead to erroneous conclusions about mental health disparities in the U.S. adolescent population. Klokgieters et al. (2021) also examined measurement invariance of the CES-D among people of Dutch, Moroccan, and Turkish origin in the Netherlands. In contrast to Perreira et al.'s (2005) study, they found that the four subscales of the CES-D were measurement-invariant among older adults of Dutch, Turkish, and Moroccan origin. Therefore, it was possible to compare the prevalence of depressive symptoms across these three ethnic groups. Buchcik et al. (2017) tested the reliability and construct validity of health-related quality of life across elderly Polish migrants, Turkish migrants, and German natives. They did not find rigorous and complete invariance and observed some differences between the groups. Therefore, comparing subscale scores of the health-related quality-of-life measure between different ethnoracial groups may be problematic.

Furthermore, important within-group differences or similarities among immigrants may be biased because of properties of the measurement scales rather than reflecting true differences or similarities. To the best of our knowledge, no one has yet tested whether and to what extent measurement equivalence exists across different subgroups of immigrants. For example, it is not clear whether recent immigrants share a similar understanding of the questions with immigrants who spent more time in the United States or U.S.-born natives, whether older and younger immigrants display measurement equivalence in their perceived health measures, or whether measurement equivalence applies for those who responded to the survey in English or Spanish. Indeed, studies providing a systematic analysis of comparability of the various health measures between immigrants and natives are lacking. Standard measures of health, developed largely for majority population samples, have been applied to immigrants without close attention to the potential impact of cultural differences on their response patterns. Therefore, further studies are needed to determine the validity of the health indicators for cross-cultural comparisons.

The Current Study

Data

The data for the analysis were obtained from the 2015–2017 IPUMS Health Surveys (formerly Integrated Health Interview Surveys) (Lynn et al. 2019). These surveys produce a harmonized data and documentation set based on material originally included in the public use files of the U.S. National Health Interview Survey (NHIS), the leading source of information on the health of the U.S. population. The IPUMS Health Surveys include many variable features not available in the original NHIS public-use files (Davern et al. 2012). The IPUMS Health Surveys data set provides detailed information on health conditions, health status, health behaviors, healthcare utilization, and insurance coverage; the data set also contains information across a rich set of individual and household characteristics, including immigrant status, sex, age, and education. We restrict our sample to individuals age 18 and older who had records in the sample adult core in 2015–2017 and who were randomly selected to respond to the Adult Functioning and Disability (AFD) section of the questionnaire. The final analytical pooled sample comprises 206,202 individuals (169,658 U.S.-born natives and 36,544 first-generation immigrants). We distinguish among four ethnoracial groups: (1) non-Hispanic White (130,691 natives and 5,644 immigrants); (2) non-Hispanic Black (21,185 natives and 2,967 immigrants); (3) non-Hispanic Asian (3,286 natives and 9,564 immigrants); and (4) Hispanic (14,496 natives and 18,369 immigrants) respondents. We thus have eight groups for the analysis: four native and four immigrant groups. We also distinguish among the following additional categories of immigrants and natives: region of birth, years since migration, language of interview, gender, and age.

Variables

We focus on the following self-perceived physical and mental health measures: (1) limitation in ADLs; (2) functional limitations; (3) self-reported disability; (4) nonspecific psychological distress (Kessler Psychological Distress Scale [K6]); and (5) perceived financial stress. These scales represent commonly used measures of self-perceived physical and mental health (Wang and Kaushal 2019). Notably, ADLs, functional limitations, self-reported disability, and perceived financial stress are part of the International Classification of Functioning, Disability and Health (ICF), which provides a standard language and framework for the description of health status (World Health Organization 2001).

Limitations in ADLs and Functional Limitations

Studies of health status and quality of life frequently assess functional disability. The concept of functional disability distinguishes basic daily activities that are necessary to be able to function daily both as an individual and in the community from other major social roles, such as work disability or social interactions (Nagi 1976). We rely on five questions measuring ADLs on a 4-point scale ranging from 1 = no difficulty to 4 = cannot do it at all, and on nine questions measuring functional limitations using a 5-point scale ranging from 1 = not at all difficult to 5 = cannot do it at all. Table 1 summarizes the five scales we examine, the questions used to measure each scale, response categories to each question, and means and standard deviations of the responses for U.S. natives and immigrants, respectively.

Self-reported Disability

The self-reported disability questions presented in the IPUMS Health Surveys use the World Health Survey's ICF as a conceptual framework. The questions focus on perceived disability defined as difficulties in functioning in any of the following domains: (1) hearing (deafness or severe hearing loss); (2) vision (blindness or severe visual impairment even with corrective lenses); (3) cognitive difficulty (cognitive impairment in memory, concentration, or decision-making); (4) physical difficulty (physical impairments causing difficulty in walking or climbing stairs); (5) mobility limitations (physical, mental, or emotional conditions lasting at least six months and inhibiting the unassisted performance of basic activities outside the home); and (6) personal care limitations (physical, mental, or emotional conditions lasting at least six months and inhibiting unassisted attending to own personal needs, such as bathing, dressing, or moving around inside the home). Information on perceived disability is an important component of health information because it shows how well an individual is able to function in daily areas of life in general. Along with traditional indicators of a population's health status, such as mortality and morbidity rates, disability has become important in measuring disease burden, evaluating the effectiveness of health interventions, and planning health policy (Üstün et al. 2010). The six questions in the scale are measured using a dichotomous scale with the response options 1 = no difficulty and 2 = difficulty in performing the behavior.

Nonspecific Psychological Distress

The K6 was developed by Kessler and colleagues (Kessler et al. 2002) as a measure of nonspecific psychological distress, using a 30-day reference period (Fleishman and Zuvekas 2007). It has been used in numerous studies of mental health, focusing on particular races or ethnic groups (Wang and Kaushal 2019). The sociological and broader social science literature has relied extensively on the K6 as a reliable measure of depressive symptoms (Kessler et al. 2002). The responses are measured on a 5-point scale ranging from 0 = none of the time to 4 = all the time.

Perceived Financial Stress

Among the array of chronic stressors that people may confront in their daily lives, perceived financial stress is one of the most pivotal (Pearlin et al. 1997). We rely on eight questions measuring financial stress on a 4-point scale ranging from 1 = not worried at all and 4 = very worried.

Immigrant Status and Ethnoracial Origin

Immigrant status pertains to individuals who were not born in the United States. To distinguish between ethnoracial groups, we include dummy variables indicating whether the respondent was non-Hispanic Black, non-Hispanic Asian, or Hispanic, with non-Hispanic White as the reference category.

Variables for Within-Group Comparisons

We include 10 regions of birth: (1) Mexico, Central America, Caribbean Islands; (2) South America; (3) Europe; (4) Russia (and former USSR areas); (5) Africa; (6) the Middle East; (7) the Indian subcontinent; (8) Asia; (9) Southeastern Asia; and (10) the United States (for native U.S.-born residents). We also divide immigrants into groups by the number of years since migration: (1) recent immigrants (up to 5 years in the United States); (2) immigrants living in the United States for 5–15 years; (3) immigrants living in the United States 15 or more years; and (4) U.S. native-born residents. In addition, we include language of interview (Spanish vs. English), respondent's age (50 or younger vs. older than 50), and gender (men vs. women) as categories across which we test for measurement invariance. (See section 1 of the online appendix for the descriptive statistics by group.)

Methods

Measurement invariance and measurement equivalence indicate “whether or not, under different conditions of observing and studying phenomena, measurement operations yield measures of the same attribute” (Horn and McArdle 1992:117). In other words, measurement invariance guarantees that the underlying latent variables measure the same phenomenon in different groups or at different times. Measurement equivalence may be tested by different methods, but the most common is multigroup confirmatory factor analysis (MGCFA; Reise et al. 1993). MGCFA analyzes unobserved (latent) constructs that are reflected by multiple observed (manifest) indicators. The relation between each indicator and its corresponding latent construct is expressed as a factor loading parameter. Furthermore, intercept parameters indicate the expected value of an observed variable when the latent variable is 0. With MGCFA, statistical analyses can compare the properties of a measure in two or more groups.

The cross-cultural methodological literature considers several levels of measurement equivalence (Davidov et al. 2014). Configural invariance refers to equivalence of the measurement structure and indicates that each latent variable (factor) has the same set of indicators in each group, the model fits the data well in each group, all factor loadings are substantial, and correlations among factors are less than 1. Configural invariance does not imply, however, that the meaning of a latent variable is the same across groups. Metric invariance additionally requires factor loadings to be equal across groups and is considered necessary to be able to compare factor covariances and unstandardized regression coefficients meaningfully. Finally, scalar measurement invariance additionally requires that intercepts of the indicators are equal across groups. When the analyses demonstrate scalar measurement invariance, we can confidently compare the means of the latent variables across groups.

The literature suggests that when measurement indicators have fewer than four or five categories, researchers should test for measurement invariance while treating the measures as categorical rather than continuous (e.g., Davidov et al. 2018). Scalar measurement invariance in this case requires that the thresholds linking the underlying latent response variables to the observed categorical variables be held equal across groups in addition to the intercepts of the latent response variables (see Millsap and Yun-Tein 2004). When this specification is supported empirically, we can confidently compare the means of the substantive latent variables across groups.

Software packages such as Mplus (Muthén and Muthén 1998–2017) provide various measures of fit that can be used to evaluate the model. These measures of fit include the comparative fit index (CFI), the root mean square error of approximation (RMSEA), and the standardized root mean residual (SRMR). We consider models with a CFI value higher than 0.90 and RMSEA and SRMR values lower than 0.08 as acceptable (West et al. 2012). We adopt a bottom-up analytical strategy that begins with the least-restrictive (no invariance) model. We then introduce restrictions on the measurement parameters (factor loadings, intercepts, and thresholds). If the model fit does not significantly decrease by imposing additional restrictions, the more-restrictive invariant model could be supported by the data. Traditional difference tests based on chi-square values are sensitive to sample size (Saris et al. 1987). We follow the recommendations of Rutkowski and Svetina (2017) for cutoff values of fit indices when testing measurement invariance with categorical indicators. Accordingly, differences between a metric and a configural model are considered irrelevant when the change in the CFI is smaller than 0.004 and the change in the RMSEA is smaller than 0.05. Differences between a scalar and a metric model are considered irrelevant when the difference in CFI is smaller than 0.004 and the difference in RMSEA is smaller than 0.01. However, we do not consider the recommendations for chi-square differences because of the very large sample size in this study.

We treat all question items as categorical because the number of categories is always five or fewer and because the univariate distributions of the responses to the question items are highly skewed in many cases (Rhemtulla et al. 2012). Indeed, a large proportion of the respondents reported that they did not suffer from any physical or mental health issues. We use the mean- and variance-adjusted weighted least squares (WLSMV) estimator (Muthén et al. 1997), which is implemented in the software package Mplus (Muthén and Muthén 1998–2017) to address the categorical character of the data. With weighted least squares, the model in question is fitted to a vector of variable thresholds and a matrix of polychoric correlations assuming multivariate normality of the underlying (latent) continuous response variables. The fit function minimizes the weighted sum of differences between the observed polychoric and model-implied correlations (i.e., the model residuals). WLSMV estimation does not require large sample sizes and produces unbiased and efficient estimates under a number of conditions (e.g., Li 2016). With no covariates in the estimated models, the treatment of missing data is equivalent to pairwise deletion (Asparouhov and Muthén 2010b).¹

Results

Descriptive Results

Table 1 reveals significant variations across immigrants and natives in the self-reported health levels. Immigrants reported better physical health (i.e., perceived disability, ADLs, functional limitations) than native-born individuals, in line with the healthy immigrant effect hypothesis. However, with regard to mental health (i.e., nonspecific psychological distress and perceived financial stress), the findings are somewhat less consistent. Finally, immigrants tended to feel worse than natives on only two parameters of psychological distress (sadness and hopelessness), whereas they felt much more worried, on average, than natives on all items assessing financial distress. This pattern is similar across all ethnic groups of immigrants and natives.

Testing for Measurement Equivalence Across Native and Immigrant Ethnoracial Groups: MGCFA

Next, we present the results of the measurement equivalence testing using the MGCFA models performed for each of the five latent variables measuring perceived physical and mental health. First, we perform our analyses both in the overall sample and across the eight population groups described earlier. The models for each of the five latent variables—limitations in ADLs, functional limitations, self-reported disability, perceived financial stress, and nonspecific psychological distress—fit the data well both for the total sample as well as for the four immigrant (non-Hispanic White, non-Hispanic Black, non-Hispanic Asian, Hispanic) and four native-born (non-Hispanic White, non-Hispanic Black, non-Hispanic Asian, Hispanic) groups (i.e., CFI ≥ 0.90, RMSEA ≤ 0.08, SRMR ≤ 0.08). The standardized factor loadings are high and exceed 0.5 in almost all models and samples.²

Given the good fit of the models to all samples, measurement invariance testing is performed in the next step. Specifically, we impose equality constraints across eight samples on the measures consecutively (moving from a less-restrictive to a more-restrictive model, as described in the Methods section). At each step, we inspect whether the deterioration in model fit is too large to accept the more constrained model. Table 2 summarizes the results of the measurement invariance tests. As mentioned earlier, CFI ≥ 0.90, RMSEA ≤ 0.08, and SRMR ≤ 0.08 are considered acceptable (West et al. 2012).

The configural model for all constructs fits the data well: the fit indices for this model are acceptable across eight sample groups for the five health measures. For example, the fit indices for limitations in ADLs are as follows: CFI = 0.993 (≥0.90); RMSEA = 0.072 (≤0.08); SRMR = 0.073 (≤0.08). These results indicate that the same model structure is given across the eight sample groups. Thus, configural invariance as a baseline model for each scale is established.

Given that configural invariance is supported, we can test for higher levels of invariance by first imposing equality constraints on the factor loadings across groups.³ Differences between a configural and a metric model are considered irrelevant when the change in the CFI is smaller than 0.004 and the change in the RMSEA is smaller than 0.05 (Rutkowski and Svetina 2017). Table 2 indicates that for limitations in ADLs, for example, the change in CFI is only 0.001 (0.994 – 0.993), lower than 0.004; and the RMSEA is 0.012 (0.072 – 0.060), lower than 0.05. Similar results are obtained for the other four constructs. Next, we impose additional equality constraints on the item thresholds across groups. Differences between a metric and a scalar model are considered irrelevant when the difference in CFI is smaller than 0.004 and the difference in RMSEA is smaller than 0.01. Table 2 indicates that the differences are irrelevant for all constructs, suggesting that the health items have both equal semantic load and comparable latent means for immigrants and natives from four ethnoracial backgrounds. Thus, we conclude that our perceived physical and psychological health measures are comparable across the eight groups of immigrants and natives in the U.S. sample and that comparisons of latent mean scores across groups are valid for all constructs.

Next, we reexamine the measurement invariance properties of the scale, controlling for potential differences in the basic demographic composition of the subgroups by regressing the latent variables on age, gender, education, and family income while retesting for measurement invariance. We also control for the year of the survey. The measurement invariance results remain the same, indicating that differences in the demographic composition of the subgroups are no threat to the comparability of the measurements of physical and mental health across native and immigrant ethnoracial subgroups.

Figure 1 illustrates the differences in the latent variables measuring perceived physical and psychological health across the eight (four immigrant and four native) groups. The estimated latent means are based on the MGCFA scalar invariance models. The latent means for the non-Hispanic White natives group are set to 0; this is the reference group. Latent means of the other groups are estimated as a difference score between their means and the mean of the reference group. Higher scores on the bars refer to higher levels of reported limitations in ADLs, self-reported disability, functional limitations, perceived financial stress, and psychological distress.

Overall, Figure 1 demonstrates that all immigrant and native groups (with the exception of Black natives) displayed better physical and psychological health scores for four of the five measured constructs. Perceived financial stress was the only health measure for which the groups under study reported worse outcomes in comparison to White natives. Interestingly, Black natives reported worse health compared with White natives across all health indicators except psychological distress.

Testing for Measurement Invariance Within Native and Immigrant Ethnoracial Groups

We conduct additional measurement invariance tests to evaluate whether our measurement invariance findings hold within native and immigrant ethnoracial groups. We assess the comparability of the perceived health measures within the immigrant and native subgroups separately across the following categories: region of birth, language of interview, years since migration, gender, and age (see sections 2–6 of the online appendix). The measures again show the highest (scalar) level of invariance, thus indicating that comparisons within these groups are meaningful.⁴ Latent mean differences across the tested groups are presented in Figures A1–A5 in the online appendix.

Discussion and Conclusions

Immigration and health researchers have long reported that at the time of their arrival in a host country, immigrants are likely to be healthier than native-born populations (Akresh 2007). However, the reasons underlying the initial health differences between immigrants and natives remain a topic for vigorous research and debate. Previous studies have focused on selectivity explanations and have not examined whether the utilized health indicators are comparable across population groups of different ethnoracial origin and immigrant groups. That is, no coherent evidence exists regarding whether the healthy immigrant effect is true or a result of culturally different perceptions of health and variability in the way survey questions are understood by respondents. Testing for measurement invariance of the scale before comparing scores across immigrant and ethnoracial groups is a prerequisite, however, to allow researchers to draw valid conclusions about similarities and differences. Indeed, differences in health may result from differences in response styles of various immigrant and ethnoracial groups or from their different understanding of the concept. At the same time, differences in response characteristics or in the understanding of the concept may conceal actual differences. Thus, in the current study, we aim to fill this gap and investigate measurement equivalence of five commonly used health indicators across different immigrant and ethnoracial groups in the United States. We utilize data from the 2015–2017 IPUMS Health Surveys (Lynn et al. 2019), which include multiple-item measures of five health indicators: nonspecific psychological distress, perceived financial stress, perceived disability, limitation in ADLs, and functional limitations. Our analyses involve two analytical methods: MGCFA and alignment optimization.

The results of the MGCFA analysis reveal support for measurement equivalence in the data. Thus, the findings imply that the health scales used in the IPUMS Health Surveys are comparable across most of the immigrant and ethnoracial groups participating in the surveys and that their latent means may be compared by immigrant status and across ethnoracial origin with confidence. Therefore, the differences across ethnoracial and immigrant groups are due to the healthy immigrant effect or some additional factors but not due to lack of invariance.

Indeed, when comparing our invariant latent scores, we reveal considerable differences in the level of mental and physical health. Specifically, our analysis affirms the robustness of the key finding of the literature: namely, that immigrants are healthier than U.S. natives and that this advantage declines with years of stay in the United States. We also find that Blacks and Hispanics are, on average, healthier than their non-Hispanic White peers.

Although this research has succeeded in establishing measurement invariance across the groups explored, it is not without limitations. First, our study examines both perceived physical and psychological health measures, but the literature discusses several other self-reported measures of health that may be subject to noninvariance across the compared groups. Second, our study is limited to the U.S. context. Although we established measurement equivalence for the data at hand, it may not necessarily hold for other time points, across more specific ethnic groups (e.g., from specific geographical areas in the world), or in other countries. Third, because the data do not contain information as to whether the native-born respondents are second-generation immigrants, we cannot examine whether measurement invariance holds across the first and second generation of immigrants compared with the native-born. Finally, the data are restricted to those aged 85 or younger and exclude institutionalized adults, and the interviews were conducted in English and Spanish only. Furthermore, we cannot distinguish among immigrant groups by reason for migration (e.g., to distinguish between refugees and economic immigrants). These factors may have affected the results and potential comparability with other data sets. Future studies could address these important issues by analyzing data for other countries, covering a diverse range of immigrant groups at different points in time, and using additional scales measuring respondents' health.

Notwithstanding these limitations, the current study is one of the first to assess the measurement comparability of various measures of health across several ethnoracial and immigrant groups systematically using state-of-the-art empirical methods. Researchers interested in comparing the health measures across ethnoracial and immigrant groups in different countries may apply similar procedures to examine whether their concepts of interest display a sufficient level of equivalence across these groups. We hope that the current study will thus help researchers in their endeavor to conduct a meaningful comparative study of immigrants' health.

Acknowledgments

The authors would like to thank Lisa Trierweiler for the English proof of the manuscript prior to its acceptance, and Laura Tesch, Teresa Artman, and Bethany Sage Curtis for copy-editing after its acceptance. We also extend our gratitude to anonymous reviewers and to the editors of Demography for helpful comments. Eldad Davidov would like to thank the University of Zurich Research Priority Program Social Networks for support during work on this study.

Notes

References