|Home | About | Journals | Submit | Contact Us | Français|
Despite lower rates of mortality and some forms of morbidity, Latinos report worse self-rated health (SRH) than Whites. These inconsistencies have raised questions about the validity of SRH for cross-ethnic comparisons and its use as a measure of health disparities. We examine whether the translation of this measure into Spanish helps explain these patterns.
We analyzed levels of SRH under different language conditions using cross-sectional data from the 2002 Chicago Community Adult Health Study and the 2003 Behavioral Risk Factor Surveillance System.
Being interviewed in Spanish was associated with significantly higher odds of rating one’s health as fair/poor in both data sets, and adjusting for language of interview substantially reduced the SRH gap between whites and Latinos. Spanish-language interviewees were also more likely to rate their health as “fair” (“regular” in Spanish) than any other response category, after adjusting for age, sex, socioeconomic position, health conditions, and other factors. The association between being interviewed in Spanish and reporting “fair”/“regular” health was strongest when contrasted against response categories representing better health (good, very good, and excellent).
The findings support the hypothesis that the translation of the English word “fair” to “regular” induces Spanish-speaking respondents to report worse levels of health than they otherwise would in English. We recommend caution in interpreting this widely used instrument—especially when making racial/ethnic comparisons—and propose experimental research using different translations of this measure to arrive at one that better equates its meaning in Spanish and English.
For more than two decades, researchers have been intrigued by studies indicating that Latinos experience equal or better health outcomes than whites, despite having lower socioeconomic status—what has come to be known as the “Latino Health Paradox.”1-8 One important exception to the apparent Latino health advantage is self-rated health (SRH). When rating their overall health status (e.g., “Would you say that your health is generally excellent, very good, good, fair, or poor?”), Latinos’ assessments are lower than those of whites and below what would be expected from more objective measures.8-15 Understanding this puzzling side of Latino population health is critical to advancing research on racial/ethnic disparities in health and addressing their root causes.16
Some scholars who have studied Latinos’ relatively poor SRH hypothesize that it is more a reflection of cultural norms than actual physical health. For example, they suggest that, all else being equal, Latinos have low SRH (1) because they somatize psychosocial and/or emotional distress as physical health conditions, or (2) because their cultural orientation discourages “boasting” about health.11,12,17 To test these ideas, researchers have analyzed the extent to which Latino-white disparities in SRH are reduced after controlling for measures that might proxy Latinos’ distinct cultural orientation toward health.8-13,17-17 Studies examining the effects of language show that those who report greater proficiency in Spanish relative to English report lower levels of SRH,11-13,17-21 and that controlling for language use tends to reduce Latino-white disparities in SRH.11-13,17-19.
This paper advances an explanation for the relatively low levels of SRH among Latinos that, while not precluding other hypotheses, presents a more parsimonious and ultimately testable proposition: that the usual Spanish translation of the response categories to the SRH question induces Spanish-speaking respondents to report worse levels of health than they otherwise would if they were answering the question in English. The response categories for the SRH question are conventionally translated into Spanish as excelente (excellent), muy buena (very good), buena (good), regular (fair), and mala (poor). Because the Spanish word “regular” connotes more positive meanings than “fair” does in English, the translation of this response option may downwardly bias estimates of Latino health status. Although other scholars have suggested similar explanations,11-13,17 this is the first study to demonstrate empirically that respondents to the conventional Spanish translation of the SRH question are far more likely to report that their health is “fair”/“regular” than those who respond to the question in English.
We analyze ethnic and linguistic differences in the pattern of responses to the SRH question with data from two surveys that have complementary strengths: the Behavioral Risk Factor Surveillance System (BRFSS) and the Chicago Community Adult Health Study (CCAHS). The CCAHS is a multi-level, stratified probability sample of 3,105 adults, ages 18 and older living in Chicago, Illinois.22 The data were collected between May 2001 and March 2003 via face-to-face interviews with one individual per household, with a response rate of 71.8 percent. The data were weighted to match the age, race/ethnicity, and sex distributions of the 2000 Census population estimates for the city of Chicago. The BRFSS is an ongoing cross-sectional health survey of non-institutionalized individuals 18 years and older; the survey is conducted at the state level via telephone using a random-digit dialing sampling methodology.23 We use the 2003 BRFSS data from the 20 states that conducted the survey in English and Spanish: Arizona, California, Colorado, Connecticut, Florida, Illinois, Indiana, Kansas, Massachusetts, Nebraska, Nevada, New Jersey, New Mexico, New York, North Carolina, Oklahoma, Rhode Island, Texas, Utah, and Washington.
We chose these two data sets for substantive and practical reasons. We first tested our hypothesis in the CCAHS and then replicated the analysis on the BRFSS because it is a large, nationally-representative data set that has been used in prior studies of Latino SRH.19,21,23 The BRFSS was administered by phone, while the CCAHS was face-to-face, and the CCAHS contains a wider range of relevant controls, as explained below.
We also analyzed the 2002 National Health Interview Survey (NHIS) but do not report the results because of concerns about its comparability with other data sets vis-à-vis language of interview and Latino SRH. First, previous research has found (and we confirmed) that Latinos’ assessments of their overall health are substantially more positive in the NHIS compared to the BRFSS and other data sets.24,25 For example, a recent comparison of the 1997 NHIS and BRFSS found that the unadjusted risk ratio of being in poor or fair health for Latinos compared to whites was only 1.15 in the NHIS, far below the 1.80 risk ratio reported for the BRFSS.24 Thus, the problem that motivates our study—the relatively low assessments that Latinos give to their health—is much less evident in the NHIS, which, although puzzling and worthy of further investigation, makes it less relevant for our research. Also, the NHIS interview protocol is unique in allowing respondents to answer the questionnaire in English and Spanish on a question-by-question basis, rather than offering the entire interview only in one language, as most surveys do. Although it is possible to identify which NHIS respondents selected a mixture of English- and Spanish-language questions, it is not possible to know which respondents answered any particular question in Spanish or English. Also, the composition of people who responded to all questions in Spanish is likely to be different in the NHIS than it is in other studies, which further limits its comparability.
We constructed a core set of nearly equivalent measures from the CCAHS and BRFSS that are used in all models. We constructed two dummy variables for language of interview, one for Spanish and a second for being interviewed in another language (in the CCAHS, the only other language available was Polish, but the BRFSS was offered in a wider-range of languages, which vary from state to state and are not explicitly documented in the data set). Race/ethnicity is measured comparably in the two data sets with four categories: Latino, non-Latino white, non-Latino black, and non-Latino other race. We also control, in all models, for comparable measures of sex, age (measured categorically with intervals of 18-29, 40-49, 50-59, 60-69, and 70 or above), and household income (measured categorically with intervals of less than $15,000, $15,000-24,999, $25,000-34,999, $35,000-49,999, and $50,000 or more). Educational attainment was measured slightly differently in the two data sets: in the CCAHS, we used a measure of years of education completed (with categories of 0-11, 12, 13-15, and 16 or more), whereas in the BRFSS we used a measure of highest degree completed (less than high school, high school, some college but no degree, and college degree or higher).
In some models, we control for an additional set of health risk factors/conditions that are nearly comparable across the two data sets, including respondent reports of being unemployed, without health insurance, a smoker, a heavy drinker (five or more drinks on at least one occasion during the last 30 days), and ever being diagnosed with asthma, hypertension, diabetes, or arthritis. We also control for categories of body mass index (less than 25, 25-29, and 30 or higher), calculated from height and weight measurements that were collected anthropometrically in CCAHS but via self-reports in the BRFSS. When a variable had missing data, we treated “missing” as a category (all of our covariates are categorical) and included a dummy variable, but we did not report these coefficients in the tables.
In supplemental analysis (not shown in the tables but reported in the text), we controlled for the following immigration-related factors measured in the CCAHS but not in the BRFSS: nativity status (foreign born); age at migration (whether a foreign-born respondent migrated before or after age 17); a dichotomous measure of linguistic acculturation (indicating whether the respondent meets the following conditions: (a) speaks and reads only in Spanish,(b) speaks only Spanish at home, and (c) speaks only Spanish with friends). We also included indicators of Latino subgroups (Mexicans, Puerto Ricans, and other Latinos).
Our analysis proceeds in two stages. First, we replicate and extend prior work on Latino SRH disparities by using logistic regression to compare Latinos and other racial/ethnic groups on the odds of reporting that one’s health is poor or fair. We begin with a model that examines racial/ethnic disparities in SRH, adjusted for sex, age, education and income. We next control for language of interview to see how much it changes estimates of Latino-white disparities in SRH, and then add controls for additional health risk factors/conditions that could confound the association between language of interview and SRH. In the second stage of the analysis, we test our hypothesis that Spanish interviewees will be more likely to rate their health as regular, all else being equal, using multinomial logistic regression models to predict the odds of responding that one’s health is fair/regular compared to other categories. All of our analyses were weighted to account for each survey’s complex design using Stata version 10/SE.
We present descriptive statistics on the CCAHS and BRFSS in Table 1, first for all respondents and then for the Latino sub-sample from both data sets. The distribution of SRH across response categories is fairly similar in the two data sets, for both the full sample and the Latino sample. For example, the unadjusted odds ratio (not shown) of being in poor or fair health for Latinos compared to whites is 2.95 (p < 0.001) in the CCAHS and 2.24 (p < 0.001) in the BRFSS. Also noteworthy is that the proportion of Latinos who were administered the questionnaire in Spanish was almost identical in the two data sets (0.44 in the CCHAS and 0.43 in the BRFSS). Reflecting particular aspects of Chicago’s population, the CCAHS has a different racial/ethnic distribution than the BRFSS (with higher proportions of Latinos and blacks and a smaller proportion of whites), and CCAHS respondents have lower levels of education than those in the BRFSS, but otherwise, the two data sets are fairly comparable on most variables.
In Table 2 we present logistic regression models predicting poor/fair SRH. The results from model 1 show that the odds of rating one’s health as poor or fair are 86 percent higher for Latinos compared to whites in the CCAHS and 66 percent higher in the BRFSS, after adjusting for sex, age, education, and income. Model 2 shows that being interviewed in Spanish is associated with significantly higher odds of poor/fair SRH, and that adjusting for language of interview narrows the estimated Latino-white SRH gap considerably, to a point where the coefficients are almost identical in the two data sets (and the CCAHS coefficient is no longer significant). In model 3 we add controls for other health risks/conditions that could confound the association between language of interview and SRH. Interestingly, we find that the Spanish language effect gets larger in both data sets after adjusting for these risk factors, and supplemental analysis found that the main reason for this increase was that Spanish interviewees reported fewer chronic health conditions; thus controlling for those conditions increases their odds of reporting poor/fair SRH. The implications of adding these controls for the Latino-white SRH gap was different in the two data sets: the gap narrowed in model 3 in the CCAHS but increased in the BRFSS.
The results from Table 2 generally replicate what has been found in prior studies regarding the role of language of interview in reducing SRH disparities between Latinos and whites. However, most prior studies have stopped here and not investigated whether the translation of the SRH response categories from English to Spanish results in a “heaping” of responses on the “fair” category among Spanish speakers after multiple controls. To test this hypothesis, we turned to multinomial logistic regression models that estimate the log odds of responding “fair” health compared to each of the other categories. Table 3 presents the relative risk ratios associated with having a Spanish-language interview on the odds of rating one’s health as fair/regular compared to the other options. Only the Spanish-language coefficients are displayed in Table 3, but model 1 adjusts for the “core” set of covariates (see Table 2, model 2), and model 2 adds controls for the additional health risk factors/conditions (see Table 2, model 3). The results show that respondents who interviewed in Spanish were more likely to rate their health as regular than any other category, even after adjusting for the full set of covariates in model 2, and in all but one case (fair vs. poor health in CCAHS, model 2), this association was significant. Also noteworthy in model 2 is that the tendency for Spanish interviewees to rate their health as regular is most pronounced when contrasted to the odds of selecting categories that represent better health (good, very good, or excellent), consistent with the idea that Spanish-language respondents may have better assessments of their health than the English word “fair” implies. In addition, the Latino-white gap in the odds of reporting “fair” health is reduced by controlling for language of interview (results not shown).
Supplemental analysis (not shown) assessed whether the association between being interviewed in Spanish and rating one’s health as regular was confounded by immigrant status or level of linguistic acculturation, measures available in the CCAHS but not in the BRFSS. Controlling for nativity status, age at migration, and the linguistic acculturation scale did not alter our results. We also examined whether the language-of-interview effect on SRH varied by nativity, age at migration, and Latino subgroup by including appropriate interaction terms in our models but found none to be statistically significant.
To help interpret the results from the multinomial models, we used the coefficients from model 1 in Table 3 to calculate the predicted probability of responding to each SRH category, and graphed these predicted probabilities in Figure 1, highlighting the comparison between Latinos who responded to the interview in English versus Spanish. The predicted probabilities that Latino-English and Latino-Spanish interviewees will rate their health as “fair” are very comparable across the two data sets: the probability of selecting “fair” for Latino-English interviewees was 0.11 in the CCAHS and 0.14 in the BRFSS, while the probability of selecting “regular” for Latino-Spanish interviewees was 0.24 in the CCAHS and 0.26 in the BRFSS. The predicted probabilities for the other categories did not match as closely across the two data sets, with the main difference being that the Latino-Spanish respondents in the CCAHS were more likely than their counterparts in the BRFSS to rate their health as excellent or very good (compared to good).
Our analysis of the CCAHS and the 2003 BRFSS supports the hypothesis that language of interview is a critical source of variation in SRH among Latinos. Consistent with prior studies,11-13,17-21 we found that Latinos who interviewed in Spanish reported worse SRH than those who interviewed in English, and controlling for language of interview significantly reduced the differences in SRH between Latinos and whites in both data sets.11-13,17-19 Moving beyond prior studies, we examined whether the translation of one response category—“fair/regular”— downwardly biases the distribution of SRH among Latinos who responded to the survey in Spanish. Our multinomial analysis demonstrated that in both samples, people who interviewed in Spanish had higher odds of reporting “fair”/“regular” health than those who interviewed in English, after controlling for a host of possible confounders, including health status, education, income, and immigrant status. The propensity for Spanish-language interviewees to select “fair”/“regular” was most pronounced when it was contrasted to the odds of selecting categories that reflect better levels of health—a finding consistent with the hypothesis that translation problems induce some Spanish-language respondents to answer regular when they actually mean to rate their health in more positive terms.
Although our findings regarding the effect of language of interview on SRH are compelling, they do not preclude other explanations for why Latinos report relatively low levels of SRH relative to whites. Cultural theories—such as the idea that a traditional cultural orientation among Latinos leads adherents to “downgrade” their health assessments or somatize physical symptoms—have been offered in previous studies to explain why first-generation immigrants and Spanish speakers have lower SRH.11,12,17
Other scholars have suggested that newer immigrants may be more likely to downgrade their ratings because their health has suffered from the stressfulness of migration and barriers to social integration or because immigrants’ reference groups shift through the process of migration and settlement.14,20 Although we cannot rule out this explanation, in supplemental analysis of the CCAHS, we took several steps to ensure that the language-of-interview effect was not confounded by other differences between immigrants and non-immigrants. We controlled for nativity status, age at migration, and linguistic acculturation, and found that that the language-of-interview effect remained strong and significant. However, the language of interview variable was highly correlated with both nativity and linguistic acculturation, making it difficult to disentangle the effects of the survey translation from other social and cultural factors that differentiate Spanish-from English-language interviewees.
Another explanation that has been offered for Latinos’ poorer health self-assessments is that SRH may reflect limited access to life opportunities and resources, as reflected in their lower levels of education.12,26,27 Indeed, education had strong and significant effects on the odds of being in poor/fair health and even on the propensity to respond that one is in “fair” health. Moreover, supplemental analysis revealed that controlling for education reduced the Latino-white gap in the odds of having poor/fair SRH almost as much as language of interview in both datasets.
One limitation of our findings is that we were unable to replicate them in our analysis of the 2002 NHIS. In our supplemental analysis of the NHIS, we found that the Latino-white SRH disparity was much smaller (OR = 1.17 in a model comparable to model 1 from Table 2), and the effect of being interviewed in Spanish was nonexistent (OR = 0.98 in a model comparable to model 2 from Table 2). We also compared descriptive statistics on Latinos from the NHIS to those from the BRFSS and CCAHS to see if there were any differences that might explain why the distribution of Latino SRH was higher in the NHIS than in the other data sets. The main differences we found were that the NHIS had a lower proportion of Latinos in general (0.10 in the NHIS, 0.19 in the BRFSS, and 0.26 in the CCAHS) and of Latinos who were interviewed only in Spanish (0.25 in the NHIS, 0.43 in the BRFSS, and 0.44 in the CCAHS). An additional 12 percent of the Latinos in NHIS were interviewed in English and Spanish, but it is impossible to determine which language was used for the SRH question.
Thus, one possible reason why Latinos in NHIS give more positive assessments of their health is that fewer of them were interviewed in Spanish. However, when we replicated the multinomial logit models from Table 3 on the NHIS, we did not find any significant effects of interviewing in Spanish, so it is not clear that increasing the number of NHIS respondents who interviewed in Spanish would change the distribution of Latino SRH. Still, with a smaller proportion of NHIS Latinos interviewing only in Spanish, it is likely that NHIS Spanish-language interviewees are selectively different from their counterparts in the other data sets, and this is a major reason why we do not report these results alongside those from the BRFSS and CCAHS.
Another reason that the distribution of SRH may vary across surveys is that it may be affected by variation in the ordering of the SRH question vis-à-vis questions on specific health conditions. For example, if a respondent has few adverse health conditions and is asked the SRH question after being asked about specific conditions, she might view her overall health more favorably than she would otherwise. Since Latinos tend to report fewer chronic health conditions, they may be particularly sensitive to the ordering of the SRH question relative to questions on chronic conditions. In fact, one experimental study found that Spanish-language interviewees reported worse SRH when the SRH question was asked before inquiring about chronic health conditions.28 In the BRFSS and NHIS, the SRH question was asked before questions on specific health conditions, whereas the order was reversed in the CCAHS (also, the specific health conditions being asked about vary across the three studies).
A final noteworthy limitation is that we did not disaggregate our estimates of the Latino-white SRH gap by Latino subgroup because such data are not available in the BRFSS, and some subgroups are small in the CCAHS. The dominant Latino subgroup in the CCAHS was Mexicans (67% of all Latinos in the CCAHS), and all of the main findings held up for Mexicans when we did such a disaggregation in the CCAHS.
The most important implication of our findings is that great caution must be taken in analyzing SRH data when the standard translation of the question from English to Spanish is used. In the future, researchers might consider alternative Spanish translations of the English word “fair,” such as “así así” or “más o menos,” which have more negative connotations than “regular.” Interestingly, this translation problem may not be limited to Spanish, as our multinomial models also detected that respondents who interviewed in other languages were more likely (albeit not significantly) to report being in “fair” health. Perhaps the English adjective “fair” (in the sense of “so so” or “moderately or tolerably good”) is complicated to translate into most other languages, especially if done somewhat literally.13,29
Still, further research—preferably with an experimental component—is necessary to determine whether differing interpretations of the SRH question in Spanish (or other languages) versus English are due to its translation, to culturally mediated understandings of this item, to immigrant social integration, or to a combination of these factors. One experiment would be to randomly assign a group of respondents interviewed in Spanish to be asked questions with an alternative translation of “fair” and to compare their responses to those who were given the conventional translation.
There is also a need for further qualitative research that investigates how Latinos interpret the meaning of the SRH question. In particular, cognitive interview techniques—asking respondents to think aloud about each question and their process for arriving at an answer— would be useful in shedding light on the role of language in shaping the interpretation of the SRH response options.30
Clearly, a better understanding of the factors causing disparities in response to SRH is necessary, given the importance of this measure as an indicator of population health and for predicting health outcomes. Until then, scholars drawing scientific, policy, or clinical implications from this measure—especially with regard to comparisons across ethnic groups— should do so with caution and awareness of this potential linguistic bias.