Impact of the Terms “Regular” or “Pasable” as Spanish Translation for “Fair” of the Self-Rated Health Question Among US Latinos: A Randomized Experiment

Abstract

Objectives. To examine measurement comparability of a Spanish version of self-rated health (SRH) with pasable as an alternative to regular for the response category “fair” in the English version.

Methods. We translated “fair” into 2 Spanish versions: regular and pasable. We implemented a split-half experiment in 3 surveys independently conducted from October 2015 to January 2016, from April to November 2016, and from August to November 2017. Within each survey, we randomly assigned Spanish-interviewed Latino respondents to 1 of the 2 SRH versions. The total sample included 3261 Latino and 738 non-Latino White adults in the United States.

Results. Spanish-interviewed Latinos reported substantively more favorable health on SRH with pasable than with regular. When pasable instead of regular was used for SRH, we observed a larger difference between respondents reporting positive versus negative SRH on objective health measures, including the frequency of doctor’s visits. Furthermore, when we accounted for correlates of health, Latino–White disparities were attenuated with pasable.

Conclusions. We recommend using pasable instead of regular for SRH Spanish translations to improve measurement equivalence in cross-lingual and cross-cultural research.

Self-rated health (SRH), an indication of a person’s perceived own health status, is measured by response to a single question: “Would you say that in general your health is excellent, very good, good, fair, or poor?” Backed by numerous studies reporting its predictive power for mortality,^1,2 SRH has become one of the most frequently asked items in health surveys as well as quality-of-life assessments.³ Considered as a reliable proxy for true health,^4–6 SRH is analyzed to monitor and compare population health^7–9 and included as a profile characteristic of important epidemiological study cohorts.¹⁰

Although SRH is used to examine racial/ethnic health disparities in the United States,^11–14 it has been questioned whether measurement comparability of SRH confounds comparisons, particularly between Latinos and non-Latinos.^12,15–18 Latinos choose “fair” more frequently than non-Latino Whites (NLWs),¹⁹ resulting in consistently less favorable assessment of SRH for Latinos than NLWs, even after controlling for relevant correlates, such as socioeconomic status and objective health conditions.^20,21 Moreover, unlike for NLWs, SRH is an inconsistent predictor of Latinos’ mortality.¹⁷

One hypothesis for this noncomparability concerns the Spanish translation of the SRH question. The translation usually reads: ¿Diría que, en general, su salud es excelente, muy buena, buena, regular, o mala? In particular, it is the English response category “fair” translated as regular in Spanish that has been suggested as a source of noncomparability.¹⁵ Some researchers suggest that regular means something better than “fair,” as regular in Spanish is “a common way of describing a level of health that is common and perhaps normal”²². Although Spanish-language questionnaires are not always available, most Spanish-version questionnaires of established surveys—such as the US National Health Interview Survey—appear to use regular for “fair.” Some surveys (e.g., the Kidney Disease Quality of Life Instrument Short Form), however, use pasable instead. Interestingly, Short Form 36 for the Multicenter Aids Cohort Study in the United States uses pasable, whereas another Short Form 36—for the Medical Outcomes Survey—uses regular. Unfortunately, there is no literature indicating why there is divergence on this translation and, more importantly, how the use of regular or pasable affects the measurement of SRH. Although both pasable and regular are used for “fair,” they have different nuances: regular indicates average, whereas pasable suggests barely acceptable.

To improve the English–Spanish comparability of SRH, we examined the use of pasable as an alternative to regular for “fair.” As “fair” in English connotes negative health, we hypothesized that the use of pasable will better align with “fair” than regular. To test this hypothesis, we implemented an experiment that randomly assigned Spanish SRH using regular or pasable in 3 separate studies that included US Latinos and NLWs. We compared the response distributions of the 2 Spanish versions of SRH. When applicable, we evaluated the concurrent validity—if pasable is better than regular, SRH with pasable should result in a stronger relationship with other health measures. Furthermore, we examined the pattern of the Latino–NLW SRH disparity by translation.

METHODS

To subject our hypothesis to various methodological contexts, we implemented the same experiment in 3 independent studies. These studies differed with respect to target populations, recruitment methods, sample characteristics and interview modes as described below.

Data

Study 1: Web survey of Spanish-speaking US Latinos.

SurveyMonkey collected data using a Web survey from October 2015 to January 2016. We recruited respondents using a Web intercept method.²³ Participants in various Spanish surveys created by SurveyMonkey users were “intercepted” from those surveys’ end pages by inviting them to our Web survey. When they agreed, these participants were directed to our survey conducted only in Spanish. Of an original 1415 participants, we retained only 1235 who identified as US Latino, were aged 18 years or older, and completed the survey for this analysis. The survey asked questions about cellphone, smartphone, and tablet ownership and usage, followed by SRH and sociodemographics; it lasted about 6 minutes.

Study 2: telephone survey of US Latinos.

We conducted a telephone survey from April to November 2016 with 913 Mexican American, Cuban American, and Puerto Rican adults in the United States and Puerto Rico. Approximately half of the participants (n = 437) came from a list of respondents of an earlier telephone survey, whereas the remainder (n = 476) were newly recruited. We randomly selected participants in both sources from a list of telephone numbers associated with individuals with 12 years or less of education and a household income of $25 000 or less. This education and income information came from commercial data and was attached to the telephone numbers, a method used for targeting population subgroups.^24,25 Eligible respondents were aged 18 to 90 years, spoke English or Spanish, and self-identified as Mexican American, Cuban American, or Puerto Rican. Furthermore, only those who showed acquiescent response tendencies on a 20-item screener were eligible. (Note that our experiment was embedded in a survey specifically designed to examine acquiescent response-style bias among Latinos.)

Unlike in study 1, we conducted study 2 in both Spanish and English. Interviews started in the language a respondent used in answering the call. During the screener interview, we asked 3 questions related to English and Spanish use for speaking with friends and family and for thinking. If a respondent indicated using 1 language more than the other on average, we assigned that language for the main interview. If a respondent used both languages equally, we randomized the main interview language. Furthermore, we asked if respondents were comfortable with the assigned language; if they were uncomfortable, we switched the language. Only 67 of 913 respondents completed the main survey in English. Because of the small sample size, this analysis excluded study 2 English interviews.

The questionnaire started with SRH, followed by sociodemographics and a series of statements designed to measure acquiescence. The main interview lasted about 37 minutes. The response rate based on American Association for Public Opinion Research response rate 5²⁶ was 49.4% for the recontact (multiplying rates from the earlier survey and this survey) and 18.7% for the newly recruited cases.

Study 3: Web survey of US Latinos and non-Latino Whites.

We conducted a Web survey from August to November 2017 with respondents pre-identified as English-dominant Latino, Spanish-dominant Latino, and NLW adults in Qualtrics online panel, which has gained popularity from the research community.²⁷ To Spanish-dominant Latinos, we sent a Spanish invitation e-mail with a link to the Spanish questionnaire; to the remainder, an English invitation was sent with a link to the English questionnaire. The questionnaire comprised a series of objective health questions about health conditions and care utilization, anchoring vignettes related to pain, mobility, sleep and affects, and detailed sociodemographic questions. SRH was asked at the end of the survey. Respondents completed the survey, on average, in 18.8 minutes. Our survey asked race/ethnicity, in addition to the pre-identified race/ethnicity provided to Qualtrics. Those whose pre-identified race/ethnicity did not match the self-reports in our survey were considered ineligible and excluded from the analysis. This resulted in a total of 1851 respondents: 738 NLWs and 1113 Latinos; among the latter, there were 720 Spanish and 393 English interviews.

Unlike studies 1 and 2, study 3 specifically included English-dominant Latinos as well as NLWs. This allowed us to examine an effect of translation when comparing NLW and Latino SRH. Study 3 asked a wide range of health-related questions, some of which served as validation measures.

Table 1 summarizes our study samples (separated into Latinos and NLWs for study 3) with respect to age, gender, education, income, nativity, and interview language, and compares them with the population data from the American Community Survey. Compared with respondents in studies 1 and 3, study 2 respondents were older, more likely to be female, and had lower education and income levels. This pattern held when we compared the study 2 sample to the population data. Compared with study 3 respondents, the study 1 sample was more educated but had lower income and was more likely to be foreign-born and to be interviewed in Spanish. Across all studies, Latinos in our samples had lower income and were more likely to be born outside of the United States compared with the general Latino population. The NLW sample in study 3 was similar to the general NLW population in age and gender, but somewhat less educated and with lower income. Although about 31.0% of the general Latino population has limited English proficiency (i.e., speaks English less than very well), our Latino samples in studies 2 and 3 had high rates of Spanish interviews, perhaps reflecting a higher level of foreign nativity than in the general Latino population.

TABLE 1—

Comparison of Study Samples and Population: United States, October 2015–November 2017

	Study 1	Study 2	Study 3		Population
	Latinos (n = 1235), No. or %	Latinos (n = 913), No. or %	Latinos (n = 1113), No. or %	Non-Latino Whites (n = 738), No. or %	Latinos, No. or %	Non-Latino Whites, No. or %
Age, y
18–34	32.3	4.4	37.5	22.2	40.9	26.2
35–44	27.4	7.1	18.4	16.0	21.3	14.7
45–54	23.9	13.0	16.6	17.2	16.9	17.5
55–64	13.5	19.1	20.0	22.2	11.1	18.3
≥ 65	2.8	56.4	7.5	22.4	9.8	23.2
Female gender	55.4	80.2	58.8	50.3	49.7	51.2
Education
< high school	10.0	51.8	5.9	5.4	33.3	7.7
High school/GED	16.7	23.0	37.2	44.7	27.4	27.7
Some college	32.4	12.3	28.8	17.9	24.1	30.1
≥ college degree	41.0	12.9	28.2	32.0	15.2	34.5
Annual income, $a
< 30 000	59.5	NA	39.9	34.4	31.6	22.7
≥ 30 000 to < 50 000	19.3	NA	26.2	19.7	21.4	17.1
≥ 50 000 to < 100 000	14.0	NA	25.7	29.3	29.9	31.0
≥ 100 000	7.3	NA	8.3	16.7	17.0	29.2
< 20 000	NA	70.9	NA	NA	NA	NA
≥ 20 000 to < 40 000	NA	19.9	NA	NA	NA	NA
≥ 40 000 to < 80 000	NA	7.5	NA	NA	NA	NA
≥ 80 000	NA	1.7	NA	NA	NA	NA
Foreign-born	93.5	90.5	47.9	3.5	34.4	3.9
Latino subgroup
Cuban	NA	33.0	NA	NA	3.8	NA
Mexican	NA	34.3	NA	NA	63.2	NA
Puerto Rican	NA	32.7	NA	NA	9.6	NA
Other	NA	NA	NA	NA	23.4	NA
Interview language: Spanish	100.0	93.2	64.7	NA	NA	NA
Speaks English < very well	NA	NA	NA	NA	31.0	1.5
Spanish interview	1235	846	720	NA	NA	NA
SRH experiment
Regular	612	417	373	NA	NA	NA
Pasable	623	429	347	NA	NA	NA

Note. GED = general equivalency diploma; NA = not available; SRH = self-rated health. Population estimates are from the American Community Survey (ACS) 2012–2017 5-year summary file. The ACS estimates that there are 159 521 916 non-Latino and 38 341 709 Latino adults aged 18 years or older. The study dates are as follows: study 1: October 2015–January 2016; study 2: April–November 2016; study 3: August–November 2017.

^aThe first set of incomes was used for studies 1 and 3 and the second set for study 2.

Self-rated health Spanish translation experiment.

We embedded a split-half experiment in all 3 surveys. Targeting only Spanish interviews, we asked a random half of respondents in each survey to indicate their SRH using a response scale of excelente—muy buena—buena—regular—mala; the other half used the scale excelente—muy buena—buena—pasable—mala. Other than regular versus pasable, the Spanish wording was identical. We did not implement an experiment for the English interviews. The sample sizes for the total sample, for Spanish interviews, and for each SRH condition are included in Table 1. We determined the Spanish wording assignment at the start of Spanish questionnaires through randomization.

Validation measures.

The experiment only indicated whether regular and pasable produced different distributions in SRH responses. To examine which translation version is “better,” we used a series of variables from study 3 to create the following 5 validation measures. The first measure was the number of diagnoses of the following 17 chronic conditions: asthma, heart diseases, hypertension, cholesterol, diabetes, chronic obstructive pulmonary disease, osteoporosis, cancer, ulcer, Parkinson’s disease, eye conditions, Alzheimer’s disease, dementia, cognitive impairment, arthritis, epilepsy, and hay fever. The second measure was the number of doctor’s visits in the past 12 months.

For the third measure, we computed the difference between “subjective age”²⁸ (as determined by response to the question “How old do you feel?”) and chronological age. To alleviate extreme values, we coded respondents who reported having more than 18 doctor’s visits (n = 45) or feeling younger than 10 years old (n = 15) as having 18 visits and feeling 10 years old, respectively. (Note that the doctor’s visit question was asked only among those with health insurance, which excluded 77 NLWs, 55 Latinos interviewed in English, and 209 Latinos interviewed in Spanish. Results of our analysis involving the doctor’s visit were consistent whether excluding those without health insurance or coding them as not visiting doctors. Therefore, we report analysis excluding these respondents.)

The last 2 measures related to mental health: presence of high psychological distress as determined by the Kessler 6²⁹ and whether respondents had taken any medicine for mental health or emotional problems in the past 12 months. We regard these as validation measures, as it is logical to assume that poorer SRH should be associated with feeling older than one’s chronological age.

Analysis Procedures

The main dependent variable was SRH, which we examined using all 5 response categories as well as a binary that collapsed “fair” and “poor” into negative SRH and the remaining categories into positive SRH. Note that when we analyzed SRH as a binary and an ordinal variable, the difference was negligible.³⁰ The primary independent variable was the Spanish translation of “fair”: regular versus pasable.

We first examined whether SRH translation versions had an effect on the distribution of SRH, using the Fisher exact test for the SRH with 5 response categories and the binary SRH. We then examined whether the translation effect was consistent across age (< 35, < 45, < 55, or < 65 years vs ≥ 65 years), gender (male vs female), education (≤ high school vs > high school) and Latino subgroup (Cuban or Mexican vs Puerto Rican; measured only in study 2).

Focusing on Spanish-language interviews from study 3, we compared those categorized into positive versus negative SRH on the number of chronic conditions, the number of doctor’s visits, the subjective and chronological age difference, the rate of high psychological distress, and whether taking medication for mental or emotional problems. To accommodate the distribution of these respective validation measures, we used the Mann–Whitney–Wilcoxon test for the first 3 factors (continuous, not normal) and the Fisher exact test for the last 2 factors (binary). We then examined the concurrent validity by SRH translation versions. This was done through comparing the differences between regular and pasable versions in each validation measure among respondents classified as having negative SRH and among those having positive SRH. We used the Dunn’s Kruskal–Wallis multiple comparison for the first 3 factors and the Fisher exact test multiple comparison with Benjamini and Hochberg adjustment for the remaining validation measures. Here, the focus was on whether the difference between respondents with negative and positive SRH followed the same pattern between regular and pasable, and, if not, which translation version produced a larger difference in the validation measures. The version with a larger difference would be considered “better” in terms of concurrent validity.

The last analysis attempted to address the major issue of SRH that stems from noncomparability between Latinos and NLWs. Using study 3 data, we fitted a multiple logistic regression that modeled positive SRH as a function of Latino ethnicity and controlled for age, gender, education, income, and all 5 validation measures. The model was fitted separately to a sample composed of NLWs, English-interviewed Latinos, and Spanish-interviewed Latinos assigned to SRH with regular and to a sample composed of NLWs, English-interviewed Latinos, and Spanish-interviewed Latinos assigned to SRH with pasable. The 2 models mimicked situations examining NLW–Latino disparities in reported SRH using regular and passable, respectively. The focus of this analysis was the coefficient on Latino ethnicity and whether Latinos reported worse SRH than NLWs consistently between 2 translation versions after we controlled for characteristics relevant to health. This analysis used casewise deletion for item missing.

RESULTS

Pasable produced different response distributions of SRH than regular. Across all 3 studies, Latinos’ health appeared better with pasable than with regular. In particular, the proportion of “fair” responses decreased with pasable, as shown in Figure 1, and that of “good” as well as “very good” responses increased consistently across studies. For example, in study 2, more than half (51.3%) of the respondents would be designated as having “fair” health using regular but only slightly more than one third (38.0%) using pasable, yielding a 13.3 percentage point difference. Overall, the translation effect on SRH responses was significant in study 2 (P < .001), but not significant in study 1 (P = .085) or study 3 (P = .205).

FIGURE 1—

Distribution of Self-Rated Health Response by Translation of “Fair” Response Category, Studies 1–3 Spanish-Language Interviews Only: United States, October 2015–November 2017

Note. P values are from the Fisher exact test of association between self-rated health responses and translation of “fair” within each study. The study dates are as follows: study 1: October 2015–January 2016; study 2: April–November 2016; study 3: August–November 2017.

The translation effect was more apparent when we used the binary SRH. In study 1, the positive SRH was 82.2% (SE = 1.5%) with regular, which was lower than 87.6% (SE = 1.3%) with pasable (P = .009 by Fisher exact test). The corresponding rates were 39.8% (SE = 2.4%) and 49.6% (SE = 2.4%) for regular and pasable, respectively, in study 2 (P = .005), and 73.2% (SE = 2.3%) and 79.5% (SE = 2.2%) in study 3 (P = .050). Positive SRH significantly increased with pasable compared with regular across studies.

Figure 2 shows positive SRH between the 2 SRH versions for all 3 studies across age, gender, education, and Latino subgroups. Positive SRH was consistently higher (by 5 to 7 percentage points) with pasable than with regular across age, gender, and education. This pattern held true for each study (results not shown). This translation effect, however, was not consistent across Latino subgroups. The largest effect was observed for Mexicans, for whom pasable resulted in positive SRH more than 20 percentage points higher than observed with regular. For Puerto Ricans, there was a 9.7-percentage-point difference; for Cubans, there was virtually no difference.

FIGURE 2—

Proportion of Positive Self-Rated Health (SRH) by Translation of “Fair” Response Category, in Studies 1–3 Spanish-Language Interviews Only, by Respondent (a) Age, (b) Gender, (c) Education, and (d) Latino Subgroup: United States, October 2015–November 2017

Note. The study dates are as follows: study 1: October 2015–January 2016; study 2: April–November 2016; study 3: August–November 2017.

Relationship Between Self-Rated Health and Validation Measures

On average, study 3 Spanish-interviewed Latinos designated with positive (“excellent,” “very good,” or “good”) SRH reported 1.1 chronic conditions, significantly lower than for those with negative (“fair” or “poor”) SRH, who reported 2.3 conditions (P < .001 by Mann–Whitney–Wilcoxon test). Positive SRH was associated with fewer doctor’s visits (3.1 vs 5.4; P < .001). Respondents with positive SRH reported feeling 4.6 years younger than their chronological age, whereas those with negative SRH reported feeling 2.5 years older than their age (P < .001). The psychological distress rate was almost twice as high among respondents reporting negative SRH as among those reporting positive SRH (34.8% vs 17.3%; P < .001 by Fisher exact test). Similarly, respondents with negative SRH reported higher rates of having taken medication for mental or emotional problems in the past 12 months (32.2% vs 11.3%; P < .001).

Appendix A (available as a supplement to the online version of this article at http://www.ajph.org) further breaks down these patterns by SRH versions. Although the relationships between SRH binary (positive vs negative health) and the validation measures examined in the previous paragraph were consistent between regular and pasable, the positive–negative SRH difference tended to be larger for pasable than for regular across all measures. For respondents with positive SRH, there were no significant differences between the 2 translation versions across all measures. By contrast, respondents with negative SRH using pasable reported significantly more doctor’s visits (P < .01), felt significantly older (P < .05), and had a significantly higher rate of having taken medication for mental or emotional problems (P < .01) than did those with regular. For example, respondents with positive SRH reported feeling 4.6 to 4.7 years younger than their actual age regardless of the translation. On the other hand, respondents with negative SRH reported feeling 5.3 years older with pasable but only 0.7 years older with regular. Overall, the positive–negative SRH differences indicate that SRH with pasable had a larger discrimination power than SRH with regular.

Relationship Between Latino Ethnicity and Self-Rated Health

Latinos in study 3 reported positive SRH (75.7%; SE = 1.3%) at similar rates as NLWs (77.2%; SE = 1.5%). When we combined Latinos interviewed in English and in Spanish using regular, the positive SRH rate decreased to 73.9% (SE = 1.6%); when we did the same with Latinos interviewed in Spanish and in English using pasable, it was 76.9% (SE = 1.6%).

We examined the effect of Latino ethnicity on positive versus negative SRH in a multiple logistic regression (Table 2). When using the SRH response category “fair” translated as regular, Latinos appeared less likely to have positive SRH than NLWs (b = –0.393; P = .019). However, this NLW–Latino difference disappeared with pasable, as Latino ethnicity was not a significant predictor (b = –0.101; P = −.549). The model fitted reasonably well for both analyses, with the area under the receiver operating characteristic (ROC) curve above 0.8 and McFadden’s R² around 0.25, and slightly better with pasable than with regular.

TABLE 2—

Multiple Logistic Regression of Positive Self-Reported Health by Translation of “Fair” Response Category, Study 3 Only: United States, August–November 2017

Covariates	Regulara (n = 1238)		Pasableb (n = 1228)
	b (SE)	P	b (SE)	P
Race/ethnicity
Non-Latino White (Ref)	1		1
Latino	−0.393 (0.165)	.019	−0.101 (0.169)	.55
Age, y
18–34 (Ref)	1		1
35–44	0.058 (0.263)	.82	0.286 (0.280)	.31
45–54	−0.605 (0.248)	.015	−0.509 (0.256)	.047
55–64	−0.587 (0.246)	.017	−0.531 (0.253)	.036
≥ 65	−0.414 (0.297)	.16	−0.136 (0.313)	.67
Gender
Female (Ref)	1		1
Male	−0.099 (0.163)	.55	−0.120 (0.169)	.48
Education
< high school (Ref)	1		1
High school/GED	0.342 (0.352)	.33	0.349 (0.363)	.38
Some college	0.492 (0.371)	.19	0.623 (0.385)	.11
≥ college degree	0.910 (0.383)	.017	1.006 (0.402)	.012
Annual income, $
< 30 000	1		1
≥ 30 000 to < 50 000	0.444 (0.205)	.031	0.265 (0.215)	.22
≥ 50 000 to < 100 000	1.029 (0.218)	< .001	0.724 (0.223)	.001
≥ 100 000	1.210 (0.309)	< .001	0.990 (0.324)	.002
No. of chronic conditions	−0.266 (0.046)	< .001	−0.259 (0.048)	< .001
No. of doctor’s visits	−0.050 (0.020)	.012	−0.078 (0.020)	< .001
Difference: subjective age (y) – chronological age (y)	−0.050 (0.007)	< .001	−0.046 (0.007)	< .001
High psychological distress	−0.710 (0.198)	< .001	−0.780 (0.200)	< .001
Taking medication for mental condition or emotional problem	−0.101 (0.202)	.62	−0.139 (0.204)	.49
Model fit
Area under the ROC curve	0.824		0.830
McFadden’s pseudo R2	0.242		0.257

Note. GED = general equivalency diploma; ROC = receiver operating characteristic.

^aAnalysis sample includes non-Latino Whites (n = 652), Latinos interviewed in English (n = 332), and Latinos interviewed in Spanish who were given regular on self-rated health (n = 254).

^bAnalysis sample includes non-Latino Whites (n = 652), Latinos interviewed in English (n = 332), and Latinos interviewed in Spanish who were given pasable on self-rated health (n = 244) with no missing on items included in the multiple logistic regression model.

DISCUSSION

Our study examined pasable as an alternative to regular in Spanish translations of the SRH response category of “fair,” and it found improved measurement properties with pasable. Notably, compared with regular, passable (1) consistently resulted in higher positive SRH among Latinos across age, gender, and education; (2) was more sensitive in discriminating those with positive versus negative SRH with respect to number of doctor’s visits, subjective age, and mental health; and (3) made the NLW–Latino SRH disparities disappear, once we controlled for health-related correlates. These findings suggest that previous NLW–Latino disparities reported in the literature may well have been artificially induced by translating “fair” into regular.

Although these results are encouraging, they should be taken with caution. First, all 3 studies we analyzed used nonprobability samples. Although probability samples offer generalizability, they are often prohibitively costly for methodological studies of ethnic–linguistic minorities such as Latinos. For this reason, we conducted 3 independent studies and implemented the same experiment in different methodological contexts. In all 3 studies, the translation effect was substantive and consistent. This offers some assurance that our findings would probably remain consistent with probability samples. Moreover, studies reporting replicated experiment effects between probability and nonprobability samples^31,32 add to this assurance. Second, when administering surveys, response time matters, and items requiring shorter time may be preferred. In our study, respondents spent 0.6 seconds more on SRH with pasable than with regular (7.3 vs 7.9 seconds; P = .070 by Mann–Whitney–Wilcoxon test). As it is not significant in a statistical sense, this small increase may not be a major hurdle for using SRH with pasable. As some argue, response latency tends to increase with the amount of information respondents consider.³³ The improved measurement properties of SRH with pasable may be a product of higher cognitive efforts by our respondents.

Our study warrants more rigorous and sophisticated cross-cultural investigations into SRH, a powerful survey question. For example, our Latino samples tended to include a higher proportion of females, those with higher education, and those with somewhat lower income than what the general Latino population data suggest. Moreover, the translation effect was not consistent across Latino subgroups. Because of data limitations, we were able to examine this translation effect only for Cuban Americans, Mexican Americans, and Puerto Ricans. Because Latino subgroups may speak different Spanish dialects, how the translation effect holds needs to be further investigated with diverse subgroups. The standard practice of survey questionnaire translation involves translating a source language into a target language.³⁴ Often, English is the source language. Our study clearly demonstrates that this may not always work, as the meaning of a simple word such as “fair” is not easily translated. The possibility of revising the current English SRH response scale into a scale that may be translated into languages besides Spanish without altering the meaning may deserve consideration to further improve its measurement comparability and validity. Broadly speaking, the transparency of questionnaires, and whether and how question wording and translation are considered in research, should be emphasized as a dissemination protocol.³⁵ As this study shows, mere changes in translation of “fair” may dramatically change conclusions about Latinos’ health.