Page 14 - IJPS-5-1

Language and self-assessed health in the U.S

the difference in the probability of being in a specific category of SAH relative to that category, all else being equal. For
instance, the probability that a naturalized citizen reports poor SAH is 0.007 points (i.e., 0.7 percentage points) lower
than the probability for a U.S. born citizen. By design, the marginal effects across categories total zero, and therefore, it
is relatively straightforward to determine into which categories of SAH an individual are more likely to be in comparison
to the average. The sample probabilities reported in the last row of the table are the predicted probability based on the
multinomial model that arises when all variables are set at their means. These are the chances of being in each SAH
category for a hypothetical individual that is the most average across all covariates in Model 3 of Table 3.
By far the largest marginal effect is for the probability of a Spanish speaker rating their health as fair. It is fully 0.155
points (i.e., 15.5 percentage points) greater than for an English speaker when controlling for citizenship, ethnicity, and
all other covariates. Spanish speakers are also much less likely to report very good or excellent health. In contrast, a
non-citizen is more likely than a U.S.-born or naturalized citizen to report excellent health, other things being equal.
On balance, citizenship status does not have an overly large impact on SAH, but in combination, excellent health is
most often reported by English speaking non-citizens in comparison to any other individuals. Ethnicity is consequential.
White respondents are by far the most likely to report very good and excellent health, while Hispanic, black, and other/
multi-racial respondents are more likely in the poor, fair, or good category, controlling for other covariates. Finally, there
is substantial variation in the probability of reporting excellent health across categories of language, citizenship, and
ethnicity.
Turning to the issue of “acculturation,” Table 5 examines whether the length of residence in the U.S. weakens
associations between language and SAH. As in Table 3, we show only the covariates or most interest, but the models
control for other variables. The sample for this table includes only those not born in the U.S.; thus, the sample size
is reduced and citizenship status includes only two categories (with naturalized as the comparison). The results show
no significant association between years living in the U.S. and SAH. Other coefficients not shown are similar to the
coefficients in the previous table. The Spanish language relates to a significantly greater likelihood of reporting fair SAH,
while differences between naturalized and non-citizens are non-significant.
Supplementary analyses were performed and due to space limitations are not reported in tabular form. First, a set of
analyses entered interaction effects into models. Some of these were statistically significant, but none substantively altered
the interpretation of the association of language and health. Second, non-Hispanics do not use Spanish for these surveys,
and as such empty cells are present. The models were tested on the population of Hispanics only, leaving ethnicity out of
the model completely. The results were substantively the same as those reported. That is, looking only at Hispanics, the
Spanish speakers are far more likely to report fair SAH than are English speakers, while non-citizens on balance report
better SAH than do naturalize or the U.S. born.
4. Discussion

This paper examines the role of survey language, English or Spanish, in SAH reporting, when taking into account
citizenship status, ethnicity, and a number of other demographic characteristics. Building on previous studies, it poses
three primary research questions: (1) Is there an association between survey language and SAH? (2) If so, does citizenship
status mediate the association between survey language and SAH? (3) Do these associations further hold when accounting
for ethnic variation? These questions are addressed by leveraging seven waves of NHANES up to the most recently
available. NHANES contains a large, nationwide, representative sample of the U.S. population. The study captures
key aspects of immigration – including nativity, survey language, and duration in the U.S. – alongside standard SES
variables in multivariate analyses. The analyses avoid pitfalls of prior studies that dichotomized or otherwise collapsed
SAH categories or employed an ordered logit analysis of the measure. Instead, our analysis uses multinomial regression
to better appreciate potential conceptual distinctions among SAH categories. Furthermore, it addresses ongoing health
measurement challenges of migrating populations.
The results confirm that survey language is an extraordinarily critical factor in SAH reporting. When surveys are
conducted in Spanish, the likelihood that a respondent will report their health as “fair” increases substantially. Overall, the
Spanish language, in addition to non-white ethnicity, has a significant negative effect on a respondent’s SAH rating. These
findings may not be surprising given the nature of structural health inequalities in the U.S., but the effect of the Spanish
language is particularly robust. When ethnicity is added to models that include language, the effect of survey language
persists. Being a non-citizen or naturalized citizen appears to have a protective health effect, but taking the survey in
Spanish and identifying as Hispanic, black, or other/multiracial markedly offset this advantage. Thus, English-speaking
white non-citizens are more likely than any other group to rate their health as excellent. Meanwhile, Spanish-speaking

8 International Journal of Population Studies | 2019, Volume 5, Issue 1