The weighted sample population included almost equal numbers of males and females with about half younger than 35 years of age (). A quarter of the population had less than nine years of education, with only 10% reporting a college degree or more. The population was also predominantly low to middle income with 56% making less than $35,000 per year. The majority were foreign born (59%), with 57% identifying their ethnicity as Mexican. Over half (54%) of the weighted sample interviewed in Spanish, and a similar percentage (54%) reported using Spanish when thinking all or most of the time, whereas about 70% reported excellent or good Spanish proficiency. also shows the cross-tabulation of Spanish proficiency and language used when thinking. Given that the majority of the population were immigrants, it was not surprising that the largest cell (42%) was persons with excellent or good Spanish and thinking in Spanish all or most of the time. More surprising was that 11% of the population reported fair or poor Spanish but thinking in Spanish all or most of the time. For self-rated physical health and mental health, 28% rated themselves in poor or fair physical health and 12% rated themselves as having poor or fair mental health.
shows mean values of the neighborhood social cohesion, family support, family cultural conflict, and friend support scales (adjusted by sex and age) by categories of self-rated physical and mental health, education, income, and language variables. Self-rated physical health had highly significant (p < 0.001) associations with both the social cohesion and friend support scale; persons with lower scale scores tended to report worse physical health. Associations with the other scales were also significant, but much less so. Self-rated mental health had highly significant (p < 0.01) associations with all four scales, with the strongest associations for family and friend support. Education was significantly related only to family and friend support. Household income was significantly associated with all scales except family cultural conflict.
The association with language is best examined by looking at the 6 category combination of Spanish proficiency and language used when thinking. This combined language measure was strongly associated (p < 0.001) with each of the scales. Persons who described their Spanish as fair or poor and yet thought in Spanish all or most of the time (11% of the population; see ) reported the lowest scores on the social cohesion, family support, and friend support scales. Persons who described their Spanish as fair or poor and thought in English and Spanish reported the highest levels of family cultural conflict. This group was mostly comprised of U.S. born (60%) and immigrants who arrived in the U.S. before age 18 (31%), likely representing individuals raised in a family where Spanish was mostly or exclusively spoken who simultaneously experience a high degree of acculturation to U.S. society outside the home.
As expected, there were strong correlations among most of the social connection variables: social cohesion was significantly correlated (p < 0.001) with family support (r = 0.18), family cultural conflict (r = −0.17), and friend support (r = 0.19). Family support was significantly correlated (p < 0.001) with family cultural conflict (r = −0.19) and friend support (r = 0.28). The only non-significant correlation was between family cultural conflict and friend support (r = −0.01), which may indicate that friend support occurs across all levels of family conflict, in some cases exacerbating or resulting from family conflict and in others concomitant with family cultural values
and each present three order logistic regressions models, one for self-rated physical health and the other for mental health. In each table, the first model includes the social connection variables and demographic variables (sex, age, marital status, nativity, and subethnicity); the second adds education and income; and the third includes language variables. Because of the high correlations among social connection scales, we dropped non-significant scales from the models.
Self-rated physical health
In the first model in , only family and friend support scales have a significant association with self-rated physical health, after controlling for demographic characteristics. Although the social cohesion scale is significant in the first model without the other scales, it is non-significant after the inclusion of family and friend support scales. This finding indicates that the association of self-rated physical health and social cohesion is likely due to the association between social cohesion and family and friend support (or an association with an unmeasured common factor). When family cultural conflict is put in the model without the other scales, it is marginally non-significant (p = 0.06).
When education and income are added in Model 2 (both significant at p < 0.001), the friend support scale becomes non-significant. The apparent effect of friend support in Model 1 was likely due to its strong association with education and income (see ). Again, social cohesion, if entered in Model 2 without family support, is significant.
Once the language variables are entered in Model 3, however, the family support scale becomes marginally non-significant (p = 0.06). Similarly, social cohesion becomes non-significant with the addition of the language variables when it is the only scale in the model. In , we saw all scales (controlled only for sex and age) significantly associated with self-rated physical health. But in the final model of , the scales are neither singly nor jointly significant. In this dataset, the strongest associations of self-rated physical health are with education, income, and language (as well as sex and age), and associations with social connection scales were only manifest when these variables were not fully controlled for.
For self-rated physical health, language of interview is not significant. However, those with poor or fair Spanish proficiency who thought mostly in Spanish or in English and Spanish equally had significant negative associations with self-rated physical health. Jointly, the Spanish proficiency and language of thinking terms are highly related to self-rated physical health (p < 0.001). Interestingly, the coefficient estimates for the sociodemographic variables change very little from Models 2 and 3, suggesting that the effects of language are independent of other predictors, notably education and income.
Self-rated mental health
In the models for self-rated mental health in , family support and family cultural conflict are significant in all three models. Friend support becomes nonsignificant in Model 3 when family support and family cultural conflict scales are included, despite being highly significant in Models 1 and 2. The spurious association between friend support and mental health seen in Models 1 and 2 (and ) is likely due to friend support being strongly associated with the stronger predictors of education and language. Social cohesion is not significant in any of the three models when the other scales are included. When social cohesion is included without any other scales, it is highly significant in Model 1 (p = 0.004), significant (p = 0.02) in Model 2, and nonsignificant in Model 3.
Spanish language of interview is negative and marginally significant (p = 0.04) in Model 3 of . As with physical health, the joint test of the Spanish proficiency and language of thinking terms are highly significant (p < 0.001). Those who think in Spanish and rate themselves as poor or fair Spanish speakers are associated with poorer self-rated mental health (p < 0.01). In contrast, those who have excellent or good Spanish proficiency and think in English or in English and Spanish equally are associated with better self-rated mental health (p < 0.01). In contrast to physical health, we see here that sex has a weaker, but still significant association; age is not significant; nativity is highly significant; and income is not significant. The nativity term shifts directionality across the models, from negative in Model 1, near zero in Model 2 and positive in Model 3.
Tests of interactions
Interactions between nativity and the social connection scale variables were analyzed in other models and found to be non-significant (data not shown). Interactions between the language proficiency and language of thinking variables and the scales were also analyzed and found to be non-significant. Hence, there was no evidence to suggest that the scales affected the outcomes differentially in the U.S. born than the foreign born or among the different language groups.
Language of interview
Since language of interview is highly collinear with the language proficiency and language of thinking variables, language of interview may interact with the other language terms in and , explaining some of the effects seen. Hence, models were fit with language of interview fully interacted by the other language terms. Since results are best seen graphically, linear regressions were used so that adjusted means could be plotted (results were statistically no different from ordered logistic regressions that were also run). for self-rated physical health corresponds to Model 3 of , and for self-rated mental health corresponds to Model 3 of . and show that within each of the six categories, language of interview makes little difference to the level of self-rated physical or mental health. and are consistent with the statistical results reported in and : language of interview was non-significant for physical health and marginally significant for mental health, whereas Spanish proficiency and language of thinking categories were highly significant (p < 0.001). In this dataset, the strong language associations seen were not primarily due to the language of interview.