Predictive Strength of Self-Rated Health for Mortality Risk Among Older Adults in the United States: Does It Differ by Race and Ethnicity?

Abstract

Self-rated health (SRH) is widely used to capture racial and ethnic disparities in health. It is therefore critical to understand whether individuals with different racial and ethnic backgrounds assess their SRH differently. Despite the high overall predictive validity of SRH for subsequent mortality, few studies paid attention to potential variations by race and ethnicity. This study examines racial and ethnic differences in the predictive validity of SRH for subsequent mortality risk among older adults (55–84) by estimating Cox Proportional Hazard models using data from the National Health Interview Surveys Linked Mortality Files (1989–2006; N = 289,432). Results indicate that SRH predicts mortality risk less well for non-Hispanic Blacks and Hispanics than non-Hispanic Whites. Three proposed mechanisms—socioeconomic status, immigration status, and cause of death—explain only a modest proportion of the variation. These results suggest that individuals from different racial and ethnic groups may evaluate their heath differently, and thus caution is necessary when using SRH to estimate racial and ethnic health disparities.

Keywords

self-rated health mortality race and ethnicity predictive validity older adults

Introduction

Accurate measurement of population health is crucial for designing and implementing policies aimed at reducing health disparities across various social groups in the United States, and self-rated health (SRH) is one of the most commonly used measures of overall health status. Leaning on their own contextual framework, respondents typically answer a single-item SRH question “How would you rate your health?” on a 5-point scale, with the response alternatives “excellent,” “very good,” “good,” “fair,” and “poor.” In addition to its simplicity, another key strength of SRH is its validity. SRH has a high predictive validity for subsequent health outcomes and mortality (Benjamins, Hummer, Eberstein, & Nam, 2004; Benyamini & Idler, 1999; Chandola & Jenkinson, 2000; DeSalvo, Bloser, Reynolds, He, & Muntner, 2006; Idler & Benyamini, 1997; Idler, Leventhal, McLaughlin, & Leventhal, 2004; Jylhä, 2009), even higher than more objective health measures such as medical diagnoses from a health professional or health assessment by clinical exam (Idler & Kasl, 1991; Mirowsky & Ross, 2003).

While the pervasiveness of SRH in the population/health literature dates back several decades, more recently the measure has attracted increasing scholarly attention for potential variations in assessing health across population subgroups. For instance, several studies reported that the predictive validity of SRH differs by socioeconomic status (SES) (Burstöm & Fredlund, 2001; Dalen, Huijt, Krokstad, & Eikemo, 2012; Dowd & Todd, 2011; Dowd & Zajacova, 2007; Huisman, Lenthe, & Mackenbach, 2007; Singh-Manoux et al., 2007; van Doorslaer & Gerdtham, 2003; Zajacova & Dowd, 2011), gender (Benyamini, Blumstein, Lusky, & Modan, 2003; Dowd & Todd, 2011), and age (Dowd & Todd, 2011; Helweg-Larsen, Kjøller, & Thoning, 2003; Thong et al., 2008; Zajacova & Woo, 2015). These studies found that while results are mixed with SES, SRH predicted mortality better among men, and younger adults compared to women, and older adults, respectively. This suggests that individuals from different population groups may form their health judgments in systematically different ways, highlighting the need to recognize these variations as they can lead to biased estimates and in turn may misrepresent health trends as well as health disparities between the subgroups.

In the present study, we contribute to this emerging literature with a focus on race and ethnicity. Racial and ethnic differences in health are an important and urgent public health issue. Because SRH evaluations are often considered as a critical predictor of subsequent health outcomes and longevity, it is imperative to assess whether respondents of different racial and ethnic groups use the SRH scale in a comparable way. The understanding of group differences in the SRH predictive validity, moreover, can provide valuable insights into how the health judgments are formed. We explore the predictive validity of SRH using mortality risk as an outcome among older adults (55+) from all major racial and ethnic groups in the United States. Utilizing data from the National Health Interview Surveys (NHIS) linked to mortality information from the National Death Index (NDI), we investigate possible variations in the predictive validity of health rating for non-Hispanic Whites, non-Hispanic Blacks, Hispanics, and Asian Americans. Additionally, we examine several mechanisms to explain the racial and ethnic differences in the predictive power of SRH for subsequent mortality.

Background

The large and persistent racial and ethnic disparities in health among U.S. adults highlight the importance of measuring the disparities accurately using simple indicators such as SRH. However, despite the increased attention to variations in the SRH and mortality across different subgroups, surprisingly only a few studies have systematically examined racial and ethnic variations in the association between SRH and mortality risk in the United States. One such study by Lee et al. (2007), using the Health and Retirement Study, found that SRH is a stronger predictor of mortality among older Whites than Blacks. Unfortunately, the data included a relatively small number of Black respondents and the authors provided no explanations for the observed patterns beyond baseline health conditions. In another study using an NHIS sample from 1986 to 1994, McGee, Liao, Cao, and Cooper (1999) applied an analysis of mortality as a function of SRH stratified for all major racial and ethnic groups. Predictive strength appeared to be greater for Whites and Asian/Pacific Islanders than for Blacks and Hispanics, but this article did not test the observed differences for statistical significance, nor did it explore possible explanations for them. While the research on racial and ethnic comparisons of the SRH-mortality association is limited, these two studies suggest that older Black adults—and perhaps also respondents from other minority groups—might show a weaker association between SRH and mortality risk compared to older White adults. Below we review relevant literature for racial and ethnic variations in the SRH evaluation process, draw implications of the findings for the predictive validity of SRH by race and ethnicity, and propose several mechanisms that may help explain the variations.

SRH Evaluation by Race and Ethnicity

Racial and ethnic differences in overall health reporting have been well documented previously. However, previous studies focused more on general reporting tendencies, such as whether respondents from different groups tend to provide optimistic or pessimistic ratings for their health. For instance, minority adults have been found to report worse SRH than Whites with similar levels of other health conditions, such as self-reported morbidity and physician assessed health (Grol-Prokopczyk, 2014; Idler et al., 2004). Other studies noted that poor SRH is more common among Blacks than Whites with comparable self-reported physical health condition even after controlling for access to health care, health behaviors, and functional health status (Ferraro 1993; Spencer et al. 2009). One explanation offered for these reporting differences is that racial minorities are more likely to be pessimistic about their health partly due to interpersonal maltreatment that they often experience as minority (Boardman, 2004).

While these studies suggest systematic differences in reporting the average health rating across racial and ethnic groups, it remains unclear whether respondents from different racial and ethnic groups differ in where they place the health rating thresholds or how they use the five levels of the SRH measure. We address this issue by examining such differences, which will manifest in statistical variation in the predictive validity of SRH on subsequent mortality risk. Our work advances the literature on racial and ethnic differences in health evaluation by aiming to answer whether individuals place the health judgment relative to one another comparably across race and ethnicity, rather than to estimate average SRH levels, and thus, our findings help glean insight into the health evaluation process in general.

Proposed Mediating Mechanisms

If racial and ethnic differences in the predictive validity of SRH exist, there are several mechanisms to produce such differences: SES differences, issues related to immigration and acculturation, and different compositions of causes of death. First, SES seems to influence the predictive strength of SRH for mortality, although the direction of the effect is mixed in previous studies. For example, the predictive validity of SRH was found to be higher among higher SES individuals than among their lower SES counterparts (Dowd & Zajacova, 2007). Other researchers, however, found the opposite: The health judgments of those with higher SES have lower predictive validity (Singh-Manoux et al., 2007). Additionally, several other studies found that predictive validity of SRH was not significantly altered by SES (Burström & Fredlund, 2001; Dalen et al., 2012; van Doorslaer & Gerdtham, 2003) or that the higher predictive validity of SRH among higher SES respondents on mortality appeared among men (Huisman et al., 2007).

The mixed findings could be attributed partly to the different samples that the studies are based on, different measures of SES, as well as other covariates included in the analysis. Dowd and Zajacova (2007) used a national sample of the U.S. population, while other studies examined populations in European countries such as Sweden (Burström & Fredlund, 2001; van Doorslaer & Gerdtham, 2003), Norway (Dalen et al., 2012), the Netherlands (Huisman et al., 2007), and France (Manoux et al., 2007). In any case, these studies provide reasons to expect that SES would influence predictive validity of SRH. Higher SES groups may factor in health conditions more comprehensively when assessing their health than their lower SES counterparts. Additionally, given the substantial differences in SES across racial and ethnic groups, it is necessary to incorporate SES as a potential confounder that could affect gross racial and ethnic patterns in SRH-mortality gradient for this study.

The second mechanism for possible racial and ethnic differences is related to acculturation and language use, issues related to immigration status. A sizeable proportion of minority populations, particularly Hispanics and Asian Americans, are first-generation immigrants. While little research has compared the predictive power of SRH on mortality by immigration status, some research suggests that acculturation may influence SRH assessment. More traditionally oriented Hispanics are somewhat more reluctant than nonimmigrants to rate their health as excellent or very good but more likely to report fair or poor (Bzostek, Goldman, & Pebley, 2007; Kandula, Laudeerdale, & Baker, 2007; Shetterly, Baxter, Mason, & Hamman, 1996). In one study examining how acculturation affects the predictive strength of SRH among Latinos, Finch et al. (2002) found that SRH appears to be a weaker predictor for subsequent mortality risk among relatively recent immigrants compared to those with higher levels of acculturation: The longer Latino immigrants stayed in the United States, the greater the predictive strength of SRH for mortality risk. Another argument related to language use is that immigrants, especially those whose primary language is not English, may be more likely to evaluate their health as relatively poor (Angel & Guarnaccia, 1989; Franzini & Fernandez-Esquer, 2004; Phillips, Hammock, & Blanton, 2005). For example, the Spanish version of the interview questionnaire was found to be significantly associated with lower SRH scores among Latino respondents compared to other respondents (Bzostek et al., 2007; Viruell-Fuentes, Morenoff, Williams, & House, 2011). We therefore expect that the longer the immigrants stay in the United States, the more similar they become to nonimmigrants in their evaluations of their own health. That is, accounting for immigration status and duration of stay in the United States may explain some of the race and ethnic differences in the SRH-mortality link. This mechanism is likely more evident for Hispanic and Asian minorities because these two groups include a higher proportion of recent immigrants than non-Hispanic Whites and Blacks.

A third mechanism concerns differences in the prevalence and severity of chronic health conditions by race and ethnicity. To the degree that the predictive strength of SRH may depend on a particular cause of death, and that the distributions of these causes differ across racial and ethnic groups, the SRH-mortality association may also differ by race and ethnicity when causes of death are taken into account. Although no study to our knowledge directly tested whether and to what extent the existing chronic health conditions account for racial and ethnic variations in the SRH-mortality link likely due to data availability, there is evidence that the predictive strength of SRH for mortality risk is higher for deaths caused by chronic conditions (e.g., diabetes and respiratory diseases) and acute conditions (e.g., infections) than for social pathologies such as accidents, homicide, and suicide (Benjamins et al., 2004; Goldstein, Siegel, & Boyer, 1984). Because some of the chronic health issues are more commonly found among non-Hispanic Blacks and Hispanics (Hummer, Benjamins, & Rogers, 2004), we may expect a higher predictive validity of SRH for the minorities with such conditions. However, given that higher death rates related to such social pathologies have been reported among some racial and ethnic minorities (Heron, 2013), the association between SRH and all-cause mortality risk may be weaker for such groups than for non-Hispanic Whites when examining the specific causes of deaths.

There are other possible mechanisms that may also explain the racial and ethnic differences, including, but not limited to, cultural differences in the conceptualization of health, social network (in terms of both the number of significant social contacts and the depth of relationships with them), and presence of chronic health conditions and physical limitations across the racial and ethnic groups. Environmental factors, such as residential segregation and neighborhood characteristics, as well as religion and religiosity also deserve attention to better understand systematic differences in self-health assessment using SRH across racial and ethnic groups. The data we use, unfortunately, do not include information regarding these potential mechanisms. We thus focus on the three mechanisms outlined above: SES, immigration status, and causes of death. These three factors have been identified in previous research as important possible factors influencing the SRH-mortality association, and they all are expected to be particularly relevant to racial and ethnic differences in the association among U.S. adults.

In the current analysis, we first describe the overall patterns of predictive validity in SRH for mortality risk in the older U.S. population to determine racial and ethnic variations in the SRH-mortality link: Is the predictive strength of SRH for subsequent mortality risk stronger or weaker for minority respondents? Then, we estimate to what extent the racial and ethnic differences in the SRH-mortality link can be explained by SES, immigration status, and different compositions of causes of death. Our study will therefore add several unique contributions to the literature on predictive validity of SRH by a systematic examination of all major racial and ethnic groups in the United States, and a test of three possible mechanisms to provide explanations for the observed patterns.

Method

Data

Data are from the NHIS-Linked Mortality Files (LMF). The NHIS is an annual cross-sectional survey that collects a wide range of information about health, demographics, and SES from a sample of noninstitutionalized U.S. population. The NHIS-LMF links adult respondents in the 1986–2004 NHIS to mortality records through December 31, 2006, in the NDI using a probabilistic matching algorithm to determine respondents’ vital status (Lochner, Hummer, Bartee, Wheatcroft, & Cox, 2008; National Center for Health Statistics, 2009). While the percentage of eligible matching varies by the survey year due to available information from the respondents, ranging from 86.1% to 98.5%, majority of all the respondents were successfully linked to mortality status (National Center for Health Statistics, 2009). This unique data source serves as one of the most comprehensive data sets currently available offering the highest quality of mortality data for the general population in the United States and has been widely used for mortality research. We used the matched NHIS surveys from 1989 to 2004 because information about immigration status, one of our key covariates, is only available from the 1989 survey onward.

We defined our analytic sample as adult respondents aged 55–84 at the time of the interview. The upper age limit of 84 was selected because the NHIS top coded age at 85 starting from the 1997 survey. We excluded a small proportion of respondents who were classified as “other” on race and ethnicity as well as those with missing data on any covariates (about 3% of the older adult respondents). Our final sample included 289,432 respondents.

Measures

The dependent variables were overall as well as cause-specific mortality risk operationalized by death event occurred during the follow-up period for up to almost 18 years. The follow-up period was calculated based on the information about a survey year, a quarter of the survey year, a year of death, and a quarter of the death year to define the number of years a respondent contributed to the duration of survival until deceased before December 2006. Those who survived during the follow-up period were right censored at the fourth quarter in 2006. For the cause-specific mortality, we selected the 10 leading causes of death: heart disease, cancer, respiratory disease, stroke, accidents, diabetes, infectious and parasitic diseases, Alzheimer’s disease, suicide, and homicide.

SRH and race and ethnicity were the key predictors. There were five response categories for SRH: excellent (reference), very good, good, fair, and poor. Race and ethnicity was classified as non-Hispanic White (reference), non-Hispanic Black, Hispanic, and non-Hispanic Asian.

All analyses incorporated key sociodemographic variables to control for possible heterogeneity in health and mortality attributable to biological and environmental characteristics: age, sex, marital status, and region of residence (hereafter region). Age was a continuous variable; sex was dichotomous (female as reference); marital status was coded as “married” (reference), “widowed,” “divorced,” “separated,” or “never married.’ Region was coded as “Northeast” (reference), “Midwest,” “West,” or “South.”

We included several variables for the three hypothesized explanations of racial and ethnic differences in the association between SRH and mortality. To test the first explanation, we included three measures of SES: education, poverty, and employment. Education was coded in one of four categories: “less than high school,” “high school,” “some college,” and “college or more” (reference). Poverty was defined at the household level and dichotomized to “above” (reference) or “below” the U.S. Census Bureau’s poverty threshold at the time of the survey, which was calculated based on total household income adjusted by family size and the number of children under age 18. Employment was dichotomized as either “employed” (reference) or “not employed,” with the latter category including respondents who were retired, unemployed, or not in the labor force. Second, to examine to what extent racial and ethnic differences in the association between SRH and mortality outcome may be explained by immigration status, we created a three-level immigration variable: “nonimmigrants” (reference), “immigrants who have stayed in the United States less than 10 years,” and “immigrants who have stayed in the United States for 10 or more years.” This variable served as a proxy for acculturation and language use because the NHIS does not consistently collect specific information on these factors. The third proposed mechanism was tested using cause-specific mortality models, which we explain further below.

Analytic Models

We estimated a series of nested Cox proportional hazard models to evaluate the association between SRH and mortality risk across racial and ethnic groups. The first model started with the two primary independent variables—SRH and racial and ethnic categories, adjusting only for demographic information (i.e., sex, age, marital status, and region). In subsequent models, we included a set of interaction terms between SRH and racial and ethnic categories and added information about SES and immigration status to test the first two explanations proposed above. Specifically, Model 2 added the interaction terms between SRH and racial and ethnic categories to Model 1. In the third model, we further controlled for SES by adding education, poverty, and employment. In the fourth model, we controlled for immigration status. In Model 5, we included both SES and immigration status with demographic characteristics adjusted. Finally, to test the last considered mechanism, we estimated additional models for the 10 leading cause-of-death categories as an outcome to explore whether or not any racial and ethnic differentials found in the all-cause mortality models are also observed in cause-specific models. Because the above mediated moderation models we used to include SES and immigration status are not applicable to the cause-specific models, we estimate models stratified by the leading causes of death to detect if racial and ethnic differences in the SRH-mortality association vary by the leading causes of death. While not formally estimating mediation of the SRH-mortality association, the stratified models allow us to indirectly capture if chronic health conditions substantially explain the racial and ethnic variations in the predictive strength of SRH in the mortality risk.

All the models were estimated using “proc surveyphreg” in SAS 9.4 to adjust for the complex sampling design of the NHIS-LMF, and the estimates reported here were properly weighted to represent the U.S. older population unless otherwise indicated. We used the “efron” option to handle ties.

Results

Descriptive Statistics

Table 1 presents the descriptive statistics. Over the course of the mortality follow-up from 1989 through 2006, nearly 30% of the sample died. The majority of the individuals appeared to be healthy, reporting their health as good or better, while nearly a quarter rated their health as fair (16%) or poor (7%). In terms of racial and ethnic composition, 83% of the sample were non-Hispanic White and 9% were non-Hispanic Black. The percentages of Hispanics and non-Hispanic Asians were 6% and 2%, respectively. As expected, there were more females (55%) than males (45%). In term of marital status, 66% of the individuals were married and 20% were widowed. The proportions of divorced and never married were relatively small (10% and 4%, respectively). A larger proportion of the individuals resided in the South (36%) than in other regions. As for the socioeconomic indicators, more than half the individuals (64%) had high school or less education, 7% lived below the poverty threshold, and 33% were employed. Finally, the table also shows that 9% of the older adults were immigrants and that most of them had resided in the United States for more than 10 years.

Table 1.

Sample Characteristics.

Variables	Unweighted Frequency	Weighted Proportion	Mean	Std Dev	Minimum	Maxmum
Mortality status
Dead	88,954	0.28			0	1
Alive	200,478	0.73			0	1
Self-rated health
Excellent	52,820	0.19			0	1
Very good	72,735	0.26			0	1
Good	92,687	0.32			0	1
Fair	49,335	0.16			0	1
Poor	21,855	0.07			0	1
Demographics
Age, in years			66.84	8.06	55	84
Non-Hispanic White	226,051	0.83			0	1
Non-Hispanic Black	33,135	0.09			0	1
Hispanic	24,781	0.06			0	1
Non-Hispanic Asian	5,465	0.02			0	1
Female	160,365	0.55			0	1
Male	129,067	0.45			0	1
Married	188,818	0.66			0	1
Widowed	58,558	0.20			0	1
Divorced/ separated	29,887	0.10			0	1
Never married	12,169	0.04			0	1
Region
Northeast	60,242	0.21			0	1
Midwest	68,868	0.24			0	1
West	56,320	0.19			0	1
South	104,002	0.36			0	1
Education
<High school	89,103	0.27			0	1
High school	104,507	0.37			0	1
Some college	48,552	0.18			0	1
College or more	47,270	0.18			0	1
Poverty
Above threshold	208,998	0.72			0	1
Below threshold	23,235	0.07			0	1
Missing	57,199	0.21			0	1
Immigration
Nonimmigrant	258,872	0.91			0	1
Immigrant	30,560	0.09			0	1
Immigrated > 10 yrs ago	3,550	0.01			0	1
Immigrated < 10 yrs ago	27,010	0.08			0	1
Insurance
Insured	260,898	0.91			0	1
Uninsured	28,534	0.09			0	1
Mortality follow-up
Follow-up period, in years			8.10	4.53	0.125	17.875

Note. N = 289,432.

The distribution of specific causes of death is presented in Table 2. Almost 80% of deaths were caused by one of the leading chronic conditions, including heart disease, cancer, respiratory disease, stroke, and diabetes. Although heart disease and cancer were the two major causes of death in all the racial and ethnic groups, there were some between-group differences as well. For example, the proportion of respondents dying from respiratory diseases appeared to be higher among non-Hispanic Whites than others. In contrast, the proportions of those dying from diabetes were highest among non-Hispanic Blacks and Hispanics. The proportion dying from stroke was higher among Asian Americans than the other groups.

Table 2.

Specific Cause of Death.

Cause of Death	All		Non-Hispanic White		Non-Hispanic Black		Hispanic		Non-Hispanic Asian
	Unweighted	Weighted	Unweighted	Weighted	Unweighted	Weighted	Unweighted	Weighted	Unweighted	Weighted
	Frequency	Percentage	Frequency	Percentage	Frequency	Percentage	Frequency	Percentage	Frequency	Percentage
Heart disease	28,285	31.37	22,712	31.18	3,762	32.58	1,577	32.81	234	29.94
Cancer	23,183	26.47	18,900	26.59	2,948	26.55	1,117	23.73	218	26.87
Respiratory disease	9,373	10.71	8,111	11.32	801	6.81	397	7.96	64	8.21
Stroke	6,190	6.81	4,955	6.74	820	7.13	351	7.21	64	8.19
Diabetes	2,935	3.23	1,979	2.80	585	5.36	338	6.83	33	4.17
Alzheimer’s disease	1,828	2.02	1,584	2.14	156	1.25	76	1.57	12	1.76
Infectious disease	1,590	1.81	1,157	1.68	323	2.90	98	2.00	12	1.62
Accident	1,577	1.80	1,335	1.85	152	1.40	78	1.76	12	1.29
Suicide	367	0.44	327	0.47	18	0.19	18	0.37	4	0.37
Homicide	66	0.07	40	0.06	21	0.19	5	0.10	0	0.00
Other causes	13,560	15.27	10,871	15.18	1,765	15.63	786	15.67	138	17.59
Total	88,954	100.00	71,971	100.00	11,351	100.00	4,841	100.00	791	100.00

Note. n = 88,954.

SRH and Mortality by Race and Ethnic Group

Table 3 presents results of the five proportional hazard models of all-cause mortality. Results of Model 1 confirmed the previous studies that SRH was a strong predictor of mortality risk: People with lower SRH scores had significantly higher risks of mortality. Compared to people with excellent health, those with very good health had 20.5% higher mortality risk (Hazard Ratio [HR] = 1.205). The mortality risks were 64.3%, 152.2%, and 344.6% higher for those whose SRH was good, fair, and poor, respectively. With SRH adjusted, racial and ethnic minorities had lower mortality risks compared to non-Hispanic Whites, net of the demographic characteristics. These results are also consistent with the earlier studies (i.e., McGee, Liao, Cao, & Cooper, 1999; Lee at al., 2007). To test the robustness of the results, we performed a sensitivity analysis by estimating a series of models of all-cause mortality as a function of SRH and basic covariates stratified by race and ethnic groups (results not shown but available on request). The results of the stratified models also indicate that the SRH effects on mortality are smaller for non-Hispanic Black and Hispanic adults, compared to non-Hispanic Whites, corroborating the findings shown in Model 1 in Table 3.

Table 3.

Proportional Hazard Models Estimating Effect of SRH on Mortality Risk.

Variables	Model 1		Model 2		Model 3		Model 4		Model 5
Variables	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio
SRH (excellent)
Very good	0.186***	1.205	0.203***	1.225	0.179***	1.196	0.201***	1.223	0.177***	1.194
Good	0.496***	1.643	0.520***	1.681	0.472***	1.603	0.519***	1.680	0.472***	1.603
Fair	0.925***	2.522	0.964***	2.623	0.886***	2.424	0.966***	2.627	0.887***	2.427
Poor	1.496***	4.466	1.570***	4.806	1.463***	4.317	1.575***	4.832	1.467***	4.336
Race/ethnicity (non-Hispanic White)
Non-Hispanic Black	−0.033**	0.967	0.356***	1.428	0.323***	1.382	0.362***	1.436	0.328***	1.389
Hispanic	−0.342***	0.710	−0.137**	0.872	−0.173***	0.841	−0.013	0.987	−0.051	0.950
Non-Hispanic Asian	−0.662***	0.516	−0.664***	0.515	−0.658***	0.518	−0.448***	0.639	−0.441***	0.644
Interactions
NHB × Very Good			−0.278***	0.757	−0.268***	0.765	−0.281***	0.755	−0.271***	0.762
NHB × Good			−0.313***	0.732	−0.289***	0.749	−0.316***	0.729	−0.293***	0.746
NHB × Fair			−0.420***	0.657	−0.405***	0.667	−0.428***	0.652	−0.413***	0.662
NHB × Poor			−0.646***	0.524	−0.634***	0.530	−0.658***	0.518	−0.648***	0.523
Hispanic × Very Good			−0.056	0.945	−0.051	0.951	−0.046	0.955	−0.041	0.960
Hispanic × Good			−0.210***	0.810	−0.207***	0.813	−0.191***	0.826	−0.188**	0.829
Hispanic × Fair			−0.332***	0.718	−0.335***	0.716	−0.311***	0.733	−0.314***	0.731
Hispanic × Poor			−0.240***	0.787	−0.238***	0.788	−0.212***	0.809	−0.211**	0.810
NHA × Very Good			0.081	1.085	0.079	1.082	0.069	1.072	0.067	1.069
NHA × Good			0.004	1.004	−0.008	0.992	0.024	1.025	0.012	1.012
NHA × Fair			0.017	1.017	0.006	1.006	0.064	1.066	0.052	1.053
NHA × Poor			−0.152	0.859	−0.181	0.835	−0.067	0.935	−0.093	0.911
Education (college or more)
Less than high school			0.186***	1.204					0.182***	1.200
High school			0.153***	1.165					0.145***	1.156
Some college			0.144***	1.155					0.139***	1.149
Below poverty (above poverty)			0.043**	1.043					0.051***	1.052
Missing on poverty			−0.096***	0.909					−0.095***	0.910
Unemployed (employed)			0.286***	1.331					0.286***	1.331
Immigration (nonimmigrant)
Immigrant, less than 10 years							−0.701***	0.496	−0.712***	0.491
Immigrant, more than 10 years							−0.250***	0.779	−0.245***	0.783
−2 Log Likelihood	1,823,403.6		1,823,171.7		1,821,975.3		1,822,747.7		1,821,555.0
Akaike Information Criterion	1,823,433.6		1,823,225.7		1,822,041.3		1,822,805.7		1,821,625.0

Note. N = 289,432. All of the models also include age, sex, marital status, and region. The values in parentheses are references. SRH = Self-rated health, NHB = non-Hispanic Black, NHA = non-Hispanic Asian.

*p < .05. **p < .01. ***p < .001 (two-tailed test).

With respect to racial and ethnic variations in the predictive strength of SRH for mortality risk, we found significant differences in Model 2. At most levels of SRH from very good to “poor,” the interaction terms were statistically significant for non-Hispanic Blacks and Hispanics (p < .001). The negative coefficients of the interaction terms indicated that the predictive strength of SRH was lower for non-Hispanic Blacks as well as Hispanics compared to non-Hispanic Whites. Asian Americans showed no significant differences in the predictive power of SRH on mortality, compared to Whites. We illustrate the results of Model 2 in Figure 1. The estimates used for the figure were based on the main effects of SRH and its interactions with racial and ethnic categories.

Figure 1.

Predicted effect of Self-rated health (SRH) on mortality risk by race and ethnicity. This figure is based on the results of Model 2 in Table 3. “ns” indicates that the difference between non-Hispanic White and Hispanic is not statistically significant. The predicted effect of SRH on mortality risk for non-Hispanic Asian is not included in this figure because of their nonsignificant results.

Figure 1 clearly depicts that the predictive validity of SRH on mortality risk differs by race and ethnic groups: SRH predicted subsequent mortality risk less accurately for non-Hispanic Blacks than for non-Hispanic Whites, net of demographic characteristics (i.e., age, sex, marital status, and region).

The differences between non-Hispanic Whites versus Black and Hispanic older adults seemed partly explained by socioeconomic conditions measured by educational attainment, poverty status, and employment status especially for the Blacks. According to the results of Model 3, the interaction terms capturing the race and ethnic differences in the SRH-mortality association became attenuated, while all previously significant coefficients still remained significant. These results suggest that the racial and ethnic differences are partially explained by SES differentials across the race and ethnic groups, and the mediating mechanism appears stronger for the Blacks. Model 4 shows the influence of immigration status on racial and ethnic differences in the association between SRH and mortality risk. Compared to the results of Model 2, the interaction coefficients for Black respondents in Model 4 remained largely unchanged. However, there was an attenuation of the interaction coefficients for Hispanics, indicating that the weaker predictive power in this group was partly attributable to immigration status. Results of Model 5, which included SES and immigration status, were also consistent in that SES differentials explained the race and ethnic variations in the SRH-mortality link especially for Blacks, while the lower predictive validity of SRH among Hispanics was substantially mediated by immigration status.

Finally, we estimated additional 10 proportional hazard models to examine the SRH-mortality association across racial and ethnic groups for leading causes of death (i.e., heart disease, cancer, respiratory disease, stroke, diabetes, Alzheimer’s disease, infectious diseases, accidents, suicide, and homicide). Differences across causes of death may contribute to racial and ethnic differences in the predictive strength of SRH for all-cause mortality risk. As mentioned earlier, rather than directly controlling for chronic conditions and other health problems in estimating the predictive strength of SRH for the overall subsequent mortality risk, the cause-specific models are designed to overcome the data limitation by utilizing information about cause-specific mortality outcomes instead.

The results presented in Table 4 showed that a statistically significant association between SRH (at least some levels) and mortality risk was found among White respondents for all the leading causes of death except homicide: Lower SRH ratings predicted higher mortality risks. There were some differences in the main effects of SRH by different causes of death. For instance, the predictive strength of SRH appeared to be stronger for diabetes and respiratory diseases than other causes, implying that diabetes and respiratory diseases were somewhat more self-recognizable and/or that people tended to weigh these health conditions (among others) more seriously when assessing the overall health status.

Table 4.

Proportional Hazard Models Estimating Effect of SRH on Underlying Cause-Specific Mortality Risk.

Variables	Heart Disease		Cancer		Respiratory Disease		Stroke		Diabetes
Variables	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio
SRH
Very good	0.24***	1.27	0.07**	1.07	0.29***	1.34	0.21***	1.24	0.49***	1.63
Good	0.59***	1.81	0.34***	1.40	0.74***	2.09	0.56***	1.75	1.10***	2.99
Fair	1.10***	3.02	0.72***	2.06	1.44***	4.22	0.92***	2.50	1.98***	7.27
Poor	1.72***	5.56	1.40***	4.06	2.32***	10.21	1.52***	4.56	2.81***	16.64
Race/ethnicity
Non-Hispanic Black	0.44***	1.55	0.33***	1.39	0.11	1.12	0.34*	1.41	1.08***	2.96
Hispanic	0.09	1.09	−0.28**	0.76	−0.45*	0.64	−0.03	0.97	0.70*	2.01
Non-Hispanic Asian	−0.51*	0.60	−0.43*	0.65	−0.27	0.76	−1.18*	0.31	0.48	1.62
Interactions
NHB × Very Good	−0.36***	0.70	−0.25**	0.78	−0.42*	0.66	−0.28	0.76	−0.33	0.72
NHB × Good	−0.38***	0.68	−0.29***	0.75	−0.59***	0.55	−0.32†	0.73	−0.60*	0.55
NHB × Fair	−0.54***	0.58	−0.45***	0.64	−0.94***	0.39	−0.39*	0.68	−0.76**	0.47
NHB × Poor	−0.82***	0.44	−0.73***	0.48	−1.30***	0.27	−0.53**	0.59	−1.13***	0.32
Hispanic × Very Good	−0.22†	0.80	0.07	1.07	−0.09	0.91	−0.14	0.87	0.09	1.09
Hispanic × Good	−0.38***	0.68	−0.16	0.85	−0.07	0.93	−0.24	0.79	−0.26	0.78
Hispanic × Fair	−0.59***	0.55	−0.30*	0.74	−0.45	0.64	−0.27	0.76	−0.57†	0.57
Hispanic × Poor	−0.36**	0.70	−0.43**	0.65	−0.42†	0.66	−0.51*	0.60	−0.19	0.83
NHA × Very Good	−0.03	0.97	−0.12	0.89	−0.10	0.91	0.55	1.74	−1.00	0.37
NHA × Good	−0.07	0.94	−0.22	0.80	−0.58	0.56	0.93†	2.55	−0.48	0.62
NHA × Fair	0.02	1.02	−0.07	0.93	−0.58	0.56	0.57	1.77	−1.47*	0.23
NHA × Poor	−0.33	0.72	−0.35	0.71	−1.22*	0.29	0.63	1.87	−0.45	0.64
n	228,763		223,661		209,851		206,668		203,413
	Alzheimer’s disease		Infectious disease		Accident		Suicide		Homicide
Variables	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio	Coefficient	Hazard Ratio
SRH
Very good	−0.11	0.90	0.20	1.22	0.04	1.05	0.34†	1.41	−0.35	0.71
Good	0.06	1.06	0.74***	2.10	0.31***	1.36	0.51**	1.67	−0.05	0.95
Fair	0.18†	1.19	1.30***	3.67	0.62***	1.85	1.11***	3.04	−0.01	1.00
Poor	0.76***	2.14	1.95***	7.03	1.17***	3.23	1.64***	5.17	−0.36	0.70
Race/ethnicity			0.96***	2.62
Non-Hispanic Black	−0.16	0.85	1.56***	4.75	−0.41	0.66	−0.37	0.69	0.46	1.58
Hispanic	−0.15	0.86	0.14	1.15	−0.61	0.55	0.19	1.21	0.87	2.38
Non-Hispanic Asian	−0.97	0.38	−0.28	0.75	−1.02	0.36	−2.67	0.07	—	—
Interactions
NHB × Very Good	−0.07	0.93	−0.53†	0.59	−0.31	0.73	−0.40	0.67	−1.40	0.25
NHB × Good	−0.30	0.74	−0.75**	0.47	0.21	1.23	−0.58	0.56	0.72	2.06
NHB × Fair	−0.45	0.64	−0.82**	0.44	0.04	1.04	−0.85	0.43	0.99	2.70
NHB × Poor	−0.38	0.68	−0.65*	0.52	−0.11	0.89	−0.89	0.41	0.10	1.11
Hispanic × Very Good	0.06	1.06	−0.09	0.92	0.18	1.19	−1.40	0.25	−0.57	0.56
Hispanic × Good	−0.30	0.74	0.53	1.71	0.21	1.23	−0.98	0.38	−7.23	0.00
Hispanic × Fair	−0.51	0.60	−0.44	0.64	−0.14	0.87	−1.75*	0.17	−1.38	0.25
Hispanic × Poor	−0.68	0.50	0.05	1.06	−0.06	0.94	−1.27	0.28	1.41	4.11
NHA × Very Good	0.50	1.64	0.53	1.69	−0.22	0.81	0.99	2.68	—	—
NHA × Good	0.21	1.24	0.22	1.25	0.16	1.18	0.15	1.17	—	—
NHA × Fair	0.72	2.06	−0.75	0.47	−0.18	0.83	0.40	1.49	—	—
NHA × Poor	−0.89	0.41	0.63	1.89	−0.51	0.60	−1.77	0.17	—	—
n	202,306		202,068		202,055		200,845		200,544

Note. The models include all covariates. The references are “excellent” for SRH and “NHW” for race/ethnicity. SRH = Self-rated health, NHW = non-Hispanic White.

^† p < .10.*p < .05.**p < .01.***p < .001 (two-tailed test).

With respect to Black–White differences in the SRH effects, these were statistically significant for the causes that were often led by chronic health conditions and lowered immune system, such as heart disease, cancer, respiratory disease, stroke, diabetes, and infectious disease: Consistent with the overall mortality risk models, the predictive validity of SRH on mortality was lower for the non-Hispanic Blacks than for the White counterparts. For other causes of death that mostly stemmed from social pathologies (i.e., accident, suicide, and homicide), the predictive strength of SRH for Blacks was not statistically different from that of Whites. For Hispanics, the only causes with systematic significant predictive strength differences vis-à-vis Whites were heart disease and cancer. The overall pattern of the weaker predictive validity of SRH on mortality risk was observed in other leading causes of death often due to chronic health conditions such as respiratory disease, stroke and diabetes; however, most coefficients were not statistically significant. These results may have to do with the fact that deaths caused by heart disease and cancer were relatively numerous compared to deaths from other causes, so the models had more statistical power to detect differences than for other causes. Nonetheless, the negative interaction coefficients between Hispanic and SRH for the cause-specific models suggest a consistently weaker predictive validity of SRH for Hispanics than their White counterparts, even if not statistically confirmed in the current analysis. The SRH-mortality link did not appear to be different by causes of death for Asians.

Discussion and Conclusion

SRH has been widely used to measure population health trends and disparities. However, if people from different subgroups of the population assess their health in systematically different ways, estimates using the SRH measure may produce a biased picture of health disparities. While several recent studies have addressed this possibility by examining the predictive validity of SRH by age, gender, and SES, few studies specifically focused on racial and ethnic differences in the SRH-mortality association. Using data from the NHIS 1989–2004 with mortality follow-up through 2006, we addressed this gap by testing how SRH predicts mortality risk for older adults across the four major racial and ethnic groups in the United States, including non-Hispanic Whites, non-Hispanic Blacks, Hispanics, and Asians. Additionally, we explored three mechanisms to explain the racial and ethnic differences: SES, immigration status, and differences in chronic health conditions causing death.

First, we found statistically significant and substantively large differences across the racial and ethnic groups in the SRH-mortality association. Among non-Hispanic Black and Hispanic respondents, health ratings appear to be less discriminating, in that the differences between excellent and lower SRH categories are smaller with respect to mortality risks than for Whites. Therefore, SRH ratings of non-Hispanic Black and Hispanic respondents are less strongly predictive of their subsequent mortality risk. We found no significant differences between non-Hispanic White and Asian respondents. This may be perhaps because our sample of Asian respondents was too small to detect existing differences with comparison to non-Hispanic White; alternatively, the Asian group is too diverse to be considered as one group and some subgroups within Asians may negate the patterns of other subgroups. Or the two groups (i.e., non-Hispanic White and Asian) may really be similar in how they form their health judgments.

Two of our three proposed mechanisms—SES and immigration status—help explain the racial and ethnic differences. For non-Hispanic Blacks, the lower SRH predictive power is partly attributable to the fact that they tend to have lower SES, which is linked to weaker predictive strength of SRH for mortality. Among Hispanics, immigration status partly accounts for their lower predictive strength. There is a greater proportion of recent immigrants among Hispanics than non-Hispanic Whites, and the SRH judgment may differ especially among recent immigrants. However, it should be noted that the racial and ethnic variations in the SRH-mortality association persisted even after controlling for SES and immigration status.

The remaining hypothesized mechanism, the different causes of death, did not appear to play a major role in the observed racial and ethnic variations. We considered the possibility that differences in cause of death could account for some of the race and ethnic variation in the SRH predictive power for all-cause mortality. We found that the cause-of-death analyses corroborated the all-cause results: While the specific coefficients varied across causes of death, in general they were larger for non-Hispanic White, compared to non-Hispanic Black and Hispanic respondents. This result suggests that much of the racial and ethnic difference reflects the influence of the underlying health-judgment process rather than differences in the distributions of the chronic health conditions causing death. This part of the analysis is different from the mediation approach we could do for SES and immigration status because we could not “control for” causes of death. We simply explored racial and ethnic variations in the SRH effects across different causes of death to detect any major differences that could indicate that the all-cause mortality findings were actually driven by only some causes of death. However, across the leading chronic causes of death, the predictive strength of SRH for mortality risk was weaker for racial and ethnic minorities, especially non-Hispanic Blacks, than for non-Hispanic Whites, even net of the sociodemographic conditions. Given this consistent pattern across causes of death, it is less likely that racial and ethnic differences in the prevalence in the leading causes of death substantially alter our main finding.

Our results raise the concern that the health disadvantages among the racial and ethnic minorities measured by SRH may not have been accurately estimated. As noted earlier, people tend to compare themselves with peers who are around the same age and in the same racial and ethnic group as themselves. Additionally, racial and ethnic minorities are more often exposed to stressful situations and maltreatment and thus disadvantaged in terms of various health outcomes, including overall life expectancy, chronic health problems, physical limitations, and mental illness. Therefore, it is important to note that racial and ethnic health disparities measured by SRH would need to be interpreted with caution as SRH is less discriminating among the minorities, especially non-Hispanic Blacks and Hispanics.

This study is not without limitations. First, all the data (except the mortality follow-up) came from a single interview; the NHIS is cross-sectional and provides no longitudinal measures of health and other covariates. In our data using NHIS-LMF, our maximum follow-up period was almost 18 years, and thus some social and economic conditions, namely, marital status, employment, and poverty, might have changed during the mortality follow-up. However, we are confident that the cross-sectional nature of our data set is unlikely to bias our findings substantially. Even if substantial changes in the conditions might have occurred in systematically different ways across racial and ethnic groups, racial and ethnic minorities would have been more vulnerable than others to unfavorable changes such as marital dissolution, unemployment, and economic hardship. Because this vulnerability is often associated with a worse SRH, possibly further weakening the predictive strength of SRH for less healthy minorities, we would argue that our estimates are rather conservative.

Another limitation is that, despite our large NHIS sample overall, sample sizes for Asian Americans were quite small, especially for the cause-of-death analyses. For example, only four Asian respondents died from suicide and none of the Asians died from homicide in our older adult sample. Even in the analyses that combined all the causes of mortality, the absence of statistical significance when comparing Asian respondents with the non-Hispanic White reference group may be due to a lack of statistical power.

Furthermore, although we attempted to account for the acculturation of immigrants by asking how long they had resided in the United States, it is possible that other questions, such as primary language use and its fluency, whether education was completed in the United States, or the quantity and quality of social interaction outside the immigrant community, may have been well served for the analysis in this regard. Similarly, in addition to the cause-specific mortality analysis, it would have been ideal to consider presence(s) of chronic health conditions, physical disability, and functional impairments for the analysis if available in the data. While they are not necessarily direct causes of death, they may explain some of unexplained racial and ethnic variations in the SRH-mortality link by allowing us to examine if and to what extent disproportionately distributed chronic health conditions across racial and ethnic groups mediate the weaker predictive validity in the SRH-mortality link among the racial and ethnic minorities.

We would also like to acknowledge that individuals’ utilization of health care in addition to their insurance coverage status could be potential mediators. Results of supplemental analysis that controlled for insurance coverage status indicated that health insurance coverage did not appear to mediate the racial and ethnic differences in the SRH-mortality gradient (results not shown but available on request). However, it remains possible that our single measure of health-care coverage did not tell us whether the respondents had recently obtained medical care through Medicare or Medicaid, the influences of their interactions with their doctors, and to what degree their utilization of health services made them aware of their health conditions and any illnesses they may have had.

Despite these data limitations, the results of this study clarify our understanding of racial and ethnic variations in the SRH/mortality gradient, underscoring the important notions of racial and ethnic variations in self-reported health assessment. We have demonstrated that among older adults, racial and ethnic minority respondents’ SRH judgments to be less predictive of their future mortality. Additionally, we identified two mechanisms that play a role in these racial and ethnic differences: Lower SES in the case of non-Hispanic Blacks and immigration status in the case of Hispanics. Our results provide a clear pattern of racial and ethnic differences in the predictive validity of SRH, which helps health professionals better understand health evaluation and health status among the minorities. The remaining unexplained differences in the gradient call for future research to explore other factors that could provide a more complete understanding of the validity of SRH for the diverse populations comprising our current society.

Footnotes

Acknowledgments

The authors thank the editor of Research on Aging, the three anonymous reviewers, Jason Schnittker, and Kelly Raley for their insightful suggestions and comments on earlier drafts of this article.

Authors’ Note

The study was presented at the Annual Meeting of the Population Association of America in 2013.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Angel

Guarnaccia

P. J.

(1989). Mind, body, and culture: Somatization among Hispanics. Social Science & Medicine, 28, 1229–1238.

Benjamins

M. R.

Hummer

R. A.

Eberstein

I. W.

Nam

C. B.

(2004). Self-reported health and adult mortality risk: An analysis of cause-specific mortality. Social Science & Medicine, 59, 1297–1306.

Benyamini

Blumstein

Lusky

Modan

(2003). Gender differences in the self-rated health-mortality association: Is it poor self-rated health that predicts mortality or excellent self-rated health that predicts survival? The Gerontologist, 43, 396–405.

Benyamini

Idler

E. L.

(1999). Community studies reporting association between self-rated health and mortality: Additional studies, 1995 to 1998. Research on Aging, 21, 392–401.

Boardman

J. D.

(2004). Health pessimism among black and white adults: The role of interpersonal and institutional maltreatment. Social Science & Medicine, 59, 2523–2533.

Burström

Fredlund

(2001). Self rated health: Is it as good a predictor of subsequent mortality among adults in lower as well as in higher social classes? Journal of Epidemiology and Community Health, 55, 836–840.

Bzostek

Goldman

Pebley

(2007). Why do Hispanics in the USA report poor health? Social Science & Medicine, 65, 990–1003.

Chandola

Jenkinson

(2000). Validating self-rated health in different ethnic groups. Ethnicity & Health, 5, 151–159.

Dalen

J. D.

Huijt

Krokstad

Eikemo

T. A.

(2012). Are there educational differences in the association between self-rated health and mortality in Norway? The HUNT Study. Scandinavian Journal of Public Health, 40, 641–647.

10.

DeSalvo

K. B.

Bloser

Reynolds

Muntner

(2006). Mortality prediction with a single general self-rated health question. Journal of General Internal Medicine, 21, 267–275.

11.

Dowd

J. B.

Todd

(2011). Does self-reported health bias the measurement of health inequalities in U.S. adult? Evidence using anchoring vignettes from the health and retirement study. Journal of Gerontology, Series B: Psychological Sciences and Social Science, 66, 478–489.

12.

Dowd

J. B.

Zajacova

(2007). Does the predictive power of self-rated health for subsequent mortality risk vary by SES in the US? International Journal of Epidemiology, 36, 1214–1221.

13.

Ferraro

K. F.

(1993). Are black older adults health-pessimistic? Journal of Health and Social Behavior, 34, 201–214.

14.

Finch

B. K.

Hummer

R. A.

Reindl

Vega

W. A.

(2002). Validity of self-rated health among Latino(a)s. American Journal of Epidemiology, 155, 755–759.

15.

Franzini

Fernandez-Esquer

M. E.

(2004). Socioeconomic, cultural, and personal influences on health outcomes in low income Mexican-origin individuals in Texas. Social Science & Medicine, 59, 1629–1646.

16.

Goldstein

M. S.

Siegel

J. M.

Boyer

(1984). Predicting changes in perceived health status. American Journal of Public Health, 74, 611–614.

17.

Grol-Prokopczyk

(2014). Age and sex effects in anchoring vignette studies: Methodological and empirical contributions. Survey Research Methods, 8, 1–17.

18.

Helweg-Larsen

Kjøller

Thoning

(2003). Do age and social relations moderate the relationship between self-rated health and mortality among adult Danes? Social Science & Medicine, 57, 1237–1247.

19.

Heron

(2013). Deaths: Leading causes for 2010 (National vital statistics reports; 62 no 6). Hyattsville, MD: National Center for Health Statistics. Retrieved July 21, 2014, from http://www.cdc.gov/nchs/data/nvsr/nvsr62/nvsr62_06.pdf

20.

Huisman

Lenthe

F. V.

Mackenbach

(2007). The predictive ability of self-assessed health for mortality in different educational groups. International Journal of Epidemiology, 36, 1207–1213.

21.

Hummer

R. A.

Benjamins

M. R.

Rogers

R. G.

(2004). Racial and ethnic disparities in health and mortality among the U.S. elderly population. In Anderson

Norman B.

Bulatao

Rodolfo A.

Cohen

Barney

. (Eds.), Critical perspectives on racial and ethnic differences in health in late life (pp. 53–94). Washington, DC: The National Academies Press.

22.

Idler

E. L.

Benyamini

(1997). Self-rated health and mortality: A review of twenty-seven community studies. Journal of Health and Social Behavior, 38, 21–37.

23.

Idler

E. L.

Kasl

(1991). Health perceptions and survival: Do global evaluations of health status really predict mortality? Journal of Gerontology, 46, S55–S65.

24.

Idler

E. L.

Leventhal

McLaughlin

Leventhal

(2004). In sickness but not in health: Self-rating, identity, and mortality. Journal of Health and Social Behavior, 45, 336–356.

25.

Jylhä

(2009). What is self-rated health and why does it predict mortality? Towards a unified conceptual model. Social Science & Medicine, 69, 307–316.

26.

Kandula

N. R.

Laudeerdale

D. S.

Baker

D. W.

(2007). Differences in self-reported health among Asians, Latinos, and non-Hispanic whites: The role of language and nativity. Annals of Epidemiology, 17, 191–198.

27.

Lee

S. J.

Moody-Ayers

S. Y.

Landefeld

C. S.

Walter

L. C.

Lindquist

Segal

M. R.

Covinsky

K. E.

(2007). The relationship between self-rated health and mortality in older black and white Americans. Journal of the American Geriatrics Society, 55, 1624–1629.

28.

Lochner

Hummer

Bartee

Wheatcroft

Cox

(2008). The Public-use National Health Interview Survey linked mortality files: Methods of reidentification risk avoidance and comparative analysis. American Journal of Epidemiology, 168, 336–344.

29.

Mirowsky

Ross

C. E.

(2003). Social causes of psychological distress. New York, NY: Aldine de Gruyter.

30.

McGee

D. L.

Liao

Cao

Cooper

R. S.

(1999). Self-reported health status and mortality in a multiethnic US cohort. American Journal of Epidemiology, 149, 41–46.

31.

National Center for Health Statistics. (2009). The National Health Interview Survey (1986–2004) linked mortality files, mortality follow-up through 2006: Matching methodology. Hyattsville, MD: Office of Analysis and Epidemiology. Retrieved July 15, 2015, from http://www.cdc.gov/nchs/data/datalinkage/matching_methodology_nhis_final.pdf

32.

Phillips

L. J.

Hammock

R. L.

Blanton

J. M.

(2005). Predictors of self-rated health status among Texas residents. Preventing chronic disease: Public health research. Practice and Policy, 2, 1–10.

33.

Shetterly

S. M.

Baxter

Mason

L. D.

Hamman

R. F.

(1996). Self-rated health among Hispanic vs. non-Hispanic white adults: The San Luis Valley Health and Aging Study. American Journal of Public Health, 86, 1798–1801.

34.

Singh-Manoux

Dugravot

Shipley

M. J.

Ferrie

J. E.

Martikainen

Goldberg

Zins

(2007). The Association between self-rated health and mortality in different socioeconomic groups in the GAZEL cohort study. International Journal of Epidemiology, 36, 1222–1228.

35.

Spencer

S. M.

Schulz

Rooks

R. N.

Albert

S. M.

Thorpe

R. J.

Brenes

G. A.

… Newman

A. B.

(2009). Racial differences in self-rated health at similar levels of physical functioning: an examination of health pessimism in the health, aging, and body composition study. Journals of Gerontology, 64, 84–94.

36.

Thong

M. S. Y.

Kaptein

A. A.

Benyamini

Krediet

R. T.

Boeschoten

E. W.

Dekker

F. W.

(2008). Association between a self-rated health question and mortality in young and old dialysis patients: A cohort study. American Journal of Kidney Diseases, 52, 111–117.

37.

van Doorslaer

Gerdtham

U. G.

(2003). Does inequality in self-assessed health predict inequality in survival by income? Evidence from Swedish data. Social Science and Medicine, 57, 1621–1629.

38.

Viruell-Fuentes

E. A.

Morenoff

J. D.

Williams

D. R.

House

J. S.

(2011). Language of interview, self-rated health, and the other Latino health puzzle. American Journal of Public Health, 101, 1306–1313.

39.

Zajacova

Dowd

J. B.

(2011). Reliability of self-rated health in US adults. American Journal of Epidemiology, 174, 977–983.

40.

Zajacova

Woo

(2015). Examination of age variations in the predictive validity of self-rated health. Journal of Gerontology, Series B: Psychological Sciences and Social Sciences, 1–7. doi:10.1093/geronb/gbv050