In contemporary California, we found a gap of nearly twenty years between the sociodemographic group with the shortest life expectancy—African-American males living in neighborhoods of the lowest SES quintile (65.3 years)—and that with the longest life expectancy—Asian/Pacific Islander females living in neighborhoods of the second-highest SES quintile (84.9 years). This gap is of comparable magnitude to those observed between developing and developed countries internationally and is of similar size to that, for example, between Bangladesh (63 years) and Japan (82 years) in 2008 (U.S. Census Bureau, 2008
) We also observed marked positive socioeconomic gradients in life expectancy among African-American and white males and females, averaging 1.4 additional years of life expectancy per quintile of increasing SES.
To our knowledge, our analysis is the first to use sub-county level measures of area SES to examine life expectancy in the US. A recent assessment using county-level SES classifications reported a socioeconomic gradient in life expectancy about half the size observed in our data: an increase of +0.7 years per quintile for all races/ethnicities combined (Singh & Hiatt, 2006
). However, a different assessment of US disparities in life expectancy using county-level data found a comparable difference of 20.7 years between the groups with high (Asian females) and low life expectancy (African-American males living in “high risk urban areas” (Murray et al., 2006
). As the authors of these studies noted, their use of county-level SES characteristics likely underestimate disparities among groups. Although we used the smallest geographic unit available from the US census (the block group averaging about 1500 residents), it is likely that some of these units were heterogeneous with respect to SES. Of course, regardless of the size of the available geographic unit, these and all analyses using area-level SES classifications may underestimate disparities occurring at the individual level. In the absence of individual-level measures, we are unable to address the likelihood of “ecologic fallacy”, when area-level effects differ from individual-level effects. Thus, our neighborhood measure implies that the observed gradients reflect an unspecified mix of individual- and neighborhood-level (e.g., contextual) factors. Individual-level SES may influence access to health care, insurance coverage, health knowledge and awareness, and the reverse effects of health on ability to work and subsequent economic achievement, all of which can be additionally modified by neighborhood-level features of the social or physical environment.
In our California data, life expectancy for African-Americans was consistently two to four years lower than that for whites across all levels of neighborhood SES. This observation, based on a large and representative US population, confirms prior reports of a mortality disparity between African-Americans and whites after accounting for various measures of SES (Lin, Rogot, Johnson, Sorlie, & Arias, 2003
). Reasons for this persistent disparity are discussed at length elsewhere (David R. Williams & Collins, 1995
) and, briefly, may involve racial differences in early life experiences or cultural norms, health impacts of factors associated with racism and discrimination, or augmented coping resulting in intensified stress among African-Americans (James, 1994
; James, Keenan, Strogatz, Browning, & Garrett, 1992
; D. R. Williams, Haile, Gonzalez, Neighbors, Baser, & Jackson, 2007
In contrast to the clear SES gradient in life expectancy observed among African-Americans and whites, there was virtually no such gradient seen among Hispanics and Asians/Pacific Islanders, with the exception of a slight positive gradient among Asian/Pacific Islander males. These results are consistent with those of numerous studies that have identified a so-called “Hispanic paradox”: that life expectancy of Hispanics in the US is greater than would be expected based on their socioeconomic position (Markides & Eschbach, 2005
). Goldman and colleagues recently elaborated on the absence of an educational gradient in health measures in populations of Mexican origin (Goldman, Kimbro, Turra, & Pebley, 2006
). This study suggests a similar paradox to that described in US Hispanics for US Asians/Pacific Islanders.
There are several possible explanations for the lack of neighborhood socioeconomic effects on life expectancy among California Hispanics and Asians/Pacific Islanders, among whom there is a large foreign-born contingent. At the individual-level, immigrants may be self-selected for good health, and be physiologically or psychologically hardier than their counterparts at home. Compared to those born in the US, foreign-born Asians/Pacific Islanders have been reported to have higher markers of some health status indicators (Frisbie, Cho, & Hummer, 2001
). Immigrants who do suffer from ill health may be more likely to return to their country of origin (Palloni & Arias, 2004
). This combination of “immigrate-when-healthy” and “emigrate-when-sick” patterns may lead to a highly selected group of healthy individuals. In addition, immigrants may continue healthful behaviors or attitudes, such as dietary or physical activity habits, associated with their country of birth, and these may counteract the negative health effects associated with lower SES in the US. Although data suggest that the generally more favorable health among immigrants wane with years in the US and increasing levels of acculturation (Frisbie et al., 2001
), the extent to which this varies with SES is unclear. Third, Hispanic and Asian/Pacific Islander groups may have more extensive family and neighborhood social networks for support (e.g., ethnic enclaves), again counteracting the negative effects of lower SES, or of living in a lower-SES neighborhood on health. Some of these patterns may be particularly pronounced among older Hispanics and Asians/Pacific Islanders, resulting in the observed cross-over with lower mortality rates among lower, rather than higher, neighborhood SES groups at advanced ages.
It is also possible that data limitations in our study masked a socioeconomic gradient in life expectancy among Hispanics and Asians/Pacific Islanders. As each of these racial/ethnic groups comprise different cultural and linguistic subgroups with heterogeneous disease profiles, this heterogeneity may have varied across SES strata. For example, SES gradients in health among Mexican-Americans could be obscured when they are combined into a single, larger group of Hispanics. A strong neighborhood-SES gradient in mortality rates at ages 20-64 years has been reported among Mexican-Americans in the US, with a magnitude similar to that observed among whites and African-Americans (Winkleby & Cubbin, 2003
). This apparent discrepancy in results could relate to differences between the US Mexican-American population and the California Hispanic population, although 77.1% of Hispanics in California were of Mexican origin according to the 2000 US census (U.S. Census Bureau, 2001
In addition to independent contributions of race/ethnicity and SES, we also noted an important interaction by age to the observed disparities in life expectancy. Differences in age-specific mortality rates among racial/ethnic and SES groups tended to converge at the older ages, suggesting that Medicare and other programs providing universal health insurance coverage to older adults may be helping to reduce health inequalities among older Americans. However, the convergence of mortality rates at older ages in our data appears to begin in the late 40s—well before the age of Medicare eligibility, perhaps implying a role for the relocation of persons from higher- to lower-SES neighborhoods after retirement. Furthermore, even among those over age 65 years, important differences were evident across race/ethnicity and SES, with lowest-SES elderly African-Americans and whites facing mortality rates 33% above the statewide average. Thus, Medicare and similar health insurance programs cannot entirely account for the reduction in health disparities among older adults, nor are they likely to completely ameliorate such imbalances. Some of the attenuation of life expectancy disparities at older ages may be due to a “healthy survivor” effect, by which individuals in certain high-risk subgroups, especially among African-Americans, whites, and those of lower SES, die at a younger age, leaving a healthier older population. For instance, young African-Americans have a higher prevalence of health-risk behaviors (Eaton, Kann, Kinchen, Shanklin, Ross, Hawkins et al., 2008
) and a higher rate of deaths due to homicide than whites, Hispanics, and Asians/Pacific Islanders (Karch, Lubell, Friday, Patel, & Williams, 2008
Despite the convergence in mortality rates with age, differences among racial/ethnic and SES groups at older ages contribute substantially to the life expectancy gap. We found that 70% of the gap in life expectancy between the most advantaged and least advantaged population groups related to differences in mortality risk after age 45 years. While significant media and public health attention has conventionally been focused on disparities in mortality risks for infants, teenagers, and young adults, our data suggest that the major contributors to disparities in overall longevity lie in smaller differential risks at older ages.
Our estimates are based on large, routinely collected sources of vital statistics and population data. However, our results should be interpreted in light of several well-known possible sources of bias in using these kinds of data. Numerator/denominator bias, which occurs whenever case counts are collected using different methods than population estimates, is a possibility. Relatedly, it is possible that race and ethnicity were classified inconsistently for US census and vital statistics resources. In the US census, people self-identify their race and ethnicity, whereas on death certificates, race and ethnicity information are entered by the person (usually the funeral director) responsible for recording the death certificate, and is done through either direct observation or questioning of relatives. This kind of misclassification may have led to Hispanic and Asian/Pacific Islander deaths being undercounted on death certificates (Rosenberg et al., 1999
; Swallen & Guend, 2003
), which may have resulted in underestimation of Hispanic and Asian/Pacific Islander mortality rates. Within these groups, there may be additionally important differences in tabulation between those born in the US and those born abroad.(Eschbach, Kuo, & Goodwin, 2006
) We attempted to correct for the underestimation of Hispanic deaths through the use of a surname list. However, we did not correct for Asian/Pacific Islander misclassification in the absence of a similarly well-tested surname-list based method of adjustment. As a result, our findings for the Asian/Pacific Islander population may understate true mortality rates and thereby represented inflated estimates of life expectancy.
The 2000 US census data are subject to error, with overall counts estimated to be 1.2 percent lower than the true population (National Research Council, 2001
), and larger undercounts for younger persons, African-Americans, Hispanics, and persons who rent their homes. By contrast, death registration is believed to be nearly 100 percent complete among all population groups. Therefore, the use of census-based denominators may result in overestimated mortality rates, and the degree of overestimation may be larger for African-Americans and neighborhoods with larger proportions of renters. Additionally, self-reporting of age has been reported to be misclassified in US census data, with older individuals more likely to overstate their age. Census age misclassification has been reported to be more common among older African-Americans, which may underestimate mortality rates among older age groups (Elo IT, 1997
). In this analysis, we were unable to adjust the census counts for undercounts and age misclassification, although some of the effects of the latter could have been mitigated by grouping together persons aged 85 years and older for analysis.
In addition, the meaning of our neighborhood measure of SES may vary by racial/ethnic group. In particular, residential segregation by race/ethnicity and factors related to SES may have produced heterogeneity of neighborhoods within the same neighborhood SES quintile. Thus, unmeasured features of neighborhoods that correlate with both neighborhood SES quintile and individual race/ethnicity may have influenced our life expectancy estimates. Second, although census-based SES measures were originally developed as a readily-available proxy for individual-level measures of SES (Berkman & Macintyre, 1997
; Geronimus & Bound, 1998
; Krieger, 1992
; Krieger & Fee, 1994
; Yeracaris & Kim, 1978
), more recent studies have shown that both individual-level and aggregate-level SES measures have independent effects on health (Chandola, Bartley, Wiggins, & Schofield, 2003
; Diez Roux, 2001
; Haan, Kaplan, & Camacho, 1987
; Kubzansky, Subramanian, Kawachi, Fay, Soobader, & Berkman, 2005
; Pickett & Pearl, 2001
; Subramanian, 2004
; Winkleby & Cubbin, 2003
; Yen & Kaplan, 1999
), that is, contextual measures can capture attributes of the neighborhood environment over and above individual-level characteristics. It is increasingly recognized that contextual SES measures account for complex relationships between individual- and neighborhood-level SES measures, and that SES on several levels can independently affect health (Winkleby & Cubbin, 2003
Our findings suggest that small-area SES is a meaningful differentiator of life expectancy disparities at least among whites and African-Americans in the state of California, which represents one-ninth of the US population and demonstrates variation in SES similar to that observed in the US as a whole (Reynolds, Hurley, Quach, Rosen, Von Behren, Hertz et al., 2005
). Quantification of the magnitude of racial/ethnic and socioeconomic health disparities should be an important goal of public health surveillance. As most public health surveillance entities rely heavily on US census data as well as statewide death, cancer, or other health registries, these agencies need to consider adding routine collection of more meaningful measures of SES, preferably an individual-level measure that can be assessed in conjunction with the neighborhood or other small-area-based measures based on residential street address. Quantifying disparities among population groups is a necessary first step to remediating them, and is also necessary to document the impact of policies and programs designed to reduce health disparities. If life expectancy is ever to be equalized among population groups in the US—a lofty goal requiring the strong collaboration of policy makers, public health professionals, health care providers, and community members—we must continue to identify more and better mediators of health disparities and incorporate their measurement into our routine surveillance.