Socio-Economic Status and Life Expectancy in the United States, 1970–1990

Abstract

The relationship between socio-economic status (SES) and life expectancy for 1970 and 1990 is examined in eight states randomly selected from each of eight of the nine census divisions in the United States. High SES populations in seven of the eight states gained additional life expectancy over low SES populations between 1970 and 1990. In the remaining state, the gap between high and low SES populations found in 1970 narrowed by 1990, but did not disappear. The findings have implications for the study of social inequality and its relationship to health outcomes. They also suggest that the United States is unlikely to meet one of two key national health policy goals, the elimination of health disparities by 2010.

Background

Using regression analysis, Swanson and Stockwell (1986) examined life expectancy in 1930 and 1980 in Ohio and found that while differences narrowed between 1930 and 1980, significant geographic variations in life expectancy persisted. They noted that these geographical areas are themselves associated with socio-economic status (SES) differentials (Darroch and Marston 1971, Duncan and Duncan 1955). Swanson (1992) extended the sub-state geographic analysis of Swanson and Stockwell (1986) to sub-state SES differentials and found that between 1970 and 1990 high SES populations in Arkansas not only experienced absolute gains in life expectancy, but also relative gains in life expectancy over low SES populations.

Along with race and gender, SES is one of the three cornerstones of social stratification in the United States (Massey 2007). It has been linked in many studies to mortality differentials (Congressional Budget Office 2008, Ezzati et al. 2008, Gortmaker and Wise 1997, James and Cossman 2006, Hummer 2005, Kitagawa and Hauser 1973, McGehee 1994, Rogers Hummer and Nam 2000, Singh and Siahpush 2006, Stockwell 1961, Stockwell 1963, Stockwell and Laidlaw 1977, Stockwell and Wicks 1984, Stockwell Swanson and Wicks 1987, Stockwell Swanson and Wicks 1988a, Stockwell Swanson and Wicks 1988b, Stockwell Goza and Balisteri 2005, and Stockwell Bedard Swanson and Wicks 1987). Because of the pervasiveness of the findings, Hummer (2005) postulated that socioeconomic differences are fundamental causes of health disparities in the United States, a point also made by Link and Phelan (1995).

Life Expectancy is arguably the single most important indicator of the general health of a population (Lamb and Siegel 2004) and it has long been documented that variations in life expectancy exist among the broad geographic divisions within the United States, as well as among individual states (Dublin et al. 1949, Glover 1921, Oosse 2003).¹ However, until the work of Stockwell and Swanson (1986), virtually nothing was known about sub-state variations.²

The years 1970 and 1990 are selected for this study because they represent what may be regarded as the “bookends” of a pivotal social policy period in the U. S., where federal support of social welfare programs was at a high point circa 1970 (via the New Deal under Roosevelt and the War on Poverty under Johnson, among other programs), but by 1990 (shortly after the end of the Reagan era) federal support was significantly smaller (Reese, 2005; Stockwell Goza, and Balisteri 2005).

The decline in federal support during this period coincided with growing SES inequality in the United States.³ Weinberg (1996) finds, for example, that the distribution of income among households was far more equal in 1970 (Gini Ratio = .394) than in 1990 (Gini Ratio = .428). Massey (2007) elaborates on this theme and finds a significant and continuing increase in social inequality since the late 1960s. Moreover, there is evidence that this increase in social inequality is associated with increased infant mortality rates among those at the low end of the SES scale (Stockwell Goza and Balisteri 2005).

Materials and Methods

In order to have a reasonable level of representation of the United States as a whole, one state from each of eight of the nine census divisions was randomly selected for this study. None of the five states from the New England Census Division was selected because the number of counties in each (Maine, Massachusetts, New Hampshire, Rhode Island, and Vermont) is insufficient for the analysis. The state with the largest number of counties, Maine, has 16, while Rhode Island, the state with the smallest number, has only five.

The random selection procedure in MS-Excel was used to select states within each of the remaining eight census divisions. As many “draws” were made as there are states in each census division so that there was an ordered set of states within each census division from which to draw. This was done because for some states, the 1970 census data available online from the U.S. Census Bureau were found to be incomplete or corrupted.⁴ California and Texas were excluded because of the so-called “Hispanic Mortality Paradox,’ which could confound the analysis in these two states.⁵ The set of states included in the analysis is listed in Table 1 by Census Division.

Table 1.

States Included in the Analysis, by Census Division

For the same reasons described by Swanson (1992), a regression-based technique is used to estimate life expectancy at birth (Swanson 1989), an earlier version of which was used by Swanson and Stockwell (1986). Until the advent of this method for estimating life expectancy, there was no reliable way to examine life expectancy for small populations because life expectancy was calculated through the construction of a life table, which has rigorous data requirements that are difficult to meet for small populations (Kintner 2004). The model used here was tested by Swanson (1989) and found to be sufficiently accurate for its use here. The model is defined as:

where

e_o

is life expectancy at birth

CDR

is the Crude Death Rate (expressed as deaths per 1000 population)

Ln(P65+)

is the natural base logarithm of the percent of the population aged 65 & over

While this model was found to work well for small populations, it has two conditions under which it can produce unreliable estimates: (1) a substantial “special” population, such as is found in a 55+ retirement community; and (2) a small population with very few deaths, such that the crude birth rate can fluctuate substantially from year to year. In terms of the former condition, a very high difference between the percent aged 65 and over at the state level and a given county warrants further examination. In terms of the second, it is advisable to not use the model if the number of deaths is much less than 50. None of the counties in this study was found to be severely impacted by the presence of large retirement populations. However, there are counties and county-equivalents that otherwise would have been included in the analysis but were excluded because the number of deaths was below 50 in one or both of the two years. These counties are listed by state in the Appendix.

As was the case in the study of Arkansas (Swanson 1992), the analytical unit consists of a county population. As also was the case of the Arkansas study by Swanson (1992), this study is an “ecological” design, one that is appropriate because life expectancy is being examined, which is a characteristic of a population, not an individual (Swanson and Stephan 2004: 764). Moreover, no attempt is made to cross the “group-level/individual-level” boundary, so the analysis itself is not subject to the “ecological fallacy” (Freedman 2002)

Mortality data needed to estimate 1970 and 1990 life expectancy values are taken from 1970 and 1990 vital statistics reports produced by the U.S National Center for Health Statistics (1974, 1994), respectively. Population data for 1970 and 1990 are taken from reports for the 1970 and 1990 censuses (U.S. Bureau of the Census 1973U.S. Bureau of the Census 1992), respectively. County populations are grouped into two sets for 1970 and 1990: (1) low SES, the 1^st quartile in terms of median household income; and (2) high SES, the 4^th quartile in terms of median household income. Median household income data are taken from a special report by the Census Bureau (no date). Because the 1970 and 1990 censuses asked for income in the preceding year, the median income data are actually for 1969 and 1989, respectively. All amounts are expressed in 1989 dollars.

To measure change in life expectancy between 1970 and 1990, a dummy variable regression model was constructed for each of the two SES populations by state:

where

e_o

is life expectancy in and 1970 and 1990 for a given SES population as found from the equation shown above

a

is the intercept (1970 mean life expectancy for the same SES population)

b

is the change in mean life expectancy between 1970 and 1990 for the SES population in question

YR

is a dummy variable for year (YR=0, in 1970; YR=1, in 1990)

The one-tailed test (p=.05) is applied to the slope coefficient, b, in each of the two equations to determine if there is a statistically significant change in life expectancy for the SES population in question. Because there is a positive correlation between life expectancy for each SES population in 1970 and 1990, the standard error is diminished. However, this effect is mediated by the extremely small sample sizes and the net result is that a given t-test is not highly subject to a Type I error. The null hypothesis is that there is no change (i.e., b=0); and the alternative hypothesis is that there is positive change (i.e., b >0). This “one-tailed” test structure is appropriate because there is evidence to indicate that, on average, life expectancy increased between 1970 and 1990 (Swanson 1992, U. S. Department of Health and Human Services 2000).

Results

The estimated life expectancy values for each of the two SES populations in 1970 and 1990, by county, are given for each state in Tables 2a through 9a and 2b through 9b, respectively. All significance results are done using a 1-tailed, two-sample t-test assuming unequal variances in the two samples.

To examine the changes in life expectancy, two dummy variable regression equations were constructed using the life expectancy values. One equation represents the change in life expectancy for low SES populations from 1970 to 1990 and the other represents the change for high SES populations from 1970 to 1990. As stated earlier, the slope coefficient in each of these models provides a measure of change in life expectancy (in years) for each of the populations. This allows for a comparison between the low and high SES populations within each state. These equations are found in for each state in tables 2c through 9c, respectively.

In Colorado, mean life expectancy for the low SES population is 72.60 in 1970 (Table 2a), while that for the high SES population is 73.73 (Table 2b), a difference of 1.13 years. This difference is not statistically significant (p = .10). In 1990, mean life expectancy for the low SES population is 76.07 (Table 2a), while that for the high SES population is 77.67 (Table 2b), a difference of 1.60 years, which is statistically significant (p < .001).

Table 2a.

1970 and 1990 Life Expectancy at Birth for Low SES Populations (By Quintile) in Colorado

Table 2b.

1970 and 1990 Life Expectancy at Birth for High SES Populations (By Quintile) in Colorado

Table 2c.

Dummy Regression and Statistical Test Results: Changes in Life Expectancy by SES Population in Colorado between 1970 and 1990

For Colorado (Table 2c), the slope coefficient (3.94) in the dummy variable regression equation for the high SES population is statistically significant (p<.001). This suggests that the high SES population did experience gains in life expectancy between 1970 and 1990 – nearly four years. The slope coefficient (3.47) in the dummy variable regression equation for the low SES population also is statistically significant (p < .001), which suggests that this population gained approximately 3.5 years in additional life expectancy between 1970 and 1990. These results confirm what is seen from the data in tables 2a and 2b by showing that high SES populations posted both absolute and relative gains in life expectancy over low SES populations between 1970 and 1990. On average, the relative gain was about half a year.

Mean life expectancy in Florida for the low SES group is 72.05 in 1970 (Table 3a), while that for the high SES group is 72.21 (Table 3b), a difference of 0.16 years that is not statistically significant (p = .145). In 1990, mean life expectancy for the low SES group is 74.06 (Table 3a), while that for the high SES group is 78.30 (Table 3b), a statistically significance (p < .001) difference of 4.24 years.

Table 3a.

1970 and 1990 Life Expectancy at Birth for Low SES Populations (By Quintile) in Florida

Table 3b.

1970 and 1990 Life Expectancy at Birth for High SES Populations (By Quintile) in Florida

Table 3c.

Dummy Regression and Statistical Test Results: Changes in Life Expectancy by SES Population in Florida between 1970 and 1990

The regression results for Florida are found in Table 3c. The slope coefficient (6.9) in the dummy variable regression equation for the high SES group is statistically significant (p<.001). This suggests that the high SES group did experience gains in life expectancy between 1970 and 1990. However, the slope coefficient in the dummy variable regression equation for the low SES group is not statistically significant (p = 0.085), which suggests that this group did not experience gains in life expectancy between 1970 and 1990. These results support the data in tables 3a and 3b by showing that high SES populations in fact posted relative gains in life expectancy in Florida over low SES populations between 1970 and 1990. As just stated, the high SES populations gained, on average, over six additional years in life expectancy while the low SES populations remained at the 1970 level, making no gains.

In Louisiana, mean life expectancy in 1970 for the low SES population is 71.06 (Table 4a), while that for the high SES population is 70.18 in 1970 (Table 4b). This difference is not statistically significant (p = 0.11). In 1990, mean life expectancy for the low SES population is 72.09 (Table 4a), while that for the high SES population is 74.19 (Table 4b), a statistically significant (p < .001) difference of 2.1 years.

Table 4a.

1970 and 1990 Life Expectancy at Birth for Low SES Populations (By Quintile) in Louisiana

Table 4b.

1970 and 1990 Life Expectancy at Birth for High SES Populations (By Quintile) in Louisiana

Table 4c.

Dummy Regression and Statistical Test Results: Changes in Life Expectancy by SES Population in Louisiana between 1970 and 1990

For Louisiana the regression results are shown in Table 4c. The slope coefficient (4.01) in the dummy variable regression equation for the high SES population is statistically significant (p<. 001). This suggests that the high SES population did experience gains in life expectancy between 1970 and 1990 – approximately four years. The slope coefficient (1.63) in the dummy variable regression equation for the low SES population in Louisiana also is statistically significant (p =. 02), which suggests that this population gained approximately 1.6 years in additional life expectancy between 1970 and 1990. The results confirm what is seen from the data in tables 4a and 4b by showing that high SES populations posted both absolute and relative gains in life expectancy over low SES populations in Louisiana between 1970 and 1990. On average, the relative gain was about 2.4 years (2.4 ≈ 4.01 – 1.63).

In 1970, there is a difference of 1.34 years in mean life expectancy for Mississippi between the two SES groups (Table 5a and Table 5b). This difference is statistically significant (p = 0.01). In 1990, there is a difference of 4.1 years in mean life expectancy between the two SES groups (Table 5a and Table 5b). This difference also is statistically significant (p < 0.001).

Table 5a.

1970 and 1990 Life Expectancy at Birth for Low SES Populations (By Quartile) in Mississippi

Table 5b.

1970 and 1990 Life Expectancy at Birth for High SES Populations (By Quartile) in Mississippi

Table 5c.

Dummy Regression and Statistical Test Results: Changes in Life Expectancy by SES Population in Mississippi between 1970 and 1990

The dummy variable regression equations found in Table 5c for Mississippi are statistically significant (p<.001) and they support the data in tables 5a and 5b by showing that high SES populations in fact posted relative gains in life expectancy over low SES populations between 1970 and 1990. The high SES populations in Mississippi gained, on average, 4.10 years in life expectancy while the low SES populations gained on average 3.36, a relative difference of .74 years between 1970 and 1990.

For Ohio, mean life expectancy for the low SES population is 70.29 in 1970 (Table 6a), while that for the high SES population is 71.18 (Table 6b), a statistically significant difference (p =.013) of 0.89 years. In 1990, mean life expectancy for the low SES population is 74.89 (Table 6a), while that for the high SES population is 77.65 (Table 6b), a difference of 2.76 years that is statistically significant (p < 0.001).

Table 6a.

1970 and 1990 Life Expectancy at Birth for Low SES Populations (By Quintile) in Ohio

Table 6b.

1970 and 1990 Life Expectancy at Birth for High SES Populations (By Quintile) in Ohio

Table 6c.

Dummy Regression and Statistical Test Results: Changes in Life Expectancy by SES Population in Ohio between 1970 and 1990

For Ohio, the regression equations are found in Table 6c. The slope coefficient (6.50) in the dummy variable regression equation for the high SES population in Ohio is statistically significant (p<.001). This suggests that the high SES population in Ohio did experience gains in life expectancy between 1970 and 1990 – approximately 6.5 years. The slope coefficient (4.60) in the dummy variable regression equation for the low SES population in Ohio also is statistically significant (p < .001), which suggests that this population gained approximately 4.6 years in additional life expectancy between 1970 and 1990. This result confirms what is seen from the data in tables 6a and 6b by showing that high SES populations in Ohio posted both absolute and relative gains in life expectancy over the state’s low SES populations between 1970 and 1990. On average, the relative gain was more than two years.

In Pennsylvania, mean life expectancy for the low SES population is 69.47 in 1970 (Table 7a), while that for the high SES population is 71.29 (Table 7b), a difference of 1.82 years that is statistically significant (p <.001). In 1990, mean life expectancy for the low SES population is 75.67 (Table 7a), while that for the high SES population is 76.29 (Table 7b), a difference of .62 years that also is statistically significant (p <.001).

Table 7a.

1970 and 1990 Life Expectancy at Birth for Low SES Populations (By Quintile) in Pennsylvania

Table 7b.

1970 and 1990 Life Expectancy at Birth for High SES Populations (By Quintile) in Pennsylvania

Table 7c.

Dummy Regression and Statistical Test Results: Changes in Life Expectancy by SES Population in Pennsylvania between 1970 and 1990

The regression equations for Pennsylvania are found in Table 7c. The slope coefficient (4.99) in the dummy variable regression equation for the high SES population is statistically significant (p<. 001). This suggests that the high SES population in Pennsylvania did experience gains in life expectancy between 1970 and 1990 – approximately five years. The slope coefficient (6.20) in the dummy variable regression equation for the low SES population in Pennsylvania also is statistically significant (p <.001), which suggests that this population gained approximately 6.2 years in additional life expectancy between 1970 and 1990. This result confirms what is seen from the data in tables 7a and 7b by showing that while the 1970 life expectancy gap was narrowed, high SES populations in Pennsylvania still had higher life expectancy than low SES populations in 1990.

Mean life expectancy for the low SES population in South Dakota is 71.03 in 1970 (Table 8a), while that for the high SES population is 72.74 (Table 8b). This difference is not statistically significant (p = 0.23). This is not surprising given the small number of counties available to test, especially in the low SES set. In 1990 (Table 8a), mean life expectancy for the low SES population is 73.00, while that for the high SES population is 76.91 (Table 8b), a difference of 3.9 years that also is not statistically significant (p =0.07). Again, this result is affected by the small sample size, especially in the low SES set.

Table 8a.

1970 and 1990 Life Expectancy at Birth for Low SES Populations (By Quintile) in South Dakota

Table 8b.

1970 and 1990 Life Expectancy at Birth for High SES Populations (By Quintile) in South Dakota

Table 8c.

Dummy Regression and Statistical Test Results: Changes in Life Expectancy by SES Population in South Dakota between 1970 and 1990

The slope coefficient (4.17) in the dummy variable regression equation for the high SES population in South Dakota is statistically significant (p<.001), as can be seen in Table 8c. This suggests that the high SES population did experience gains in life expectancy between 1970 and 1990 – approximately four years. The slope coefficient (1.97) in the dummy variable regression equation for the low SES population in South Dakota is not statistically significant (p =.54), which again is due largely to the small sample size for the low SES set. If one ignores the statistical test for the low SES set (Table 8a) and assumes that the model is a good approximation of its change in life expectancy, then it would appear that with a gain of about two years, the low SES populations in South Dakota have gained half as many additional years as did the high SES populations, on average (Table 8b).

In 1970, mean life expectancy for the low SES population in Washington (Table 9a) is 72.15, while that for the high SES population is 72.42 (Table 9b), a negligible difference of 0.27 years that is not statistically significant (p = 0.35). In 1990, mean life expectancy for the low SES population is 76.14 (Table 9a), while that for the high SES population is 77.32 (Table 9b), a difference of 1.18 years that also is not statistically significant (p = 0.07). As was the case in South Dakota, these results are affected by the small sample size.

Table 9a.

1970 and 1990 Life Expectancy at Birth for Low SES Populations (By Quartile) in Washington

Table 9b.

1970 and 1990 Life Expectancy at Birth for High SES Populations (By Quartile) in Washington

Table 9c.

Dummy Regression and Statistical Test Results: Changes in Life Expectancy by SES Population in Washington Between 1970 and 1990

The dummy regression equations for Washington are found in Table 9c. The slope coefficient (4.90) in the dummy variable regression equation for the high SES population is statistically significant (p<.001). This suggests that the high SES population in Washington did experience gains in life expectancy between 1970 and 1990 – approximately 4.9 years. The slope coefficient (3.99) in the dummy variable regression equation for the low SES population also is statistically significant (p < .001), which suggests that this population gained approximately four years in additional life expectancy between 1970 and 1990. This result confirms what is seen from the data in tables 9a and 9b by showing that high SES populations in Washington posted both absolute and relative gains in life expectancy over low SES populations between 1970 and 1990. On average, the relative gain was more than one year.

In summary, high SES populations are found in seven of the eight states to have gained not only additional years of life expectancy absolutely, but also relative to low SES populations in seven of the eight states in the sample. The relative gains ranged from half a year to six years, with an overall average of about two years. Pennsylvania was the single state where the life expectancy gap between high SES and low SES that existed in 1970 (about 1.8 years) had narrowed by 1990 (to .62 years).

In five of the eight states, there was no statistically significant difference in 1970 life expectancy between the state’s low and high SES populations. However, by 1990, all of the eight states had statistically significant differences in life expectancy between their low and high SES populations. This was the case even in Pennsylvania, which had a moderate, but statistically significant, difference in 1970 and a negligible, but statistically significant, difference in 1990.

Finally, in 14 of the 16 dummy variable regressions, a statistically significant slope coefficient was found. The only two where this was not found were both for low SES populations, one in Florida and the other in South Dakota. This suggests that there may have been no change in life expectancy for these two populations, a tentative conclusion supported by their very low coefficients of determination, which are r² = .12 and .03, respectively.

Discussion

Before summarizing the findings, it is useful here to discuss two issues that may come to mind in regard to the analysis, which basically is comprised of examining cross-sectional data at two points in time. This type of analysis is a natural approach given that a period-based measure of life expectancy measure is being used, which is itself cross-sectional in nature. The first issue is comprised of fertility and migration, which along with mortality, form the components of population change. All three of these components of population change affect the high and low SES populations examined here between 1970 and 1990. Migration is particularly salient. The second issue is the social mobility of a given population, where it might be upward or downward. Both of these issues can interact with one another, so it is useful to address them together.

In regard to the first issue, consider that a low SES (county) population in 1970 could be impacted by migration such that by 1990 this (county) population is no longer in the low SES group. Similarly, a high SES population in 1970 could be impacted by migration such that it is no longer in this group by 1990. The issue of social mobility for a given population can be viewed as a variation on the first issue, whereby it is not the coming and going of people that affect changes in SES but, rather, income changes specific to the population. For example, it could be the case that a major source of employment comes into being in a county subsequent to 1970 that improves the household incomes of the residents. Similarly, a major source of employment in 1970 may no longer be in the county in 1990 and, as such, household incomes are depressed. Finally, it can be through a combination of migration and social mobility that the county populations change in terms of SES groups between 1970 and 1990.

It is for precisely these reasons that the analysis reported here does not attempt to follow a population in a given SES set in 1970 into 1990. Instead, the analysis examines the low and high SES populations as found at each of the two time points. This is entirely appropriate because the period (cross sectional) life table needs to be matched with cross-sectional (period) SES groups. By doing this, the analysis remains on track to its goal – the association of SES with a period-based measure of life expectancy. It simultaneously provides a tractable solution to the issue of dealing with temporal effects such as upward and downward social mobility and migration.

Turning to a summary of the results, with the exception of Pennsylvania, the results found here are consistent with those found in Arkansas (Swanson 1992) in that seven states of the eight states analyzed here saw high SES populations gain additional years in life expectancy relative to low SES populations between 1970 and 1990. These results, in turn, are generally consistent with those reported by the Congressional Budget Office (2008), Ezzati et al. (2008), and Singh and Siahpush (2006).

The reasons for these absolute and relative gains by high SES populations may be due to what is observed by Stockwell, Goza, and Balistreri (2005) in regard to infant mortality rates, namely that income inequality has been increasing in the United States since 1970, the beginning of a trend in which many social welfare programs were cut back by the federal government.⁵ The fact that the relative gains varied by state, with southern states having larger relative gains than states outside the south, suggests that state-level programs may have ameliorated federal program cuts and in Pennsylvania’s case, overcome them. In this regard, it is of interest to examine median household incomes (as noted earlier, all amounts are expressed in 1989 dollars) in these states (U.S. Census Bureau no date). These data are found in Table 10.

Table 10.

1970 and 1990 Median Household Income by State (in 1989 dollars)*

In 1970, Mississippi ranked last (50^th) among the states in terms of median household income ($16,432), while Louisiana ($20,576) and Florida ($22,559) ranked 44^th and 37^th, respectively and South Dakota ($20,299) ranked 47^th. It would have been difficult for these states to make up for lost federal monies with state funds. In 1990, Mississippi ($20,136) again ranked last (50^th), with Louisiana ($21,949) and South Dakota ($22,503) ranked 47^th and 46^th, respectively and Florida ($27,483) ranked 28^th. In regard to Ohio, it ranked 12^th in 1970 with a median household of $29,203 that was not only much higher income than any of these states, but also higher than that in Colorado ($26,509), Pennsylvania ($26,902), and Washington ($28,718). However, Ohio was only one of three states (the other two being Michigan and Montana) that experienced a decline in median household income between 1970 and 1990, dropping from 12^th highest at $29,203 in 1970 to 25^th in 1990 with a median household income of $28,706. While median household income in Ohio was declining, it was increasing in Colorado (from $26,509 to $30,140), Pennsylvania, (from $26,902 to $29,069) and Washington ($28,718 to $31,183). It would have been difficult to implement or maintain social welfare programs in Ohio, given its income decline as would have been the case in Florida, Louisiana, Mississippi, and South Dakota, given their low 1970 incomes. However, all else being equal, it would not have been so difficult in Colorado, Pennsylvania, and Washington, with their relatively high median household incomes in 1970 and their increases by 1990. This may account for the fact that the life expectancy gap narrowed in Pennsylvania, while in Colorado and Washington, the high SES populations posted the lowest relative gains in life expectancy over the low SES populations - patterns different than those observed in Ohio, Florida, Louisiana, Mississippi, and South Dakota. These results also suggest that spatial inequalities in health outcomes exist across the United States.

SES is one of three primary mechanisms of social stratification in the United States (Massey 2007) and it has been found to have a broad range of health access and health outcomes in the United States (Gortmaker and Wise 1997, James and Cossman 2006, Hummer 2005, McGehee 1994, Stockwell Goza and Balistreri 2005). Income is not the only dimension of SES, but it is arguably the most significant in the United States (Massey 2007) and as observed by Riley (2001: 143), it serves as “…an enabling factor, which boosts or diminishes life expectancy, depending on how it is spent.” Riley (2001) also observes that moderate investments in health care in the modern era can produce striking advances in life expectancy. The key in accomplishing this is to have investments in health care follow a social strategy that improves patient access to doctors and nurses rather than technical medicine (Riley 2001: 120). This observation is particularly salient for the United States, virtually the only “modern” country lacking universal healthcare. It also is salient for individual states.

These findings are of practical interest because in its “Tracking Healthy People 2010” report, the U. S. Department of Health and Human Services (2000) cites the elimination of health disparities by the end of this decade as one of its two key goals. Clearly, the findings that these disparities increased between 1970 and 1990 in seven of the eight states do not bode well for meeting this goal nationally by 2010, whether in spatial terms, SES terms, or a combination of the two. The findings also provide support for the argument by Stockwell, Swanson, and Wicks (1988a, 1988b) that declining relative standards in living for the low and lower middle SES populations along with the imposition of national policies that limited their health care were likely to be factors contributing to a lack of narrowing of mortality differentials between them and high SES populations subsequent to 1970. Singh and Siahpush (2006) also explicitly state that widening SES gaps in life expectancy may be related to increasing temporal inequalities in the material and social living conditions, both in absolute and relative terms. The Congressional Budget Office (2008), however, does not mention the possibility that decreased federal funding of social welfare programs may be at least part of the reason for the widening life expectancy gap it found between low and high SES groups. Rather than listing any structural factors, it lists as possibilities, four individual-level factors: (1) smoking; (2) Obesity; (3) self-management of disease; and (4) health lifestyles and use of health care.

As stated earlier, the finding that SES-based life expectancy differentials tended to increase during the period 1970 to 1990 is consistent with the literature. Unfortunately, the widening differentials represent a major break with the past. Shortly before his death in 1969, the pioneering actuary, demographer, and biostatistician, Mortimer Spiegelman (1968) wrote that that gender, race, spatial, and SES mortality differentials in the United States had been narrowing since 1940. He concluded his paper by stating that (national) leaders were now responsible for seeing that adequately staffed and functioning health facilities were conveniently accessible to the public in order for these trends to continue. He clearly believed that the country’s leadership would shoulder this responsibility because he foresaw that even smaller mortality differentials were in the country’s future. Unfortunately, from the standpoint of national policy, it appears that his optimism was misplaced.

What are the next steps in examining the relationship between SES and life expectancy? First, the results here suggest that state supported program can play a role in offsetting federal cuts to social welfare programs. This needs to be examined more carefully across the states using relevant budget information. Second, in addition to extending the analysis to other states, which would provide more of an idea on the role of state governments in maintaining social welfare and other programs for low SES populations, it will be important to examine how race interacts with SES and life expectancy. Swanson and Stockwell (1988) found, for example, that while race moderated the geographic association with life expectancy in Ohio, the association was not spurious. Swanson and McGehee (1996) found similar results in regard to race and SES in Arkansas, where between 1970 and 1990: (1) High SES Black populations gained more than three additional years of life expectancy over Low SES Black populations; and (2) High SES White populations gained more than 0.5 years of life expectancy over Low SES White populations. It is expected that similar moderating effects of race will be found elsewhere, but this is a working hypothesis to be tested. Similarly, it will be important to examine differentials by Hispanic and no-Hispanic ethnicity. This effort will be challenging in some states because of the “Hispanic Mortality Paradox” (Hummer et al. 2007), but if it can be sorted out, determining the effect of ethnicity on life expectancy in conjunction with SES should be valuable for both substantive and policy reasons.

There are factors beyond income that play a role in access to health care and health outcomes. (Rogers Hummer and Nam 2001, Kirby 2008). One area for further research would be to examine some of these other factors, such as educational attainment and health insurance, in regard to life expectancy.

Finally, with the arrival of census 2010 it will be useful to see if the trends found between 1970 and 1990 changed subsequently to 1990. Such an examination should include race (and ethnicity), for as McGehee (2005) notes, the economic restructuring that has occurred in the United States and the accompanying increase in income inequality, has disproportionately affected minorities. Including these factors would provide a current picture on where the U.S. stands in terms of meeting the “Healthy America 2010” goal of achieving the elimination of health disparities, not only spatially and by SES, but also by one of the other two cornerstones of social stratification in the United States, race (and ethnicity).