U.S. Census Bureau

Variations in State Mortality From 1960 to 1990

By Monique Oosse

Population Division
U.S. Census Bureau
Washington, D.C. 20023

May 2003

Population Division Working Paper Series No. 49

Disclaimer

This paper reports the results of research and analysis undertaken by the U.S. Census Bureau staff. It has undergone a Census Bureau review more limited in scope than that given to official Census Bureau publications. This report is released to inform interested parties of ongoing research and to encourage discussion of work in progress.


Abstract

In order to evaluate a fundamental assumption in state population projections for the United States, this paper assesses trends in mortality levels as indicated by changes in state life expectancy at birth and at age 65 from 1960 to 1990. The most recent state population projections assume that existing differences between state mortality levels will continue through the projection period. Inter-quartile ranges, coefficients of variation, and the Dispersion Index were calculated from National Center for Health Statistics life table data to determine whether state life expectancies became more or less similar to each other over the four time periods. There is some evidence of converging life expectancies from 1960 to 1980, but the trend appears to have reversed by 1990 when most of the indicators of variation in state mortality returned to the 1960 levels. Though there is little evidence of systematic convergence or divergence for the nation from 1960 to 1990, some general regional trends are identified. Given the number of exceptions to the patterns of mortality changes in every region, inconsistencies in the national trends, and the limited data for races other than White, this evaluation concludes that the assumption currently used for state population projections is the most appropriate assumption.


Table of Contents

  1. Introduction
  2. Background
  3. Data
  4. Methods
  5. Results
    1. Variation in White Life Expectancy
    2. Variation in Nonwhite Life Expectancy
    3. State Rankings of White Life Expectancy
    4. State Rankings of Nonwhite Life Expectancy
    5. Linear Regression Estimates for State Life Expectancies
    6. Variation in Life Expectancy by Regions
  6. Summary and Conclusions
References

Figures

Figure 1. Life Expectancy by Age for the District of Columbia and Hawaii: 1989-1991
Excel (35k) | PDF (59k)
Figure 2. State Life Expectancy at Birth for White Males in the Midwest Region: 1960 to 1990
Excel (25k) | PDF (72k)
Figure 3. State Life Expectancy at Age 65 for White Females in the South Central Division: 1960 to 1990
Excel (23k) | PDF (67k)
Figure 4. State Life Expectancy at Birth for Nonwhite Females in the South Central Division: 1960 to 1990
Excel (23k) | PDF (65k)
Figure 5. State Life Expectancy at Age 65 for Nonwhite Females in the South Atlantic Division: 1960 to 1990
Excel (30k) | PDF (67k)

Tables

Table 1. Measures of Variation in State Life Expectancy at Birth and at Age 65 for White Males and Females: 1960 to 1990
Excel (16k) | PDF (74k) | CSV (1k)
Table 2. Measures of Variation in State Life Expectancy at Birth and at Age 65 for Nonwhite Males and Females: 1960 to 1990
Excel (16k) | PDF (75k) | CSV (1k)
Table 3. Rankings of State Life Expectancy at Birth for White Males: 1960 to 1990
Excel (18k) | PDF (73k) | CSV (2k)
Table 4. Rankings of State Life Expectancy at Birth for White Females: 1960 to 1990
Excel (18k) | PDF (73k) | CSV (2k)
Table 5. Rankings of State Life Expectancy at Birth for Nonwhite Males and Females: 1960 to 1990
Excel (18k) | PDF (73k) | CSV (2k)
Table 6. Rankings of State Life Expectancy at Age 65 for White Males: 1960 to 1990
Excel (18k) | PDF (73k) | CSV (2k)
Table 7. Rankings of State Life Expectancy at Age 65 for White Females: 1960 to 1990
Excel (17k) | PDF (66k) | CSV (2k)
Table 8. Rankings of State Life Expectancy at Age 65 for Nonwhite Males and Females: 1960 to 1990
Excel (17k) | PDF (66k) | CSV (2k)
Table 9. Slope Estimates for Linear Fit of 10-Year Changes in Life Expectancy at Birth by States for White Males and Females: 1960 to 1990
Excel (19k) | PDF (50k) | CSV (2k)
Table 10. Slope Estimates for Linear Fit of 10-Year Changes in Life Expectancy at Age 65 by States for White Males and Females: 1960 to 1990
Excel (19k) | PDF (50k) | CSV (2k)
Table 11. Slope Estimates for Linear Fit of 10-Year Changes in Life Expectancy at Birth by States for Nonwhite Males and Females: 1960 to 1990
Excel (17k) | PDF (73k) | CSV (2k)
Table 12. Slope Estimates for Linear Fit of 10-Year Changes in Life Expectancy at Age 65 by States for Nonwhite Males and Females: 1960 to 1990
Excel (16k) | PDF (73k) | CSV (2k)
Table 13. Coefficients of Variation in Life Expectancy at Birth and at Age 65 within Census Regions or Divisions for White Males and Females: 1960 to 1990
Excel (18k) | PDF (80k) | CSV (2k)
Table 14. Coefficients of Variation in Life Expectancy at Birth and at Age 65 within Census Regions or Divisions for Nonwhite Males and Females: 1960 to 1990
Excel (18k) | PDF (82k) | CSV (2k)

Appendix A

Appendix A-1. Life Expectancy at Birth by Sex and Race: 1959-1961
Excel (19k) | PDF (50k) | CSV (3k)
Appendix A-2. Life Expectancy at Birth by Sex and Race: 1969-1971
Excel (20k) | PDF (50k) | CSV (3k)
Appendix A-3. Life Expectancy at Birth by Sex and Race: 1979-1981
Excel (20k) | PDF (51k) | CSV (3k)
Appendix A-4. Life Expectancy at Birth by Sex and Race: 1989-1991
Excel (20k) | PDF (51k) | CSV (3k)
Appendix A-5. Life Expectancy at Age 65 by Sex and Race: 1959-1961
Excel (19k) | PDF (51k) | CSV (3k)
Appendix A-6. Life Expectancy at Age 65 by Sex and Race: 1969-1971
Excel (20k) | PDF (51k) | CSV (3k)
Appendix A-7. Life Expectancy at Age 65 by Sex and Race: 1979-1981
Excel (20k) | PDF (51k) | CSV (3k)
Appendix A-8. Life Expectancy at Age 65 by Sex and Race: 1989-1991
Excel (20k) | PDF (51k) | CSV (3k)

Population Division Working Paper Series


I. Introduction

The U.S. Census Bureau periodically produces state population projections for the 50 states and Washington, DC. The most recent projections released in 1996 used a cohort-component method to project the state populations by adding future births, subtracting future deaths, and adding future net movement between states and to or from other countries. These components were projected by age, sex, race, and Hispanic origin for each state. The projections based future births, deaths, and domestic and international migration on various historical and national trends and assumptions (see Campbell 1996 and Campbell 1997).

This paper evaluates the assumption used in this projection that the current differences between state mortality levels by age, sex, race, and Hispanic origin would remain constant over the projection period. This assumption had a substantial impact on projected mortality trends by states, but historical trends have not been examined in detail to determine if the assumption is justified. In fact, there is some evidence that mortality trends may converge toward national averages.

To evaluate the accuracy of the mortality assumption for U.S. state population projections, this research assesses how the relative mortality levels of states1 changed from 1960 to 1990. Section II presents related research on state, rural-urban, and international differences in mortality. Section III describes data and limitations of the Nation Center for Health Statistics (NCHS) decennial life tables used for the analysis. Section IV discusses the statistics used to evaluate the overall patterns of change in state mortality from 1960 to 1990. Section V reports the results for three measures of variation in state life expectancies for the nation and discusses trends in mortality changes within Census Regions. The life expectancy changes for individual states are also described in the results section. Finally, Section VI summarizes the conclusions drawn from this research and offers recommendations for projecting state mortality changes.


II. Background

Most of the literature addressing relative changes in U.S. mortality over time is limited to single time periods, specific causes of death, or rural-urban differences across the nation or within individual states. Several studies report consistent regional patterns for mortality levels in the 1980s and 1990s with mortality levels highest in the South. Data from 1988 to 1992 show that mortality rates from all causes were highest in the East South Central Division2 and mortality rates were also relatively high in the Pacific region for women (Pickle et al. 1996). The regional pattern of mortality is most influenced by the geographic distribution of heart disease and lung cancer, the leading causes of death. For the period of 1986 to 1990, NCHS vital statistics data show that lung cancer death rates were higher in the South and most of the Midwest and lower in the Northeast and West (Edmondson 1994). Circulatory death rates were also highest in the South (Schneider et al. 1997).

National studies report that mortality in urban areas is higher than rural areas and is lowest in suburbs (Clifford and Brannon 1985). For example, much of the excess mortality from ischemic heart disease (IHD) in New York in 1980 to 1987 was associated with the higher rates of IHD mortality in urban centers, particularly New York City (McNutt et al. 1994). These differences may be related to differential access to health care, different lifestyles, and exposure to environmental hazards.

The question of interest regarding change over time within the United States is examined more directly in two studies that are limited to one cause of death in one state or rural-urban differences. Lung cancer research by Pommerenke and colleagues (1994), for example, found that within South Carolina lung cancer mortality for all men ages 40 years and over rose from below the national average in 1953 to 1962 to above the national average in 1978 to 1987. A national study of rural-urban differences in mortality from 1950 to 1980 showed that the absolute rural-urban mortality differential declined over the time period, but the relative differential increased (Clifford and Brannon 1985).

There is also some research on the convergence of mortality trends at the international level. A detailed analysis of mortality patterns in Africa and Latin America describes regional convergence during periods of rapid life expectancy improvements in the 1930s to 1950s. From the 1950s to 1970s, the life expectancy differences among nations in many of the developing regions remained relatively stable. The Latin American and Caribbean countries actually experienced some divergence in mortality levels from the 1960s to 1970s (Gwatkin 1980).

White (2002) evaluated annual changes in life expectancy for 21 industrially developed countries over four decades. Among other findings, lower life expectancy in 1955 was associated with higher average annual increases in life expectancy at birth from 1955 to 1996. This tendency to converge toward the mean occurred primarily before 1980, and there has been relatively low variance in life expectancy since 1980. The rate of life expectancy change for these developed countries was generally between its previous rate of change and that of the entire group. A similar study for Japan concludes that the slowing of Japanese mortality decline corresponds to a convergence with the mortality levels of other developed counties (Wilmoth 1998).

An evaluation of about 184 United Nations countries reports declining differences in the range of life expectancy at birth from 1950-55 to 2000 (Wilson 2001). The inter-quartile range of life expectancies for these countries declined from 26 years in the early 1950s to 8 years in 2000. This study also showed convergence of fertility rates over the same time period.

These findings suggest that geographic differences in mortality may change over time rather than remain constant as assumed in the state projections. Factors that influence changes in life expectancy within the United States include increasing mobility and communication, decreasing differential access to health care, nearly universal access to vaccines against childhood diseases, improved public health services (White 2002), and changes in income inequality (Lochner et al 2001). It is hypothesized that these changes reduce the differences in mortality levels between states and that state mortality rates converge, or become more similar to each other, over time. A convergence toward national mortality levels may also be associated with approaching the biological limits of mortality (Coale 1996, White 2002). This hypothesis proposes that states with above average life expectancy will experience slower life expectancy gains and states with below average life expectancy will experience relatively faster life expectancy gains and "catch up" to the states with higher life expectancies.

The alternative hypothesis and the assumption currently used in state population projections is that relative differences in state mortality levels will remain in the future. This may be due to the wide variety of factors that influence mortality levels, such as education, income, environmental conditions, public health services and access to health care, and these may continue to differ across states.

Similar studies of the convergence or divergence of state fertility rates over time provide the general framework for this research. O’Connell (1981) calculated changes in the relative variation (coefficient of variation) among states from 1940 to 1977 for the total fertility rates and the average number of lifetime births expected. The results did not show a consistent pattern of convergence or divergence of fertility over time for individual states, however, South and non-South differences declined substantially over this time period. Heaton (1998) also found no consistent trends in the changes in fertility for states relative to the nation and each other over time.


III. Data

The average number of years lived, or life expectancy at birth, is a composite measure of the mortality experience of a population at a given time period. Life expectancy at birth captures the higher risk of mortality at infancy and early childhood as well as the degree to which mortality risk increases at older ages. The average number of years remaining for people who survive to a specific age eliminates variations caused by the heavily weighted impact of infant and child mortality and death in early adulthood. Comparisons of life expectancy at age 65 are commonly used to indicate different levels of mortality from chronic, degenerative diseases of old age rather than the more diverse mortality experience of the entire population.

Figure 1 shows examples of life expectancy trends from birth to age 110 for Hawaii and the District of Columbia in 1990. The small increase from life expectancy at birth to life expectancy at age 1 for the District of Columbia is explained by the relatively high risk of death during the first year of life compared to later in childhood. After age 65, life expectancy begins to decline at slower rates. The differences in life expectancy are generally smaller at age 65 than at birth. Life expectancy at birth is used in this research to capture state differences in overall mortality and life expectancy at age 65 is used to capture the more limited mortality differences among states at the ages where most deaths occur.

NCHS produced a series of U.S. Decennial Life Tables from state level mortality data for the three years surrounding the decennial years 1960, 1970, 1980 and 19903, as well as earlier periods not included in this study (NCHS 1975b, 1985-86, 1998b). These years were selected because they follow the substantial mortality declines in the first half of the twentieth century due to the widespread use of antibiotics and vaccines (NCHS 1987b) and they are most relevant to projecting future mortality. In each series U.S. and state life tables were calculated for single years of age up to 110 for the total population, males, females, Whites, All Other Nonwhites (referred to as Nonwhites in this paper), and the four sex-race groups. In 1980 and 1990, life tables were also calculated for Blacks, Black males, and Black females within the All Other Nonwhite group.

NCHS made several adjustments to the life tables and source data:

  1. The life table calculations used birth registration data for the population under age two instead of Census data for all four time periods because infants were under-enumerated in Census population data (NCHS 1967, 1975a, 1987a, 1998a).

  2. To correct for birthday preferences of 1900 in the 1960 Census, the Nonwhite population ages 55 to 64 was redistributed in the two relevant five-year age groups (55 to 59 years and 60 to 64 years) (NCHS 1967).

  3. Sample Census data were used to construct 1960 life table populations for ages 85 to 100 because complete 1960 Census tabulations were only available through age 85 (NCHS 1967).

  4. The 1970 and 1980 Census populations required adjustments to the centenarian population and the population other than White or Black due to overestimates of each (NCHS 1975a, 1987a).

  5. The 1990 Census population included modifications by age and race that differed from the 1970 and 1980 adjustments (NCHS 1998a).

  6. The 1970, 1980, and 1990 life tables used Medicare death data for ages 85 and older (NCHS 1975a, 1987a, 1998a).

  7. Trend smoothing was applied to address some of the variations caused by differential coverage, age misreporting, and small denominator populations (NCHS 1967, 1975a, 1987a, 1998a).

Though the NCHS life tables were based on Census data and birth and death registration data, no effort was made to adjust these data where they were believed to be incomplete (NCHS 1967, 1975a, 1987a, 1998a). Census coverage rates varied by state, race, and age and differed for the four time periods. The estimated net undercount generally declined from 1960 to 1990 and was 3.1 percent in 1960, 2.7 percent in 1970, 1.2 percent in 1980, and 1.8 percent in 1990 (Hogan and Robinson 1993, Robinson and West 2000). For all four Censuses the undercount for Blacks was 3.7 to 4.4 percent higher than for the other races combined. The net undercount was particularly large for Black children and adult men. Due to race and age compositions and other factors, Census coverage differed by state and region as well. The estimated 1990 Census undercount in the West and South Regions was higher than in the Northeast and Midwest Regions. The undercount was also higher for non-urban areas than urban areas (Hogan and Robinson 1993).

Evaluations of birth registration data have shown that under-registration, though less than one percent by the 1960s, was historically larger for the Black population than other populations. The death registration data used to create the NCHS life tables were largely complete also, but may have contained errors due to misreported age and race (Robinson and West 2000).

There are well-established historical differences in life expectancy for males and females and Whites and Nonwhites (see Appendices A-1 through A-8). For example, U.S. White male life expectancy at birth was 72.7 years in 1990, about seven years lower than the White female life expectancy of 79.5 years. The Nonwhite female life expectancy of 75.4 years was also substantially lower (about four years) than the White female life expectancy. While life expectancy differences between the sexes have generally increased, differences between the Black and White populations have decreased over the past several decades. In order to make comparisons of state mortality that are not biased by differences in race and sex distributions across states, comparisons must be made separately for White males, White females, Nonwhite males, and Nonwhite females.

NCHS did not compute state life expectancies for all four sex-race groups for every state because some states did not contain enough registered deaths to produce reliable and stable estimates of mortality rates for the three-year period4. For Nonwhites, life expectancy was calculated for 25 states in 1960 and 40 states in 1990. For Whites, life expectancy was computed for every state and for every year except Alaska and Hawaii in 1970. The 48 remaining states and the District of Columbia are referred to as the "selected states" for the White population in this paper.

Long-term patterns of state life expectancy can only be analyzed for Whites and Nonwhites because Black life expectancy was not available in the NCHS data for 1960 and 1970. However, life expectancy differences between states and changes in individual states over time are biased because there are different proportions of Blacks, Asian and Pacific Islanders, American Indians, and other groups included in the Nonwhite classification and because these populations have very different life expectancies. The proportion of the Nonwhite population who are Black decreased for most states from 1960 to 1990 and Asian and Pacific Islanders made up increasingly larger proportions of Nonwhites.

In order to evaluate at least some of the available data on Nonwhite life expectancy, states were selected for the analyses if at least 80 percent of the Nonwhite population was Black, as enumerated in each of the 1960 to 1990 decennial Censuses. Though this set of states is not entirely free of biases due to differential racial composition, it represents a fairly homogeneous population of Black males and females. The twenty states that met the selection criteria for all four time periods are Alabama, Arkansas, Delaware, District of Columbia, Florida, Georgia, Indiana, Kentucky, Louisiana, Maryland, Michigan, Mississippi, Missouri, North Carolina, Ohio, Pennsylvania, South Carolina, Tennessee, Virginia, and West Virginia. However, the number of Nonwhite deaths for at least one sex group in Delaware and West Virginia did not meet the NCHS standards to compute life tables for 1960 and 1970 and those two states were excluded from the analyses. The 18 remaining states are referred to as the "selected states" for the Nonwhite population in this paper.

The state Figures for the White population are also biased by combining the mortality experiences of both the Hispanic and non-Hispanic populations. However, there are no separate state-level data on these two populations from 1960 to 1990, so Hispanic and non-Hispanic effects cannot be separated in this evaluation.


IV. Methods

Three measures of variation are used to determine whether state life expectancies are becoming more or less similar to each other or the national average over time: 1) inter-quartile range; 2) coefficient of variation; and 3) Dispersion Index. The measures of variation are calculated for life expectancy at birth and at age 65 separately for White males, White females, Nonwhite males, and Nonwhite females for all the selected states combined and within Census Regions and the Southern Divisions. The inter-quartile range measures the difference between the life expectancies for states at the 25th and 75th percentiles. The coefficient of variation and the Dispersion Index are both summary measures of the variation in state life expectancy that can be compared over the four time periods.

The coefficient of variation is the standard deviation of the differences between state life expectancies and the mean life expectancy for the selected states divided by the mean life expectancy for the selected states (Blalock 1979, O’Connell 1981). The mean for the selected states is used in order to treat the life expectancies of all states equally, regardless of population size.

The Dispersion Index is an indicator of variation that does not rely on state differences from the average life expectancy. In situations where the mean life expectancy is dominated by one or a few outliers among the states, the Dispersion Index should provide a better measure of the overall spread of the data than the coefficient of variation. The Dispersion Index is based on the Root Mean-Squared Deviation (RMSD), which

is the square-root of mean-square deviation (MSD), a measure of variation that is computed independently of the mean. That is, unlike variance, MSD is computed by averaging the sum of squared differences among all distinct pairs of observations instead of averaging the sum of squared differences from the mean (Galal and Qureshi 1997: 1894).

The Dispersion Index is the ratio of the RMSD for each time period to the first time period of interest, 1960, designated as the base year. The index gives a relative measure of change in the RMSD, but does not produce results that are uniquely different from it.

The inter-quartile range, the coefficient of variation, and the Dispersion Index show the variation or dispersion in state life expectancy values change from 1960 to 1990. If the indicators decline in magnitude over time then state life expectancies generally converged toward the national average. If the indicators increase in magnitude over time, states became increasingly different from each other and the national average. There may be no consistent pattern across the four time periods, in which case the 1996 state population projection assumption that states maintain their mortality levels relative to the nation is probably the best alternative.

In addition to the summary measures, trends for individual states are assessed by changes in life expectancy ranks from 1960 to 1990 and comparisons of the linear fit or average amount of change in life expectancy. States are ranked in order of life expectancy values and how their positions shift over time roughly indicates convergence or divergence. Slopes generated from the linear fit of life expectancy over the four time periods are used to determine whether the average rate of change in life expectancy for individual states is higher or lower than the average rate of change in life expectancy for the United States.


V. Results

  1. Variation in White Life Expectancy

    Table 1 presents the inter-quartile range, coefficient of variation, and Dispersion Index, as calculated for the selected states, for life expectancy at birth and age 65 for White males and White females. There is no uniform pattern of convergence or divergence from one decade to the next. For White male life expectancy at birth, the inter-quartile range increases from 1.0 year in 1960 to 1.6 years in 1990. The Dispersion Index also shows some increasing variation from 1960 to 1990 with the exception of 1980. In contrast, the coefficient of variation remains relatively constant throughout the four time periods.

    For White female life expectancy at birth, there is little change in all three measures of variation across the four time periods. For example, the coefficient of variation only ranges from 0.009 to 0.010 from 1960 to 1990.

    For White male life expectancy at age 65, the summary measures decline in value from 1960 to 1980, suggesting a pattern of convergence, and increase in value from 1980 to 1990, suggesting divergence. A similar reversal in the trend is recorded by the Dispersion Index for White female life expectancy at age 65, but the inter-quartile range and coefficient of variation show convergence across the entire time period. One explanation for the apparent convergence of life expectancy levels from 1960 to 1980 and the divergence from 1980 to 1990 may be the increasing diversity of the White population in some states, such as the growing Hispanic population in the South and West. However, these findings are also consistent with a cyclical pattern of convergence and divergence.

    Though there are few differences over time in the summary measures of variation, the relative positions of individual states in the range of White life expectancies may have changed completely between 1960 and 1990 (see Section V.C. below).



  2. Variation in Nonwhite Life Expectancy

    Table 2 presents the inter-quartile range, coefficient of variation, and Dispersion Index for Nonwhite male and female life expectancy at birth and at age 65 for the 18 selected states. Similar to White male life expectancy at birth, the measures of variation for Nonwhite male life expectancy at birth do not show clear patterns of convergence or divergence. The inter-quartile range suggests convergence from 1970 (1.8 years) to 1990 (1.4 years) while the other two measures show declining variation from 1960 to 1980 with relatively large increases in variation from 1980 to 1990. The coefficient of variation and Dispersion Index for 1990 are larger than for 1960, suggesting that the convergence of state life expectancy between 1960 and 1980 had little lasting effect on the overall levels of differences in state mortality.

    The trends for Nonwhite female life expectancy at birth are similar to those for Nonwhite male life expectancy at birth except the increases in the measures of variation from 1980 to 1990 are somewhat smaller, suggesting a general trend toward convergence. For example, the Dispersion Index is 1.00 for 1960 (by definition), 0.50 for 1980, and 0.65 for 1990.

    For life expectancy at age 65, all three measures of variation for Nonwhite males and females generally show steady declines in variation from 1960 to 1990. For example, the coefficient of variation for Nonwhite males decreased from 0.046 for 1960 to 0.020 for 1990. The convergence hypothesis is most consistently supported by the results for the two Nonwhite groups.



  3. State Rankings of White Life Expectancy

    Given the mixed pattern of convergence and divergence described by the three measures of variation, a closer examination of the trends in life expectancy for individual states and regions is warranted. Tables 3 through 8 show the state rankings for life expectancy at birth and at age 65 by sex and race group. While there are some general regional patterns, there are also a number of exceptions to the patterns and the results do not support a single conclusion of convergence or divergence. It is important to note that due to the small differences in life expectancies, the precise order of state rankings should not be considered reliable.



    For White male life expectancy at birth (see Table 3):



    For White female life expectancy at birth (see Table 4):



    For White male life expectancy at age 65 (see Table 5):



    For White female life expectancy at age 65 (see Table 6):



  4. State Rankings of Nonwhite Life Expectancy

    Tables 7 and 8 show the state rankings for life expectancy at birth and at age 65 for the Nonwhite population in the selected states. As with White life expectancy, there are some general regional patterns, but there are also a number of exceptions to the patterns and the results do not support a single conclusion of convergence or divergence. With only 18 states selected for the analysis of Nonwhite life expectancy, the regional trends are better described by North-South and East-West differences.



    For Nonwhite male life expectancy at birth (Table 7):



    For Nonwhite female life expectancy at birth (Table 7):



    For Nonwhite male life expectancy at age 65 (Table 8):



    For Nonwhite female life expectancy at age 65 (Table 8):



  5. Linear Regression Estimates for State Life Expectancies

    Tables 9 through 12 summarize the state life expectancies for four time periods into a single measure of linear fit, or slope estimate. While the linear fit ignores some of the fluctuation in life expectancy by averaging the change over time, it allows for a systematic evaluation of relative change. A linear fit was calculated for each state and compared to the corresponding value for the nation. States that are identified as improving slower or faster than the U.S. may be converging toward or diverging from the national trends, depending on the actual levels of life expectancy. State slopes that are relatively close to the U.S. slopes are considered to be equivalent to the U.S. These are the states for which the population projection assumption regarding mortality, that states maintain their mortality levels relative to the nation over time, is most accurate. Slope values were calculated for life expectancy at birth and at age 65 for the four sex-race groups. The results are consistent with the general conclusions about the state rankings as discussed above.

    White male life expectancy at birth improved slowest in Missouri and all the states in the West South Central and East South Central divisions (see Table 9). States in the South with smaller than average changes also had lower than average life expectancies throughout all four time periods. This combination essentially describes divergence from the U.S. mean from 1960 to 1990. In contrast, many states in the Northeast and West had better than average improvements in White male life expectancy at birth. The states with lower life expectancies in 1960 converged with and sometimes exceeded the national average by 1980 or 1990. Most of the Midwest states experienced average changes in White male life expectancy at birth similar to the average change at the national level.

    The geographical distribution of states with slower increases in life expectancy at birth for White females is similar to that of White males. In addition to most states in the South, California, Kansas, and Nebraska had lower than average improvements in White female life expectancy at birth. More rapid increases in life expectancy at birth occurred for New Mexico, North Dakota, South Dakota, Washington, D.C. and most Northeastern states. Given their lower life expectancies in 1960 and their higher rates of change, some of these states converged toward the national average from 1960 to 1990 and some rose above the national average by 1990.

    As seen with the results for life expectancy at birth, improvement in White male life expectancy at age 65 was slower than average in many of the Southern states (see Table 10). Combined with some of the higher life expectancies in 1960, the slower increases in life expectancy produced convergence toward the U.S. average eventually followed by decline in relative life expectancy in the South, as shown by the changes in ranks above. Many of the Midwest states, such as Nebraska, Kansas, Indiana, and Missouri, also had relatively small average improvements in life expectancy at age 65. Several states in the Northeast and a few of the states in the West had larger improvements in hite male life expectancy at age 65 than the national average. The relatively large improvements in life expectancy in the Northeast generally produced convergence toward the U.S. mean.

    Several states in the South and West have smaller slopes for White female life expectancy at age 65, including Oregon, Nevada, California, Louisiana, Mississippi, Georgia, Oklahoma, Arkansas, and Texas. The states in the West generally converge toward the U.S. mean life expectancy for White females, while the states in the South moved toward the U.S. mean from 1960 to 1980 and then continued to fall in rank and diverged below the mean by 1990. The average change in life expectancy at age 65 for the Northeast and Midwest was higher than the change for the U.S. resulting in convergence toward the mean.

    In addition to the slopes for Nonwhite life expectancy for the entire U.S., slopes were calculated for the average of the 18 selected states, weighted by their corresponding Census populations, to provide more meaningful comparisons. Among the states selected to represent the Black population, Nonwhite male life expectancy at birth rose slower in Washington, D.C., Pennsylvania, and the Midwest (see Table 11). The more western of the Southern states, Arkansas, Louisiana, and Mississippi, also experienced smaller average changes than the national average. These findings are consistent with the changes in ranks discussed above and with a convergence toward the national average because these states had relatively higher life expectancies in 1960. The life expectancy in Kentucky and the South Atlantic Division states increased faster than the U.S. average. These states all had lower life expectancies in 1960 and thus converged toward the U.S. mean and some states even surpassed the mean U.S. life expectancy at birth by 1990. Over the period of 1960 to 1990, Pennsylvania and the Midwest states reversed positions with the South Atlantic Division and the life expectancy differentials, as indicated by the measures of variation discussed above (Section V.B.), actually increased from 1980 to 1990.

    The geographic distribution of states with slowly increasing life expectancy at birth for Nonwhite females closely resembles that of Nonwhite males. Most of the Midwest and more western of the Southern states had life expectancies with smaller than average changes than the nation. The Nonwhite female life expectancy for Maryland, Virginia, and South Carolina increased faster than the national average. As with Nonwhite male life expectancy at birth, the measures of dispersion for Nonwhite females indicate convergence from 1960 to 1980 and divergence between 1980 and 1990.

    The trends in Nonwhite male life expectancy at age 65 are very different from the corresponding trends for life expectancy at birth (see Table 12). Most of the Northern half of the country, including Virginia, Kentucky, and Missouri, had life expectancies at age 65 that improved faster than the Southern half of the country. By 1990, only Arkansas, Florida, and Mississippi had above average life expectancy at age 65 along with the majority of the Northern states. The combination of smaller improvements in life expectancy for Southern states that had above average life expectancy in 1960 and larger improvements for Northern states that had below average life expectancy in 1960 is consistent with the measures of variation that indicate convergence of Nonwhite male life expectancy at age 65 from 1960 to 1990 (Section V.B.).

    Life expectancy at age 65 for Nonwhite females improved more slowly in the southernmost states of Arkansas, Louisiana, Florida, Georgia, and South Carolina than the average for the states selected for this analysis. More rapid advances in life expectancy for Nonwhite females occurred for Kentucky, North Carolina, Virginia, Maryland, and Washington, D.C. As with Nonwhite males, life expectancy at age 65 in these states generally converged toward the Nonwhite female national average from 1960 to 1990.



  6. Variation in Life Expectancy by Regions

    The coefficient of variation was calculated by Census Regions or Divisions because the measures of variation as calculated for the entire U.S. do not capture the regional patterns identified by examining life expectancy changes for individual states. This measure was selected because it is largely consistent with the Dispersion Indices calculated for all states and because the inter-quartile range would lose meaning if applied to only a few observations. The coefficient of variation includes the number of observations in its formula and there are about twice as many states in the South as in the other Census Regions. Therefore, separate calculations were produced for the South Atlantic Division and the combined states of the East South Central and West South Central Divisions. For some Nonwhite statistics, Pennsylvania, the only Northeast state of the selected group, was included with the states in the Midwest.

    Table 13 shows the coefficients of variation for White life expectancy at birth and at age 65 by Census Regions and Southern Divisions. White male and female life expectancy at birth for states in the Midwest show no measurable change in variation from 1970 to 1990. These states did not experience large changes in rank relative to other states and had average changes between the four time periods that were similar to those for the nation as a whole. Figure 2 shows how the life expectancy changes in the Midwest states for White males roughly parallel each other and the changes for the U.S.

    The coefficients of variation for states in the South Central Division decline in value for both White male and female life expectancy at birth, suggesting some convergence of life expectancy trends from 1960 to 1990. The coefficients of variation fluctuate over time for the Northeast, West, and South Atlantic Division. With relatively small ranges of values within these regions, it could be argued that the levels of variation were about the same over the four time periods.

    The coefficients of variation for White male and female life expectancy at age 65 for the Midwest varied more than the corresponding values for life expectancy at birth (Table 13). However, the values fluctuate and there is no consistent pattern across time. In general, life expectancy at age 65 for White males in the Northeast tended to diverge, while life expectancy for White females tended to converge from 1960 to 1990.

    The variation in White male life expectancy among states in the West Region and South Atlantic and South Central Divisions declined from 1960 to 1980, but increased from 1980 to 1990. This also occurred for White female life expectancy at birth for the South Atlantic Division. There is no consistent pattern in the coefficient of variation for White females in the West, but the values consistently decrease over the four time periods in the South Central Division, indicating convergence within the region. Figure 3 shows the narrower range of life expectancy at age 65 for White females in the South Central states in 1990 compared to 1960.

    Table 14 shows the coefficients of variation for Nonwhite male and female life expectancy at birth and at age 65 by Census Region and the Southern Division. For Nonwhite male life expectancy at birth, the coefficients of variation show some convergence or decreasing variation from 1960 to 1980 for the Midwest, the Midwest and Northeast combined, the South, and the South Central Division. All the coefficients of variation for 1990, however, are larger than the coefficients of variation for 1980, suggesting a reversal of the convergence within regions. This also occurred within all regions for Nonwhite female life expectancy at birth except for the South Central Division where the coefficient of variation shows convergence across all four time periods. Figure 4 traces the declining variation in Nonwhite female life expectancy at birth for the South Central Division.

    The states in the South Atlantic Division had increasing variation from 1970 to 1990 for Nonwhite male life expectancy at birth. This result is largely influenced by the decline in life expectancy for Washington, D.C. from 1980 to 1990. The range of values for Nonwhite female life expectancy for the South Atlantic Division is much smaller than for males and does not indicate a pattern of increasing or decreasing variation.

    The coefficient of variation for both Nonwhite male and Nonwhite female life expectancy at age 65 indicates consistent trends toward convergence within all regions. Many of the values decline by more than half from 1960 to 1990. An example of the dramatic decreases in variation for Nonwhite females in the South Atlantic Division is shown in Figure 5. As found with the measures of variation for all of the selected states combined (see Section V.B.), the evidence for convergence of Nonwhite life expectancy at age 65 is the most consistent in this research.


VI. Summary and Conclusions

This paper assesses trends in the relative changes in state life expectancy from 1960 to 1990 for White males, White females, Nonwhite males, and Nonwhite females. Three measures of variation, the inter-quartile range, Dispersion Index, and coefficient of variation, generally show that state life expectancy at birth and at age 65 converged toward the national average from 1960 to 1980 for all four sex-race groups. However, in many cases there was a reversal of the convergence trend between 1980 and 1990 where the measures indicate that the variation in state life expectancy at birth for all four sex-race groups increased. In some cases the measures of variation for 1990 are as large as or larger than for 1960. The trend reversal also occurred for White male and female life expectancy at age 65, but the 1960 to 1980 decreases in variation continued through 1990 for Nonwhite male and female life expectancy at age 65.

Though the measures of variation do not yield a single conclusion about whether life expectancy differences among states became smaller or larger from 1960 to 1990, there are some fairly consistent regional patterns. States in the Northeast experienced relatively large improvements in White life expectancy over the four time periods. Western states also recorded relatively large increases in life expectancy compared with other states, but only for White males. States in the South generally had smaller life expectancy gains among Whites and life expectancy changes for states in the Midwest kept pace with the national average.

With sufficient data on Nonwhite life expectancy available for only 18 states, the analysis of the regional trends from 1960 to 1990 for Nonwhites is more limited. The changes in life expectancy at birth were generally smaller in the Midwest and the more western of the Southern states while the changes were larger in the South Atlantic Division. In contrast, the pace of the change in life expectancy at age 65 was higher for the northernmost of the Southern states than that for the Midwest and the other Southern states.

Despite the trends identified by this research, there are several reasons why it would be difficult to justify projecting mortality rates that differ by state or region. First, there are exceptions to the patterns for every region and some of those exceptions are states with large populations, such as California and New York. Second, there are insufficient data for the Nonwhite population for much of the period from 1960 to 1990, so there is little basis for projecting different changes in Nonwhite mortality for the majority of the states. Third, the state life expectancies for the White population apply to Hispanics and non-Hispanics and the state life expectancies for the Nonwhite population apply to Blacks, Asian and Pacific Islanders, American Indians, and other Nonwhite groups. These populations may have very different life expectancies and may experience different changes in life expectancy over time. Consequently, the trends identified in this paper are biased to the degree that population distributions by race and Hispanic origin differ across states.

Finally, many of the measures of variation for both the nation and regions indicate convergence from 1960 to 1980 and trend reversals from 1980 to 1990. These findings offer little support for the hypothesis that the levels of state mortality become more similar to each other over time. These results may be interpreted as a cyclical pattern with a peak in life expectancy convergence around 1980, but predicting the degree and timing of state mortality changes based on only part of a cycle is not advisable. Updated state life tables for 1999-2001 may clarify the trends, but NCHS life tables will not be available for some time.

Attempting to incorporate the trends described in this analysis into mortality projections for individual states would require broad assumptions and would likely introduce serious flaws into the results. Lacking a better alternative, the assumption that the levels of state mortality relative to national mortality are constant is the most defensible for state population projections produced in the near future.


1All references to "states" include all 50 states of the U.S. and the District of Columbia, except where noted.

2Census Regions and Divisions are: Northeast Region, New England Division - Maine, New Hampshire, Vermont, Massachusetts, Rhode Island, Connecticut; Northeast Region, Middle Atlantic Division - New York, New Jersey, Pennsylvania; Midwest Region, East North Central Division - Ohio, Indiana, Illinois, Michigan, Wisconsin; Midwest Region, West North Central Division - Minnesota, Iowa, Missouri, North Dakota, South Dakota, Nebraska, Kansas; South Region, South Atlantic Division - Delaware, Maryland, District of Columbia, Virginia, West Virginia, North Carolina, South Carolina, Georgia, Florida; South Region, East South Central Division - Kentucky, Tennessee, Alabama, Mississippi; South Region, West South Central Division - Arkansas, Louisiana, Oklahoma, Texas; West Region, Mountain Division - Montana, Idaho, Wyoming, Colorado, New Mexico, Arizona, Utah, Nevada; West Region, Pacific Division - Washington, Oregon, California, Alaska, Hawaii.

3Decennial years refer to the three-year period for which the life expectancies are calculated: 1960 (1959-1961); 1970 (1969-1971); 1980 (1979-1981); and 1990 (1989-1991).

4The number of deaths required for the standards of accuracy was 2000 in 1960, 1600 in 1970, and 700 in 1980 and 1990.

5If included, Hawaii would rank first in 1980 and 1990.


References

Blalock, Jr., Hubert M. 1979. Social Statistics. 2nd edition. New York: McGraw-Hill, Inc.

Campbell, Paul R. 1996. "Population Projections for States by Age, Sex, Race, and Hispanic Origin: 1995 - 2025." PPL-47. U.S. Census Bureau, Population Division.

Campbell, Paul R. 1997. "Population Projections: States, 1995 - 2025." Current Population Reports, P25-1131. U.S. Census Bureau, Population Division.

Clifford, William B. and Yevonne S. Brannon. 1985. "Rural-Urban Differentials in Mortality." Rural Sociology. 50(2): 210-224.

Coale, Ansley J. 1996. "Age Patterns and Time Sequence of Mortality in National Populations with the Highest Expectation of Life at Birth." Population and Development Review. 22(1): 127-135.

Edmondson, Brad. 1994. "Mind if I Breathe?" American Demographics. 16: 60.

Galal, Osman M.. and Azhar K. Qureshi. 1997. "Dispersion Index: Measuring trend Assessment of Geographical Inequality in Health - The Example of Under-Five Mortality in the Middle East/North African Region, 1980-1994." Social Science and Medicine. 44(12): 1893-1902.

Gwatkin, Davidson R. 1980. "Indications of Change in Developing Country Mortality Trends: The End of an Era?" Population and Development Review. 6(4): 615-644.

Halli, Shiva S. and K. Vaninadha Rao. 1992. Advanced Techniques of Population Analysis. New York: Plenum Press.

Heaton, Laura. 1998. "Fertility in the United States: Is State-Level Fertility Converging to the National Level?" U.S. Census Bureau unpublished research.

Hogan, Howard and Gregg Robinson. 1993. "What the Census Bureau’s Coverage Evaluation Programs Tell Us About Differential Undercount." Paper presented at the U.S Census Bureau 1993 Research Conference on Undercounted Ethnic Populations, May 5-7, 1993.

Lochner, Kim, Elsie Pamik, Diane Makuc, Bruce P. Kennedy, and Ichiro Kawachi. 2001. "State-Level Income Inequality and Individual Mortality Risk: A Prospective, Multilevel Study." American Journal of Public Health. 91(3): 385-391.

McNutt, Louise-Anne, David S. Strogatz, F. Bruce Coles, and Laura J. Fehrs. 1994. "Is the High Ischemic Heart Disease Mortality Rate in New York State Just an Urban Effect?" Public Health Reports. 109(4): 567-570.

National Center for Health Statistics. 1967. Methodology of the National, Regional, and State Life Tables for the United States: 1959-61. Washington, DC: Department of Health, Educations, and Welfare.

National Center for Health Statistics. 1975a. Methodology of the National and State Life Tables for the United States: 1969-71. Rockville, Maryland: Department of Health, Educations, and Welfare.

National Center for Health Statistics. 1975b. Some Trends and Comparisons of United States Life-Table Data: 1900-1971. Rockville, Maryland: Department of Health, Educations, and Welfare.

National Center for Health Statistics. 1985-1986. State Life Tables, Alabama-Wyoming, U.S. Decennial Life Tables for 1979-81. Vol. II, Nos. 1-51. Hyattsville, Maryland: Department of Health and Human Services.

National Center for Health Statistics. 1987a. Methodology of the National and State Life Tables: 1979-81. Hyattsville, Maryland: Department of Health and Human Services.

National Center for Health Statistics. 1987b. Some Trends and Comparisons of United States Life-Table Data: 1900-1981. Hyattsville, Maryland: Department of Health and Human Services.

National Center for Health Statistics. 1998a. Methodology of the National and State Life Tables: 1989-91. Hyattsville, Maryland: Department of Health and Human Services.

National Center for Health Statistics. 1998b. U.S. Decennial Life Tables for 1989-91, State Life Tables. Vol. II, Nos. 1-51. Hyattsville, Maryland: Department of Health and Human Services.

National Center for Health Statistics. 1999. Some Trends and Comparisons of United States Life-Table Data: 1900-1991. Hyattsville, Maryland: Department of Health and Human Services.

O’Connell, Martin. 1981. "Regional Fertility Pattern in the United States: Convergence or Divergence?" International Regional Science Review. 6(1): 1-14.

Pickle, Linda W., Michael Mungiole, Gretchen K. Jones, and Andrew A. White. 1996. Atlas of United States Mortality. Hyattsville, MD:DHHS Publication.

Pommerenke, Forrest A., Robert W. Miller, Sudhir Sirvastava, and Susan P. Ackerman. 1994. "Targeting Cancer Control: The State Cancer Control Map and Data Program." American Journal of Public Health. 84(9): 1479-1482.

Robinson, J. Gregory and Kristen K. West. 2000. "Demographic Analysis." Encyclopedia of the U.S. Census, p. 168-168. Margo J. Anderson, editor. Congressional Quarterly Press: Washington, DC.

Schneider, Dona, Michael R. Greenberg, and Lisa Li Lu. 1997. "Region of Birth and Mortality from Circulatory Disease among Black Americans." American Journal of Public Health. 87(5): 800-804.

Shryock, Henry S. and Jacob S. Siegel. 1976. The Methods and Materials of Demography. Orlando, Florida: Academic Press, Inc.

White, Kevin M. 2002. "Longevity Advances in High-Income Countries, 1955-96." Population and Development Review. 28(1): 59-76.

Wilmoth, John R. 1998. "Is the Pace of Japanese Mortality Decline Converging Toward International Trends?" Population and Development Review. 24(3): 593-600.

Wilson, Chris. 2001. "On the Scale of Global Demographic Convergence 1950 - 2000." Population and Development Review. 27(1): 155-171.

Yeager, Kimberly K., Robert F Anda, Caroline A. Macera, Ralph S. Donehoo, and Elaine D. Eaker. 1995. "Sedentary Lifestyle and State Variation in Coronary Heart Disease Mortality." Public Health Reports. 110(1): 100-102.


[PDF] or PDF denotes a file in Adobe’s Portable Document Format. To view the file, you will need the Adobe® Reader® Off Site available free from Adobe.