Mortality in the United States: Historical Estimates and Projections

Mortality in the United States: Historical Estimates and Projections

COVID-19 Demographics

Mortality in the United States: Historical Estimates and Projections

Duncan Haystead · · 162 min read
Mortality in the United States: Historical Estimates and Projections

Improvements in U.S. life expectancy stalled in the 2010s, years before the COVID-19 pandemic produced a spike in mortality. We present new estimates of historical patterns in mortality by socio-demographic group and projections of U.S. mortality through 2060.


Key Points

  • We find mortality improvement has slowed across the population, with substantial heterogeneity across socio-demographic groups. Notably, working age mortality among high-school graduates rose by around 16% from 1996 to 2019 while working age White mortality had almost no net improvement over the period and rose by a little under 10% from 2010 to 2019. Meanwhile, working age Black and Hispanic mortality fell by nearly 25% and 20%, respectively, from 1996 to 2010, before stagnating.

  • We estimate that the COVID-19 pandemic increased overall mortality by around 20 percent in 2020 and 2021, with around a 40% increase in mortality among Hispanics adults and an over 25% increase in mortality among working age adults without any college education.

  • We project life expectancy at birth to increase to 82.2 years by 2060. Life expectancy for White and low-educated working-age adults is expected to improve over the next 10 years but stagnate long-term. Mortality among elderly adults is expected to continue to improve, but at a slower rate than in the second half of the 20th century.

Background

Across high-income countries, the 20th century was marked by substantial reductions in mortality and increases in life expectancy. Yet at the end of the first quarter of the 21st century, it is clear the trends of the 20th no longer prevail. The rate of improvement in all-cause mortality has fallen, reflecting a slowdown in the pace of medical and technological innovation that drove mortality improvement in the late 20th century. The shift has been especially sharp in the U.S., where progress on overall mortality stalled around 2010 and then reversed for several years in the mid-2010s as life expectancy declined. The slowdown has not been uniform across socio-demographic groups, however, and countervailing trends by age, race, and educational attainment have emerged in recent decades. Characterizing the recent path of U.S. mortality therefore requires a granular level of socio-demographic detail.

In this brief, we present PWBM’s estimates of mortality trends in recent decades and projections out to 2060. Because official mortality estimates provide very limited demographic detail and are not fully comparable over time, PWBM developed its own estimates of historical mortality rates. Drawing on the complete universe of individual death records in the U.S., we estimate annual mortality rates by single year of age, sex, race, educational attainment, and marital status since 1996.

Mortality rates at this level of socio-demographic detail cannot be calculated directly due to a lack of high quality data on population (the denominator in the mortality rate calculation) and extreme variability in measured rates for such small populations. To address these issues, we model an “underlying” mortality profile for each socio-demographic group, which evolves smoothly across age and time. For each year and age, the observed mortality rate reflects the stochastic nature of the underlying mortality process. The corresponding estimated underlying mortality rate is the expected value of the mortality process. We treat the COVID-19 pandemic as a unique and persistent deviation from underlying mortality and account for it separately from the main mortality process.

Historical Patterns in Mortality from 1996 to 2019

Recent trends in overall U.S. mortality are obscured by the aging of the population, which raises aggregate mortality even if mortality rates at a given age are unchanged. To distinguish trends in age-specific mortality from the effects of population again, we calculate adjusted mortality rates that hold the age and sex distribution of the population constant over time. Unless otherwise noted, all mortality rates presented in the section are age-sex adjusted.

Figure 1 compares the unadjusted (dashed line) and age-sex adjusted (solid line) underlying all-cause mortality rates. We compute adjusted rates using the composition of the population in 2010 so the two are equal in that year.1

The age-sex adjusted rate – which removes the impact of aging – fell throughout the 1996-2019 period. However, the rate of decline slowed significantly after 2010: in the decade and a half before 2010, adjusted mortality fell 1.3 percent annually on average; in the decade after 2010, it fell at a rate of less than 0.7 percent per year.

Figure 1: Age-Sex Adjusted and Unadjusted Mortality Rate (Ages 0-95)

Notes: The age-sex adjusted mortality rate is based on the age and sex composition of the population in 2010. The unadjusted mortality rate is based on the actual population in each year.
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

The slowdown in overall mortality improvement around 2010 was not uniform across the population. Mortality increased for some socio-demographic groups, stagnated or slowed for others, while some experienced no meaningful change. Drawing on our detailed group estimates of underlying mortality, we document the divergent trends by age, sex, race, and education that produced the overall slowdown.

Mortality by sex and age: Figure 2 plots all-cause mortality rates by sex for three age groups: young (under 25), working age (25-64), and elderly (65 and over).2 Note that in this figure and in others in this brief, the y-axis scale is different in each panel.

Mortality among working-age adults declined steadily in the 1990s and 2000s, falling slightly faster for men than women, but this trend reversed in the early 2010s and working-age mortality began rising for both men and women. The increase slowed in the late 2010s but working-age mortality rates have stagnated since.

Figure 2: Age-Sex Adjusted Mortality Rate by Sex and Age Group

Figure 2
Figure 2

Notes: The age-sex adjusted mortality rate is based on the age and sex composition of the population in 2010.
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

Among the elderly, all-cause mortality improved throughout 1996 to 2019, though some deceleration is evident over the course of the 2010s. Mortality fell significantly faster for elderly men over this period; the mortality rate gap between men and women 65 and over shrunk from 27 percent in 1996 to 15 percent in 2019.

The slowdown in overall mortality improvement seen in Figure 1 is the product of these two trends: the rise in working-age mortality and the slower rate of improvement in elderly mortality after 2010.

Mortality by race: Figure 3 compares underlying mortality among adults (ages 25 and over) by race and age group.3 Trends in working-age mortality worsened after 2010 across all race groups, but to different degrees. Among White working-age adults – who make up the largest share of the working-age population – mortality began rising in the early 2010s, accounting for much of the increase in overall working-age mortality seen in Figure 2. At the same time, working-age Black mortality stagnated after 2010, ending decades of steady improvement from very high levels, while working-age Asian and Hispanic mortality shifted from modestly declining to stagnant. Among working-age American Indians and Alaska Natives (AIAN), mortality was rising before 2010 and accelerated after. Working-age AIAN mortality rose nearly one third from 1996 to 2019, but because the AIAN share of the population is small this dramatic rise only had a small effect on overall mortality.

Figure 3: Age-Sex Adjusted Mortality Rate by Race and Hispanic Origin and Age Group

Figure 3

Notes: The age-sex adjusted mortality rate is based on the age and sex composition of the population in 2010. Race and Hispanic origin groups are single race, non-Hispanic or any race, Hispanic origin.
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

All-cause mortality among the elderly improved from 1996 to 2019 at similar rates across race groups. While some deceleration is apparent in the late 2010s, trends in elderly mortality by race generally continued uninterrupted throughout this period.

Mortality by educational attainment: Figure 4 plots the evolution of underlying mortality by educational attainment. Among both working-age and elderly adults, mortality is substantially lower for those with educational attainment above a high school degree. However, trends over the 1996-2019 period have a more complex relationship to education: working-age mortality declined among the most educated and the least educated, while rising for those in between. Among the elderly, mortality rates declined regardless of educational attainment, with non-high school graduate mortality declining at the fastest rate.

Figure 4: Age-Sex Adjusted Mortality Rate by Education and Age Group

Figure 4

Notes: The age-sex adjusted mortality rate is based on the age and sex composition of the population in 2010.
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

The relationship between educational attainment and changes in working-age mortality is largely a product of diverging trends by race within education groups. Figure 5 shows working-age mortality by race at different levels of educational attainment. The mortality decline among working-age adults without a high school degree was driven both by rapidly falling Black mortality in the 2000s and by a rise in the Hispanic or Asian share of this population – groups that have substantially lower mortality rates. These trends more than offset the rise in White and AIAN mortality among the least educated.

Figure 5: Age-Sex Adjusted Working Age Mortality Rate by Race and Hispanic Origin and Educational Attainment

Figure 5
Figure 5
Figure 5

Notes: The age-sex adjusted mortality rate is based on the age and sex composition of the population in 2010. Race and Hispanic origin groups are single race, non-Hispanic or any race, Hispanic origin.
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

Across all race groups, working-age mortality for those with a bachelor’s degree or above generally declined throughout the period, while mortality among high school graduates or those with some college increased after 2010.4 Race and education groups with worsening mortality pre-2010 generally had accelerating increases in mortality rates over the 2010s; conversely, those with decreasing or stagnant mortality were generally more robust to the influence of this period.

Excess Mortality During the COVID-19 Pandemic

Mortality increased sharply in 2020 with the emergence of the COVID-19 pandemic, with continuing effects in the years since. To separate the impact of COVID-19 from trends in underlying mortality, we estimate “excess” deaths associated with the pandemic. Excess all-cause mortality is a more complete measure of the pandemic’s impact than reported deaths attributed to COVID-19. It describes the deviation of actual all-cause mortality from expected all-cause mortality, which is based on recent trends. Excess mortality captures COVID-19 deaths that were misattributed to a different cause of death (due to a lack of testing or inconsistent reporting practices) as well as deaths from other causes that are indirectly attributable to the pandemic. We estimate excess deaths by detailed socio-demographic group.

Figure 6: Cumulative Monthly Excess Deaths, March 2020-December 2022

Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

Figure 6 plots our estimates of total monthly excess all-cause deaths from March 2020 through December 2022. Table 1 compares estimates of excess and expected deaths for different subperiods since March 2020. Over the full period, we estimate nearly 1.4 million excess deaths occurred. Around 1.2 million of these deaths occurred in the first two years of the pandemic, with the first year (March 2020-February 2021) containing the greatest number of excess deaths.

Table 1: Excess & Expected Deaths and Relative Excess by Period

Excess Deaths Expected Deaths Relative Excess
March 2020-December 2022 1,392,410 8,193,541 -
March 2020-February 2021 612,132 2,905,511 21.1%
March 2021-February 2022 561,332 2,896,139 19.4%
March 2022-December 2022 218,945 2,391,892 9.2%

Source: Penn Wharton Budget Model

Previous research has highlighted wide disparities in the pandemic’s impact on different socio-demographic groups. To compare the excess mortality across the population, we calculate the percentage relative excess deaths (sometimes called the “P-score”), defined as the ratio of excess deaths to expected deaths multiplied by one hundred. We refer to this measure as the “relative excess.” Groups with higher relative excess experienced a greater rise in mortality relative to pre-pandemic trends.

To assess the relative “protective effect” of socio-demographic characteristics during the first two years of the COVID-19 pandemic, we compare relative excess by age group and education in Figure 7 and relative excess by race and education in Figure 8. Across age and race groups, the mortality impact of COVID-19 was greatest among adults without a secondary education. Higher educational attainment was correlated with lower excess mortality, especially among younger adults and Hispanics.

Figure 7: Relative Excess All-Cause Mortality by Educational Attainment and Age Group, March 2020-February 2022

Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

In general, relative excess was lower at older ages, as shown in Figure 7. While the elderly were more likely than working-age adults to die from COVID-19, the relative increase in mortality risk was smaller for the oldest adults given their high baseline mortality rates.

Controlling for educational attainment, relative excess mortality by race was lowest for non-Hispanic White adults. Figure 8 shows that relative excess was substantially higher among Hispanic and AIAN adults and somewhat higher among non-Hispanic Black adults.

Figure 8: Relative Excess All-Cause Mortality by Educational Attainment and Race and Hispanic Origin (Age 25 and older), March 2020-February 2022

Notes: Race and Hispanic origin groups are single race, non-Hispanic or any race, Hispanic origin.
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

Projected Mortality through 2060

PWBM projects mortality by socio-demographic group based on underlying trends from the mid-1990s to 2022, after removing the effects of COVID-19. We model mortality as a combination of a long-term process reflecting trends over the entire period and a short-term, transitory process that accounts for the increase in mortality in the 2010s. We project the transitory component so that it has no permanent effect on long-term mortality levels. We then incorporate an additional permanent effect accounting for the expected long-term influence of endemic COVID-19.

PWBM projects that life expectancy at birth will rise to 82.2 years by 2060, as shown in Figure 9. Life expectancy is projected to recover quickly from pandemic-era decline and improve at an accelerated rate through 2030, reflecting the end of the transitory slowdown that began around 2010. Life expectancy continues to rise after 2030 but at a more modest rate of improvement. As we discuss in more detail below, our projections are comparable to official projections from government agencies (see Table 3).

Figure 9: Projected Period Life-Expectancy at Birth

Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

The life expectancy projection in Figure 9 reflects PWBM’s projections of the population, in addition to projected trends in mortality. We project mortality rates by socio-demographic group, which means the presentation of forecasts at any aggregated level is sensitive to the relative sizes of these groups. Shifts in population composition over time can obscure the underlying trajectories of socio-demographic group-specific rates.

To foreground projected changes in group-specific mortality rates, we compute aggregate mortality rates based on a fixed socio-demographic population composition. For comparability between historical and projected mortality rates, we use the socio-demographic composition of the population in 2019, the latest historical estimate prior to the pandemic. Unless otherwise noted, all mortality rates presented below are adjusted to the distribution of the population in 2019 by age, sex, race, education, and marital status.

Figure 10: Historical and Projected Mortality Rate (Ages 0-95), Adjusted for Population Composition

Notes: The population composition-adjusted mortality rate is based on the estimates of 2019 population by age, sex, educational attainment, race, and marital status
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

The composition-adjusted overall mortality rate, shown in Figure 10, is projected to continue its long-term decline at a modest rate over the next several decades. The accelerated decline in the 2020s reflects the projected end of the transitory mortality process. After this, the projection is driven entirely by the estimated long-term component for each socio-demographic group. Generally, the long-term forecast reflects the patterns of the fitting period: stagnant working-age mortality and modest improvement among the elderly.

Figure 11: Historical and Projected Mortality Rate by Sex, Adjusted for Population Composition

Notes: The population composition-adjusted mortality rate is based on the estimates of 2019 population by age, sex, educational attainment, race, and marital status
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

Projected mortality by sex and age: Our projections imply a slow closing of the mortality gap between men and women, as men’s mortality rate improves at a faster rate than women’s. Figure 11 plots composition-adjusted mortality by sex. We estimate an overall reduction in men’s all-cause mortality of 11% by 2060 relative to 2019, compared with only a 5% reduction in women’s mortality. The expected end of the short-term process in the 2020s leads to more rapid mortality improvement for men than women, undoing the transitory increase which saw mortality rise faster among men than women during the 2010s (see Figure 2).

Figure 12 Historical and Projected Mortality Rate by Sex and Age Group, Adjusted for Population Composition

Figure 12

Notes: The population composition-adjusted mortality rate is based on the estimates of 2019 population by age, sex, educational attainment, race, and marital status.
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

Most of the projected improvement in working-age mortality over the next 35 years is expected to come from the recovery of gains lost during the 2010s. Over the long-term, working-age mortality improvement levels off among men and women and composition-adjusted mortality rates stagnate, as shown in the left panel of Figure 12. By contrast, elderly mortality continues to improve in both the near- and long-term. This agrees with historical trends during our fitting period, which saw mortality improvements stagnating among working-age adults while continuing among the elderly (see Figure 2).

Figure 13: Historical and Projected Mortality Rate by Education and Age Group, Adjusted for Population Composition

Figure 13

Notes: The population composition-adjusted mortality rate is based on the estimates of 2019 population by age, sex, educational attainment, race, and marital status.
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

Projected mortality by educational attainment: Figure 13 plots composition-adjusted mortality by educational attainment for working-age adults and the elderly. Among working-age adults, we project that mortality will continue improving at a modest rate for the most educated and the least educated, while continuing to rise for those in between.

Compared with other working-age adults, those with at least a bachelor’s degree experienced less of a slowdown in mortality improvement prior to the pandemic, and we expect their long-running trend to continue. Mortality also continues falling for working-age adults without a high school degree. This group increasingly comprises Hispanic and Asian immigrants, whose mortality rates are considerably lower than those U.S.-born adults who did not complete high school (see Figure 5, which shows mortality rates for all working-age Hispanics and Asians without a high school degree; mortality among immigrants tends to be even lower).

Overall mortality among those with just a high school degree or Associate’s degree is projected to continue rising in the coming decades, primarily due to steadily rising long-term White mortality within this group (see Figure 14 and its discussion).

Across education levels, those aged 65 and over are projected to experience continued mortality improvement, with the greatest gains concentrated among the elderly with no high school degree. As in the working-age population, gains among the least educated are driven by projected declines in corresponding Hispanic and Asian mortality.

Figure 14: Historical and Projected Mortality Rate by Race and Age Group, Adjusted for Population Composition

Figure 14

Notes: The population composition-adjusted mortality rate is based on the estimates of 2019 population by age, sex, educational attainment, race, and marital status. Notes: Race and Hispanic origin groups are single race, non-Hispanic or any race, Hispanic origin.
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data.

Projected mortality by race: Figure 14 shows the projected evolution of underlying working-age and elderly mortality rates by race. Despite a slight projected improvement initially – resulting from the recovery of 2010s-era losses – working-age White mortality continues to rise over the long-term as smaller improvements among the college-educated are outpaced by moderate mortality increases among less-educated counterparts. Working-age Black mortality is projected to continue its considerable rate of recent improvement, leading to a reversal of the historical disparity with White mortality for the first time by the mid-2050s (conditional on the 2019 population composition to which these rates are adjusted). Working-age Hispanic and Asian mortality rates improve slightly from already low levels. We project that working-age AIAN mortality, which rose steadily throughout the historical period, levels off but remains elevated.
Comparison with Official Projections

Our projections of life expectancy at birth in 2024 and 2054 are similar to projections from government agencies.5 Table 2 compares our projections with those of the Congressional Budget Office (CBO) and Social Security Administration (SSA), showing that our estimates are closest to SSA’s. We attribute our slightly more conservative results to projected rises in middle-aged mortality among those with low educational attainment, which down-weights improvements in survival probability among the elderly.

Table 2: PWBM Forecasted Life Expectancy at Birth, Compared with CBO and SSA

Year PWBM CBO (Official) SSA (Official)
2019 78.4 - -
2024 78.9 78.4 78.9
2054 81.8 82.2 81.9

Source: Congressional Budget Office, Penn Wharton Budget Model

Compared with SSA’s ‘intermediate forecast’, we estimate similar male life expectancy at birth and more conservative female life expectancy at birth, as shown in Table 3. Our more conservative estimate of female life expectancy over the long-term reflects the slow rate of improvement in female mortality over our relatively recent fitting period.

Table 3: PWBM Forecasted Life Expectancy at Birth by Sex, Compared with SSA

Year PWBM Male (at birth) PWBM Female (at birth) SSA Male (Intermediate) SSA Female (Intermediate)
2024 76.4 81.6 76.4 81.3
2060 80.4 84.0 80.3 84.6

Source: Social Security Administration, Penn Wharton Budget Model

CBO projects the change in mortality after 2025 using the average rate of decline from 1950 to 2019, assuming an idiosyncratic period effect of 2010s that continues until 2025. The influence of the longer fitting period can be seen in Figure 15, which compares PWBM and CBO projections through 2060 in terms of age-sex adjusted mortality (based on the 2019 population). PWBM’s choice of a shorter fitting period (beginning in the mid-1990s versus 1950) reflects the mounting evidence that US mortality experience has entered a new era. Substantial portions of the working-age population have seen increases in mortality sustained since at least the late 1990s and there have been important divergences in trends by race and education in recent decades. Our projection includes these effects but excludes the more rapid preceding 20th century declines, which often derived from medical or technological advances that have slowed or one-time public health improvements. As a result, we obtain a more conservative forecast of mortality than CBO.

Figure 15: PWBM Historical and PWBM and CBO Projected Mortality Rate, Adjusted for Age-Sex

Notes: The age-sex adjusted mortality rate is based on the age and sex composition of the population in 2019.
Source: Penn Wharton Budget Model estimates from National Vital Statistics System and Census Bureau data and Congressional Budget Office Demographic Projections



This analysis was produced by Duncan Haystead under the direction of Alex Arnon and the faculty director, Kent Smetters. Mariko Paulson prepared the brief for the website.

  1. We use the population in 2010 because it is the most recent Decennial Census year before 2020, when the population and population measurement were affected by the COVID-19 pandemic.  ↩

  2. The working-age population is often defined to include younger ages; for example, the OECD defines working age as ages 15 to 64. We focus on the working-age population 25 and older in this brief because we will examine trends by educational attainment within the working-age population, and a person’s ultimate level of education is highly uncertain before age 25.  ↩

  3. We combine measures of race and Hispanic ethnicity to form five race/ethnicity groups: Hispanic, non-Hispanic White (White), non-Hispanic Black (Black), non-Hispanic Asian and Pacific Islander (Asian), and non-Hispanic American Indian or Alaska Native (AIAN).  ↩

  4. An exception to decline mortality among college graduates is AIAN, which reversed its gains in the late 1990s by the end of the 2010s.  ↩

  5. To put our projections by socio-demographic group on comparable basis with other forecasters (who do not publish composition-adjusted mortality rates) we incorporate other elements of our population projections.  ↩