National Academies Press: OpenBook

Assessing Knowledge of Retirement Behavior (1996)

Chapter: 6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT

« Previous: 5 RETIREMENT AGE AND RETIREMENT INCOME: THE ROLE OF THE FIRM
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

6
Assessing Forecasts of Mortality, Health Status, and Health Costs During BabyBoomers' Retirement

Ronald D. Lee and Jonathan Skinner

The U.S. economy may soon stagger under the weight of the elderly baby boomers, who are expected both to live much longer than earlier cohorts of the elderly and to fuel continued growth in health care costs. Recent projections of life expectancy suggest that the Social Security Administration may be under considerable strain to support the nearly threefold growth by 2040 in the number of people over age 65 (Lee and Carter, 1992). Many of these elderly will be in nursing homes; Schneider and Guralnik (1990) predict ''there may be two to three times as many individuals aged 85 years and above in nursing homes in 2040 as there are individuals aged 65 years and above in nursing homes today!" Combined with projected increases in the population of disabled elderly is the rapid growth in health expenses per elderly person. The Health Care Financing Administration (HCFA) forecasts that nearly one-third of gross domestic product will be spent on health care by 2030 (Burner, Waldo, and McKusick, 1992). Auerbach and Kotlikoff (1994) predict that future generations will be required to pay in taxes 82 cents per dollar of income to support currently legislated Social Security and Medicare benefits. It is possible, of course, that the elderly might be expected to pay more out of pocket. But calculations by Bernheim (1994) suggest that, if anything, most baby boom families are saving too little for their retirement.

Ronald Lee's research for this paper was supported by National Institute on Aging grant AG11761-01A1. The authors are grateful to David Cutler, John Bound, Alan Garber, Bert Kestenbaum, Nancy Maritato, S. Jay Olshansky, Joshua Wiener, and panel members for helpful suggestions.

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

An alternative view is much more optimistic about retirement prospects for the baby boom generation. Disability and morbidity will continue to become more compressed, leading to healthier years later in life (Manton, Stallard, and Liu, 1993a; Manton, Corder, and Stallard, 1993) as well as to a secular increase in the average retirement age. The Social Security tax base may be buoyed by immigration and increased fertility rates. The economic demands of higher health care costs will be offset by productivity gains and higher income levels; projections from the ICF-Brookings model, for example, anticipate the percentage of elderly (65+) requiring Medicaid coverage for long-term care to decline by 2018 (Wiener, Illston, and Hanley, 1994). Another projection of long-term health care costs predicts that nursing home expenses, as a fraction of median income, will actually decline by the year 2030 (Zedlewski and McBride, 1992). As a recent Business Week cover story concluded, "The elderly are more vital than before. Americans can afford to grow old. And they will grow old gracefully" (Farrell, 1994, p. 68).

Figuring out which of these two scenarios is correct is clearly crucial for forming policies to prepare for the next century. If the retiring baby boom generation will drag down the American economy by 2020, then government policies designed to smooth the projected health and Social Security costs are likely to be most effective now, while the baby boomers are nearing the peak of their earning capacity. Conversely, a government program designed to save against a nonexistent crisis can disrupt the saving and retirement plans of the generation it was designed to help.

In this paper, we attempt to identify the major factors that account for these very different predictions, and we suggest how these discrepancies can be reconciled. We focus on data requirements that may be useful, or even necessary, to piece together the puzzle of how health and mortality trends will affect retirement income security in 2020. We also stress, however, that much of the work that remains to be done is not simply gathering or linking more data. Instead, the task of reconciling the two divergent views of baby boom retirement must involve more consensus about the interpretation of the data or more generally developing modeling strategies that are more likely to hold long-term predictive power. For example, as we show below, a large part of the difference in projections of health costs depends on alternative assumptions about the extent to which the relative price of health care will rise over the next 40 years. Projections of this type are based on past data, but it is not clear whether the past 25 years reflect a long-term trend or a transition to a new steady state in which medical care prices are stabilized. Improvements in mortality do not result from the passage of time, but rather from the influence on biological processes of changes in health care interventions, lifestyle choices, medical technology, epidemiological processes, and so on, and the evolution of these is not well understood (see Warner, 1993). Similarly, changes in income and health care costs depend on many influences, including policy decisions to be taken in the future, that are difficult to predict

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

over the long run. In other words, many of the problems to be surmounted involve the modeling or interpretation of the existing data, rather than shortfalls in the data themselves. These problems are more intractable (and divisive) than simply collecting better data and relate fundamentally to the intrinsic difficulties in forecasting very complex economic systems.

We focus on four issues related to projecting how mortality, health status, and health costs will affect retirement income over the next 30 to 50 years. The first general issue is how rapidly mortality will decline. There are a wide range of mortality projections. Which statistical approaches hold the greatest promise for long-term projections necessary to maintain the financial viability of the Social Security system? We suggest a number of research approaches, using existing data, that may improve our ability to assess the predictive power of competing models of mortality. Finally, we discuss how mortality projections of specific ethnic groups, or of the very elderly, can be improved.

The second issue is, what will be the health-disability status of the elderly in the next 30 to 50 years? There is considerable debate about the growth in the number of disabled or frail elderly as the consequence of many more people living past age 85. This is a crucial question both for baby boomers setting aside resources for future illness and for future Medicare and Medicaid expenditures. Despite the controversy over the future progress of morbidity and disability, there is surprising agreement in projections of the nursing home population in 2020. Since a consensus doesn't necessarily mean that these predictions are correct, we also consider some possible strategies for better measuring long-term changes in patterns of disability.

The third issue is a related question: given the predicted health-disability levels, how will per-person costs for a given health-disability level evolve in the next 30 years? Will health care costs continue to grow at historical rates, or will they converge to a rate commensurate with wage growth? Determining which of these scenarios is correct is crucial in deciding whether baby boomers will enjoy a plentiful or a strapped retirement. The answer is not likely to be found solely by extrapolating the past 30 years of information. We argue that structural predictions require better information about whether changes in health care costs are the consequence of economic and political policies or are generated by a residual called "technological progress." This question cannot be resolved simply by collecting more data, but must come through improved modeling strategies.

Finally, we consider how these various factors might be expected to affect retirement income—after Social Security payments are received and after health costs are spent—for the baby boom generation in the coming 40 years. For example, how might extended life span affect the ability of the Social Security Administration to pay benefits? Such a question requires information not just about the elderly population, but about the working population in 2020 as well. If disability does decline over time, would retirement ages be extended, providing more earned income and placing less strain on households' nest eggs? Finally,

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

most of the projections relate to aggregate or per capita spending or utilization. But it is very likely that different socioeconomic groups would fare differently under predicted changes in income and in health spending. We know that people with lower socioeconomic status have substantially higher rates of disability. We also know that income growth for this group has been lagging behind aggregate growth rates (or even declining). Yet the nexus of disability, income, and wealth accumulation, especially for lower income households, is not well understood.

HOW RAPIDLY WILL MORTALITY DECLINE?

To determine whether the Social Security and private pension funds held by the baby boom generation are actuarially sound, it is necessary to have good forecasts of mortality so that one knows how long participants in a plan are likely to draw benefits. A distinct but related problem for a pay-as-you-go pension program is that one also needs forecasts of the population in the ages that qualify for benefits. Forecasting the elderly population decades ahead requires forecasting the mortality not only of the elderly but also of the younger adults who, if they survive, will become elderly. For long-term forecasts, one must deal with the entire age distribution of mortality, from infancy on up. We consider both problems here. Population forecasts for the age group 65+ over a time horizon of 65 years evidently depend primarily on forecasts of mortality, and fertility does not enter in. However, to some degree they depend on forecasts of the rate and age distribution of immigration as well: in 1990, 8.6 percent of the elderly population was foreign born, according to the Census bureau. In this paper, we will largely ignore the immigration issue and focus on forecasts of mortality. The reader is also referred to a critical review of the topic by an interdisciplinary group convened by the National Academy of Sciences/Institute of Medicine whose views are summarized in Stoto and Durch, 1993.

Mortality Decline in the United States During the 20th Century

The pace of mortality decline in the United States during the 20th century has varied, as shown by Table 6-1, and there is one subperiod, 1954–1968, in which the age-standardized male death rate actually rose.1 Overall, the figures in the table do not give a strong impression of either an accelerating or a decelerating rate of decline. However, it is useful to consider a hypothetical population in which each age-specific death rate declines at its own constant exponential rate. Life expectancy would rise at a slowing pace because death rates at younger ages would approach zero, and increasingly the deaths averted would be those of the elderly, who would not live many more years in any case. Thus there is a built-in tendency for life-expectancy gains to decelerate even if each age-specific rate continues to decline steadily. Nothing, of course, says that death rates cannot begin to decline more rapidly at older ages, but unless there is a break with

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

TABLE 6-1 Annual Rate of Decline in Age-Adjusted U.S. Death Rates, by Sex for Selected Periods

Period

Males

Females

1900–1936

0.8

0.9

1936–1954

1.6

2.5

1954–1968

-0.2

0.8

1968–1990

1.5

1.4

NOTE: This is the rate at which the crude death rate would have declined for a population with the age distribution of the 1990 U.S. population subject to the age-specific death rates of each period.

SOURCE: Social Security Administration (1992).

historical trends, life-expectancy gains are bound to slow. Indeed, life expectancy increased by about 18 years from 1900 to 1944, but by only about 10 years from 1944 to 1988; Lee and Carter (1992) forecast it to rise by only 6.5 years in the 44 years from 1988 to 2032.

Comparisons of Forecasts

Many people rely on mortality projections by the Bureau of the Census and/ or the Office of the Actuary of the Social Security Administration (SSA). These agencies generally forecast less rapid declines in mortality, and smaller gains in life expectancy, than do other recent mortality forecasts available today. Their forecasts are roughly consistent with the views of some authors, such as Fries (1980, 1989) and Olshansky, Carnes, and Cassel (1990), who argue that future life expectancy is bounded at around 85 years for the general population (sexes combined).

Other authors argue that life expectancy as high as 100 years may be obtainable in the not too distant future. For example, Manton, Stallard, and Tolley (1991) calculate a lower bound to attainable future life expectancy of 95 or 100 years, if people were to adopt optimal lifestyles, and also claim that such levels are already exhibited by some special subpopulations with particularly healthy lifestyles, such as Mormon high priests. Ahlburg and Vaupel (1990) foresee the possibility of such high life expectancies by 2080 by extrapolating the rapid rates of mortality decline that obtained in the United States in the 1970s. Obviously a U.S. life expectancy of 95 or 100 years would have important implications for pension systems of all kinds.

In work described more fully below, Lee and Carter(1992) use extrapolative

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

time series methods combined with a simple model of the age distribution of mortality to forecast that life expectancy will rise to about 86 years by 2065 (with a 95% probability interval of 81 to 90 years), or to 84.3 by 2050. These forecasts implied a life-expectancy gain that was twice that forecast by the Census Bureau and SSA at the time, and is still twice as great as the SSA forecasts and substantially higher than those of the Census Bureau.

These large differences in forecasted levels of mortality lead, of course, to correspondingly large differences in forecasts of the number of elderly. For example, point forecasts for the number of people over age 85 in 2050 vary from 18 million in the Census projections to 41 million by Manton, Stallard, and Tolley. Furthermore, the forecast by Manton, Stallard, and Tolley lies far outside the high-low bracket given by Census.

The SSA forecasts consider trends in 10 groups of causes of death. At the start of the forecast period, the death rates from each cause are assumed to continue to decline at the exponential rate observed during a 20-year base period. These initial rates of decline are then merged into ultimate rates of decline that are assumed for each of the cause groups based on an assessment of various factors believed to influence the rate of decline for each cause in the long run. The ultimate rates of decline are fully in effect about 25 years into the forecast. (This discussion is based on Social Security Administration, 1992.)

The forecasts that result from this approach imply a sharp slowing of the rates of decline of mortality at all ages, relative both to the previous two decades and to longer run historical trends, measuring from the start of almost any decade back to 1900. Table 6-2 shows the difference between long-run historical trends and the rates of decline assumed in the SSA forecast. These differences are great at the youngest ages, where they imply that the SSA death-rate forecasts would be about five times as high as the simple trend-extrapolated rates. The differences diminish with age and are least for rates at 65+. For men in these ages, the difference is negligible, but for women it is considerable. More detailed calculations show that at the younger old ages, in the 60s and 70s, the SSA forecasts rates for females that are 60 percent to 70 percent higher than simple trend extrapolation would suggest. If instead we compare the forecasts to the average rate of decline from 1968 to 1988 for the total age-adjusted death rate, 1.49 percent per year for males and 1.56 for females, the contrast with the SSA forecasts is even greater. There is nothing intrinsically wrong with forecasting that mortality will decline more slowly in the future than it has in the past, and SSA evidently believes it is right to do so, based on its cause-specific analysis.

Other Methods of Forecasting Mortality Change

From a demographic point of view, the problem of forecasting mortality has two aspects that may be usefully separated conceptually, and which are in fact often separated procedurally. First, one must deal with the complexity of the age

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

TABLE 6-2 Average Annual Rate of Decline in Mortality for Base Period Versus SSA Forecast, by Age and Sex, Percent per Year

Age Group

1900–1988 (base period)

1988–2066 (forecast period)

Forecast Rate - Base Rate

Ratio of Forecast to Trend Extrapolation in 2068

Male

0–14

3.25

1.21

-2.04

4.9

15–24

1.54

0.65

-0.89

2.0

25–64

1.09

0.71

-0.38

1.3

65+

0.52

0.54

0.02

1.0

Total

0.95

0.60

-0.35

1.3

Female

0–14

3.39

1.24

-2.15

5.3

15–24

2.52

0.61

-1.91

4.4

25–64

1.59

0.61

-0.98

2.1

65+

0.95

0.55

-0.40

1.4

 

SOURCE: The first two columns of data are taken directly from Table 4 in Social Security Administration (1992:9). The third column is the second minus the first. The last column is calculated as exp(-78* entry from previous column). It represents the ratio of the SSA forecast of mortality levels in 2068 to the death race in 2068 that would result from extrapolating the historical trend from the period 1900–1988.

distribution of mortality, somehow reducing the dimensionality of the problem so that it is not necessary to forecast the many age-specific rates separately and independently. Second, one must forecast the level of mortality and decide how the level is to be characterized and measured. For example, Statistics Canada first prepares a forecast of life expectancy at birth, taking this as the measure of level, and then determines how to allocate death rates by age in a manner consistent with the prior forecast of life expectancy. For other approaches, such as modeling health status as the outcome of dynamic disease processes and changing risk factors, this distinction is less useful. We will first discuss the problem of age distribution and then that of level.

In recent work, there have been two approaches to dealing with age distribution. In the functional parametric approach, a complicated nonlinear function of age with up to nine parameters is fit to the age profile of mortality for a given year or series of years. Changes in the level of mortality then come from changes in the parameters. In practice it may be desirable to hold most of the parameters fixed and capture changes over time through variations in just three key parameters (see McNown and Rogers, 1989, 1992). Forecasts of level and age distribution are obtained by modeling the sample period variations in these key parameters and then forecasting them.

A different approach uses so-called relational methods. In this approach, a

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

standard age profile of mortality is established, representing a central tendency in the shape of the age distribution. New mortality schedules representing different levels of mortality are then generated by some transformation of the standard, where the transformation may be characterized by one or two parameters. The Lee-Carter (1992) method is of this form. In it, the model:

is fit to the sample period age-time-specific mortality rates, mx,t. Here exp(ax) can be viewed as the standard schedule, and the coefficient bx describes how this standard is transformed to generate new age schedules of rates when kt, the index of mortality level, varies. Time series methods can be used to model the sample period variation in kt, and the estimated model can then be used to forecast kt. From these forecasts, forecasts of mx,t can be recovered using the equation. In most applications of this model, kt is well modeled as random walk with drift. Gomez de Leon (1990), using exploratory data analysis on Norway's extensive historical mortality data, selected this model from among a variety of simple models to represent patterns of change, based on a variety of criteria including goodness of fit.

This model makes some strong assumptions. If the model fit perfectly in the sense that all errors ex,t were zero, then any age-specific death rate could be expressed as a linear function of any other, and the correlations among them would all be unity. Autocorrelations for each rate would equal the autocorrelation of kt. Lee and Carter construct probability intervals for the mortality forecasts generated in this way, and for forecasts of period life expectancy. Figure 6-1 plots base period estimates and forecasts for kt, while Figure 6-2 plots life expectancy since 1900 and its forecast derived from that of k. It is notable that whereas the time path of historical life expectancy is decelerating (the gains in the first half of the period were twice as great as those in the second half of the period), the trend in k is roughly linear, and the gains in the two subperiods were equal. Also note that while rates of mortality decline reported in Table 6-1 were quite variable across subperiods, the decline in k, which indexes the log of the level of mortality, appears quite regular.

The Lee-Carter forecasts foresee substantially larger gains in life expectancy than do the SSA forecasts: about 10 years versus about 5 years. We have seen that the SSA forecasts assume a substantial slowing of mortality decline. The Lee-Carter forecasts come close to assuming that historical trends will continue, although this is more nearly true for some age groups than for others. For ages over 60, for complicated reasons, the forecasts published in Lee and Carter (1992) are for more rapid decline than the average rates for 1900 to 1987 and come closer to the average rates of decline for 1930 to 1987.

Both the Lee-Carter and Rogers-McNown approaches draw on time series analysis and ARIMA type models for forecasting the level of mortality. Another

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

FIGURE 6-1 Comparison of mortality forecasts to 2065, based on data from 1900–1989 (dots) and from 1933–1989 (solid), with 95 percent confidence bands. NOTE: Both forecasts are (0,1,0). The forecast from 1900 has a dummy for the influenza epidemic; see text.

approach has been to develop a standard trajectory for life expectancy based on the historical record for many populations and to incorporate the pronounced tendency for life expectancy to rise more slowly when it is at high levels than when it is at low levels. This approach has often been used quite successfully by international agencies.

While there is certainly room for further work on extrapolative models (see, e.g., suggestions made below), current work appears to be pursuing logical directions, and there is no reason to expect that major new initiatives in this area would have a high payoff. The real question, we believe, is whether these extrapolative methods currently yield the best possible forecasts, or whether other models, incorporating more structural information about the complicated biological processes leading to disease and death, might yield superior forecasts. One possibil-

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

FIGURE 6-2 U.S. life expectancy and forecasts (95 percent confidence intervals with and without uncertainty from trend term). NOTE: The forecasts employ a (0,1,0) model with an influenza dummy estimated on mortality data from 1900 to 1989. The 95 percent confidence intervals are shown with and without uncertainty from drift.

ity is to apply extrapolative methods to cause-specific data, which permits relevant medical and biological outside information to be introduced to some degree, as is done by SSA. The development and estimation of explicit statistical models relating disease, disability, and mortality to individual behaviors and risk factors, as will be further discussed later, is another possibility drawing on deeper information. Still another approach is to consider the history of mortality change in terms of epochs of progress against particular kinds of diseases and, in so doing, to identify the likely future direction and pace of progress.

David Cutler has suggested the following illustrative periodization for U.S. mortality since 1850: (1) 1850–1880: Primitive medical knowledge and poor sanitary conditions yield high mortality, particularly for children. (2) 1880–1920: Rapid mortality declines reflect improvements in water supplies and general

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

sanitary conditions. (3) 1920 to 1950: Continuing dramatic progress against infectious diseases result from the development of antibiotics and other effective treatments. (4) 1950–1970: The end of dramatic gains against communicable diseases and slow progress against chronic and degenerative diseases result in more slowly declining mortality. (5) 1970-present: Reductions in cardiovascular mortality result from life-style improvements and more effective treatment.

What does this approach suggest for future mortality trends? There are a number of possibilities: (1) Ever more money is spent on limited-applicability major surgery for cardiovascular diseases, and modest reductions occur in mortality largely owing to lower levels of smoking. (2) Expensive treatments are curtailed with little effect on mortality, and cost growth slows. (3) The genetic revolution leads to earlier diagnosis and treatment of disease, leading to substantial mortality improvements and slower growth in health costs. (4) Dramatic progress is made against chronic and degenerative diseases through behavioral lifestyle changes, through new genetic interventions, and through progress in treating cancer, with uncertain implications for costs. (5) Newly emerging infectious diseases such as AIDS, new antibiotic-resistant strains of old diseases, and diseases caused by worsening environmental conditions lead to rising mortality, with uncertain effects on costs.

This range of possible future scenarios shows the difficulty with the approach. A case could be made for each. How is the forecaster to choose among them? The record since 1900, as summarized by the mortality index plotted in Figure 6-1, shows a surprisingly regular pattern of decline, despite such major breakthroughs as the development of antibiotics. Forecasts using exactly the same method, but with starting dates ranging from 1900 to 1960, yield virtually identical forecasts of mortality and probability bounds for the United States (see Lee and Carter, 1992); a starting date of 1970 yields forecasts of more rapid mortality decline, but starting in 1980 gives results quite similar to those from earlier starting dates. This consistency suggests that historical periods of relative progress and stagnation tend to average out and do not lead to turning points in the underlying trend of mortality.

Evaluating the Uncertainty of Mortality Forecasts

Considerable progress has been made in describing the uncertainty of mortality forecasts in recent years. In a notable paper, Alho and Spencer (1990) develop probability intervals for the SSA mortality forecasts. They find that below age 20, the SSA high-low bounds are narrower than empirical 95 percent probability intervals; from 40 to 64, they are wider; and for other ages, they are about equal to the empirical bounds. In interpreting these results, it is important to keep in mind that these bounds refer to the probability distribution for mortality in any given single year, not for the general level of mortality over the forecast horizon. Lee and Carter (1992) provide probability intervals describing the un-

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

certainty arising in their extrapolative method. The difficulty is that these quantifications, based on time series analysis, do not allow for specification error or for the possibility of larger breaks in patterns than have occurred in the historical past. Some may find this approach to assessing uncertainty persuasive, and indeed it is common in statistical forecasting. Others, however, believe that because of genetic research or lifestyle modification, such structural breaks may be imminent and that therefore the probability intervals based on time series models are too narrow (see Stoto and Durch, 1993).

We do not know how to attach a probability to attaining a life expectancy of 100 by 2030. The Lee-Carter model assigns this outcome a positive probability, but one that is vanishingly small. However, the mere fact that reputable and well-informed scientists believe that such an outcome is a real possibility, and one that should be contemplated in formulating policy, suggests that it does not have negligible probability (Stoto and Durch, 1993). It is very important that this probability be better assessed. Unfortunately, we do not have any concrete suggestions to make on how to do it.

It is also possible, of course, that the extrapolative methods may greatly overstate future mortality declines. If these methods were used to extrapolate backwards before 1900, the result would imply that during the 19th century mortality declined rapidly from impossibly high levels at its beginning. In reality, however, mortality levels were fairly stable throughout the 19th century (see Haines, 1994). In other words, there was an abrupt acceleration of the pace of mortality decline some time around the beginning of the 20th century, and the Lee-Carter model was fit only to the years after the acceleration took place. It therefore overstates the support for a constant rate of mortality decline and understates the variance in rate of decline. In sum, it is best to view the Lee-Carter probability intervals on mortality forecasts as lower bounds on the uncertainty of the forecasts.

One possible reason for slower mortality decline in the future is the emergence of new diseases such as AIDS. The current impact of AIDS mortality on life expectancy at birth is not large on average, although it is substantial for some subpopulations. Census has incorporated AIDS mortality explicitly into the 1992 projections. In the middle projection assumption for mortality, ''the incidence of AIDS is projected to increase linearly until the turn of the century. After 2000, mortality from AIDS will slowly decrease, returning to the 1990 level of AIDS mortality by 2050" (Bureau of the Census, 1993:xxvi; see Campbell, 1991, for the analysis underlying these assumptions). Under the Census scenario for low gains in life expectancy, which incorporates somewhat more pessimistic AIDS projections, AIDS mortality would reduce life expectancy for males by 0.9 years, and for females by 0.2 years, for an average reduction of about 0.5 years, where these figures apply to the year 2005 and later.2 For an alternative analysis that is critical of these methods but that leads to lower projected increases in AIDS mortality, see Bloom and Glied(1992).

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

We will make three kinds of suggestions for advancing work in the area of mortality forecasts: straightforward research using existing data and established methods; data collection; and modeling.

Some Specific Research Suggestions Using Existing Data and Established Methods

It would be useful to reexamine the studies of life expectancy in special subpopulations such as Mormon high priests, as reviewed by Manton, Stallard, and Tolley (1991). The data on the subpopulations do not actually contain much information about death rates at very old ages. Mortality rates are imputed at old ages by multiplying a standard mortality schedule (which does cover mortality at very old ages) by the standard mortality ratios estimated in each study. It is possible that the use of U.S. mortality rates at old ages as a standard has led to questionable results. It would be a simple matter to recompute these using as the standard either a reliable mortality schedule, such as that for Sweden, or the Kestenbaum (1992) rates for the United States. We would then learn whether the apparently very high life expectancies for these subpopulations were artifacts of the estimation method.

By analyzing the past record of the United Nations and the Census Bureau in forecasting population size, Stoto (1983) derived a measure of the standard error of the implied forecasts of growth rates. These ex post standard errors have been widely used to attach prospective probability intervals to population forecasts, although some strong assumptions are evidently required to do so. The Stoto intervals have been compared with those arising from Lee and Tuljapurkar's (1994) stochastic population forecasts (to be discussed below) and found to be similar. We believe it would be useful to apply the Stoto method more broadly to variables produced in demographic forecasts other than average population growth rates or equivalently future population sizes. In particular, the method could be applied to the following:

  • Assessing the performance in forecasting mortality: by the United Nations for individual countries, and by the U.S. Census Bureau and Social Security actuary. Standard errors based on some evaluation metric could then be compared with probability intervals from Lee and Carter (1992) and Alho and Spencer(1985, 1990), and perhaps to McNown, Rogers, and Little (1995).

  • Assessing the performance of forecasts of dependency ratios and comparing these with high-low brackets from official forecasts and with the Lee and Tuljapurkar (1994) intervals.

A last suggestion is that the McNown-Rogers and the Lee-Carter models be fitted and tested on data from populations with high-quality data at older ages, such as Sweden, France, and Japan (see recommendations in Stoto and Durch,

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

1993). In sum, there are various dimensions along which demographic forecasts can be assessed using existing data. These assessment approaches may allow researchers to better reach a consensus on future mortality trends.

Recommended Improvements in Data

Mortality data for older ages in the United States are believed by many to be seriously deficient, to the extent that analyses based on them are thought to be flawed. For example, Kannisto(1994:14–16) reports on old-age mortality data for 32 populations, almost all of developed countries. Based on diagnostic tests on the data, he ranks the U.S. data as "weak" along with those from Canada, Chile, and New Zealand (for the Maoris), and rejects them as too unreliable for further analysis. Others believe the U.S. data to be usable for many purposes. It is important that the quality of these data be more convincingly established (Coale and Kisker, 1990; Bennett and Olshansky, 1994; Kestenbaum, 1992; Himes, 1994). Ideally, of course, one would want to have accurate measures of mortality at very old ages.

SSA prepares estimates and forecasts of the mortality of the U.S. population. It uses registration and census data for death rates up to age 64, but for ages 65 to 95, it uses Medicare records. These have two advantages: first, documentation of age is required at enrollment, and second, the numerators and denominators come from the same very large data set and are therefore less vulnerable to errors arising from age misreporting. Above age 95 a mathematical formula is used to impute mortality rates because the problem of phantom Medicare enrollees becomes increasingly severe. Kestenbaum (1992) reviews these procedures and data sources, and shows that the HCFA Medicare data files that are generally used contain many errors not present in the master-beneficiary-role files of the SSA. He tracks down the sources of errors, and confirms through a matching analysis that the master-beneficiary-role file data for Medicare Plan B enrollees (excluding the railroad-retirement subpopulation) are highly consistent with death certificate data for Texas and Massachusetts. His analysis, and new estimates of mortality rates in extreme old age, are for 1987 only. He finds that rates based on the HCFA data files are reasonably accurate until age 97 or so, but that after that they increasingly understate death rates calculated more carefully from the master-beneficiary-role files. Kestenbaum's work shows that it should be possible to construct improved estimates of mortality at older ages, at least for the last two decades or so.

Despite the fact that SSA does not use the NCHS data for older ages, a recent paper (Bennett and Olshansky, 1994) finds that employing adjustments suggested by Coale and Kisker (1990) together with adjusted rates calculated by Kestenbaum (1992) would reduce current estimated life expectancy at age 65 by 0.5 years, reduce the projected population over age 65 in 2050 by 3.3 million relative to the SSA forecasts (and by far more relative to the Census forecasts),

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

and "significantly influence the projected solvency of the trust funds of the United States" (Bennett and Olshansky, 1994). This study used the standard SSA assumptions, while simply adjusting the initial levels of mortality.

Medicare data, analyzed along the lines pioneered by Kestenbaum (1992), appear to offer the best route to developing satisfactory measures of oldest old mortality for the United States, possibly as far back as the late 1960s (but see a cautionary note in Elo and Preston, 1994:13, on the quality of age reporting in these data and other problems with them). Establishing this data set merits high priority and should be relatively inexpensive. The following specific steps would be likely to yield significant benefits in more accurate measures of old-age mortality:3

  • Kestenbaum's (1992) procedures should be used to estimate death rates for the older population back to the earliest date feasible, possibly to the late 1960s. These death rates should routinely be calculated on an annual basis. Care should be taken with records originating earlier in the history of the system, when documentation of age for enrollees was less stringent.

  • There should be systematic reconciliation and balancing of the Medicare records at HCFA and in SSA's master-beneficiary-role files. HCFA misses many death records that these files have. Some reconciliation is now done for ages over 95; this should be done for other ages as well.

  • SSA should make better use of outside information, such as death certificates, which are now often discarded if they do not prove to be administratively useful, even if they might have valuable statistical uses.

  • Algorithms could be developed to impute deaths in some cases, based on such information as age and years elapsed since last use of Medicare.

  • The master-beneficiary-role files should be checked to ensure that there is no more than one record per person enrolled, although this is not expected to be a major problem.

One might wonder why it is necessary to be so concerned with the mortality of the oldest old or the extremely old. Currently, only about 32 percent survive from birth to age 85 in period life tables, and only 6 percent survive from birth to 95. But according to the Lee-Carter forecasts, by 2030 half of births will be surviving to 85 (in period life tables), and 16 percent will be surviving to 95. By 2060, these figures will be 60 percent and 25 percent. Our current annual life tables from NCHS lump all deaths at ages over 85 together into the open-ended category. Clearly, pension and Social Security liabilities will be strongly affected by what happens to mortality after age 85.

Even after the Kestenbaum adjustments, the United States exhibits lower mortality at older ages than is observed in other populations of Europe and Japan with high-quality data. Canada, with data that are somewhat better than those of the United States, shares this deviant pattern. Is there a special North American

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

age pattern of mortality? If so, why? Or are there further defects in the U.S. mortality data even for death rates at ages 70 to 79? These are important questions, since researchers seeking results applicable to the United States, and frustrated by the poor quality of U.S. data, frequently turn to international data of higher quality. This research area deserves attention (see Himes, 1994; Himes, Preston, and Condran, 1994; and Bennett and Olshansky, 1994).

Mortality data for the period before 1933 in the United States are available, but of uncertain quality (Alter, 1990). Some problems arise from the usual source of age exaggeration at old ages; others arise from the limited number of states reporting registered deaths during this period. These data are useful for statistical studies of mortality change in the United States and for extrapolative methods. It would be worth using systematic methods to inflate the rates for death-registration states to national totals (it is not clear how this is now done in available data sets), and to assess the reliability of these data carefully, particularly for mortality at older ages, building on Alter's work.

Mortality data by race/ethnic group are of suspect quality. Some studies have found that high proportions, on the order of 30 percent, of the deaths of members of some ethnic groups are misclassified as occurring to non-Hispanic whites, based on the classification of these deceased individuals in the census that provides the denominators for calculating death rates. The Bureau of the Census (1993:ix) did special calculations of mortality by race/ethnicity, estimating the life expectancies reported in Table 6-3.

Many people would be surprised to learn that estimated life expectancy is 1.8 years greater for Hispanics than for non-Hispanic whites; and 1.6 years greater for American Indians than for non-Hispanic whites. Observable health conditions and access to medical care seem inconsistent with these data. It is also very striking that the life expectancy for Asians is fully 7 years greater than for white non-Hispanics; this is an enormous difference, and the Asian life expectancy might be the highest observed anywhere in the world for a subpopulation this

TABLE 6-3 Census Estimates of Life Expectancy by Race/Ethnicity, 1993

Ethnic Group

Life Expectancy at Birth

Hispanic

79.0

White, not Hispanic

76.6

Black, not Hispanic

70.2

American Indian, not Hispanic

76.3

Asian, not Hispanic

82.9

 

SOURCE: U.S. Bureau of the Census (1993).

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

large. So there are reasons to doubt the accuracy of these estimates, and the misclassification of ethnicity on death certificates is a plausible source of error.

Yet it is far from clear that these estimates are wrong. Elo and Preston (1994) have conducted a careful review of recent estimates derived from a variety of kinds of data sources, including longitudinal survey data in which race/ethnicity is held fixed, Social Security data linked to death records, and conventional census and vital registration data. They conclude that Asian-Pacific Islander mortality is substantially below that of non-Hispanic whites and that Hispanic mortality is also lower than that of non-Hispanic whites. They thus confirm the direction of the differentials found by the Census Bureau.

It is possible that the Medicare data discussed earlier could be useful in this context. Unfortunately, in the past the only race/ethnic information gathered was white, black and other, and this was for the primary beneficiary rather than for the actual person (who might, for example, be a surviving widow of a different race/ethnicity). The data are now being reclassified according to own race/ethnicity. Beginning in 1980–1982, information was gathered along the now-standard lines (white, black, Asian, North American Indian, Hispanic), but only new applicants for Social Security cards or those applying for replacement cards supplied the information. There is soon to be a mailing to about 2.5 million beneficiaries to elicit more detailed race/ethnic information on a voluntary basis. It appears, therefore, that in a couple of years it may be possible to calculate much-improved race/ethnic mortality rates for the elderly. This would be a major advance, and it is important that the data analysis actually be carried out. It is also important that other approaches be pursued, such as the revealing analysis by Preston, et al. (1994) of black-white mortality differentials based on attempted linkage of data from the Census Bureau, Vital Registration, and Social Security.

Finally, it is not clear that forecasting by cause of death is preferable to forecasting overall mortality. However, if forecasting by cause of death is to be done, then it is necessary to improve the cause-of-death data, particularly at older ages, and to take into account multiple causes of death (see Stoto and Durch, 1993).

Recommended Research on Modeling and Methods

In principle, it should be possible to improve extrapolative forecasts by bringing to bear knowledge about medical progress in specific areas. Experience to date, largely from the use by Social Security of experts on specific causes of death, has not been encouraging. One might be tempted to carry out a mortality forecast disaggregated by cause of death using other means. Whenever this is done in an extrapolative mode, one is likely to get a lower forecast of life expectancy gains since the more rapidly declining causes of death necessarily come to claim a smaller and smaller share of total mortality while the share of the slowly declining causes of death grows. Not infrequently, mortality from some causes is

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

increasing absolutely, for example, death rates for some cancers. In this case, the effect is even more pronounced. The decline of the aggregate, being a weighted average of cause-specific rate, slows (Wilmoth and Preston, 1994), and if any cause-specific rate is actually increasing, then in the very long run the aggregate forecast will be for rising mortality overall. To point out that this is so is not to say that it is wrong. Furthermore, extrapolative forecasts need not extrapolate constant exponential rates of change for cause-specific death rates forever: asymptotes could be imposed, or shares could be forecast rather than levels.

An alternative approach is to explicitly model and estimate the disease processes (at least in general terms) leading to disability, recovery, or death. This is the approach taken by Manton and his collaborators. Their variables include risk factors, such as lifestyle behaviors like smoking, drinking, and exercise. By exogenously varying these risk factors, they can simulate the effects of policies or behavioral changes on mortality and disability. In addition to such analytic simulations, mortality projections can be generated by forecasting each of the risk factors, perhaps by extrapolating observed trends from longitudinal data sets. Such an approach is attractive in many respects. At the same time, it suffers from difficulty in obtaining long time series of the relevant variables, perhaps from some instability in the forecasts and perhaps from some difficulty in obtaining probability intervals. Another problem is that uncertainty about long-term trends in mortality is simply pushed back onto uncertainty about the underlying risk factors. There may be new patterns of behavior, for example, that could worsen long-term survival trends.

HEALTH STATUS PROJECTIONS AND THE COMPRESSION OF DISABILITY

The burden of the baby boom retirement cohort will depend not just on their total number, but also on their health. In this section, we consider the difficulties in forecasting whether people in the next century can look forward to an active, healthy retirement or a relatively frail and inactive one.

The Compression Debate

The original Fries (1980) article hypothesized a biological limit to average human life expectancy at around 85 years in a population.4 Declining prevalence rates of morbidity and disability would lead to a gradual "compression" of morbidity and disability, which would in turn reduce the costs of providing health care and ancillary services for large numbers of chronically ill elderly people. The previous section discussed more recent evidence strongly suggesting that the "old-old," those over age 85, are expected to be among the fastest growing segments of the population for the foreseeable future. Even if there is a compression of morbidity over the individual life cycle, there may be a rapid expansion of

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

the numbers of elderly frail people in the population requiring outpatient or institutional care, given the extremely rapid growth that is forecast for the population age 85 and over (e.g., Schneider and Guralnik, 1990).

One problem with predicting how disability will evolve over the next 40 years is the difficulty in establishing trends from historical data. The reason is that there is no unique definition of disability; objective medical measures such as blood pressure are imperfectly correlated with functional ability, while more subjective self-reported measures of disability may be functions of economic factors such as disability insurance benefits or may evolve over time according to changing social norms. Below, we consider the empirical evidence on secular changes in four types of measurement for disability, ranging from objective (blood pressure, body mass index) to subjective, self-reported disability.

Objective health measures have the advantage of being comparable over time, although they are only imperfectly correlated with a more inclusive definition of disability. Waidmann, Bound, and Schoenbaum (1995), for example, compare the incidence of men and women with systolic blood pressure at or above 140 mmHg between 1971–1975 and 1976–1980 using successive waves of the National Health and Nutrition Examination Surveys (NHANES I and NHANES II). They found a decline in the prevalence of high blood pressure of 14 percent for men and 19 percent for women. (The decline was less pronounced when they included people taking medical treatment for hypertension as part of their "high blood pressure" group.) During this same time period, however, self-reported hypertension increased by about 5 percent. These data illustrate one important pitfall with self-reported disability measures. Better medical treatment is likely to identify or categorize more types of disability, leading to higher self-reported levels of disability even if the average level of severity is declining in the population due to treatment. Extrapolating such self-reported trends is problematic.

Another easily measured indicator of health status is the body mass index, the ratio of weight to height. Costa (1994) compared the pension medical records of Civil War veterans in 1900 with data for men of a similar age from the National Health Interview Survey (NHIS) in 1985–1991. She found first that the body mass index was a surprisingly stable predictor of labor force participation across the two data sources. Second, she found a substantial increase in the average body mass index between 1900 and the 1980s. Using labor force participation as an indicator of functional health, her estimates suggest that had the Civil War veterans enjoyed the same distribution of body mass index as men of similar ages in the 1980s, their labor force participation would have been 6 percentage points higher. These objective measures of disability point towards a secular improvement in at least a limited measurement of health status.

The second general classification of disability relies on the incidence of specific diseases. For example, comparing the medical records of the same Civil War veterans early in the 20th century with medical records of World War II

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

veterans in 1983 suggests dramatic declines in the prevalence of musculoskeletal, digestive, and circulatory diseases (Fogel, Costa, and Kim, 1993). These results imply a long-term secular compression of disability, at least among the "young-old" male veterans in their sample.

Liebson et al. (1992), however, caution that many of the studies of morbidity compression suffer from systematic data problems. There may be changes in the designation of the disease or in the likelihood of a physician's diagnosing a particular problem prior to death. Most studies do not include autopsy information that could inform whether the individual had a particular disease that was never diagnosed and hence did not appear in a retrospective study of health care records. Liebson et al. (1992) relied on a community-based sample of people living near Rochester, Minnesota, with comprehensive standardized data going back to 1907 that provided better medical information than most studies of this type. Examining specific disease categories, they found little evidence favoring long-term (20-year) compression for stroke or coronary heart disease.5

A third approach to measuring disability is to measure changes in functional ability, typically measured by activities of daily living, such as eating, dressing, or bathing, and instrumental activities of daily living, such as light housework, meal preparation, or money management. Manton, Stallard, and Liu (1993a, 1993b), and Manton, Corder, and Stallard (1993) for example, used longitudinal data, the National Long Term Care Survey (NLTCS) in 1982, 1984, and 1989 to estimate transition matrices among different levels of disability, where the levels of disability depended on how many activities of daily living the respondent reported difficulty in performing. Using a variety of statistical approaches, they found a secular decline in age-adjusted disability during the period of analysis.

Their model is quite general and allows transitions both into and out of disabled states over time. By using changes over a short period of time in a consistent longitudinal data set, they avoid many of the problems with changing definitions of morbidity and mortality noted by Liebson et al. (1992). However, the disadvantage is that their data are relatively short term, and the transition matrices may not accurately capture the "low frequency" trends in disability that are required for long-term projections. For example, Manton, Stallard, and Liu (1993b) forecast disability levels in the year 2020 based on the transition matrix estimated using 1982 and 1984 data.

The final approach to measuring disability is to use self-reported health assessments. Even though these are is in some respects the most direct way to measure the state of disability, they exhibit dramatic short-term fluctuations. For example, Wolfe and Haveman (1990) show that predicted disability rates for a male older white widower changed from 9.3 percent in 1962 (based on unweighted data) to 33.4 percent in 1973, down to 23.6 percent in 1980, and back up to 32.6 percent in 1984. Waidmann, Bound, and Schoenbaum (1995) used Census Bureau data to show that the percentage of men aged 45 to 64 reporting they were unable to work rose from 6.3 percent in 1970 to 9.7 percent in 1980 and

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

fell back to 8.8 percent in 1990. For women in the same age group, the corresponding percentages were 8.8 percent in 1970, 11.1 percent in 1980, and 9.4 percent in 1990. Given the evidence discussed above on more objective measures of underlying health, it seems likely that such changes are the consequence of more than just shifts in the underlying level of health status.

One possibility for the rising levels of disability during the 1970s was that as life expectancy was improving, those who in earlier decades would have died were now surviving but in a frail and disabled state. However, Waidmann, Bound, and Schoenbaum (1995) find little evidence to support this view. They find the magnitude of the changes in disability levels too large to be accounted for by the rise in the population of the people who are now surviving to later ages. A more plausible explanation for the variation in self-reported disability is shifts in the eligibility and generosity of the federal disability insurance program.

The federal government expanded eligibility and benefits from the early 1960s through the mid-1970s, and the number of workers receiving disability insurance grew from 0.5 million in 1960 to 2.9 million in 1980 (U.S. Congress, House, 1994:61). However, by the late 1970s and early 1980s, eligibility was curtailed sharply; the percentage of successful applications dropped from 46 percent in 1977 to only 29 percent in 1982 (U.S. Congress, House, 1994:60). Since the early 1980s, both the percentage of successful applications and the total number of those receiving disability insurance have risen, so that in 1993, 3.7 million disabled workers received benefits. Clearly, the pattern in self-reported disability, the rise during the 1970s and the dip during the 1980s, matches the secular path of disability benefits (e.g., Wolfe and Haveman, 1990), and there is a close correspondence between the magnitudes of changes in disability-insurance beneficiaries and patterns of self-reported disability (Waidmann, Bound, and Schoenbaum, 1995).

In sum, there are a wide range of variables that measure different aspects of disability. While objective measures, such as the body mass index and the presence of hypertension, allow for accurate comparisons over time, they may also be imperfectly correlated with the "true" level of disability. On the other hand, self-reported assessments of difficulty in daily functions and in the ability to work provide a clearer picture of who in the population is disabled, but the social norm of disability may itself evolve over time, making secular comparisons, and long-term predictions more problematic. Finally, it is possible that in the longer term, improvements in infrastructure, technology, travel, or job flexibility could affect what society construes as "disability," or even the link between underlying morbidity and activities of daily living. While the debate over the compression of disability is still not entirely resolved, the existing evidence suggests a trend towards lower levels of disability among the elderly population.

If the debate over compression is not entirely resolved, most forecasts of nursing home patients—the most obvious symptom of endemic disability in the elderly population—do not exhibit a great deal of variation, even for projections

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

going out 30 years. Manton, Stallard, and Liu (1993b) predict 3.2 million people over age 65 in nursing homes in 2020, Wiener, Illston, and Hanley (1994) predict 3.6 million by 2018, while Schneider and Guralnik (1990) suggest about 2.6 million for the middle Census Bureau projection for 2020 or about 3.2 million for the high Census projection.6

The close similarity of these predictions does not mean that they are accurate. Even to the extent that more sophisticated prediction models account for trends in levels of disability, there may also be changes in the demand for nursing home patients if the underlying health status is held constant. A large fraction of the elderly disabled are cared for in the community, often by their children. And as Soldo and Freedman (1994) emphasize, there is a great potential for substitution among different sources of care for the disabled elderly, ranging from care by their children, to community-based assistance programs to institutionalization in, for example, nursing homes. Of the estimated 5.5 million elderly people who are disabled, only 24 percent are in institutions, with the remaining three-quarters cared for by family members or public programs (Soldo and Freedman, 1994).7 As the baby boom generation retires, there may be fewer potential family caregivers for this group, given the smaller average size of families (through both fewer children and a greater proportion of divorced households). Alternatively, there may be changes in the financing of outpatient home care services for the disabled. Relatively modest proportional changes in the composition of care for the 76 percent of disabled elderly who are noninstitutionalized will have a much larger proportional impact on the size of the institutionalized population. By the same token, changes in medical conditions deemed ''appropriate" for nursing home care (e.g., Berg et al., 1970) can shift, leading to further variation in nursing home bed demand.

Strategies to Modeling Disability

There are different strategies to modeling how changes in the distribution of disability levels will evolve over the next 40 years. As noted above, the easiest approach is simply to take the current fraction of disabled among those, say, between ages 85 and 90, and multiply it by the number of people predicted to be alive in the same age group. In some cases, one simply uses these projections to predict the number of people who will need hip replacements, or the population of nursing homes (e.g., Schneider and Guralnik, 1990). Overall cost forecasts can also be calculated by appropriate multiplication of the disabled population base times the average cost for this group. This approach ignores potential changes in the incidence of disability or in the age composition of disability.8

A more complicated approach is to focus on the underlying level of activities of daily living or instrumental activities of daily living as the disability "state" and predict these underlying levels without tracking the actual type of disease (e.g., Manton, Stallard, and Liu, 1993a, 1993b). This method is still nonstructural

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

in that it makes the assumption that transition probabilities into and out of disability states is a function solely of a person's current disability and not of the underlying disease. By allowing for a transition matrix among different levels of disability, however, this method provides a very general model of disability that can be used for predicted trends in overall disability levels.

A more structural approach is to focus on specific diseases that are associated with differing periods of morbidity and disability (National Research Council, 1988:102). For example, Alzheimer's disease is generally associated with longer periods of disability. Cancer has a lengthy morbidity, but a relatively short period during which the patient is disabled. One might expect that if the incidence of Alzheimer's disease and cancer were to change over the next 20 years, the transition probabilities of the type estimated by Manton, Stallard, and Liu (1993b) might change as well. To the extent that we have information on changes in the relative shares of each disease category, this approach yields a more accurate prediction of future average levels of morbidity and disability. It also requires more structural knowledge about the subsequent morbidity or disability experiences of people who would have been afflicted by heart disease.

This approach holds promise in allowing more detailed disease-specific information about trends to be used in generating overall levels of morbidity and disability. However, the modeling requirements are also substantially larger because they require making inferences about transitions among disability levels for, say, the individual who avoided having a stroke in 1995 because of improvements in medical treatment or lifestyle. Is this person systematically different from the average individual in the same age group? Multidimensional models that allow for interactions among various levels of disability have large data requirements, but also allow for a great deal of flexibility (see Manton and Stallard, 1994).

Recommendations for Improving Modeling and Data Collection

Data requirements expand with the complexity of the model of disability. To make any inferences about changes over time in the prevalence of disability, one requires a representative sample of initial respondents and careful follow-up to document transitions both into and out of disability states. The NLTCS, for example, began with screening calls to 35,790 people based on Medicare records; these people included a group with disabilities, an institutionalized group, and a control of noninstitutionalized nondisabled. Because the initial sample was representative of the underlying population, valid transition probabilities between 1982 and 1984 (and subsequent years) could be made after correcting for nonresponse bias. The Health and Retirement Survey (HRS), currently in its second wave, will also provide detailed information about disability transitions over the next 15 years, as will the Asset and Health Dynamics Among the Oldest Old (AHEAD) survey, which focuses on older people.9

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

One shortcoming of existing studies, such as the NLTCS or the Longitudinal Study on Aging is that even when the sample is drawn from a representative population, the number of people in the sample who enter a nursing home is small. Of that group, the percentage who remain for more than 1 year is fewer still. While the number of people who remain in nursing homes for long periods is relatively small, they account for a large fraction of costs, so that accurately forecasting this group is crucial.

Some surveys attempt to overcome this problem by oversampling those with chronic health problems. For example, the National Long Term Care Channeling Evaluation sampled on the basis of observed risk of being institutionalized. The problem with this approach is that the sample is not representative of the population, so that inferences cannot be made for the purpose of forecasts. The National Nursing Home Survey provides no information about people who aren't admitted to nursing homes and provides only a cross-sectional snapshot of nursing home admissions.10

The second potential problem with existing data sets is nonresponse. We have argued above that long-term longitudinal panels are necessary to elicit low-frequency changes in the compression of disability over time. Yet even a low nonresponse rate of 3 percent to 4 percent annually can translate into losing 35 percent to 40 percent of the sample after 15 years. It therefore seems clear that cross-sectional samples of health, such as the NHIS, will still be of value in providing benchmarks of long-term trends in disability given that the questions asked are comparable over long periods of time.

Third, the HRS will provide detailed information for the next 15 years on a cohort of people initially in their 50s. By 2010, we should have a good handle on the retirement and disability experiences of this group (as well as older cohorts surveyed in AHEAD study). To assess long-term trends among the "young-old," however, requires detailed information about later cohorts, such as the group now aged 41 to 50. In other words, just knowing how a particular cohort fares does not in itself provide information for assessing long-term trends in disability and health needs; one requires survey information on a comparison cohort group. Comparing people in their 70s from the first AHEAD wave, with the HRS waves in the 2010, when those sampled will also be in their 70s, would provide a good measure of changes over time in disability. A survey that follows cohorts now in their 40s, to be fielded beginning in 2002, would improve long-term estimates of secular trends in disability.

In sum, the major problem with estimating long-term trends in disability is likely to be getting transition matrices that are sufficiently long term to reflect the low-frequency variation in rates and levels of disability. Coupled with the data requirements, however, is the problem of interpretation: will the notion of disability in the year 2020 be similar to that today? These issues will be partially resolved as some of the newer data sources, such as the HRS and AHEAD, continue to provide information on the retirement and disability experiences of their samples.

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

PREDICTING HEALTH CARE COSTS

Predicting the overall level of health care costs in 2020 and beyond is a daunting task. Total costs can be separated into two parts, prices and quantities. The quantity of "real" health care services in the future will depend on the size and age composition of the population, the underlying health status (or disability levels) of the age categories, the utilization of health services for a given level of health status or disability, and the intensity of such health services. The price of this service will depend on the general inflation rate over the next 25 years and the change in the relative price of health care.

The previous sections have focused on the size and composition of the population and the underlying health status or disability level of that population. We are not particularly concerned with the inflation rate for the next 25 years. So in this section, we focus on the remaining unknown variables: the level of health services (for a given underlying health status and age composition) and the relative price of health care in terms of other (nonmedical) goods.

Predicting Real Health Care Services by Demographic Group

The first step is to use current data to get a benchmark of how real quantities of health care services might be expected to evolve over time with changes in the age composition and in disability levels.11 In general, health care expenditures rise by age and by disability or health status (e.g., Manton, Stallard, and Liu, 1993a; Chulis et al., 1993). To the extent that expenditures in a given year are valid measures of health care services, we can make inferences about the use of health care services by examining the evidence on health care expenditures.12

Empirical evidence suggests that it is important to distinguish between the real health services of those who survive to a given age and those who do not. Health costs are much higher in the months prior to death (Lubitz and Prihoda, 1984). Consequently, some of the measured increase of average health costs with age is actually due to the fact that in older age groups, death rates are higher, and therefore the mix of those surviving and those dying is increasingly tilted towards those dying (i.e., those requiring higher average levels of health services). Falling mortality may reduce unconditional per capita health service use at every age by reducing the age-specific proportion who are near death, and hence reducing the proportion of those with high demand for health services.

There is some controversy over how health care services near death vary by age or disability state. One study shows that among a group of retired workers in their 60s, median health expenditures for people who died between 1982 and 1984 was $8,012 for people with some functional limitations, but $11,846 for those without limitations (McCoy et al., 1992). In other words, health care services near death were higher among the healthier population than for the less healthy. Of course, we observe only Medicare claims in this study, and not the

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

full range of health expenses for the two groups. Determining how changes in the prevalence of disability among the population affect health services utilization requires better information about differences in health care services over the life cycle by level and type of disability. There is some evidence that Medicare expenditures on older people (i.e., ages 85+) near death are below those for younger people (e.g., Temkin-Greener et al., 1992), but there is potential for newer studies, such as the HRS linked with Medicare data, or the Medicare 20 percent sample, to provide a better picture of the dynamics of health care use among these different groups.

Predicting Health Care Quantities and Prices for the Next Century

Even if we know how the level of health care services varies by different attributes of the population, a much harder problem is to predict how real health services and prices will evolve over the next 30 to 40 years. Official predictions by HCFA of national health expenditures, expressed as a ratio of gross domestic product, to the year 2030 are presented in Figure 6-3. Predicted growth is substantial, with overall national health spending reaching 26.5 percent of gross domestic product by 2020 and 32 percent by 2030. While spending nearly one-third of gross domestic product on health care may at first appear implausible, the

FIGURE 6-3 National health expenditures as a percent of gross domestic product, 1980–2030. SOURCE: Burner, Waldo, and McKusick (1992).

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

TABLE 6-4 Accounting for Increases in Real per Capita Health Spending, 1980–2030

 

Share of Annual Real Growth Due to

Category

Annual Real Growth

Utilization

Intensity

Change in Relative Prices

Physician Services

 

1980–1990

4.6

7.3

40.0

52.7

1990–2000

6.5

12.3

35.1

52.6

2000–2030

3.6

9.9

28.8

61.3

Inpatient Hospital

 

1980–1990

6.4

-91.0

138.1

52.9

1990–2000

6.5

-7.4

59.9

47.6

2000–2030

4.1

34.2

24.4

41.4

Nursing Homesa

 

1980–1990

5.2

2.4

70.0

27.6

1990–2000

6.6

18.7

54.2

27.1

2000–2030

3.8

37.5

44.6

17.9

NOTE: Column 1 shows the annualized real change in total health spending for each decade (or decades). This column reports real changes in total hospital expenditures (rather than just inpatient expenditures). Columns 2 to 4 calculate the contribution of the particular factor (utilization per capita, intensity, or relative price) as a percent of total real per capita change. For the three decades 2000 to 2030, the proportional changes are averaged for the purpose of this table.

a Excludes intermediate care facilities for the mentally retarded.

SOURCE: Burner, Waldo, and McKusick (1992).

forecast is based conservatively on the trend of real health care spending that stretches back 30 years.

What part of this overall increase is caused by changes in quantities, and what part by changes in relative prices? Burner, Waldo, and McKusick (1992) break down their estimates further; some of their results are shown in Table 6-4. In this table, we exclude changes owing to population growth and general price level changes, and focus on the relative importance of the three factors mentioned above—utilization, intensity, and changes in relative health care prices—in explaining the overall costs of real per capital health care expenses.

Real per capita expenditures on each category of health care selected above (physician, inpatient hospital, and nursing home) are expected to rise at a roughly 6-1/2 percent annual rate through the rest of this century before stabilizing at roughly 4 percent annually from 2000 to 2030. Of that increase, approximately half is attributed to a change in the relative price of health care for physician and inpatient care, and less than one-third to nursing homes. Although inpatient

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

hospital services experienced a large relative decline in utilization rates, with a corresponding increase in intensity, during the 1980s, utilization growth is predicted to rebound and to account for about one-third of hospital cost growth during the first 30 years of the next century. By contrast, growth in nursing home costs is predicted to arise largely from an increase in the age composition of the population and changes in intensity of care.13

Predicting changes in relative prices and real quantities is a difficult task. In the estimates presented above, the positive increases in both the real quantity of health services and the real price are suggestive of increasing demand for health services, with a corresponding movement along the supply curve. If there is an error in the prediction of health care utilization, for example, it will tend to be magnified in the prediction of total costs. If demand for health care is less than predicted, then the implied increase in the real price of health care will also be less than predicted, leading to even greater cost savings. Alternatively, if there is a movement along the demand curve for health (i.e., higher prices lead to lower quantity consumed), the growth in health care costs will be attenuated.

One component of the increase in predicted real health services, particularly for nursing home care, is the increased total demand for medical care as the baby boom generation ages.14 Do countries with more rapidly growing elderly populations experience more rapid health care costs? To test this proposition, Getzen (1992) compared a cross section of countries that belong to the Organization for Economic Cooperation and Development (OECD) since 1960. He found little evidence that the fraction of the population over age 65 was correlated with health spending as a share of gross domestic product, or that the change in the share of the population over age 65 was correlated with the change in the share of health spending in gross domestic product. In other words, his results suggest that the predicted impact on health care costs of the aging baby boom generation is likely to be overstated because increased demand will be offset by an "adjustment to budget realities" through a reduction in utilization (for a given age group), intensity, or prices.15

A less sanguine prediction for future fiscal burdens comes from Auerbach and Kotlikoff (1994). They use HCFA predictions to suggest that under current policies, current fiscal policies "may require a net lifetime tax rate of 82 percent on future generations of Americans." And as they note, the assumption of rising health care costs through 2030 contributes greatly to their striking result. When they assume that health care costs are stabilized at their 1994 levels, so that health spending changes only because of demographic shifts, the lifetime tax burden increases only from 37 cents per dollar for a person born in 1960 to 45 cents per dollar for future generations (Auerbach and Kotlikoff, 1994:34).16 In other words, the increasing relative price of health care is responsible for increasing the future generations' tax burden from 45 to 82 cents.

One may still wonder how an increase in Medicare and Medicaid payments by the government can account for raising the lifetime tax from 45 cents per

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

dollar to 82 cents per dollar, given that total government spending on health care is predicted to rise by only about 17 percent of gross domestic product through 2030 (Burner, Waldo, and McKusick, 1992). The magnitude of the Auerbach-Kotlikoff result can be explained partially by the structure of the policy experiment. In their model, people born in 1992 are locked in to the current tax policies, so they pay just 36 cents per dollar of lifetime income, even though they spend most of their life in a regime in which health care is extremely costly (recall that this generation is only 38 by the time health care costs stabilize at 32% of gross domestic product in 2030). Hence the unfortunate babies born in 1993, and in subsequent years, are handed an enormous debt burden made even worse by the fact that debt has been accumulating at the real rate of interest assumed equal to 6 percent.

The Wiener, Illston, and Hanley (1994) projections relate solely to long-term care, but their assumptions are quite different. Instead of a long-term trend growth rate relative to gross domestic product, as in the HCFA projections, they assume that nursing home reimbursement rates simply keep pace with wage growth. What difference does this assumption make? According to the HCFA projections, annual nursing home plus home health care expenditures are projected to increase by $181.5 billion by 2020, in 1993 dollars. By contrast, Wiener, Illston, and Hanley project an increase of only $93 billion (to the year 2018). For evaluating the retirement prospects of the baby boom generation, it matters crucially whether one views the 30-year trend in health spending as a transition to a new steady state or as a trend destined to continue (until gross domestic product is exhausted!).

Conceptual Issues in Modeling Health Care Cost Projections

The problem of forecasting the general price of health care costs cannot be resolved simply with more historical data on prices. The relative price of health care is determined by market, institutional, and government forces, and the fact that many factors have contributed to its rise in the past does not mean that these factors won't be stabilized, or even reversed, in the future. The problem is that we do not have a good model of how health care costs are determined.

For example, it is likely that the evolution of government health costs is quite different from that of private health costs. So far, the U.S. government has provided health entitlements such as Medicare and Medicaid, which have expanded rapidly in recent decades despite various attempts to contain health care costs. It will be difficult to know a priori whether the government will continue to provide health care assistance as an entitlement or switch to global budgeting, as have most OECD countries (see Wolfe and Moran, 1993). In the HCFA projections, for example, there are only modest shifts in the composition of health care spending between the private and the public sector. But should the government switch to global expenditure caps, for example, both the price of health care

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

and the real quantity of health care services would be strongly affected. To make informed predictions about the expected value of future health care expenditures therefore requires some knowledge of the probability that the federal government will impose global expenditure caps, but such a prediction presupposes that we have a model for the likelihood of the government's legislating comprehensive health care reform with expenditure caps.

Private costs are determined at a substantially more decentralized level and depend on the preferences of employees and their employers, as well as on tax and regulatory considerations affecting the design of health care benefits. The evolution of such factors is also hard to envision. Whether the shift toward managed care will continue and, if so, in what form (and with what cost saving) is difficult to predict based on current information (see Employee Benefit Research Institute, 1995). Generating a prediction for the trend in the composition of employee-provided health care is problematic; placing confidence intervals on such a prediction would be nearly impossible.

One might also expect substantial interaction between private and public spending. For example, cutbacks in Medicaid or Medicare reimbursements could be partly passed along by hospitals and health care providers to the private sector. Alternatively, U.S. government legislation of health care reform along the lines of the Clinton plan would change the balance between private and public spending on health as well as blur the distinction between the two sectors.

Below, we discuss some recent studies that have focused on the determination of government-financed health service and price determination. As noted above, government programs such as Medicare and Medicaid directly affect just one part of the health care sector. However, the centralized and more comprehensive sources of data from the federal government make such programs easier to evaluate for research purposes.

Some recent efforts to account for changes in overall Medicare spending have focused on cross-sectional variation in physician density and practice patterns (e.g., Holahan, Dor, and Zuckerman, 1990). Other analysts view the increase in health costs as reflecting the taste of consumers for more technologically complex health care (Newhouse, 1992). The view that technology is driving health care costs certainly receives some support from the evidence on continued growth in costs, despite cost-cutting programs such as the prospective payment system. However, McClellan (1994) suggests that the prospective payment system did little to stem the tide of health care costs and may actually have contributed to continued growth in hospital costs. He argued that diagnostic related groups are related not to a particular diagnosis, but to a means of treatment. In other words, a heart attack treated by coronary artery bypass graft falls into a different diagnostic related group than a heart attack treated nonsurgically. He showed that within 5 years of the start of the prospective payment system, nonsurgical (and less costly) diagnostic related group admissions dropped by 50 percent, while surgical diagnostic related group admissions grew by roughly 50

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

percent. In other words, the growth in the overall price index, at least for the 1980s, may be affected by government reimbursement policies, which are, of course, policy variables subject to change. This approach, which stresses analysis on a relatively micro scale to elicit dynamic trends in health costs, holds promise for understanding the determination of health care costs.

A related issue is, under what scenario would "the populace" more generally allow the projected continued growth in health care spending? Auerbach and Kotlikoff (1994) predict that between 1992 and 2029, health care costs for a 65-year-old will rise from 13 percent to 29 percent of income, leaving average nonmedical consumption essentially unchanged. As was shown in Table 6-4, much of the increase in spending is the consequence of increased utilization and intensity of services, not simply a change in the relative price of a given medical procedure. But it would seem unlikely that health costs would be allowed to claim an additional 16 percent of income without yielding anything of value to these future retirees. That is, if health costs continue to claim a larger fraction of gross domestic product, it could be that technological progress would provide sufficient improvement in health status that people would be willing to spend the additional resources on health care (rather than on non-health-care goods). Looking just at consumption net of medical expenses might not provide an accurate assessment of overall welfare.

In sum, typical forecasting approaches that project forward trends in individual price series are unlikely to predict future price trends successfully. Understanding the factors that determine health care costs (and quantities) will improve the accuracy of price predictions. Unfortunately, much work remains to be done in understanding what factors are important in determining health care costs and how they might be expected to evolve over the next 30 years.

HOW WILL CHANGING MORTALITY, HEALTH, AND HEALTH COSTS AFFECT RETIREMENT INCOME SECURITY?

One of the primary issues facing policy makers is how changes in mortality and health status and health costs might be expected to affect the income security of baby boom retirees. We have not yet considered the feedback effects of how such demographic and health changes might affect retirement income. In the subsections below, we look at a number of factors that are important in piecing together the puzzle of how the long-term trends discussed above affect retirement security.

The Effect of Changes in the Demographic Structure on Income Security for the Elderly

It is important for policy makers to know the degree of uncertainty that is associated with the forecasts that they use. Important decisions to raise both

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

payroll tax rates and retirement age have been taken based on very long-term forecasts of the old age dependency ratios. But how seriously should these forecasts be taken? How uncertain are they?

Traditional Approaches

The traditional method for formulating population forecasts is scenario-based. The forecaster chooses what are believed to be the most likely or perhaps the statistically ''expected" trajectories for fertility, mortality, and migration, including age distributions, and then generates forecasts by using well-known accounting identities. In choosing vital rate trajectories, the forecaster could certainly use any of the extrapolative methods just described. This method is quite straightforward, except for the relatively minor point that the population projection based on the expected values of stochastic rates does not equal the expected value of the population forecast based on stochastic vital rates; in practice the difference appears to be small enough to ignore (Lee and Tuljapurkar, 1994).

The real difficulty with the scenario-based approach comes when it is used to convey information about the degree of uncertainty associated with a population forecast. This is done by constructing two alternative trajectories for each rate, one above and one below the medium or preferred trajectory. The forecaster then combines these trajectories into high, medium, and low scenarios in ways that depend on the purpose of the forecast. For example, the Census Bureau combines high fertility and low mortality into a high scenario, and low fertility and high mortality into a low scenario. This gives the greatest range for the population growth rate and the future population size. However, since high fertility tends to make the population young, while low mortality tends to make the population old, the high and low scenarios do not cover a correspondingly broad range for the old age dependency ratio. Furthermore, because the scenarios assume that fertility is always high or low and mortality is always high or low, the Census Bureau's approach ignores all possible trajectories in which there is fluctuation. Yet strong fluctuations like the baby boom and baby bust have occurred in the past, and they lead to more extreme outcomes than the scenarios. Also, a little thought will verify that a scenario-based forecast cannot possibly yield probability intervals that are consistent among the different items that are forecast since variations in annual numbers or single age groups tend to be offsetting when the numbers or groups are summed to form larger population aggregates—broader age groups, or total population size, for example. So the basic method used by virtually all official forecasts to convey information about uncertainty is deeply flawed and largely incapable of informing policy.

Stochastic Population Forecasts

Because plans for Social Security are made with an extremely long lead

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

time, and because policy changes might have major effects on both the macro-economy and the life-cycle planning of individuals, it is particularly important to take into account the degree of certainty with which long-term forecasts should be viewed. Considerable progress has been made in recent years in producing stochastic population forecasts with consistent and meaningful probability intervals for all items forecast (Alho and Spencer, 1985; Davis, 1988; Alho, 1992; Pflaumer, 1988; McNown and Rogers, 1992; Lee and Tuljapurkar, 1994). The approach of Lee and Tuljapurkar (1994) is to formulate models of age-specific fertility (see Lee, 1993) and age-specific mortality (Lee and Carter, 1992), with fertility and mortality each driven by a single time-varying parameter that is modeled and estimated using standard techniques of time series analysis. Migration could in principle be handled in exactly the same way, but was actually taken as deterministic at the level assumed in Census Bureau forecasts. These fitted models of stochastically varying rates are then taken as inputs to the population renewal process, and the propagation of error is carefully tracked, with all variances and covariances taken into account. Drawing on results in Tuljapurkar (1990), Lee and Carter derive and calculate analytic results for a quartic approximation to find both the expected value of the forecast and the probability intervals. Similar results can be obtained using Monte Carlo methods (repeated stochastic simulations) based on the stochastic models of vital rates.

FIGURE 6-4 Projections of the population aged 65 and over and aged 85 and over from Lee and Tuljapurkar (1994) (solid lines), the Social Security Administration (S), and the Bureau of the Census (C).

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

FIGURE 6-5 Projections of old and oldest-old dependency ratios from Lee and Tuljapurkar (1994) (solid lines), the Social Security Administration (S),and the Bureau of the Census (C).

Figure 6-4 shows forecasts derived in this way for the population 65+ and 85+, with 95 percent probability intervals. The figure also shows SSA forecasts for the same quantities in 2040 and 2070, and Census Bureau forecasts for 2050, including high, medium and low variants for both. Note the very different indications of uncertainty given by the Census Bureau and SSA in each case. Figure 6-5 shows forecasts of the old age dependency ratio, that is (population 65+)/ (population 20–65), and the oldest old dependency ratio in a similar format. Note that whereas before, the Census brackets were substantially wider than the SSA brackets, in this case they have become substantially narrower, and very implausibly so. That is because the Census Bureau arrays low mortality with high fertility in their high scenario, for example. The Census forecasters are very aware of this problem, but it is inescapable. One could choose different scenarios from among the large set offered by the Census Bureau to get a different contrast, but there is no consistent or logical way to do this. Nor, it is important to add, is there any way to attach any kind of probability to these brackets, or to those of SSA.17

To a first approximation, and other things equal, payroll tax rates for Old Age, Survivors and Disability Insurance benefits must vary in proportion to the old age dependency ratio. It is because of the projected increases in this ratio that in the 1980s it was decided to begin to raise payroll taxes in order to accumulate

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

a substantial reserve before the retirement of the baby boom generation, and to slowly raise above 65 the age at which full retirement benefits could be received. According to the point forecasts in Lee and Tuljapurkar (1994), the old age dependency ratio will increase by 34 percent by 2020 and by 98 percent, a virtual doubling, by 2050 (see Figure 6-5). These figures are only slightly higher than those of the Census Bureau and SSA, and they are daunting. But is it appropriate to make long-term policy decisions based on such expected increases in old age dependency? According again to Lee and Tuljapurkar, the 95 percent probability interval is bounded below in 2020 by a 22 percent increase, and in 2050 by a 28 percent increase—very different from the expected doubling. The intervals are bounded above by a 46 percent increase in 2020 and a 167 percent increase in 2050, suggesting heavy pressure indeed on payroll taxes and the Trust Fund. Yet this range of possibilities most likely understates the width of the range with 95 percent probability coverage, since some kinds of uncertainty were not taken into account.

Figure 6-6 shows the corresponding forecasts for the total dependency ratio, or (population ≤20 + population ≥65)/(population 20 to 65). Now it is the turn of the SSA forecasts to have an implausibly tiny bracket, particularly in 2040. This is because the SSA obtains its low or "optimistic" forecast by combining high mortality with high fertility, thus getting the lowest old age dependency ratio. But

FIGURE 6-6 Projections of the total dependency ratio from Lee and Tuljapurkar (1994) (solid lines), the Social Security Administration (S), and the Bureau of the Census (C).

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

the high fertility also produces many child dependents, which raises the total dependency ratio.

If we consider the Lee-Tuljapurkar intervals for the total dependency ratio, the uncertainty is still very large. However, because child dependency and old age dependency generally move in opposite directions, the possibilities are muted and the bracket is narrowed. For example, in 2050 the upper bound of the old age dependency ratio was 2.1 times as great as the lower bound. For the total dependency ratio, however, it was only 1.6 times as great. Any sort of weighting of the age groups for tax yield and for program costs is easily incorporated into the analysis, and corresponding probability intervals can be derived.

The Lee-Tuljapurkar probability intervals do not reflect all sources of uncertainty. In particular, they do not reflect uncertainty arising from measurement error (data quality) or from the possibility of model misspecification and structural change beyond the range occurring historically during the base period. Nonetheless, these probabilistic forecasts represent an important step forward in conveying to policy makers the degree of uncertainty associated with demographic forecasts.

The stochastic population forecasts shown in Figures 6-4 to 6-6 do not simply produce new scenarios based on upper and lower probabilistic forecasts of the vital rates. Some researchers have made population forecasts of this kind (McNown and Rogers, 1992). However, such an approach suffers from most of the problems of the traditional scenario-based forecast, and no probabilistic interpretation is possible for the results. For example, the high population size forecast produced in this way would be based on fertility adhering to the upper bound of every individual percent probability bound (p), which for a horizon of T years would only have probability 1− (1−p)T. The stochastic population forecasts shown in Figures 6-4 and 6-6 instead treat fertility and mortality as random variables at every iteration of the forecast, keeping track of all the numerous variances and covariances. Similar results can also be obtained by Monte Carlo procedures, based on many hundreds of stochastic simulations of population trajectories, where the vital rates at each iteration are based on the estimated models, plus random shocks with appropriate covariance structures.

It is important to realize that the probability bounds consequently mean something very different in Figures 6-4 to 6-6 than do the scenario-based high, medium, and low bounds of traditional projections. This is not simply because a probability is attached to the former and not to the latter. The probability bounds indicate estimates of the range in which the quantity is 95 percent likely to be found in the specific period in question—here an average over a 5-year interval.18 The upper bound on the scenario projection does not indicate a bound for a single year or for a 5-year average, unless vital rates are thought to be perfectly correlated over time. It is, in fact, difficult to give any interpretation to their bounds in this context. No one would suggest that fertility would exactly and consistently follow the high trajectory and mortality the low trajectory over the entire span of

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

the forecast, yet that is the scenario underlying the high projection of the Census Bureau, for example.

Consider the important problem of calculating the probability that the Social Security reserve fund will fall to zero before a given date, under a given deterministic set of tax and benefit schedules and a set of deterministic economic assumptions. This calculation could not be done in any meaningful sense using the scenario-based projection, but can be carried out using the stochastic population forecast. To do this one would not derive the implications of following the upper and lower 95 percent bounds for the relevant quantities, which would be inappropriate. The correct calculation requires taking into account the full structure of covariances from year to year and across ages in all the relevant demographic quantities. One way to do this would be to use the Monte Carlo approach sketched above and simply to calculate the reserve fund for each year under each simulation and then, from these results for hundreds of simulations, calculate the distribution of the levels of the reserve fund in each year. This kind of calculation has not previously been possible. It illustrates how stochastic population forecasts can support a quantum increase in information for planning in the face of demographic uncertainty.

There are no new recommendations for data under this heading. Data are needed to model the vital rate processes, but once this is done, population totals are generated by recursive accounting identities. It would take us too far afield to consider the adequacy of data as a basis for forecasting fertility and migration.

So far as methods and models are concerned, however, research is badly needed on how to use these increasingly available stochastic demographic forecasts in policy formation (Tuljapurkar, 1992). How can the probability intervals for demographic quantities be combined with estimates of uncertainty for forecasts of economic variables? Health care costs? When can independence be assumed? If it cannot, what then? What should loss functions be?

Finally, we would like to call attention here to an earlier recommendation for Stoto-like analyses of past performance in forecasting dependency ratios and other measures, so that standard errors can be compared with those from the stochastic forecasts.

The Impact of Improving Patterns of Disability on Delaying Retirement

If the incidence of disability declines during ages associated with retirement (60 to 70), we might expect that retirement will be delayed, increasing the financial security of people planning their retirement. Health and disability, and particularly the transition from good health to a disabled state, have been well established in the literature as important determinants of retirement. The estimate from one structural model (Gustman and Steinmeier, 1986) suggests that the magnitude of the disability coefficient is equivalent to an increase of 4 years

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

in age. In other words, a person who is not disabled would be predicted to retire 4 years later than an equivalent person with a disability. If the proportion of the working population with a disability were to decline by 10 percentage points—a substantial decline—the average retirement age would increase by less than 5 months. In other words, estimates from cross-sectional data on retirement behavior do not suggest large changes in retirement patterns, and hence in retirement income security, as the consequence of lower projected disability rates in the future.

It is well established, however, that cross-sectional estimates of variables may not provide good predictions of time series changes. For example, employers may be more willing to employ (or re-employ) older workers if the employers expect that a smaller fraction of their elderly workers will become disabled while working at their firm. As noted earlier, the self-reported perception of disability may be subject to variation as the consequence of government or social policy towards older workers (e.g., Waidmann, Bound, and Schoenbaum, 1995; Wolfe and Haveman, 1990).

Uncovering how work and retirement will evolve in the long term is a difficult problem. The current trend appears to be towards earlier retirement, not later retirement, suggesting that the secular changes in underlying health status that might be expected to delay retirement in the future are currently being offset by other factors, perhaps related to disability insurance among others. To detect the evolution of retirement decisions requires a long-term panel of people nearing retirement, coupled with detailed information on activities of daily living and instrumental activities of daily living measured in a consistent way over time. The Retirement History Survey (RHS) is a reasonably long-term panel, but it occurred during dramatic changes in social insurance policy towards the elderly, such as an increase in real Social Security benefits. The HRS again shows great promise in shedding light on this important issue, because it begins with families nearing retirement. Nevertheless, one would also like to have a comparison cohort with objective questions that could be matched with the HRS or an earlier survey.

The Distribution of Health Expenses Among Out-of-Pocket, Private Insurance, and Public Insurance

Out-of-pocket health expenditures are an important component of financial security at retirement. These out-of-pocket expenditures are determined both by the general level of health care expenditures and by the fraction of total expenditures paid out of pocket by the household. Of course, someone must pay for even the health expenses not paid out of pocket, so changes in the financing of government (or perhaps private) health insurance may also affect the financial security of retirees. In this section, we first document recent secular trends in the fraction of total expenditures paid out of pocket. Second, we consider how changes in the

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

share of health spending accounted for by out-of-pocket costs (including private insurance premiums) might affect the indirect as well as the direct costs of health care for baby boom retirees. Finally, we consider data issues in measuring out-of-pocket expenditures.

Most projections of health costs assume a relatively stable fraction of government and private cost sharing. For example, Wiener, Illston, and Hanley (1994) predict an increase in the number of nursing home patients who spend more than 40 percent of income and assets on nursing home care, from 36 percent in 1993 to 39 percent in 2018 assuming current Medicaid policy. This predicted amount is quite modest, especially given the 25-year horizon. While not directly comparable, recent evidence on overall out-of-pocket expenses suggests that the burden of health care costs for the elderly has been rising substantially in the past several decades. Between 1977 and 1987, the percentage of elderly households (age 65+) who spent at least 20 percent of their income on out-of-pocket expenses rose from 7.0 percent to 10.6 percent (Taylor and Banthin, 1994). These figures, based on the National Medical Expenditure Surveys (NMES) in 1977 and 1987, do not even include nursing home expenses, which are likely to be substantial for a small fraction of the population.

The mix between out-of-pocket and government or private insurance payments affects the share of retiree health benefits that are actually paid by the cohort of retirees. Suppose, for example, that the out-of-pocket expenditure share declines, with the government picking up the shortfall. As a consequence, tax revenue needs to be larger to pay for the additional government share of health expenditures.19 The increased share of government spending and taxation would then transfer resources from younger taxpayers (who foot the higher government share through their tax payments) to older generations (who benefit from the increased government spending for their medical bills).

By contrast, increases in private health care spending would be more likely to be borne by the recipients, either directly through out-of-pocket expenses or indirectly through higher insurance premiums. Of course, health insurance provided by employers for their retired employees may entail some redistribution across cohorts, as rising health care costs of retired employers are implicitly paid through higher insurance costs of current workers. Still, the prospects for intergenerational transfers are much more pronounced in the public than in the private sector.

Despite the importance of out-of-pocket expenditures for assessing prospects for retirement security, we know surprisingly little about their characteristics from current data sources. Covinsky et al. (1994), for example, reported that of 2,661 seriously ill patients who survived their hospital stay, 31 percent lost all or most of their savings as a consequence of their illness. However, this is a very specific sample of ill patients, which makes inferences about the prevalence of catastrophic expenses difficult. The NMES is cross-sectional, and cannot provide a good picture of how these costs evolve over time. For example, one might

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

expect that out-of-pocket health expenses would pose a greater problem to future retiring baby boomers if catastrophic costs tended to be persistent. In other words, if households are randomly struck with out-of-pocket expenses equal to 20 percent of income, they can smooth consumption by drawing down assets or other contingencies. However, if only a few households are subject to 20 percent health expenses in every year, then these families would be subject to considerable financial distress. But there are few if any longitudinal studies of out-of-pocket health expenses. Feenberg and Skinner (1994) did use truncated panel tax data to infer the time series properties of out-of-pocket expenses in 1968–1973, but their study was limited to people who itemized in each year. More recent longitudinal data with detailed out-of-pocket expenses would be extremely useful.

Another problem with existing data is the weak link between nursing home and noninstitutionalized cost data. Both the 1977 and 1987 surveys, for example, exclude all nursing home costs from their measure of out-of-pocket expenses (although the 1987 study did make some efforts to statistically link the nursing home data with the noninstitutionalized data). Hence there is no data source that provides a general picture of the overall risk in out-of-pocket health care costs.

The Distribution of Retirement Income

Most of the estimates for future health status and health costs have focused on aggregate or average levels. Few have accounted for the fraction of people who would be deemed disadvantaged because of poor health and low levels of income and wealth. Wiener, Illston, and Hanley (1994) are notable for providing measures of the number of nursing home patients expected to be receiving Medicaid or the percentage incurring very high costs relative to income. They can provide these projections because their model simulates a large number of "people" through disability and income generators, allowing a detailed description of the distribution of health and income realizations.

Some additional factors may be necessary to capture changes in the distribution of retirement income security. Projections typically assume a homogeneous growth in wage rates among all individuals. Yet wage growth rates of lower educated people are falling behind the average growth rates, and in many cases are even negative in real terms (see Levy and Murnane, 1992). It is likely that this group would face a very difficult retirement in the face of rising health care costs. The government Medicaid burden might also be substantially larger than forecast, given the likelihood of many households falling short of a sufficient level of income and wealth to support nursing home costs.

Similarly, socioeconomic status tends to be strongly correlated with levels of disability. For example, Figure 6-7, taken from House et al. (1990), shows the number of chronic conditions as a function of age for the lowest and highest socioeconomic class. At ages 55 to 64, for example, the average number of

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

FIGURE 6-7 Average number of chronic conditions by socioeconomic status and age.

SOURCE: House et al. (1990:App. A).

chronic conditions for the upper socioeconomic status group is less than half the number of conditions for the lower socioeconomic group. Of course, some part of this correlation may be the consequence of poor health conditions reducing earning capacity (Preston and Taubman, 1994). The authors attempted to partially correct for this by requiring both low income and low education attainment for the lowest socioeconomic status, and both high income and high education for the highest socioeconomic status. Still, disentangling the correlation between poor health levels and income is an important research topic that can be potentially addressed with longitudinal data on past earnings and current disability patterns.

Lower income households tend to face substantially higher levels of disability, laggard growth in income, and low levels of wealth relative to their income (Hubbard, Skinner, and Zeldes, 1995). This is the group most likely at risk from higher levels of out-of-pocket health care costs or cutbacks in Social Security payments. However, there are no projections for this group that account for the correlation among lower wages, poor health, lower wealth levels, and lower life expectancy (Preston and Taubman, 1994).

Another shortcoming of most projections involving health care costs is that either assets are assumed to grow exogenously (Wiener, Illston, and Hanley, 1994) or saving rates are assumed to be held constant (Auerbach and Kotlikoff, 1994). Yet changes in the composition and magnitude of health care costs (particularly those out of pocket) may be expected to affect the accumulation behavior of households. For example, in Hubbard, Skinner, and Zeldes (1994, 1995), uncertainty about out-of-pocket health costs causes most people to save more in response to additional risk. In their model, however, those with a reasonable

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

chance of becoming eligible for Medicaid in the event of a bad health shock (or Aid to Families with Dependent Children [AFDC] in the event of a bad shock to earnings) save less because of asset-based means testing. Because welfare programs typically restrict eligibility to people with less than $3,000 in assets, any personal saving above that amount is effectively taxed at a 100 percent rate in the event an individual requires long-term Medicaid or AFDC assistance. In their simulation model, such asset-based means testing programs reduce saving for lower income households.

In sum, projections about retirement security often focus on average levels of health care expenditures within quite broad demographic groups. However, the design of appropriate public policy is probably more concerned with the retirement outcomes of a specified group of people, perhaps those in the bottom quintile or quartile of the income distribution. Hence developing methods for predicting the financial security of specific demographic groups may be beneficial for policy purposes.

It may be possible to learn more about the correlation among saving, income growth, disability, and health care use with the ongoing project to match up data from the Panel Study of Income Dynamics with Medicare data. HRS and AHEAD, with their linked data of health, assets, and income, can provide a better picture of the group of elderly who are likely to receive the lowest levels of income and assets, and experience (perhaps) the greatest levels of disability.

CONCLUSION

The economic well-being of the baby boom generation during retirement will depend crucially on the evolution of mortality rates, disability, and health care costs. Whether longer life expectancy will strain Social Security and private pensions, or whether rising health care costs and many frail elderly will place a large burden on the economy, would be useful to know now, when there is still time to prepare for potentially large government and private expenses in the future. In this paper, we have examined a number of issues related to predicting health and mortality in the next century and have identified why different approaches to prediction or estimation have sometimes yielded such different estimates. In some cases, the controversies about predictions can be addressed by better use of existing data sources or by additional data collection.

An underlying theme of this paper is that many of the real questions about retirement income security cannot be answered simply by collecting more data or by running more complex estimation procedures. In the case of mortality, the crucial question is whether the observed trend in mortality rate reductions will be sustained for the next 50 years. By the same token, the fundamental question about health care costs is whether cost increases from the past 30 years might be expected to persist until the year 2030. Determining the answers to these ques-

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

tions is difficult and requires conceptual and theoretical advances as well as more and better data.

NOTES

1.  

However, these dates were presumably chosen not randomly but to maximize the variation, and so the marked contrasts in rates should be interpreted with caution.

2.  

In the forecast for low gains in life expectancy, basic death rates are assumed to stay constant at their 1990 levels, and these are increased by AIDS death rates, which are assumed to increase linearly to 2005 and then to remain constant throughout the forecast period (Bureau of the Census, 1993:xi). Therefore the impact of pessimistically projected future increases in AIDS mortality on life expectancy can be calculated by comparing this projected life expectancy with that in 1990.

3.  

These are based on personal communication with Bert Kestenbaum.

4.  

Individuals, of course, might live longer but the average across individuals would be bounded in this way.

5.  

Kaplan (1991), in a study of Alameda County, concluded that age-specific prevalences of some chronic conditions might be expected to rise over time.

6.  

Zedlewski and McBride (1992) predict 3 million nursing home residents in 2010 and 4.3 million in 2030, so a midpoint of about 3.6 million in 2020 would be roughly consistent with the estimates in the text.

7.  

Obviously, average levels of disability in nursing homes are higher than levels among the disabled in the community. Still, there is a 35 percent chance that a widow with two children and 5 to 6 activities-of-daily-living limitations will have as her primary caregiver one of her children (Soldo and Freedman, 1994).

8.  

Guralnik (1991) shows that people tend to be more impaired in the 2 or 3 years before death and that the costs of such impairment of those about to die increase with age. Therefore, as mortality declines and deaths are increasingly shifted toward older ages, the costs arising from this impairment prior to death will also tend to rise. Note, however, that in the transition from one mortality regime to another, there will be a countervailing tendency for deaths to drop below their steady-state value and therefore for costs to diminish.

9.  

Also see National Research Council (1988) for a detailed discussion of data sources and limitations.

10.  

Research by Alan Garber and Thomas MaCurdy has focused on combining data from the NLTCS and the National Nursing Home Survey, although the units of observation in the two samples are obviously different.

11.  

We focus here on direct medical costs. A more general definition of costs would include the indirect costs of lost work and informal caregiving.

12.  

Since expenditures are price multiplied by quantity, one can make inferences about the average use of health care services among demographic groups from cost data if one is willing to assume that the price ''indexes" facing the different demographic groups are similar. In other words, when we use cost data to make inferences about relative health services use, we are making the assumption that if people aged 65 and above have expenditures that are twice those of people under age 65, this older group is receiving twice the real quantity of health care services.

13.  

Utilization per capita apparently comprises both changes in visits per person in a given age group and changes in the overall distribution of different ages.

14.  

Some part of the increase in relative prices can also be traced to the increased demand for real health services as the age composition of the population changes.

15.  

As Getzen notes, variation in increases in health care costs across countries may have swamped the effects that were due to population changes alone.

16.  

Of course, even an increase of 8 cents per dollar of lifetime income (37 cents versus 45 cents)

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

   

is a substantial increase in the tax burden of future generations and is caused largely by the pay-as-you-go nature of the current Social Security and Medicare programs (see Kotlikoff, 1992).

17.  

Analysis by the Census Bureau of the past performance of its forecasts, along lines pioneered by Stoto, (1983), does give some idea of the probability coverage to assign to the forecasts of population size, but these cannot be applied to any other demographic measure except population growth rates, and certainly not to dependency ratios.

18.  

Because some cancellation of variations will occur over the 5-year interval, this probability interval will be narrower in percentage terms than the corresponding interval for a single year or for a point in time.

19.  

Of course, other government spending could be cut to pay for the higher share of government health spending. But were the share of government health spending lower, taxes could also be lower, if other government spending programs were held constant.

REFERENCES

Ahlburg, D., and J. Vaupel 1990. Alternative projections of the U.S. population. Demography 27(4):639–652.

Alho, J.M. 1992. The Magnitude of error due to different vital processes in population forecasts. Pp. 301–314 in D. Ahlburg and K. Land, eds., Population Forecasting, a special issue of The International Journal of Forecasting 8(3).

Alho, J.M., and B.D. Spencer 1985. Uncertain population forecasting. Journal of the American Statistical Association 80(390):306–314.

1990. Error models for official mortality forecasts. Journal of the American Statistical Association 85:609–616.

Alter, G. 1990. Old Age Mortality and Age Misreporting in the United States, 1900–1940. Paper presented at the annual meetings of the Population Association of America, Toronto, Ontario.

Auerbach, A.J. and L.J. Kotlikoff 1994. The United States' Fiscal and Saving Crises and Their Implications for the Baby Boom Generation. Report to Merrill Lynch & Co. (February).


Bennett, N.G., and S.J. Olshansky 1994. The Impact of Adjustments to Official Mortality Schedules on Projected Age Structure and the Future of Age-Entitlement Programs in the United States. Unpublished manuscript. Department of Sociology, Yale University, New Haven, Conn.

Berg, R.L., F.E. Browning, J.G. Hill, and W. Wenkert 1970. Assessing the health needs of the aged. Health Services Research 5(Spring):327–331.

Bernheim, B.D. 1994. The Merrill Lynch Baby Boom Retirement Index. Merrill Lynch & Co.

Bloom, D., and S. Glied 1992. Projecting the number of new AIDS cases in the United States. Pp. 339–366 in D. Ahlburg and K. Land, eds., Population Forecasting, A Special Issue of the International Journal of Forecasting 8(3).

Bureau of Census 1993. Population Projections of the United States, by Age, Sex, Race, and Hispanic Origin: 1993–2050. Current Population Reports, P-25-1104, by Jennifer Cheeseman Day. Washington, D.C.: U.S. Department of Commerce.

Burner, S.T., D.R. Waldo, and D.R. McKusick 1992. National health expenditures projections through 2030. Health Care Financing Review 14(1):1–29.

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

Campbell, P.R. 1991. Projected AIDS-Age-Sex-Race/Ethnicity Specific Death Rates for the Period 1990 to 2005. Unpublished paper. The Population Projection Branch, U.S. Bureau of the Census, Washington, D.C.

Chulis, G.S., F.J. Eppig, M.O. Hogan, D.R. Waldo, and R.H. Arnett 1993. Health insurance and the elderly: Data from MCBS. Health Care Financing Review 14(3):163–181.

Coale, A.J., and E.E. Kisker 1990. Defects in data on old-age mortality in the United States: New procedures for calculating mortality schedules and life tables at the highest ages. Asian and Pacific Population Forum 4(1):1–32.

Costa, D.L. 1994. Health and Labor Force Participation of Older Men, 1900–1991. NBER Working Paper #4929. Cambridge, Mass.: National Bureau of Economic Research.

Covinsky, K.E., L. Goldman, E.F. Cook, R. Oye, N. Desbiens, D. Reding, W. Fulkerson, A.F. Connors, Jr., J. Lynn, and R.S. Phillips 1994. The impact of serious illness on patients' families. SUPPORT investigators. Study to understand prognoses and preferences for outcomes and risks of treatment. JAMA 272(23):1839–1844.


Davis, W. 1988. Calculation of the Variance of Population Forecasts. Bureau of the Census, Statistical Research Division Report Series, SRD Research Report Number: Census/SRD/RR-88/20. Washington, D.C.: U.S. Department of Commerce.


Elo, I.T., and S.H. Preston 1994. Racial and Ethnic Differences in American Mortality at Older Ages. Paper prepared for the Workshop on Racial and Ethnic Differences in Late Life in the United States, Committee on Population, National Research Council.

Employee Benefit Research Institute 1995. The Future of Employment-Based Health Benefits. Special Report and Issue Brief No. 161. Washington, D.C.: Employee Benefit Research Institute.


Farrell, C. 1994. The economics of aging: Why the growing number of elderly won't bankrupt America. Business Week (September 12):60–68.

Feenberg, D., and J. Skinner 1994. The risk and duration of catastrophic health care expenditures. Review of Economics and Statistics 76(November):633–647.

Fogel, R.W., D.L. Costa, and J.M. Kim 1993. Secular Trends in the Distribution of Chronic Conditions and Disabilities at Young Adult and Late Ages, 1860–1988: Some Preliminary Findings. Unpublished manuscript. University of Chicago.

Fries, J.F. 1980. Aging, natural death, and the compression of morbidity. New England Journal of Medicine 303:130–136.

1989. The compression of morbidity: Near or far? The Milbank Quarterly 67:208–232.


Getzen, T.E. 1992. Population aging and the growth of health expenditures. Journal of Gerontology: Social Sciences 47(3):S98–S104.

Gomez de Leon, J. 1990. Empirical DEA Models to Fit and Project Time Series of Age-Specific Mortality Rates. Unpublished manuscript. National Council on Population (CONAPO), Mexico, D.F., Mexico.

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

Guralnik, J.M. 1991. Prospects for the compression of morbidity: The challenge posed by increasing disability in the years prior to death. Journal of Aging and Health 3(2):138–154.

Gustman, A.L., and T.L. Steinmeier 1986. A structural retirement model. Econometrica 54(3):555–584.


Haines, M. 1994. The Population of the United States, 1790–1920. Working Paper No. 56, Historical Factors in Long Run Growth. Cambridge, Mass.: National Bureau of Economic Research.

Himes, C.L. 1994. Modern patterns of mortality in Sweden, Japan, and the United States. Demography 31(4):633–650.

Himes, C.L., S.H. Preston, and G.A. Condran 1994. A relational model of mortality at older ages in low mortality countries. Population Studies 48(2):269–292.

Holahan, J., A. Dor, and S. Zuckerman 1990. Understanding the recent growth in Medicare physician expenditures. Journal of the American Medical Association 263(12):1658–1661.

House, J.S., R.C. Kessler, A.R. Herzog, R.P. Mero, A.M. Kinney, and M.J. Breslow 1990. Age, socioeconomic status, and health. The Milbank Quarterly 68(3):383–411.

Hubbard, R.G., J. Skinner, and S.P. Zeldes 1994. The importance of precautionary motives in explaining individual and aggregate saving. Carnegie-Rochester Conference Series on Public Policy 40(June):59–126.

1995. Social insurance and precautionary saving. Journal of Political Economy 103(2):360–399.


Kannisto, V. 1994. Development of Oldest-Old Mortality, 1950–1990: Evidence From 28 Developed Countries. Odense Monographs on Population Aging 1. Odense University Press.

Kaplan, G.A. 1991. Epidemiologic observations on the compression of morbidity. Journal of Aging and Health 3(2):155–171.

Kestenbaum, B. 1992. A description of the extreme aged population based on improved Medicare enrollment data. Demography 29(4):565–580.

Kotlikoff, L.J. 1992. Generational Accounting. New York: The Free Press.


Lee, R.D. 1993. Modeling and forecasting the time series of U.S. fertility: Age distribution, range and ultimate level. International Journal of Forecasting 9(2):187–202.

Lee, R.D., and L. Carter 1992. Modeling and forecasting the time series of U.S. mortality. Journal of the American Statistical Association 87(419):659–671.

Lee, R.D., and S. Tuljapurkar 1994. Stochastic population forecasts for the U.S.: Beyond high, medium and low. Journal of the American Statistical Association 89(428):1175–1189.

Levy, F., and R.J. Murnane 1992. U.S. earnings levels and earnings inequality: A review of recent trends and proposed explanations. Journal of Economic Literature 30(3):1333–1381.

Liebson, C.L., D.J. Ballard, J.P. Whisnant, and L.J. Melton 1992. The compression of morbidity hypothesis: Promise and pitfalls of using record-linked data bases to assess secular trends in morbidity and mortality. The Milbank Quarterly 70(1):127–154.

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

Lubitz, J., and R. Prihoda 1984. The use and costs of Medicare services in the last 2 years of life. Health Care Financing Review 5(3):117–131.


Manton, K.G., L.S. Corder, and E. Stallard 1993. Estimates of change in chronic disability and institutional incidence and prevalence rates in the U.S. elderly population from the 1982, 1984, and 1989 National Long Term Care Survey. Journals of Gerontology 48(4):S153–S166.

Manton, K.G., and E. Stallard 1994. Medical demography: Interaction of disability dynamics and mortality. Pp. 217–278 in L.G. Martin and S.H. Preston, eds., Demography of Aging . Washington, D.C.: National Academy Press.

Manton, K.G., E. Stallard, and K. Liu 1993a. Forecasts of active life expectancy: Policy and fiscal implications. The Journals of Gerontology 48(Special Issue):11–26.

1993b. Frailty and forecasts of active life expectancy in the United States. In K.G. Manton, B.H. Singer, and R.M. Suzman, eds., Forecasting the Health of Elderly Populations. New York: Springer-Verlag.

Manton, K.G., E. Stallard, and H.D. Tolley 1991. Limits to human life expectancy: Evidence, prospects, and implications. Population and Development Review 17(4):603–638.

McClellan, M. 1994. Why Do Hospital Costs Keep Rising? Hospital Reimbursement, Hospital Production, and the Puzzles of Medicare's Prospective Payment System. Unpublished manuscript.

McCoy, J.L., H.M. Iams, M.D. Packard, and J. Shapiro 1992. Health of retired workers: Survival status and Medicare service use. Health Care Financing Review 13(3):65–76.

McNown, R., and A. Rogers 1989. Forecasting mortality: A parameterized time series approach. Demography 26(4):645–660.

1992. Forecasting cause-specific mortality using time series methods. International Journal of Forecasting 8(3)413–432.

McNown, R., A. Rogers, and J. Little 1995. Simplicity and complexity in extrapolative population forecasting models. Mathematical Population Studies 5(3):235–257.


National Research Council 1988. The Aging Population in the Twenty-First Century: Statistics for Health Policy. Washington, D.C.: National Academy of Sciences.

Newhouse, J. 1992. Medical care costs: How much welfare loss? Journal of Economic Perspectives 6(3):3–22.


Olshansky, S.J., B.A. Carnes, and C. Cassel 1990. In search of Methuselah: Estimating the upper limits to human longevity. Science 250:634–640.


Pflaumer, P. 1988. Confidence intervals for population projections based on Monte Carlo methods. International Journal of Forecasting 4:135–142.

Preston, S.H., I.T. Elo, I. Rosenwaike, and M. Hill 1994. African American Mortality at Older Ages. Results of a Matching Study. Unpublished Manuscript. Population Studies Center, University of Pennsylvania.

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

Preston, S.H., and P. Taubman 1994. Socioeconomic differences in adult mortality and health status. Pp. 279–318 in L.G. Martin and S.H. Preston, eds., Demography of Aging. Washington, D.C.: National Academy Press.


Schneider, E.L., and J.M Guralnik 1990. The aging of America: Impact on health care costs. Journal of the American Medical Association 263(17).

Social Security Administration 1992. Life Tables for the United States Social Security Area, 1990–2080. Actuarial Study No. 107, SSA Pub. No. 11-11536. Washington, D.C.: U.S. Department of Health and Human Services.

Soldo, B.J., and V.A. Freedman 1994. Care of the elderly: Division of labor among the family, market, and state. Pp. 195–216 in L.G. Martin and S.H. Preston, eds., Demography of Aging. Washington, D.C.: National Academy of Press.

Stoto, M. 1983. The accuracy of population projections. Journal of the American Statistical Association 78(381):13–20.

Stoto, M.A., and J.S. Durch 1993. Forecasting survival, health, and disability: Report on a workshop. Population and Development Review 19(3):557–581.


Taylor, A., and J. Banthin 1994. Changes in Out-of-Pocket Expenditures for Personal Health Services: 1977 and 1987 . AHCPR No. 94-0065, National Medical Expenditure Survey Research Findings 21. Agency for Health Care Policy and Research, Public Health Service. Rockville, Md.: U.S. Department of Health and Human Services.

Temkin-Greener, H., M.A. Meiners, E.A. Petty, and J.S. Szydlowski 1992. The use and cost of health services prior to death: A comparison of the Medicare-only and the Medicare-Medicaid elderly population. The Milbank Quarterly 70(4):679–701.

Tuljapurkar, S. 1990. Population dynamics in variable environments. Lecture notes in Biomathematics (85) Springer-Verlag.

1992. Stochastic population forecasts and their uses. Pp. 385–392 in D. Ahlburg and K. Land., eds., Population Forecasting, A Special Issue of the International Journal of Forecasting 8(3).


U.S. Congress, House 1994. 1994 Green Book: Overview of Entitlement Programs. Committee on Ways and Means, WMCP 103-27. Washington, D.C.: U.S. Government Printing Office.


Waidmann, T., J. Bound, and M. Schoenbaum 1995. The Illusion of Failure: Trends in the Self-Reported Health of the U.S. Elderly. NBER Working Paper #5017. Cambridge, Mass.: National Bureau of Economic Research.

Warner, H.R. 1993. Molecular biological approaches to understanding aging and senescence. Pp. 307–365 in K.G. Manton, B.H. Singer, and R.M. Suzman, eds., Forecasting the Health of Elderly Populations. New York: Springer-Verlag.

Wiener, J.M., L.H. Illston, and R.J. Hanley 1994. Sharing the Burden: Strategies for Public and Private Long-Term Care Insurance. Washington, D.C.: The Brookings Institution.

Wilmoth, J.R., and S. Preston 1994. Are Mortality Projections Always More Pessimistic When Disaggregated by Cause of Death? Paper presented at the 1994 Annual Meetings of the Population Association of America, Miami, Fla.

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×

Wolfe, B.L., and R. Haveman 1990. Trends in the prevalence of work disability from 1962 to 1984, and their correlates. The Milbank Quarterly 68(1):53–80.

Wolfe, P.R., and D.W. Moran 1993. Global budgeting in the OECD countries. Health Care Financing Review 14(3):55–76.


Zedlewski, S.R., and T.D. McBride 1992. The changing profile of the elderly: Effects on future long-term care needs and financing. The Milbank Quarterly 70(2):247–275.

Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 195
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 196
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 197
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 198
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 199
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 200
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 201
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 202
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 203
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 204
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 205
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 206
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 207
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 208
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 209
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 210
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 211
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 212
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 213
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 214
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 215
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 216
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 217
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 218
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 219
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 220
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 221
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 222
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 223
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 224
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 225
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 226
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 227
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 228
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 229
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 230
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 231
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 232
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 233
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 234
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 235
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 236
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 237
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 238
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 239
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 240
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 241
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 242
Suggested Citation:"6 ASSESSING FORECASTS OF MORTALITY, HEALTH STATUS, AND HEALTH COSTS DURING BABY BOOMERS' RETIREMENT." National Research Council. 1996. Assessing Knowledge of Retirement Behavior. Washington, DC: The National Academies Press. doi: 10.17226/5367.
×
Page 243
Next: 7 A FRAMEWORK FOR ANALYZING FUTURE RETIREMENT INCOME SECURITY »
Assessing Knowledge of Retirement Behavior Get This Book
×
Buy Paperback | $55.00 Buy Ebook | $43.99
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

This book brings together in one volume what researchers have learned about workers, employers, and retirees that is important for formulating retirement income policies. As the U.S. population ages, there is increasing uncertainty about the solvency of the Social Security and Medicare systems and the adequacy of private pensions to provide for people's retirement needs. The volume covers such critical behaviors as workers' decisions to retire, people's choices of saving over consumption, and employers' decisions about hiring older workers and providing pension and health care benefits. Also covered are trends in mortality, health status, and health care costs that are key to projecting the likely costs and effects of alternative retirement income security policies and a strategy for combining data and research knowledge into a policy modeling framework.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    Switch between the Original Pages, where you can read the report as it appeared in print, and Text Pages for the web version, where you can highlight and search the text.

    « Back Next »
  6. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  7. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  8. ×

    View our suggested citation for this chapter.

    « Back Next »
  9. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!