Lifetime Income, Longevity and Social Security Progressivity

Studies | Retirement

No. 342
Tuesday, December 11, 2012
by Liqun Liu, Andrew J. Rettenmaier, and Thomas R. Saving

Life expectancy has increased for most Americans in general over time, and life expectancy at retirement has been rising for decades. This has resulted in lengthening periods of retirement and higher associated spending on Social Security and Medicare. However, the longevity gains are significantly larger for men with relatively high lifetime earnings than those with relatively low lifetime earnings.

Executive Summary

Life expectancy at retirement has been rising for decades. This has resulted in lengthening periods of retirement and higher spending on Social Security and Medicare.

Life expectancy has increased for both the rich and poor, but the longevity gap between them has widened over time. This study relates county-level life expectancy estimates at the age of 65 to average Social Security benefits for retired workers. Social Security benefits are used as a proxy for the retirees’ relative lifetime earnings. Because the Social Security benefit formula is based on workers’ highest 35 years of indexed earning, average benefits can reasonably be used to estimate the relative lifetime earnings of retirees in each county.

Social Security benefits are used to rank counties’ retirees as lower or higher in the distribution of lifetime income. The average life expectancy at age 65 is then calculated for the men and women in the counties found to be at the lower and upper end of the income distribution. For example:

Between 1970 and 2009, life expectancy for men at the age of 65 rose about 32 percent in counties around the 10th percentile, or low end, of lifetime income distribution.

In contrast, life expectancy for men at the age of 65 rose about 43 percent in counties around the 90th percentile, or high end, of lifetime income distribution.

Thus, by 2009, men in high income counties lived 1.1 years longer on average than the men in low income counties.

The difference in 2009 is smaller for women as is the increase in life expectancy from 1970 to 2009 across all women.

These findings based on county-level income and life expectancy data are  consistent with other studies:  life expectancy at age 65 has grown over time for all retirees, has grown more rapidly for men than for women, and women’s life expectancy continues to exceed men’s. Further, life expectancy at the higher end of the lifetime earnings distribution has grown more rapidly than at the lower end of the distribution — more so for men than for women.

Given that life expectancy is growing more rapidly at the high end of the income distribution that at the low end, the question then is, should Social Security be reformed such that the full retirement age is higher for high income workers than for lower income workers, as some have suggested?

To address this question, the study next analyzes the degree to which the program’s progressivity is affected by differential mortality. For this part of the study, individual level earnings and Social Security benefit data are combined with published longevity estimates for men born in different years who are in the top and the bottom half of the earnings distribution. The study finds that once longevity differences are accounted for, the program’s progressivity is lessened, relative to estimates based on average longevity estimates by birth year. However, even for the most recent group of new retirees analyzed, the program remains progressive. Further, the study finds that within birth years, the program redistributes from high to low earning workers, after accounting for income-related differences in longevity.

Income-related differences in longevity must be considered in framing reforms given that the relative progressivity of Social Security is lessening through time. Reforms that reduce future system spending are necessary due to the system’s persistent and growing cash flow deficits. Increasing program revenues is difficult at this time due to forecast government-wide deficits. Combining an increase in the retirement age for all workers, given the general increase in longevity, with an adjustment to the benefit formula that increases the progressivity, given the income-related relative increase in longevity, would result in a program that could be financed in the long run at the current tax rate and would retain the program’s relative lifetime progressivity. Income-related, or progressive, retirement ages would be difficult (or impossible) to implement given the difficulty in determining which groups of workers would be eligible for the progressive treatment.

Increasing the progressivity of the benefit formula, on the other hand, would allow an explicit change based solely on a retiree’s lifetime earnings. Progressive price indexing is a reform by which low-income workers’ past earnings continue to be indexed by average wages whereas high income workers earnings would be indexed by price level changes. Progressive price indexing would further address concerns about the growth in total benefit payments while retaining and perhaps enhancing the system’s progressivity.

Introduction

Life expectancy has increased for most Americans in general over time, and life expectancy at retirement has been rising for decades. This has resulted in lengthening periods of retirement and higher associated spending on Social Security and Medicare. However, in a frequently referenced study in the Social Security Bulletin, Hilary Waldron found that the longevity gap between the rich and poor has widened. In other words, the longevity gains are significantly larger for men with relatively high lifetime earnings than those with relatively low lifetime earnings. For example, consider two representative male Social Security-covered workers with average earnings in the top half and the bottom half of the earnings distribution, respectively. The representative upper income worker, conditional on survival to age 60, would be expected to live 1.2 years more than the representative lower income worker for the 1912 birth cohort,18.9 versus 17.7 years, but the gap of 1.2 years increases to 5.8 years for the 1941 birth cohort, 25.4 versus 19.6 years.1 

        The widening gap in longevity between high and low income workers has important implications for reforms that aim to restore the fiscal soundness of elderly retirement programs such as Social Security and Medicare, through reducing the projected growth in their expenditures. One specific Social Security reform that has been discussed is further increasing the normal retirement age at which workers can retire and receive full retirement benefits. Because the worsening financing gap in the Social Security program is partially caused by increasing population longevity, the widening longevity gap raises the question of whether the full retirement age should be raised more for high-earnings workers than for low-earnings workers.2 Closely related is the question of whether the progressivity of the Social Security program is reversed when the widening longevity gap is used to calculate net lifetime Social Security benefits for various earnings groups.3

        This study uses county-level data on mortality and various measures of lifetime earnings to investigate the relationship between longevity and lifetime earnings and how this relationship has changed over time.4  While this exercise confirms that there is a definite trend of a widening longevity gap between high and low earning males, but it also reveals that the income-related gap for females is smaller and has only  grown in recent years. This suggests that the dynamics of the income-longevity relationship are dissimilar across genders, making a reform that indexes the full retirement age uniformly to lifetime earnings difficult to implement for the general population.5

We then use earnings-specific longevity estimates to calculate net lifetime benefits from Social Security for various earnings groups. The program continues to be a progressive, but it has become less progressive for recent retirees. Finally, we compare alternative policy options that could make the program more progressive. Specifically, we compare the proposal to make the full retirement age earnings-specific with the one that would make benefit formula even more progressive (the progressive pricing index option) but would maintain a uniform full retirement age.

        We conclude that the progressive pricing index option performs reasonably well in increasing/maintaining the progressivity of the Social Security program, is a less radical change than the earnings-specific normal retirement age option, and is easier to implement than the latter change. Moreover, making the full retirement age dependent on a group’s average longevity may have unwelcome ethical and political consequences. For example, while high-earnings individuals on average live longer and give rise to the suggested higher earnings-specific normal retirement age, some high-earnings individuals may have known health issues and expect to die before their expected longevity based on their earnings.. On ethical grounds, it would be hard to justify preventing high-income individuals in ill-health from receiving full retirement benefits at the same age as their healthier, lower earnings counterparts. Politically, implementing an earnings-specific normal retirement age could open the floodgate for claims of favorable group-specific normal retirement ages from certain demographic groups that are known to have shorter longevity based on factors such as sex, race, smoking status for example.

County-Level Income and Longevity

Numerous studies have identified a positive relationship between income and life expectancy, whether based on annual income or lifetime income. Annual income, particularly at early ages and the ages just before retirement, may underestimate lifetime income. At young ages, lower income may reflect investments in education while lower income at older ages may indicate early retirement due to declining health. The use of lifetime earnings is often suggested to overcome these kinds of concerns.

 However, even with lifetime earnings, the causal relationship between income and longevity can be difficult to determine. A positive relationship between lifetime income and life expectancy could indicate that health is a normal good and as income rises greater investments are made in health care and fitness. Alternatively, individuals in better health, or whose own assessment of their longevity is higher, may invest more in education and have higher lifetime earnings as a result. This would also account for a positive relationship between longevity and income, though the causality is reversed. Further, lifetime income is only available for older individuals, limiting its use as an indicator of longevity to older ages. Regardless of the direction of causality, there is a positive relationship between lifetime income and longevity and there is growing evidence that relationship has become stronger over time.

Several other aspects of the growing gap in income-related longevity, particularly for men, deserve discussion. Namely, the increase in the gap between men with high and low incomes may well reflect the growing gap in lifetime incomes as we move from older to younger birth years.6  With a constant income elasticity of demand for life-extending health interventions, a growing gap in lifetime income would result in a widening longevity gap. Further, the longevity gap would also result if lifetime earnings have simply become a better signal of worker’s underlying longevity potential over time. Either way, when inferring whether Social Security has become more of less progressive it is important to recognize how changes in the wage structure may drive the observed reduction in the system’s progressivity.

County-Level Death Rate Data. In this section we explore the changing relationship between income and longevity using county-level data. The county-level mortality data are from the Compressed Mortality Files (CMF) provided by the Centers for Disease Control that have been used in a wide variety of studies.7  Our interest here is on life expectancy at the age of 65 which we estimate based on the death rate data.8  The death rate data is from the cross-section of county residents in each year and thus the life expectancy estimates are period life expectancies as opposed to cohort-level life expectancies. The period life expectancies assume that the death rates at the ages above 65 remain constant at the rates that exist in the year calculated. That is, they assume no improvement in mortality at later ages for members of a particular birth year.

In contrast to county-level mortality data from which period life expectancies are estimated, cohort life tables project improvements in life expectancy as members of a birth year progress to higher ages. While cohort life tables are appropriate if the analysis is centered on individuals born in the same year, the period life expectancies used here are a better match for Social Security benefits, the primary income measure we use to proxy for lifetime earnings. The mortality data are matched to three measures of county-level income:  per capita income, nontransfer income per potential worker in the county and Social Security benefits per retiree. Per capita personal income and nontransfer income variables are from the Bureau of Economic Analysis (BEA).9  The Social Security benefits per retiree are from the Census Bureau.10

Social Security Benefits as a Proxy for Lifetime Income. Because Social Security benefits are based on workers’ highest 35 years of earning they are a good proxy of lifetime earnings. Sorting workers of the same age by their Social Security benefits results in an approximate ordering of the workers’ lifetime earnings, with a few caveats related to the timing of when they claimed their benefits. The county averages used here also provide an approximate ordering of retirees’ average lifetime incomes, but with additional caveats. Each county’s average Social Security benefit per retired worker is based on benefits for male and female workers, spans all retirees in the county and thus includes all birth cohorts, and is affected by the different ages at which retirees initially claimed their benefits. Even with these qualifications, Social Security benefits provide an approximate ordering of the counties by the lifetime earnings of the population for which we have mortality data.11 The other income metrics provide contemporaneous income for the broader population in the county and result in a different ranking of the counties.12

Figure I depicts the conditional county-level life expectancies at the age of 65 for men and women in counties at the 10th, 50th, and 90th percentiles of the distribution of retired workers’ average benefits. The percentile locations are determined by weighting the counties’ income and longevity statistics by the populations ages 65 and above.13 

Life Expectancy for Men. Consider the life expectancies for men in the counties at the 10th, 50th and 90th percentiles. Conditional life expectancy is rising across all percentiles, with life expectancy at the 90th percentile growing more rapidly than at the 10th percentile.

At the 90th percentile in the county-level data, life expectancy rose from 12.6 to 18.1 years, for an increase of 5.5 years.

However, at the 10th percentile, life expectancy rose from 12.7 years to 17.0 years for a gain of 4.3 years.

Thus, over the past four decades, male life expectancy at the age of 65 rose by 44 percent in the counties at the 90th percentile in the lifetime earnings distribution and by 34 percent in counties at the 10th percentile of the lifetime earnings distribution.

Life Expectancy for Women. Figure I also presents the life expectancies for women in the counties at the same locations in the Social Security benefits distribution. These are all higher than those for men, though male life expectancy at age 65 is rising more rapidly than for women. We also see that the differential between the series at the 90th and 10th percentiles is smaller for women and has not been uniformly positive for as many years as in the case for men.

Along with the series derived from the county-level data we also present the period life expectancies at age 65 from the 2012 Trustees Report. The series reflecting life expectancy at the 50th percentile are quite similar to the series for men and women from the 2012 Social Security Trustees Report.14 These series again confirm the relatively faster gains in longevity for men than for women over the past 40 years. Various explanations have been posited for the relatively faster growth in life expectancies for men at age 65 including the advances in health care interventions that address diseases with higher prevalence among men, such as the advances in treating heart disease. Other explanations suggest the less rapid growth in female life expectancy gains is a consequence of the potential adverse effects on longevity of increased labor market participation. But even with the faster gains for men, women’s period life expectancy in 2009 at age 65 of 19.8 years was 2.4 years, or 14 percent longer than men’s.

Table I compares the life expectancies at age 65 for men and women when the three different income measures are used to rank the counties and determine the percentiles. The contemporaneous income measures – per capita income and non-transfer income per potential worker – result in a different ranking than when the counties are ranked by average Social Security benefits. These different rankings produce slightly greater differentiation between the 90th and 10th percentiles for both men and women. The contemporaneous income measures are more indicative of the annual income of the working-age population in a county and while not as representative as Social Security benefits of retirees’ own resources, contemporaneous income proxies for community-wide resources.

The 90-10 differentials in county life expectancies for men and women when Social Security benefits are used to determine the locations in the income distribution are depicted in Figure II. Beginning in 1982 we see the differential begins to widen for men and women and again the differential for men is more pronounced than for women. The smaller income related life expectancy differentials at age 65 for women, using the aggregate county-level income, are consistent with some of the findings in other studies. While our average Social Security benefits for all retired workers in a county are multiple steps removed from the individual retiree’s Average Indexed Monthly Earnings (AIMEs) used by Courtney Monk, our qualitative results are consistent with the finding that women’s life expectancies are weakly associated with their own AIMEs and that when men and women are pooled the association is further diminished.15

This brief examination of county-level longevity gains is consistent with other studies that have been based on individual level-data. For example, we have seen that longevity gains for men conditional on reaching 65 years of age have been greater among those who had higher lifetime earnings than among those who had lower income. This has resulted in a growing differential in longevity. There is evidence of growth in the income-related longevity differential among women, but the differential in longevity between women residing in counties with high and low Social Security benefits is smaller than the differential in longevity among the men in those counties.

How the Longevity Gap Affects Social Security’s Progressivity

The preceding analysis of county-level data provides additional evidence that the longevity gains for men over the last four decades can be distinguished by lifetime income. Consequently, the progressivity of Social Security will be diminished.

How Much of a Difference Does Longevity Make? We use a large individually-based data set coupled with the differential longevity estimates reported in Waldron’s 2007 article.16 The individual level earnings and Social Security benefit data is from the 2004 Benefits and Earnings Public-Use File (BE-PUF),17 which includes earnings and benefit information on a random 1 percent of all Social Security beneficiaries as of December 2004. The BE-PUF is also used by Gopi Shah Goda, John B. Shoven and Sita Nataraj Slavov (2011) to examine the effects of differential morality on the progressivity of the Social Security program.18 They take a quite different empirical approach than we do. They analyze the outcomes that would result if the circa 2008 Social Security benefit structure and tax rate had been in place historically for all of the birth years they examine.

We examine the expected net benefits and money’s worth ratios based on the realized behavior of the men in the BE-PUF. Because we make use of Waldron’s life expectancy estimates we restrict the sample in the same way. Specifically, the sample is limited to men, and is further restricted based on earnings between the ages of 45 and 55. Waldron identifies life expectancy in the top and bottom half of the earnings distribution where an individual’s annual earnings between ages 45 and 55 are measured relative to the national average wage in each year. Only non-zero earnings are included in the average ratio for each worker, and thus, men who had no earnings between the ages of 45 and 55 are also dropped from the sample. We add one more restriction –that the retirees are entitled to their benefits as a retired worker based on their own earnings history. Our ultimate sample includes 130,029 men born between 1914 and 1939.19 

Our sample is limited to individuals who survive to 2004 and, thus, the older the cohort the greater is the degree to which the sample is biased toward survivors. The sample used by Waldron (2007) spans members of the birth-years 1912 to 1941 conditional of their reaching the 60 years of age. Waldron’s sample members are then tracked up to 2001 with the life expectancy estimates in the top and the bottom halves of the earnings distribution based on the relative likelihoods of survival to ages above 60. We utilize Waldron’s life expectancy estimates for members of each birth cohort, conditional on surviving to their respective ages in 2004.20 

With these life expectancy estimates we then calculate expected lifetime benefits for each retiree in our sample using their actual monthly benefit as reported in the BE-PUF. Benefits are assumed to start in the year each retiree first claims Social Security and end in the year 2004 plus the life expectancy conditional on the retiree’s age in 2004 that we estimated from Waldron’s study. The assigned life-expectancy depends on whether the retiree was in the top or the bottom half in the average income distribution (spanning income between the ages of 45 and 55) as defined in Waldron’s study. We also estimate each retiree’s lifetime benefits assuming they survive to the cohort specific average conditional life expectancy in 2004, also derived from Waldron’s study. Benefits are converted to 2004 dollar using the CPI-W and a constant real discount rate of 3 percent is used to calculate the value of the benefits at the age of 65.

Lifetime cumulative taxes are also calculated for each retirees by applying the historical OASI tax rates to his earnings record. As with the benefit calculations, taxes are converted to real 2004 dollar and accumulated to the age of 65 for each retiree.21  Net lifetime Social Security benefits for the top and bottom halves of the income distribution assuming differential as well as the same life expectancies are depicted in Figure III. The series illustrate some common findings.

As expected, members of the early birth years are net beneficiaries of the Social Security program and the net benefits decline as the program matures with later participants receiving lower net benefits and ultimately negative net benefits, though negative net benefits are only found here for the retirees in the top half of this quite restrictive sample. However, the results based on the average life expectancy and averaged across the top and bottom halved of the income distribution result in negative net benefits beginning with the 1934 birth year assuming the 3 percent real discount rate which is consistent with many earlier studies and annually updated estimates provided by the Social Security Administration’s Office of the Actuary.22 The relatively rapid decline in the net benefits for birth years 1917 to 1921 is associated with the 1977 Social Security reforms that corrected the overindexation of benefits that were put in place five years earlier.23

That the net benefits in the top half of the income distribution are higher than the net benefits for workers in the bottom half for the early birth year and then lower for the later birth years is also of interest. This indicates that direction of redistribution toward lower income workers began with workers born in the 1920s, at least among the survivors represented in this restricted sample. The results for the restricted sample, particularly for the older cohorts are biased toward survivors.

Of primary interest for this study, though, is the degree to which differential life expectancies change the estimated net benefits when compared to the estimates using the same life expectancy for both halves of the income distribution. The two estimates of net benefits accruing to the retirees in the bottom half of the income distribution are about the same due to the similarity of the average life expectancies from the 2004 Trustees Report and Waldron’s estimated life expectancies for the retirees in the lower half of the distribution. The similarity results from the selection of a relatively healthier population. The net benefits for the top half of the distribution are however affected by the different life expectancy estimates.

When the same life expectancy is used for all retirees, the net benefits in the top half of the distribution are negative beginning with the 1933 birth year, but when the differential life expectancy estimate is used they are negative beginning with the 1934 birth year. For workers in the top half of the income distribution who are born in 1939, the negative net benefits are ($60,159) when the average life expectancy of 16.75 years is used but when the higher differential life expectancy of 21.05 years is used to calculate lifetime benefits, the negative net benefits are increased almost $40,000 to ($20,360). With the differential mortality adjustment the program continues to redistribute from higher income to lower income workers within a generation, but the magnitude is reduced substantially.

The net benefit estimates above indicate the direction of the “redistribution” in Social Security, while the “progressivity” is typically identified by the net lifetime tax rates experienced by individuals who had different lifetime earnings. The net lifetime tax rate (or net lifetime benefit rate) is identified as the ratio of net lifetime taxes (benefits) to lifetime earnings.

Money’s Worth Ratios. An alternative estimate of progressivity that provides the same relative ordering as net lifetime benefits (taxes) is the ratio of benefits to taxes paid or the money’s worth ratio. Figure IV depicts the money’s worth ratios again for the two halves of the income distribution using the same life expectancy for all retirees and using the differential life expectancies. The inferences based on the money’s worth ratios are similar to those from the net benefits. However, the money’s worth ratios indicate that the program has been progressive with higher money’s worth ratios for lower income than for higher income workers regardless of the life expectancy estimates. But again we see that once the differential mortality estimates are used the program’s progressivity is diminished relative to the uniform longevity estimates.

 For the youngest birth cohort born in 1939, the money’s worth ratio in the top half is about 0.81 when the same life expectancy is used for all retirees. This means that lifetime benefits are equal to 81 percent of the taxes paid by the workers in the top half. However, when the income adjusted life expectancy is used to calculate lifetime benefits for this birth year, the money’s worth ratio rises to 0.94. While still below 1, this compares to the money’s worth ratios for workers in the bottom half of the distribution of 1.05 and 1.08, based on differential and the same life expectancies, respectively.

Policy Options that Address the Effects of Differential Mortality on Progressivity

Social Security is already experiencing cash flow deficits that are expected to grow larger over time. Closing the financing gap in Social Security requires tax increases or benefit cuts or both. Because there are many competing demands for tax revenues due to the overall funding problems of the federal government and a higher tax level will further impede the efficiency and competitiveness of nation’s economy, benefit reductions for future retirees relative to the currently scheduled benefits will be necessary and are preferred to further payroll tax increases.

Reforms to increase the early and full retirement age have been frequently suggested as a way to reduce Social Security lifetime benefits. A relatively new policy question is whether the retirement age should be increased more for individuals with higher lifetime income since there is evidence that high income individuals generally live longer than low income individuals. The concern is that given differential mortality, uniform increases in the retirement age will reduce the expected retirement period of low income workers by a greater percent than the retirement period of high income workers.

 Our analysis in the preceding sections generates two major findings that have implications for answering this question, which leads to the general conclusion that a reform to index the full retirement age to lifetime earnings is hard to implement and even after considering the longevity advantage of the high earners, Social Security is still progressive in the preceding exercise due to the redistributive benefit formula.

The first major finding concerning the relationship between lifetime earnings and longevity suggests that this relationship is a complex one, influenced by factors such as gender, race and family structure.24 As a result, it is questionable to index the retirement age to individual’s lifetime earnings alone, disregarding the information about gender, race, family structure, and other aspects with longevity implications. For example, some high-earnings individuals may have known health issues and expect short retirement periods. It would be hard to justify on ethical grounds that they could not retire to receive full retirement benefits at the same time a healthier though lower earnings counterpart could. Indeed, implementing an earnings-specific retirement age may encourage other population groups – based on gender, race, health status – that have shorter life expectancies to request earlier full retirement ages.

The second major finding from is that while incorporating the widening longevity gap associated with income into analysis indeed reveals an advantage for the high income individuals, this advantage, up to this point in time has not reversed Social Security’s progressivity.25 By both of the two leading measures of income redistribution, net lifetime benefits and the money’s worth ratio, considering income-specific longevity does not change the fact that Social Security carries an implicit intragenerational redistribution from high lifetime earners to low lifetime earners.26 Specifically, as Figure III shows, for the bottom half of the income distribution of all birth cohorts before 1940, the net lifetime benefits are positive regardless of the life expectancy used. In contrast, for the top half of the income distribution, the net lifetime benefits are significantly lower than those for the bottom half for the recent retiring cohorts, with the net lifetime benefits for the top half becoming negative for the cohorts from 1934 to 1939 even when the differential life expectancies are used to calculate benefits.27 This means that, except for the early birth cohorts, Social Security redistributes intragenerationally from high to low income individuals. The picture becomes even clearer when comparing the money’s worth ratios. As Figure IV shows, for all the cohorts considered, the money’s worth ratio is higher for the bottom half of the income distribution than for the top half of the income distribution. That Social Security still redistributes from high to low income individuals even after taking into account the longevity advantage of the high income suggests the redistributive benefit formula continues to produce a progressive program.

However, reforms are likely to come that address the funding shortfalls. A combination of reforms has been suggested. Together, a general, across the board increase, in the full and early retirement age and progressive price indexing would address the growth in total benefit payments while retaining and perhaps enhancing the system’s progressivity. Progressive price indexing reforms basically retain the currently scheduled replacement rates for workers who have lower lifetime earnings but reduce the replacement rates for higher earners such that the highest earners’ benefits grow with price growth rather than the higher wage growth rate. The uniform increases in the early and full retirement ages avoid the problems associated with a “progressive” retirement age while progressive price indexing more precisely addresses the income related progressivity of the program.

Conclusion

As a group, recent retirees will enjoy longer periods of receiving Social Security benefits than their predecessors due to the gains in longevity at the age of 65. This is true even with the modest increases in the full retirement age that were part of the 1983 Social Security reforms. However, as we have seen, the gains in longevity at the higher end of the lifetime earnings distribution have outpaced the gain at the lower end of the distribution, particularly for men.

The first part of the paper examined 40 years of county-level mortality data at ages 65 and above. Consistent with other studies based on individual level data, we saw that longevity gains for men in counties with higher Social Security benefits among the retired outpace the gains in counties with lower Social Security benefits. The same relative gains for men were also found when the counties were ranked by the contemporaneous income measures of per capita income and nontransfer income per potential worker. The income related county-level longevity differential at age 65 is less pronounced for women.

In the second half of the paper we examined how differential morality affects Social Security’s progressivity. Pairing individual level earnings and realized benefits for men from the BE-PUF to Waldron’s (2007) estimates of differential mortality in the top and bottom halves of the income distribution, we find that the program’s redistribution from high to low earners grows as we move from early to more recent retirees when we use either the income-related life expectancies or the uniform life-expectancies within birth years. However, the estimated redistribution is significantly reduced when the income-related mortality estimates are used. In terms of progressivity, we see that progressivity declines when the income-related mortality estimates are used to calculate lifetime benefits. However, the program, at least up to the 1939 birth cohort, remains progressive for men in this particular exercise limited to considering only retirement benefits and OASI taxes for survivors.

The growing gap in longevity gains related to lifetime income has led some observers to suggest tying the full retirement age to lifetime incomes. As we have suggested, such a change would be problematic on numerous fronts. Further, Social Security’s expenditures now exceed its revenues and the annual deficits are projected to grow substantially in all future years. A uniform increase in the early and full retirement age indexed to future longevity gains combined with progressive price indexing can be used to accomplish the dual goals of reducing future expenditure growth while at the same time enhancing the program’s progressivity.

Endnotes

1. See Hilary Waldron, “Trends in Mortality Differentials and Life Expectancy for Male Social Security-Covered Workers, by Socioeconomic Status,” Social Security Bulletin, Vol. 67, No. 3, 2007, pages 1-28. The widening longevity gap between income groups is consistent with the observed increases in longevity gap between the well-educated and the less-educated. See Ellen Meara, Seth Richards, and David M. Cutler, “The Gap Gets Bigger: Changes in Mortality and Life Expectancy, by Education, 1981-2000,” Health Affairs, Vol. 27, No. 2, March/April 2008. Also see S. Jay Olshansky et al., “Differences in Life Expectancy Due to Race and Educational Differences Are Widening, and Many May Not Catch Up,” Health Affairs, Vol. 21, No. 8, August 2012.

2. For such a proposal of using earnings-specific normal retirement age to enhance the progressivity of Social Security, see Courtney Monk, John A. Turner and Natalia A. Zhivan, “Adjusting Social Security for Increasing Life Expectancy: Effects on Progressivity,” Center for Retirement Research at Boston College, CRR Working Paper  2010-9, 2010.

3. Julia Lynn Coronado, Don Fullerton and Thomas Glass, in “The Progressivity of Social Security,” The B.E. Journal of Economic Analysis and Policy, Vol. 11, Issue 1,  2011, investigate by how much of the progressivity in Social Security is reduced by incorporating mortality rates that differ by a range of lifetime income measures.

4. County-level mortality data are explored by Majid Ezzati et al, “The Reversal of Fortunes: Trends in County Mortality and Cross-County Mortality Disparities in the United States,PLos Medicine,” Vol. 5, Issue 4, April 2008, pages 557-568. They find that, between 1983 and 1999, mortality inequality across the U.S. counties steadily increased. They focus on the contributions of specific diseases and consider the changing distribution county-level mortality in general — not conditioned on income. In contrast, the focus here is the relative levels of lifetime earnings.

5. At the conclusion of the study, we discuss policy options that account for the general increase in longevity and the changing nature of differential mortality. Courtney Monk et al. “Adjusting Social Security for Increasing Life Expectancy: Effects on Progressivity,” proposes adjusting the full retirement age by the maximum average indexed monthly earnings for married couples.

6. Wojciech Kopczuk, Emmanuel Saez and Jae Song identify the transitory and permanent components of changing inequality in “Earnings Inequality and Mobility in the United States: Evidence from Social Security Data Since 1937,” Quarterly Journal of Economics, February 2010, pages 91-128.

7. In addition to Majid Ezzati et al, (2008) a sampling of other studies using the Compressed Mortality Files (CMF) include Angus Deaton and Darren Lubotsky, “Mortality, Inequality and Race in American Cities and States,” Social Science and Medicine, Volume 56, 2003, pages 1,139-1,153 and Oliver Deschênes and Michael Greenstone, “Climate Change, Mortality, and Adaption: Evidence from Annual Fluctuations in Weather in the U.S.,” American Economic Journal: Applied Economics, Volume 3, October 2011, pages 152-185.

8. The CMFs are available at http://wonder.cdc.gov/mortSQL.html. United States Department of Health and Human Services (US DHHS), Centers for Disease Control and Prevention (CDC), National Center for Health Statistics (NCHS), Compressed Mortality File (CMF) on CDC WONDER Online Database. For the present analysis, the county-level deaths and population data for men and women 65 and above for the years 1968 to 2009 are used to estimate conditional life expectancy at age 65. A crosswalk between death rates and life expectancies at age 65 is derived from Social Security Administration’s period life tables for each year from 1968 to 2009. The Social Security Administrations’ period of life tables were consistent with the 2007 Trustees Report. Death rates in the age cells, 65 to 74, 75 to 85, and 85 and above are also available in the CMF, and their use would provide for a more precise life expectancy estimate, however, this would limit the available data due to restrictions on cell sizes in the more detailed age categories.

9. Total nontransfer personal income in each county is denominated by the population between the ages of 20 and 64 to arrive at nontransfer income per potential worker. The county age-group population data are from the Census Bureau’s county population estimates. http://www.census.gov/popest/data/historical/index.html. The BEA data series (CA1-3) and (CA35) are available at the following address: http://www.bea.gov/iTable/index_regional.cfm.

10. The Census Bureau’s county-level Social Security data are available from 1970 to 2010 and are archived at: http://www.census.gov/support/USACdataDownloads.html#SPR. The Social Security Administration also provides county-level benefit payment data and recipient counts for the years 1999 to 2011. These data include the preferred per capita benefits for men and women in each county who are 65 and above – the same population as mortality data, but given that we examine the income/mortality relationship back to 1969, the Census data is used. The Census data includes benefits paid per retiree (not for the 65 and above population exclusively) from 1990 to 2010 and as well as Old Age, Survivors and Disability Insurance (OASDI) payments per beneficiary back to 1970, but intermittently for the years 1970, 1975, 1980 and 1985). The pre-1990 data are limited to the per beneficiary OASDI payments. Per capita retiree benefits are estimated for the missing years based on within county regressions for the years 1990 to 2009 in which the dependent variable is the ratio of per capita retiree benefits to per capita OASDI benefits and the independent variable is the year. The predicated ratios are used to estimate benefits per retiree in 1970, 1975, 1980 and 1985 and then the intervening years are interpolated by county.

11. Ultimately, the counties defined by the BEA are those that are used here. Data for numerous independent cities in Virginia and counties are aggregated together to conform to the BEA definitions. Alaska and Hawaii are excluded from the current analysis. Also counties with missing morality data due to insufficient numbers, in any of the years between 1969 and 2009, are excluded.

12. For example, the population weighted correlation coefficient between the county average Social Security benefits of retired workers and per capita income in 2009 was 0.60. The correlation coefficient between average Social Security benefits and nontransfer income per potential workers was slightly higher are 0.63. The correlation coefficients have also declined over time.

13. The percentile life expectancies are actually averaged across the ten percentiles centered respectively on the 10th, the 50th, and 90th percentiles, with for example the average life expectancy from the 5th to the 15th presented as the 10th percentile.

14. The period life expectancies from the Social Security Administration are presented in 2012 Trustees Report, Table V. A3. Aggregate data from National Vital Statistics Reports, Vol. 59, No. 9, September 2011, indicate the same patterns for men and women.

15. See Courtney Monk et al. “Adjusting Social Security for Increasing Life Expectancy: Effects on Progressivity,” page 10. Also see James E. Duggan, Robert Gillingham and John S. Greenlees, “Mortality and Lifetime Income Evidence from U.S. Social Security Records,” International Monetary Fund Staff Papers, Vol. 55, No. 4, pages 566-594, 2008, for their decomposition of the relative longevity of women based on whether they were dual eligible based on their own and/or their spouse’s earnings records. Lastly, see Julian P. Cristia, “Rising Mortality and Life Expectancy Differentials by Lifetime Earnings in the United States,” Journal of Health Economics, Vol 28, 2009, pages 984-995, for additional results on the differences in the income mortality relationship for men and women over the years 1983 to 2003.

16. See Hilary Waldron, “Trends in Mortality Differentials and Life Expectancy for Male Social Security-Covered Workers, by Socioeconomic Status.”

17. The data files are available at: http://www.ssa.gov/policy/docs/microdata/earn/index.html.

18. Gopi Shah Goda, John B. Shoven and Sita Nataraj Slavov, “Differential Mortality by Income and Social Security Progressivity,” in David Wise, editor, Explorations in the Economics of Aging  University of Chicago Press, Chicago, IL, pages 189-208.

19. The BE-PUF includes annual earnings Social Security taxable earnings between 1951 and 2003 for each individual in the sample and a summary measure of average taxable earnings between 1937 and 1950. One issue that must be addressed in using these data is that earnings are capped at the taxable maximum each year affecting the reported earnings of a significant share of workers during the 1960s and 1970s ranging from 30 to as high as 60 percent of our sample. To calculate the average ratio of a worker’s earnings to the national average, earnings must be estimated for the workers who have capped wages. We use a two-stage procedure to estimate earnings above the taxable maximum. Censored normal regressions are run in each year with the explanatory variables that identify a worker’s earnings in surround years. The rankings of the relative predicted earnings above the taxable maximum are then matched to earnings above the taxable maximum from the Current Population Surveys (CPS) in corresponding years after 1967, and estimated from the CPS for early earlier years. As noted by Hilary Waldron, “Trends in Mortality Differentials and Life Expectancy for Male Social Security-Covered Workers, by Socioeconomic Status,”who used a different imputation procedure, the main purpose of the imputations is to sort individuals into two groups —workers in the top and bottom half of the average earnings distribution between the ages of 45 and 55.

20. These estimates, reported in Tables 4 and 5 of Hilary Waldron, “Trends in Mortality Differentials and Life Expectancy for Male Social Security-Covered Workers, by Socioeconomic Status,” are used to interpolate conditional life expectancies for the1914 to 1939 birth cohorts at their respective ages in 2004. The estimates in Table 5 are the cohort life expectancies consistent with the 2004 Trustees Report and they provide the basis for our estimates that use the same life expectancy for all members of a birth-year.

21. Taxes payable on the earnings between 1937 and 1950 are estimated by applying average tax rate in those years to the aggregate earnings estimate from the BE-PUF. Also, benefit tax “rates” are also estimated in each year for each half of the income distribution and are netted out of the annual benefit payments.

22. See Michael Clingman, Kyle Burkhalter, Alice Wade and Chris Chaplain, “Internal Real Rates of Return under the OASDI Program for Hypothetical Workers,” Actuarial Note No. 2011.5, and “Money’s Worth Ratios under the OASDI Program for Hypothetical Workers,” Actuarial Note No. 2011.7, Social Security Administration, Office of the Chief Actuary,  2012. These Actuarial Notes identify that between birth years 1920 and 1930 the money’s worth ratio is less than one for single men and women.

23. For analysis of how the 1977 changes to the benefit formula affected labor supply see Alan B. Krueger and Jӧrn-Steffen Pischke, “The Effect of Social Security on Labor Supply: a Cohort Analysis of the Notch Generation,” Journal of Labor Economics, Vol 10, No 4, October 1992, pages 412-437.

24. Our finding is that while the longevity gap between high- and low-earnings males has significantly widened over time, the longevity gap between high- and low-earnings females has not. This result is consistent with a similar finding in Courtney Monk et al., “Adjusting Social Security for Increasing Life Expectancy: Effects on Progressivity,” that while life expectancy is strongly correlated with AIME for men, it is only weakly correlated for women, and when pooling the genders the correlation disappears. Monk et al. mention two reasons for their finding. First, a low-earnings woman may be married to a high-earnings man and, second, women have higher longevity than men in general.  

25. Jeffrey B. Liebman in “Redistribuiton in the Current U.S. Social Security System,” The Distributional Aspects of Social Security and Social Security Reform, University of Chicago Press,   and Coronado et al in “The Progressivity of Social Security,”  find that recognizing income-adjusted longevity offsets the progressivity in Social Security. In contrast, Amy Harris and John Sabelhaus, in “How Does Differential Mortality Affect Social Security Finances and Progressivity?” Working Paper Series, Congressional Budget Office, May 2005. find that differential mortality does not play a significant role in determining progressivity in the Social Security program.

26. Liu and Rettenmaier, in “Social Security Outcomes by Racial and Education Groups,” Southern Economic Journal ,Vol. 69, No. 4, April, 2003,pages 842-864, show that the two measures do not always move in the same direction.

27. For both the top and the bottom halves, the net lifetime benefits are declining over time because Social Security redistributes intergenerationally from later to earlier cohorts. The intergenerational redistribution in Social Security also explains why the net lifetime benefits are higher for the top halves than for the bottom halves for the earliest cohorts.


Read Article as PDF