Early Social Security Claiming and Old-Age Poverty

I. Introduction

The Social Security program is the largest single expenditure of the U.S. federal government, totaling $1 trillion in 2018. It is also an important source of long-term fiscal imbalance. Due to the aging of the population, falling fertility, and a declining rate of productivity growth, Social Security currently faces a 75-year actuarial deficit equivalent to 2.78 percent of payroll (Board of Trustees, Federal Old-Age and Survivors Insurance and Federal Disability Insurance Trust Funds 2019). For this reason, there are constant conversations among policymakers about reforms to the program. There is no shortage of ideas for making incremental changes, from raising the cap on payroll taxes, to investing some of the trust fund in stocks. More significant changes include raising the payroll tax rate or raising the Full Benefit Age (FBA) from its current level of 67. For each of these proposals, the welfare implications are fairly clear—the main debates are political.

A more radical change in the Social Security program would be to raise the Early Eligibility Age (EEA), currently 62. While age 67 is also known more colloquially as the “normal” retirement age in the United States, there is nothing normal about it. By far the most common age for claiming Social Security benefits is 62, when retired-worker benefits first become available. Thus, an increase in the EEA would affect the timing and amount of benefits for a large number of older Americans. A number of OECD countries recently have increased the early entitlement age in their old-age pension systems, including Australia, Austria, Germany, and the United Kingdom, among others.

While we know that raising the EEA would reduce early retirement, we do not know the direction of the welfare effect because the excess in claiming at age 62 in the United States is difficult to reconcile. As Diamond and Gruber (1999) emphasized, this spike cannot be explained easily by simple financial incentives, which are roughly actuarially fair for the population of individuals as a whole at that age. Rather, there are a number of competing explanations. The first is liquidity constraints—individuals would like to claim earlier than 62, but are unable to because it is illegal to borrow against Social Security benefits. The second is that there is significant heterogeneity in mortality rates that make the financial incentives actuarially fair for the population as a whole, but actuarially unfair for some population subgroups, who then rationally choose to claim early (Hurd, Smith, and Zissimopoulus 2004). The third are behavioral failures— either due to misunderstanding (Brown et al. 2013), misoptimization (from, for example, cognitive limitations or financial illiteracy), or quasi-hyperbolic preferences (Diamond and Köszegi 2003), individuals do not appreciate that by taking benefits at age 62, they are lowering their monthly check until they die. The problem is that these alternative explanations have directly opposite predictions for the welfare impacts of raising the EEA. The first suggests that it would lower welfare by tightening liquidity constraints, the second suggests that it could lower welfare for subgroups of the population that had above-average mortality, and the third suggests that it would raise welfare by decreasing “mistakes.”

There is no study that comprehensively evaluates these trade-offs, and we do not either. But what we offer in this work is a piece of the puzzle: a demonstration of the impact on long-run income of early benefit claiming. If individuals who claimed benefits early did not see a diminishment of their later-life income, for example, through other sources of income or smoothing through savings (Milligan 2014; Bronshtein et al. 2018), then it would suggest little welfare cost to allowing them to claim benefits earlier. Conversely, if early claiming is associated with reduced income, it raises a potential trade-off from early claiming. This is especially true if the reduced income occurs among those with the lowest incomes.

This problem cannot be addressed by simply comparing those who do and do not claim Social Security benefits early, as these groups may differ in many ways that impact their later living circumstances. What is needed is an exogenous shift in the availability of early retirement options.

The introduction of the EEA into the Social Security program in 1961 provides that exogenous variation. Men born in 1896 and earlier moved through their early 60s prior to the 1961 Amendments to the Social Security Act. They could first claim benefits at age 65, the full benefits age (FBA). Men born in 1897 and later were affected by the law change. Those born in 1897 could claim as early as age 64, those born in 1898 could claim as early as 63, and those born in 1899 and later could claim as early as age 62. With early claiming, age-65 benefits were actuarially reduced by 5/9ths of a percent for each month of payable benefits before attainment of age 65. For a man claiming at age 62, this represented a 20 percent reduction in benefits, compared to claiming at age 65. The cohort variation in the availability of early eligibility allows us to analyze its impact on financial well-being while controlling for both age and time trends.

We begin with a description of the 1961 Amendments and other relevant changes to the Social Security system around that time, then move on to the empirical analysis, which focuses on men born 1885–1916, a roughly 15 years-of-birth window around the pivotal birth cohorts affected by the 1961 Amendments. These men attained their FBA and made their retirement decisions in the 1950s, 1960s, and 1970s. We then use data from the 1968–2001 March Current Population Surveys (CPS) to measure each cohort’s income and poverty trajectories into old age. We use a regression framework to estimate the impact of the reduction in the early entitlement age on income and poverty, controlling for both age and calendar year effects, as well as a broad set of potentially confounding factors.

There are four primary findings. First, reductions in the early entitlement age in the long run lowered the average claiming age by 1.4 years, which lowered Social Security income for male-headed families in retirement by 2.4 percent at the mean and 6 percent at the 25th percentile of the Social Security income distribution. Second, the increase in early claiming was associated with a decrease in total income, but only for the lower half of the income distribution. As a result, there was a sizeable increase in elderly poverty. In particular, the introduction of early claiming raised the elderly poverty rate by about one percentage point. Third, for the 1885–1916 cohorts, the implied elasticity of poverty with respect to Social Security income for male-headed families is −1.6. Finally, there appears to be heterogeneous effects by age. Early claiming is associated with lower poverty rates for those younger than 70, but progressively higher poverty rates for men older than 70. Overall, we find that the introduction of early claiming was associated with a reduction in income and an increase in the poverty rate in old age for male-headed households.

We then extend the analysis to examine impacts on mortality, the gold standard for well-being in the health literature. We use data on the annual mortality rate by single year of age from 65 through 98 for each sample birth cohort from the Human Mortality Database (2019), and examine the impact of early claiming on the mortality rate. While, on average, early claiming is associated with modestly lower mortality rates in retirement, there are heterogeneous effects by age. Early claiming has a U-shaped association with the annual mortality rate across ages: higher mortality rates for those under age 70, lower mortality rates for those in their 70s and mid-80s, and then progressively higher mortality rates among the oldest old.

Overall, these findings do not prove that there were negative welfare consequences from introducing the EEA. In particular, individuals may have been making a rational decision to trade off leisure and health for income. But they do rule out the notion that early claiming was a clear, unambiguous welfare improvement, especially among the oldest old, and motivate further analysis of the welfare consequences of this key policy parameter.

The paper is organized as follows. The next section briefly relates our analysis to the previous literature. Then Section III gives background on the 1961 Amendments. Section IV describes the CPS data and discusses cohort trends in income and poverty, and Section V lays out the regression framework. Both of these sections draw on the organization, exposition, and methodology developed in a companion set of papers (Engelhardt and Gruber 2006; Engelhardt, Gruber, and Perry 2005; Engelhardt 2008). Section VI presents the empirical results, and Section VII concludes.

II. Previous Literature

Our analysis is most closely related to six strands of the empirical literature on Social Security. The first is a large, long-standing literature that has examined the role of Social Security in retirement in the 1960s and 1970s. Reviewed in detail in Feldstein and Liebman (2001), it includes notable contributions by Crawford and Lilien (1981), Moffitt (1987), and Rust and Phelan (1997), among many others. This body of work generally concluded that decreasing the age of eligibility to 62 for men not only increased Social Security claims, but also led to earlier departures from the labor force, with a spike emerging in the retirement hazard at age 62. Related work in other contexts includes the option-value approach of Stock and Wise (1990) to analyze the timing of claims to private pensions, and that of Gruber and Wise (1999), who documented similar evidence on the impact of early claiming in Social Security programs in a wide variety of OECD countries.

The second strand is comparatively more recent and has used both structural models and reduced-form approaches to examine the impact of early and delayed claiming on labor supply, saving, and retirement income adequacy. This includes work by Blau (2008); Gustman and Steinmeier (1986, 2008, 2015); Coile et al. (2002); Sass, Sun, and Webb (2013); Shoven and Slavov (2014); and Shoven, Slavov, and Wise (2017), among many others.

The third strand has examined the short- and long-run impact of benefit increases and cuts, respectively, from the 1971 and 1977 Social Security Amendments that generated the so-called Social Security “notch.” These amendments changed real benefits, while leaving the FBA and EEA unchanged at (then) 65 and 62, respectively. This includes work by Krueger and Pischke (1992), who found little evidence of an impact on labor supply, and Engelhardt and Gruber (2006); Engelhardt, Gruber, and Perry (2005); and Engelhardt (2008), who found substantive long-run impacts on old-age poverty, shared living arrangements, and home ownership.

A fourth, more recent strand has estimated the short-run impact of the 1983 Social Security Amendments that increased the FBA from 65 to 67, while leaving the EEA intact at 62, which effectively generated benefit cuts. This includes work by Mastrobuoni (2009) and Behagel and Blau (2012), who examined the impact on labor supply and claiming behavior, and Duggan, Singleton, and Song (2009), who examined the spillover effects into the federal disability insurance (DI) program.

The fifth strand consists of a set of a current empirical studies that examine the short-run impacts of recent increases in the EEA that varied by birth cohort on labor force participation, poverty, and program participation of older individuals in other countries. This includes Atalay and Barrett (2015), who studied women in Australia, Staubli and Zweimüller (2013) and Manoli and Weber (2016), who studied men and women in Austria, Cribb and Emmerson (2017), who studied women in the United Kingdom, and Rabate and Rochut (2019), who studied men and women in France. Taken together, these studies show that for individuals close to the EEA (that is, in their late 50s and early 60s) raising the EEA generates significant increases in participation in the labor force and other gateway social insurance programs (for example, disability and unemployment insurance, with results differing somewhat across countries). Cribb and Emmerson (2017) showed that raising the EEA actually raised the poverty rate in the short run because the increases in earnings and other program income were not enough to offset the loss of old-age pension benefits.

Finally, there is a large literature in economics, demography, and sociology that documents an inverse relationship between income and mortality, which might suggest that the reduction in income for cohorts that claimed early might have found its way into higher later-life mortality for these cohorts. However, Finkelstein and McKnight (2008) documented a secular decline in mortality rates that was roughly parallel for younger vs. older elderly from the 1950s through the 1970s, with little evidence of differential mortality based on age across this time. In addition, Snyder and Evans (2006) examined the impact of income reductions from the Social Security notch on mortality. They found that the notch led to a reduction, not an increase in mortality.

Instead of focusing on the labor market impact of early eligibility like many of the studies above, our main contribution is to examine the long-term impact of early claiming on well-being, as measured by family income, poverty, and mortality. We do so by examining the introduction of the EEA for men in the early 1960s, which differed by birth cohort, and then trace the trajectories of income, poverty, and mortality as the affected cohorts progressed into old age.

III. Legislative History

The 1950s, 1960s, and 1970s were central to the formation of today’s Social Security program.¹ In the early 1950s, age 65 was the earliest an individual could receive any type of benefit based off career earnings. At that point, someone could claim retired-worker benefits equal to 100 percent of the primary insurance amount (PIA) based on their covered earnings history. The 1956 Social Security Amendments granted women the opportunity to claim actuarially reduced benefits as early as age 62 (Schottland 1956).

Early claiming lowered the age-65 benefit for women by 5/9ths of 1 percent for each month of payable benefits prior to age 65 and applied both to retired-worker and wife’s benefits. For a woman claiming early on her 62nd birthday, this amounted to a 20 percent reduction in benefits. The 1956 Amendments took effect for calendar year 1957. The 1956 Amendments also introduced the disability insurance (DI) program, in which insured workers with a demonstrated need could draw benefits prior to age 65.

The Social Security Amendments of 1961 extended early claiming on the same basis to men (Cohen and Mitchell 1961). Men born in 1897 and later were affected by the law change: those born in 1897 could claim as early as age 64; those born in 1898 could claim as early as 63; those born in 1899 and later could claim as early as age 62. Overall, the 1961 Amendments induced age-by-cohort variation in the eligibility for early claiming: men born in 1897–1899 became (partially or fully) eligible for early claiming in a manner that depended on age. Those born 1900 or later were eligible to claim early at all ages from 62–64. Those born prior to 1897 were not eligible for early claiming. This is the central variation that is used in the empirical analysis below.

Figure 1 illustrates the initial take-up of early claiming. It plots the change in the age distribution of claims for new retired-worker benefits for men across the focal cohorts affected by the 1961 Amendments, based on data drawn from various issues of the Social Security Administration’s Annual Statistical Supplement (United States Social Security Administration 1950–2001). Men in the 1895–1896 cohorts were not eligible for early claiming, and the majority claimed at the FBA of 65. The age distribution shifted systematically toward early claiming with the 1897, 1898, and 1899 cohorts, eligible at ages 64, 63, and 62, respectively. For the 1900 cohort, about one-third of men claimed at the FBA of 65 and one-fifth at the EEA of 62.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Fraction of New Retired-Worker Awards to Men by Claiming Age and Year of Birth

IV. Data Construction and Year-of-Birth Trends

The analysis focuses on men born 1885–1916, a roughly 15 years-of-birth window around the pivotal cohorts affected by the 1961 Amendments (Engelhardt, Gruber, and Kumar 2020). We do not include women in the analysis for three reasons. First, most married women from similar cohorts claimed benefits on their husband’s earnings history. Second, for widows and divorcees, the CPS did not ask about the year of birth for ex- or deceased husbands, so we cannot assign these women to the correct EEA for their husband’s birth cohort. Finally, for never-married women, who would have claimed on their own earnings histories, the CPS sample sizes were too small for reliable analysis.

One concern, which will become apparent in the description of the regression framework below, is that cohorts in the latter half of this sample window experienced rapidly rising generosity of benefits for workers claiming at all ages. In particular, prior to 1971, Congress adjusted benefits on an ad hoc basis to account for the impact of inflation. Because of persistent high inflation in the late 1960s, the 1971 Social Security Amendments sought to codify inflation adjustment into the calculation of benefits on an automatic basis. However, the 1971 law inadvertently introduced the double indexation of benefits by both allowing the bend points in the PIA formula to be indexed to price inflation and basing Average Monthly Earnings (AME), the lifetime earnings measure at that time, on nominal wages, which already had real wage and inflation components. With high inflation in the 1970s, this led to a rapid increase in benefits for each subsequent retiring birth cohort. This pattern of rising benefits ended with the 1977 Amendments, which eliminated double indexation by changing the PIA formula and introducing Average Indexed Monthly Earnings (AIME). Since those born in 1916 already would have attained the early retirement age of 62 in 1978 when the law went into effect, the 1977 Amendments grandfathered all individuals born in 1916 and earlier under the old benefit structure. For those born in 1917–1921, the so-called “notch” generation, a new, less generous benefit structure was phased in.

In order to hold the benefit structure constant, as much as possible, the Social Security “notch” cohorts that began with men born in 1917 are excluded from the sample. Then, a cohort-specific measure of general benefit generosity is used in the regression specification below to control directly for changes in benefits that are not due to the introduction of early claiming per se.

In addition to rising benefits prior to the notch, the 1966 Amendments to the Social Security Act narrowly targeted supplemented benefits for individuals 72 and older who were not fully insured for benefits. This provision was known as the special Age-72 Benefit (Schobel 1983) and was paid to a comparatively small number of individuals in 1966, a number that rapidly declined in subsequent years. The number of quarters of coverage required for eligibility for this supplement differed by year of birth. In the empirical analysis below, we control for eligibility for this benefit.

Men born 1885–1916 attained their FBA and made their retirement decisions in the 1950s, 1960s, and 1970s. To follow these cohorts into old age, the empirical analysis focuses on data from the 1968–2001 March CPS on elderly male-headed families aggregated into age-by-calendar year cells. The questions in the March CPS are about income earned in the previous calendar year, so that the income data from the 1968–2001 surveys can be used to measure income and poverty status in 1967–2000. By the end of the sample period in 2000, men born in 1916 would have been 84, and men from older cohorts would have been even older or deceased.

The sample construction follows the same methodology as in Engelhardt, Gruber, and Perry (2005); Engelhardt and Gruber (2005, 2006); and Engelhardt (2008), and it begins with the CPS microdata. There, an elderly “family” is defined as a male head age 66 or older, his spouse, and any of his children living with the family and under the age of 18. Over time, the CPS has provided more disaggregated questions on income sources, and, for some types of income, has changed the wording of questions. The most disaggregated income measures are used to construct the two key measures of income in the preferred specifications below: family Social Security income and total income. The former is the sum of Social Security retirement income across all members in the family; the latter is the sum of income from all sources across all members in the family. This measure is used to determine whether the family is above or below the federal poverty threshold. Finally, since the central variation in Social Security benefits from the introduction of the EEA is by year of birth, then poverty and income measures are collapsed into age-by-calendar year cells, which are the same as year-of-birth cells. Both income measures are then deflated into real 2001 dollars using the all-items Consumer Price Index (CPI). Table 1 shows sample means of the key outcome and explanatory variables used below.

The solid line in Figure 2 shows real mean Social Security income for male heads by year of birth based on the CPS data. Social Security income rises rapidly for men born 1885–1896 and then falls for those born 1897–1902, which coincides with the cohorts first eligible for early claiming under the 1961 Amendments. This peak (1885) to trough (1902) raw decline in real Social Security income for men was 5.3 percent. After 1902, Social Security income continues a rapid rise through 1916, the end of the sample cohorts. The dashed line in the figure shows real mean Social Security income for family members who are not the male head. Their income continues to trend upward fairly smoothly across the cohorts, suggesting that part of the decline in Social Security income for men may have been offset at the family level by rising Social Security income for other family members.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Mean Social Security Income for Male Heads and Other Family Members by Year of Birth

To account for changing family size across birth cohorts, the solid line in Figure 3 shows real mean family Social Security income by year of birth, which is the sum of the two sources depicted in Figure 2, adjusted for family size by the OECD equivalence scale. This is the focal measure of Social Security income used in the empirical analysis below. It rises for men born 1885–1896 and then displays a less dramatic decline for men 1897–1900. This occurs because of the offset from other family members shown in Figure 2. After 1900, Social Security income continues to rise through 1916, the end of the sample cohorts. The dashed line in the figure shows real median family total income, which trends fairly smoothly upward across these cohorts.² There does not appear to be a strong correlation in the figure between total family and Social Security income.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Mean Social Security and Median Family Income per OECD Equivalent for Male-Headed Elderly Families by Year of Birth

Figure 4 plots real mean family Social Security income per OECD equivalent and the percentage of male-headed elderly families with total income below the federal poverty threshold (for the appropriate family size), the standard measure of absolute poverty. The poverty rate declines sharply from 42 percent for men born in 1885 to 20 percent for men born in 1896. For the 1897–1900 cohorts, the poverty rate rises above that for the 1896 cohort; after 1900, the poverty rate falls gradually as Social Security income rises. In contrast to Figure 3, there is stronger visual evidence here for a relationship between the early claiming policy change and poverty.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Mean Social Security Income per OECD Equivalent and Absolute Poverty of Male-Headed Elderly Families by Year of Birth

Unfortunately, this measure of poverty has a number of well-known limitations. First, it holds constant the standard of living. Second, it adjusts for price inflation, but not for real wage growth. Finally, it does not measure the depth of absolute deprivation. Consequently, Figure 5 shows real mean Social Security income per OECD equivalent and a relative measure of poverty: the percentage of male-headed elderly families with income less than 40 percent of the median income of nonelderly families. Both elderly and nonelderly income are adjusted by the OECD equivalence scale. Nonelderly families are defined as those headed by someone 25–54 years old. The measure is constructed for elderly families by single year of age in each calendar year of the CPS and then collapsed to get a year-of-birth average, which is plotted in the figure. The results for the relative poverty rate are very similar to those for the absolute poverty rate.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

Mean Social Security Income per OECD Equivalent and Relative Poverty of Male-Headed Elderly Families by Year of Birth

Finally, Figure 6 shows the relationship between real mean Social Security income per OECD equivalent and the annual mortality rate for men aged 65 through 98 years old, based on annual mortality data by single year of age and year of birth from the Human Mortality Database (2019). There is a strong inverse correlation between Social Security income and male mortality. However, the slowdown in the decline in mortality after the EEA is phased in continues for the younger cohorts, even after Social Security begins to rise rapidly again, so that, overall, it is difficult to identify a simple cohort effect in the figure from access to early claiming.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6

Mean Social Security Income per OECD Equivalent and the Male Mortality Rate by Year of Birth

V. Regression Framework

Overall, these year-of-birth figures suggest that cohorts of men who were first eligible for early claiming had lower actual Social Security income and higher poverty rates after retirement and into old age. A fundamental challenge in interpreting these patterns as causal is that most of the variation in Social Security benefits that identifies differences in elderly family Social Security income across years of birth is time series in nature. Omitted variables that are correlated with changes in poverty rates, mortality, and Social Security, and trending over time, might explain these patterns equally well, leading to a fundamental identification problem. For example, lifetime earnings are affected by aggregate productivity and human capital accumulation that have changed across time and cohorts. As a key determinant of Social Security benefits, these changes would find their way into observed Social Security income. At the same time, federal poverty thresholds are inflation adjusted, but not average-earnings adjusted. Thus, poverty rates would have been expected to have declined for successive birth cohorts as productivity, human capital accumulation, and real lifetime earnings rose. Hence, gains in productivity and human capital could simultaneously account for a rise in Social Security income and a decline in poverty across years of birth.

To attempt to circumvent this and identify causal impacts on poverty and mortality, the analysis moves to the following regression framework. Let a index age of the male head and t index the calendar year; then the econometric specification can be written as (1)

The dependent variable S is the age-by-calendar year mean family Social Security income; u is an error term. The focal explanatory variable is EEA, the early entitlement age for men in that cohort. For men born in 1896 or earlier, the EEA was 65. For men born in 1897, the EEA was 64. For men born in 1898, the EEA was 63. For men born in 1899 and later, the EEA was 62. ξ is a vector of dummy variables for single year of age; ψ is a vector of dummy variables for calendar year. The age dummies control for differences across age groups in the outcome measure; the year dummies control for any general time trends in the outcome measure. The central objective is to get a consistent estimate of β, the impact of a one-year increase in the EEA on Social Security income. Controlling for age and calendar year, the estimate of β is identified by cross-cohort variation in eligibility for early claiming induced by the 1961 Amendments.

The vector X includes controls for cell means of educational attainment of the head (high school diploma, some college, and college or advanced degree), marital status (married, widowed, and divorced), white, and veteran status. These account for any other trends in cohort characteristics that might be correlated with both the legislative changes in benefits determination and actual income. In addition, there are controls for important changes that varied by cohort: a dummy if men in that year of birth were eligible for Medicare at age 65, a dummy if men in that year of birth were eligible for DI prior to age 65, and the quarters of coverage required for that cohort to qualify for the special Age-72 supplemental Social Security benefit.

As Figures 2 and 3 indicated, Social Security income was trending upward across the sample cohorts. So as to prevent these more general benefit increases from confounding the estimates of β, Equation 1 includes a polynomial function (f) of , a measure of the simulated real family Social Security income a man would have gotten had he claimed benefits at age 65. For a synthetic unmarried male beneficiary from a given cohort, is constructed as follows. Let B_c(y_c, k) be the primary insurance amount (PIA), which varies according to the benefit structure B_c applied to year of birth c, potential claiming age k, and earnings history y. First, an earnings history y was constructed for each cohort c. A baseline earnings history for the 1916 birth cohort—the last year of birth in the sample—was constructed from various issues of the Social Security Administration’s Annual Statistical Supplement. In particular, SSA published median male earnings by five-year age groups on an annual basis. These earnings were used to assign median earnings at age 22 (from the median earnings for ages 20–24 in 1938), age 27 (from median earnings for ages 25–29 in 1943), etc., in five-year intervals. Then a linear trend in earnings was assumed within these five-year intervals to get earnings by single year of age for the 1916 cohort. This method was used through age 60, and earnings were assumed to grow with inflation for ages 60–69.³ Importantly, the earnings history constructed for the 1916 cohort, denoted as , was then used for all birth cohorts, and the CPI was simply used to adjust the earnings profile for inflation for earlier and later cohorts. Therefore, all cohorts were assigned the same real earnings trajectory by construction.

Second, the constructed earnings histories for each cohort were put into the SSA’s ANYPIA benefit calculator, which calculates the PIA at retirement, given a date of birth, date of retirement, and earnings history. In this case, ANYPIA calculates . Unfortunately, the actual day and month of birthdates are not observed in the CPS. Because the timing of cost-of-living adjustments has varied across calendar years, for the purposes of calculating simulated benefits, birthdays of June 2 were assigned in the particular year of birth, and it is assumed that men retired and claimed benefits in June of the year in which age 65 is attained.

These PIAs, expressed in real 2001 dollars, are shown in Figure 7. The variation in PIA at age 65, even conditional on constant earnings histories, is readily apparent in the figure. The PIA rises for early cohorts, then is roughly constant in real terms, followed by a rapid rise for those born after 1903 due to ad hoc benefit adjustments, and then subsequently ramps up quickly through 1916 due to double indexation.

• Download figure
• Open in new tab
• Download powerpoint

Figure 7

Primary Insurance Amount for Claiming at Age 65 by Year of Birth for 1916 Cohort Median Male Earnings

Finally, the Social Security Administration periodically increased nominal benefits to adjust for inflation. To obtain a value for the PIA, which is measured at the time of retirement, in a future calendar year in which a cohort is observed in the CPS sample, all “cost of living adjustments” (COLAs) to which a beneficiary was entitled from the time of retirement until the date of interview, Π, were incorporated to produce a real, expected Social Security monthly benefit. This was multiplied by 12 to convert it to an annual income amount and then adjusted by the OECD equivalence scale θ to yield (2)

A synthetic married man from a given cohort was assigned 150 percent of this measure, (3) under the assumption the wife would claim on the husband’s earnings. Then the cohort simulated Social Security income for a family with a male head claiming at age 65, , was formed as a weighted average of Equations 2 and 3, with the weights equal to the share of unmarried and married male heads in each age-by-calendar year cell, respectively.

The dashed line in Figure 8 shows by year of birth. The solid line in the figure is actual real mean Social Security income per OECD equivalent. Its trajectory tracks very closely from 1885–1896 and 1905–1916, indicating that much of the upward trend in family Social Security income across cohorts is due to general benefit increases. In Equation 1, enters flexibly as a quartic function, so that the estimate of β will be identified from the cross-cohort variation in the early entitlement age, independent of benefit generosity. This is illustrated in the figure by divergence in the series for birth years 1897–1900, when early claiming is introduced.

• Download figure
• Open in new tab
• Download powerpoint

Figure 8

Simulated and Actual Mean Social Security Income per OECD Equivalent for Male-Headed Elderly Families by Year of Birth

VI. Estimation Results

Table 2 shows grouped ordinary least squares (OLS) estimates of β, where the weights are based on the age-by-calendar year cell sizes. There are 658 cells in the estimation sample, which range from ages 66 to 90.⁴ Each column shows selected estimates from a separate regression with a different measure of real mean Social Security income as the dependent variable. Row 1 of Column 1 shows the estimate of β in Equation 1 excluding the vector X of other controls, when the dependent variable is the Social Security income of the male head. Controlling for age and calendar year effects, , which says that an decrease of one year in the early entitlement age lowered Social Security income by $359. Based on the standard error of $44, clustered by year of birth and shown in parentheses, the null hypothesis of β = 0 can be rejected in favor of the alternative β > 0 at conventional significance levels.

To get a sense of the size of this effect in economic terms, note that relative to mean Social Security income of $7,560 for men born prior to 1897 (shown in the third panel of the table), and hence unaffected by the 1961 Amendments, this estimate implies that a one-year reduction in the EEA lowered income by 4.8 percent. The weighted-average reduction in claiming age below 65 for the 1916 cohort was actually somewhat larger, at 1.4 years.⁵ Using 1.4 years as a measure of the long-term reduction in claiming age implies that early claiming reduced Social Security income for men by 6.4 percent, shown in the first column of the second to last row of the table. We note that this estimated reduction is smaller than what would have been expected based on a weighted average of the actuarial reduction factors from claiming at 62, 63, and 64, respectively, of approximately 9 percent (that is, 9% = 0.315 × 20% + 0.153 × 13.3% + 0.103 × 6.7%). In fact, the 95 percent confidence interval around our 6.4 percent estimated impact excludes 9 percent. Although we present evidence below that lower income workers were more likely to respond to early claiming, so that the expected effect should be a little smaller than 9 percent, it is an open question as to why our estimate in Column 1 is smaller than a pure weighted average of the actuarial reduction.

Column 2 shows the estimate from the same specification, but with mean Social Security income of family members other than the male head as the dependent variable. Controlling for age and calendar year effects, , which says that a decrease of one year in the early entitlement age for the male head was associated with higher Social Security income for other members of the family by $65. Based on the standard error of $20, this impact is statistically different from zero at conventional levels of significance.

Relative to the sample average Social Security income for others of $1,724, this estimate implies that a one-year reduction in the EEA was associated with higher income of others by 3.8 percent. This suggests that some of the reduction in income for male heads was crowded out by an increase in income of other family members. Indeed, Column 3 shows the estimate of β when the dependent variable is Social Security income for all family members combined. Controlling for age and calendar year effect, a decrease of one year in the early entitlement age for the male head lowered family Social Security income by $294. This estimate implies that early claiming reduced Social Security family income by 4.3 percent.

Columns 4 and 5 show the same estimates, but for our preferred measure of income: family Social Security income adjusted by the OECD equivalence scale to account for trends across cohorts in family size. In Column 4, controlling for age and calendar year effects, a decrease of one year in the early entitlement age for the male head was associated with lower family Social Security income by $183. Column 5 adds the vector X of other controls. The estimated impact of early claiming is smaller: a decrease of one year in the early entitlement age lowered Social Security income by $116. Relative to mean family Social Security income per OECD equivalent of $6,668, this estimate implies that a one-year reduction in the EEA lowered income by 1.7 percent and that the average actual reduction in the early claiming age reduced Social Security income by 2.4 percent.

In addition to the estimates in the first row, for all specifications we report randomized p-values in square brackets in the final row of the table by randomly assigning EEAs while clustering by year of birth (Heβ 2017). If EEA had a significant effect, we should expect the impact of randomly assigned placebo EEAs to be centered around zero, with the estimated actual effect in the lower or upper tails of the distribution of estimated placebo effects based on 1,000 replications. Using the empirical CDF of placebo effects, p-values are obtained as the probability of observing placebo effects at least as extreme in absolute value as the actual estimated effect. In Table 2, randomized p-values are close to zero and even lower than the ones implied by the conventional standard errors.

To examine how changes in the EEA affected the distribution of Social Security income, Table 3 repeats the richest specification from Column 5 of Table 2, but with the dependent variable, S, measuring selected percentiles of the distribution of Social Security family income per OECD equivalent. The estimates suggest the bulk of the impact of changes in the EEA were on the lower to middle part of the distribution, with small (and statistically insignificant) impacts in the tails. A one-year reduction in the claiming age was associated with a reduction in Social Security income by 5.8 percent at the 25th percentile of the Social Security income distribution and 2.6 percent at the median.

Table 4 shows a parallel set of reduced-form estimates for total family income per OECD equivalent. At the mean, a one-year reduction in the EEA is associated with a reduction of total income of $107, which is not statistically significant. Looking at the distribution, there is a highly significant effect on income at the 25th percentile of $180, which is 2.4 percent of income at that point in the distribution. The effect then becomes negative and insignificant at higher quantiles. Taken together, these results indicate that in the lower half of the income distribution there is no crowd out of Social Security income by other sources of income, so that the reductions in Social Security are translated to total income. But at the upper end of the income distribution, we do not see such evidence, albeit with sufficient imprecision that we cannot rule out the same results as for Social Security income.

To estimate the reduced-form impacts on poverty and mortality, Table 5 shows the reduced-form grouped OLS estimates of ρ from a specification isomorphic to Equation 1: (4)

In Column 1, the dependent variable, P, is a measure of absolute poverty: the fraction of families below the federal poverty line. The focal parameter estimate in Row 1, , indicates that a one-year decrease in the EEA raised the poverty rate about seven-tenths of a percentage point. Relative to an average poverty rate of 27.8 percentage points for cohorts prior to early claiming (shown in Row 2), this estimate represents a 2.5 percent reduction in the poverty rate (that is, −0.0071/0.278 = −0.025) and a 3.4 percent reduction based on the long-run actual reduction in the early claiming age (Row 4).

In Column 2, the dependent variable is relative poverty, measured as the fraction of male-headed elderly families with total family income below 40 percent of the median income of nonelderly families. The estimate suggests that a one-year reduction in the EEA was associated with an increase in relative poverty of around seven-tenths of a percentage point, which is marginally significant. Relative to the mean relative poverty rate of 32.9 percentage points (Row 2) for men born in cohorts ineligible for early claiming, this represents a 2.2 percent increase in the relative poverty rate (Row 3) and a 2.9 percent increase based on the long-run actual reduction in the early claiming age (Row 4).

Column 3 shows the estimate of the reduced-form effect of eligibility on the annual male mortality rate, using the same specification as in Equation 4. The estimate indicates that a one-year reduction in the EEA was associated with a small decrease in the mortality rate of a tenth of a percentage point. This effect is statistically different from zero at conventional significance levels and suggests that reductions in elderly income from early claiming reduced the mortality of older men. This effect is counterintuitive, but mirrors the finding for the benefits notch in Snyder and Evans (2006). These effects are also nonlinear with respect to age, as we show below.

The first-stage estimates from Equation 1 can be combined with the reduced-form estimates from Equation 4 to yield instrumental variable (IV) estimates of the impact of Social Security income on poverty: (5) where EEA is excluded from Equation 5 as the instrument. Column 1 of Table 6 shows the IV estimate of φ for absolute poverty. The estimate and indicates that a $1,000 increase in Social Security income is associated with a 7.5 percentage point reduction in the poverty rate. Relative to mean Social Security income, this can be interpreted as an elasticity of the absolute poverty rate with respect to Social Security income of −1.7, shown in the second row. Column 2 shows similarly large elasticity estimates for relative poverty. Column 3 shows the IV estimate of φ for mortality. The estimate is and indicates that a $1,000 increase in Social Security income is associated with a just under one percentage point increase in the mortality rate. Relative to mean Social Security income and mortality, this can be interpreted as an elasticity of the mortality rate with respect to Social Security income of 0.4.

We end with Table 7, which shows selected parameter estimates from an expanded version of Equation 4 that allows the impact of early claiming eligibility to vary by calendar age. The key question is whether the impact of early eligibility on poverty is uniform through old age or concentrated at particular ages. To explore this, we allow the impact of early eligibility to vary quadratically with calendar age: (6)

The first three rows of Column 1 show the reduced-form grouped OLS estimates of ρ₁, ρ₂, and ρ₃, respectively, from Equation 6, for absolute poverty.⁶ The fourth row shows the p-value for the test of the null that the EEA has no effect on poverty (ρ₁ = ρ₂ = ρ₃ = 0). The last row shows the p-value for the test of the null that the effect of the EEA is uniform by calendar age (ρ₂ = ρ₃ = 0), which can be just rejected that the 5 percent level of significance. The solid line in Figure 9 plots the associated estimated marginal effect of the EEA on absolute poverty. The impact is nonlinear with age—the availability of early claiming lowers the poverty rate for those under age 70, but then increases the poverty rate at older ages. The long-dashed line shows a similar pattern for relative poverty based on an isomorphic set of estimates for that outcome in Column 2 of Table 7. Column 3 in the table shows estimates from the same specification, but for the mortality rate as the outcome, the data for which again run from age 65 through 98. The dashed line in Figure 9 plots the associated estimated marginal effect of the EEA on mortality. The impact is nonlinear with age, but in a way that differs from the poverty trajectory. The availability of early claiming is associated with a higher mortality rate for those under age 70, a lower mortality rate for those in their 70s to mid-80s, and then finally a higher mortality rate for the oldest old.

• Download figure
• Open in new tab
• Download powerpoint

Figure 9

Estimated Nonlinear Effect of One-Year Reduction in the Early-Entitlement Age, by Calendar Age from Quadratic Specification

VII. Summary and Caveats

This work estimated the impact of the early entitlement age on later-life financial well-being and mortality by tracing the income and poverty status of different birth cohorts of men who had access to different potential claiming ages due to the Social Security Amendments of 1961. Reductions in the early entitlement age in the long run lowered the average claiming age by 1.4 years, which lowered family-size-adjusted Social Security income for male-headed families in retirement by 2.4 percent at the mean and 6 percent at the 25th percentile of the Social Security income distribution. Early claiming also was associated with a decrease in total income at the bottom of the income distribution and an increase in the elderly poverty. For the 1885–1916 cohorts studied here, the implied elasticity of poverty with respect to Social Security income for male-headed families is −1.6, similar to elasticities estimated for male- and female-headed families pooled over the 1885–1933 cohorts in Engelhardt and Gruber (2005, 2006), using variation in benefit levels based on the Social Security notch. The impacts of early claiming on poverty rise as men progress into old age.

These findings are tempered by a number of caveats. First, early claiming was introduced for men in a period of rapidly rising benefit generosity at all ages. These rising real benefits may have substantially offset the specific declines in retirement income from early claiming—that is, early claiming might have had larger impacts on poverty in the absence of these general benefit increases. This is important since future changes to entitlement ages will be done in a very different environment governing the trajectory of benefits. Second, the introduction of early claiming for men in 1961 happened a long time ago; labor markets and retirement behavior of older workers have changed dramatically since then. Any prospective changes in the EEA are unlikely to share these features of the broader economic and health environment facing older Americans. Third, we examined whether there might be important heterogeneity in response for different subgroups of the elderly (by marital status, education, etc.), but were unable to draw firm conclusions given, in some cases, modest sample sizes. This is an important direction for further analysis.

Second, the interpretation of the mortality effects is complicated. The impact of early claiming on mortality was positive near retirement and among the oldest old, but negative for men in their 70s through mid 80s. Although somewhat speculative, one potential interpretation is that work itself has positive health benefits, so that early retirement has an immediate negative impact on health and consequent increase in mortality. However, as more time passes after retirement, the reduction in income from early claiming leads to a reduction in mortality, as in Snyder and Evans (2006), for the bulk of men who die in their 70s and 80s. Since health is such an important component of well-being, teasing out the exact mechanisms behind these effects is an important avenue for future research.

Finally, although income poverty is an important policy metric, it is not the only measure of well-being. In addition to health and changes in material well-being, such as nondurable consumption and housing, men who claimed benefits early also consumed more leisure than those who claimed later, compensating them to at least some extent for the loss in income from the actuarial reduction.

As noted in the introduction, there are multiple hypotheses for “excess” claiming at age 62. The results presented here are most consistent with the time inconsistency hypothesis: that claiming at age 62 represents for some a mistake that leads to lower incomes later in life. This would either suggest restrictions on earlier retirement or perhaps more rapidly increasing benefits around the age of early retirement to offset excessive early claiming. But absent data on the valuation placed on the excess leisure that went along with lower incomes, it is not yet possible to interpret this as a pure test of the time inconsistency hypothesis. Further work is needed to assess whether those men who claimed early (and their family members) and ended up in retirement and old age with less income and more impoverished regret the decision to have claimed early. Analysis of these types of issues is important for a richer understanding of the welfare implications of changes to Social Security claiming.