Can Informational Interventions Be Effective Policy Tools? An Initial Assessment of the Social Security Statement

Introduction

Recent research suggests that retirement-related informational interventions may influence worker behavior. For example, when workers at a university were provided with retirement-income projections and plan enrollment information, they increased their annual contributions to employer-provided retirement accounts (Goda, Manchester, and Sojourner 2014). Offering employees of credit unions an online financial education program increased reported retirement plan participation, emergency savings, and budget use (Collins and Urban 2016). Sending informational nudges to state government workers approaching retirement age resulted in increased contributions to their retirement savings plans (Clark and others 2017).

In this study, I assess the effect of providing information about an important benefit that most workers will receive—the Social Security retirement benefit. The Social Security Statement (or, simply, the Statement) is a major outreach initiative of the Social Security Administration (SSA). It provides workers with estimates of the benefits they can expect to receive if they claim at age 62 (the earliest age of eligibility), their full retirement age (FRA), or age 70 (when delayed retirement credits stop accruing). When SSA began mailing earnings and benefit statements to workers in 1995, one of the primary purposes was to provide workers with information on their Social Security benefits to help them plan their financial futures. The Statement has been widely acknowledged as one of the federal government's most important public communications (Jackson 2005). It stands as the largest customized mailing ever undertaken by a federal agency (SSA, n.d.)

Providing this information is important because the calculation of Social Security retirement benefits depends on complex rules that are not easy to understand. Those rules include factors such as individuals' marital histories, their work histories, and the age at which they choose to receive benefits. Many individuals are not aware that the benefit amount is affected by the age at which they claim. Claiming before FRA results in permanently reduced monthly benefits, and delaying claiming until after FRA permanently increases the monthly amount. The Statement provides detailed information on important program aspects, such as the reduction in benefits for early claiming, that might alter claiming behavior.

My research suggests that receipt of the Statement has a significant effect on the age at which workers claim their retirement benefits. Previous research found that receipt of the Statement increases knowledge of Social Security (Mastrobuoni 2011; Smith and Couch 2014a). This study is the first to find that receipt of the Statement has a statistically significant effect on behavior in addition to its effect on knowledge. I find that receipt of one Statement has no significant effect on the likelihood of claiming benefits at age 62, but it does significantly increase the likelihood of claiming at ages 64 and 65. In addition, I find that the receipt of two or more Statements has a statistically significant negative effect on the likelihood of claiming at age 62 and a statistically significant positive effect on the likelihood of claiming at ages 64 and 65. I also find that the effect of Statement receipt varies by race, sex, and earnings level; and that receiving a Statement has statistically significant effects on the likelihood of employment at ages 62 through 70.

The effects of two policy changes coincided with the implementation of Statement mailings. The first policy change established higher FRAs for workers born after 1937. Although it was enacted in 1983, this change began to affect workers reaching the earliest eligibility age of 62 in 2000, the same year the Statement was first sent to all eligible workers aged 25 or older. The FRA for each birth cohort from 1938 through 1943 is 2 months higher than that of the preceding cohort.¹ With higher FRAs, the permanent reduction for a monthly benefit claimed at age 62 increases. Such additional reductions are likely to encourage workers to claim later in order to receive a larger monthly benefit.

The second policy change eliminated the retirement earnings test (RET), effective 2000, for Social Security retired-worker beneficiaries once they attained FRA. The RET reduces monthly benefit amounts for individuals who claim benefits, continue to work, and earn more than a specified limit; until 2000, it applied even after FRA. Eliminating a reduction in benefits at FRA would likely increase the prevalence of claiming at FRA. My study addresses these two policy changes in different ways, discussed later.

My analysis draws on the rich administrative data in SSA's Continuous Work History Sample (CWHS). The CWHS is a 1 percent sample that allows me to examine Social Security benefit claiming behavior from 1975 through 2007. I also exploit the fact that SSA initially limited Statement mailings to selected age groups of older workers in a given year. These phased mailings allow me to distinguish recipients from nonrecipients and thereby assess the effect of one-time and multiple Statement receipt.

Background

This section summarizes the history of the Statement and reviews previous research on its effects.

Statement Implementation

The Statement traces its origins to the Omnibus Budget and Reconciliation Act of 1989, which amended the Social Security Act to require SSA to issue estimated-benefit and earnings-history statements beginning in 1995.² The legislation mandated that SSA mail these statements annually to workers aged 60 or older in fiscal year 1995 and, in fiscal years 1996 through 1999, to workers turning 60 during those years. The legislation also required SSA to send annual estimated-benefit and earnings-history statements to all eligible workers aged 25 or older beginning in fiscal year 2000. “Eligible workers” were defined as those with a Social Security number and having wages or net earnings from self-employment. SSA accelerated the legislated mailing schedule to include increasingly younger recipients during fiscal years 1996 through 1999.³ As shown in Table 1, this modification enabled the agency to increase the volume of mailings gradually over time.

Workers born in 1937 or earlier would have received one Statement, in either fiscal year 1995 or fiscal year 1996, before attaining age 62. Workers born in 1938 or later would have received at least two Statements before attaining age 62: one in the 1990s, as a member of one of the age groups designated to receive Statements; and the second (and any subsequent ones) beginning in fiscal year 2000, when Statements were mailed to all eligible workers aged 25 or older.

SSA staggered the Statement mailings throughout each year, with workers receiving their Statements about 3 months before their birthdays. In fiscal year 2010, the last full fiscal year of mailings to all eligible workers aged 25 or older, SSA sent more than 151 million Statements. That worked out to about 12.5 million Statements mailed each month and about 420,000 delivered each day. Fewer Statements have been mailed in the fiscal years since 2010. In March 2011, the agency suspended Statement mailings for budgetary reasons. Beginning in September 2014, SSA resumed Statement mailings, targeting workers reaching ages 25, 30, 35, 40, 45, 50, 55, and 60 or older in the year; but effective January 2017, mailings were restricted to workers aged 60 or older who had not created an online my Social Security account, which provides access to the Statement electronically.

Prior to SSA's Statement mailings, the only way for workers to find out what they would receive if they claimed benefits was to call SSA's toll-free phone number for information or make an appointment to visit an SSA field office. For several years, the Statement provided information on benefits automatically and annually to all eligible workers, freeing them from potentially waiting on the phone or going into a field office.

Legislation determines the content of the estimated-benefit and earnings-history statements. The Omnibus Budget and Reconciliation Act of 1989 specified that Statements must contain the worker's earnings history and Social Security and Medicare taxes paid; estimated retirement benefits payable if claimed at the earliest eligible retirement age (62), FRA, and age 70; estimated disability and survivor benefits payable on the worker's earnings record; and a description of the benefits payable under Medicare. The Social Security Protection Act of 2004 further mandated that the Statement include sections on the Windfall Elimination Provision (WEP) and the Government Pension Offset (GPO) beginning in 2007. The WEP and the GPO may reduce Social Security benefits for workers and spouses, respectively, if the workers receive pensions from employment not covered by Social Security.⁴ Over time, the design, content, and placement of information in the Statement have undergone slight changes. Appendix A presents a facsimile Statement from 2006.⁵

Statement Effect on Knowledge and Behavior

Smith and Couch (2014a) analyzed the Statement's effect on the Social Security knowledge of younger workers, using data from surveys commissioned by SSA. That study compared workers who did not receive a Statement in 1998 with those who did receive a Statement in 2001. The authors compared the percentages of respondents who correctly answered each of three questions about Social Security program aspects and three questions addressing Social Security benefits. Both before and after Statement receipt, younger workers' knowledge was stronger in program-level aspects than in benefit-specific aspects. In 2001, after distribution of the Statement, close to 90 percent of younger workers knew about SSA's programs, the financing of program benefits, and the relationship between benefits and earnings. However, even after Statement receipt, only about 70 percent of respondents knew that FRAs are higher for members of later birth cohorts, and less than 50 percent knew that benefits are inflation-indexed. Smith and Couch concluded that the gap in knowledge about benefits poses potential risks for the retirement security of younger workers and suggested several ways SSA could direct their outreach efforts.

Biggs (2010) used data from the 1994 through 2008 waves of the Health and Retirement Study (HRS) to analyze the effectiveness of the Statement in improving the public's knowledge of their Social Security retirement benefit levels. The HRS asks participants either to estimate their future retirement benefit amounts (if they have not yet claimed) or to report their actual benefit levels (if they are current beneficiaries). Biggs measured the accuracy of the estimated benefits when compared with realized benefits before and after the 1995 implementation of the Statement. He found that the initial mailings of the Statement did not result in more accurate estimates of retirement benefit levels but suggested that receipt of additional Statements might do so.

Mastrobuoni (2011) used data from the 1992 through 2000 HRS waves to study how the receipt of one's first Statement affected Social Security knowledge and the retirement behavior of workers aged 55 to 70. The author used HRS questions about expected retirement age and expected Social Security benefits (for workers) and retirement age and benefit levels (for retirees). Following respondents over time, Mastrobuoni compared the expectations with the actual outcomes for those who received a Statement and those who did not. He found that older workers who had not previously contacted SSA about their benefits were 20 percentage points more likely to be able to provide an estimate of their future retirement benefits if they had received a Statement than if they had not. However, he found no statistically significant changes in benefit claiming ages following receipt of the Statement. Mastrobuoni stated that his results called into question the likelihood that an informational intervention can affect behavior. He concluded that the information provided in the Statement was not sufficient to change workers' claiming patterns and called for more research on the Statement's effect on behavior.

In another related study, Liebman and Luttmer (2015) conducted a randomized field experiment that provided information on Social Security (not duplicative of the information in the Statement) to 2,500 workers aged 60 to 65. The authors provided information on longevity, how benefits increase with claiming age, and the effect of the RET. They examined whether this informational intervention affected employment and benefit claiming. The intervention included an informational mailing and a 15-minute online tutorial. A year later, respondents answered a follow-up survey. The authors found that the individuals who received the intervention, which pointed out potential advantages of working longer and delaying claiming, were 4.2 percentage points more likely to be working at the 1-year follow-up. They found no statistically significant effect of the information on the likelihood of claiming benefits.

My research extends Mastrobuoni's study by examining the effect of receiving multiple Statements, rather than just one, on behavior. My data source, the CWHS, has much larger samples than the HRS and covers the years 1975 through 2007. Thus, I am able to study a larger number of individuals over a longer period. I extend the Liebman and Luttmer (2015) study by looking in depth at the effect of one aspect of their intervention—an informational mailing, which in this case is the Social Security Statement.

Data and Study Variables

This study uses longitudinal data from the CWHS, a major source of Social Security program statistics and workforce data. It is the oldest major longitudinal sample data source in the Federal statistical system. It evolved from the recordkeeping system created to meet the requirement in a 1939 amendment to the original Social Security Act that eligibility for benefits be determined based on quarters of coverage, an earnings-based measure of employment duration.

The CWHS is a 1 percent sample of the agency's administrative data. It can be described as a stratified-cluster probability sample of all possible Social Security numbers (SSNs). The population from which the 1-percent sample is selected consists of the 1 billion possible nine-digit SSNs. The SSNs are stratified geographically (with place of application for the SSN indicated by the first three digits) and chronologically by date of SSN assignment (the fourth and fifth digits). The last four digits in the SSN are random serial numbers. The actual sample is selected on the basis of specified numbers in positions six through nine of the SSN (Smith 1989).

The CWHS is selected from workers whose earnings are reported to SSA. Thus, the CWHS contains Federal Insurance Contributions Act (FICA)-covered summary earnings from 1937 to the present; and annual detailed earnings, Medicare taxable compensation, and total compensation from 1978 to the present. It also contains demographic information such as birth and death dates, place of birth, race, and sex. Most relevant for this project, it contains information related to the administration of SSA's retirement and disability programs, such as benefit application and entitlement dates, benefit amounts, and payment status. Once selected for inclusion in the CWHS, an individual remains in the sample for life.

The CWHS is well suited to address my research questions about the effect of Statement receipt on benefit claiming and employment because it provides a large and accurate source of longitudinal data on Social Security retirement benefits and on earnings, which are used to determine employment status. It also contains the exact date of entitlement for Social Security benefits. The exact date is important because this project examines the effect of Statement receipt on age at claiming of Social Security benefits. There are some disadvantages, however, to the use of administrative data. For example, the CWHS has no information on health status, education, or family characteristics. Of concern for this project, it also contains no information on Statement receipt.

Demographic and Economic Variables

Because the administrative data do not include many demographic or economic variables, this study uses dummy variables to account for other factors that might influence the age at which workers claim their Social Security retirement benefits. For instance, I use dummy variables to control for year effects. Year dummies, in theory, capture demographic and economic changes in the year of benefit claiming that might have affected the claiming decision. Year dummies can also capture policy and program changes that were occurring in the year when benefits were claimed.

Policy and Program Variables

As mentioned earlier, two notable policy changes took effect in 2000, when the Statement was first mailed to all eligible workers aged 25 or older. Workers who reached age 62 (the earliest age of eligibility to claim retired-worker benefits) in 2000 were members of the first birth cohort whose FRA is higher than 65. Also beginning in 2000, the RET was eliminated for beneficiaries on attaining their FRA.

Variable Representing Receipt of the Statement

Because there is no variable in the administrative data indicating whether a worker received a Statement, this study assumed receipt or nonreceipt based on birthdates. Statement receipt was assumed for any individual of an age cohort that was scheduled to receive one in the implementation phase from fiscal year 1995 through fiscal year 1999, and for all individuals in fiscal year 2000 and later when all eligible workers aged 25 or older were sent one.⁶ For example, workers born 1939–1944 are assumed to have received a Statement in fiscal year 1997, when the agency plan called for mailing one to all eligible workers aged 53 through 58. According to the distribution schedule, workers born from 1933 through 1937 would have received just one Statement before age 62, during the implementation phase. Workers born in 1938 or later would have received two or more Statements before age 62. For workers born before 1960, one of the Statements would have been received during the implementation period. To account for receipt of the Statement, this study created a variable that was set equal to 1 if workers would have received a Statement based on their date of birth and 0 otherwise.

Depending on the estimation model, workers who received a Statement were placed in one of three groups:

• those who received at least one Statement before age 62 and were born from 1933 through 1945,
• those who received only one Statement before age 62 and were born from 1933 through 1937, or
• those who received two or more Statements before age 62 and were born in 1938 or later.

Workers who did not receive a Statement before age 62 were born from 1913 through 1932. This group is the same across all models.

Variables Representing the Varying FRAs

The differing FRAs for workers in the 1938 and later birth cohorts first began to affect benefit-eligible workers in 2000. The Social Security Act of 1935 had established a universal FRA of 65. In 1983, Congress enacted amendments—including the introduction of gradually increasing FRAs for later birth cohorts—to extend the financial stability of the Social Security program. The FRA rises in 2-month increments for successive birth cohorts, beginning with the 1938 cohort and reaching 66 for the 1943 cohort—thus affecting workers reaching age 62 in 2000 through 2005. When retirement benefits are claimed prior to a worker's FRA, they are reduced by an actuarially fair amount to account for the additional time over which they will be received. The maximum monthly benefit reduction for early claiming is larger for workers with higher FRAs. For example, claiming at 62, the earliest age of eligibility, results in a 20 percent reduction in “full” monthly benefits (that is, benefits claimed at FRA) for a worker whose FRA is 65 but a 25 percent reduction for a worker whose FRA is 66. The larger monthly benefit reductions faced by workers with higher FRAs might result in declines in pre-FRA claiming if workers want to ensure a certain level of monthly benefits in retirement. Thus, as the higher FRAs have begun affecting retirement-age workers, we might expect to see declines in claiming at the earlier eligibility ages of 62 and 63 and increases in claiming at ages closer to FRA, such as 64 and 65.

This study uses a dummy variable for each FRA affecting the study sample members, from 65 and 2 months to 66. These dummies are represented as follows:

• FRA65_2_it = 1 if the individual was born in 1938, for whom the FRA is 65 and 2 months; else 0.
• FRA65_4_it = 1 if the individual was born in 1939, for whom the FRA is 65 and 4 months; else 0.
• FRA65_6_it = 1 if the individual was born in 1940, for whom the FRA is 65 and 6 months; else 0.
• FRA65_8_it = 1 if the individual was born in 1941, for whom the FRA is 65 and 8 months; else 0.
• FRA65_10_it = 1 if the individual was born in 1942, for whom the FRA is 65 and 10 months; else 0.
• FRA66_it = 1 if the individual was born during 1943 through 1954, for whom the FRA is 66; else 0.

The Effect of Eliminating the RET at FRA

The Senior Citizens Freedom of Work Act of 2000 eliminated the RET for Social Security beneficiaries who have attained FRA. The RET reduces monthly Social Security benefits for current beneficiaries with work earnings exceeding specified amounts; however, any benefits withheld are credited back once the beneficiary attains FRA, resulting in a permanent increase in monthly benefits. The elimination of the RET at the FRA affected workers aged 65 or older in our study. Evidence suggests that the 2000 RET reform resulted in beneficiaries claiming benefits earlier than they would have without the reform (Olsen and Romig 2013). Therefore, we should expect to see some increase in benefit claiming at age 65. The overall effect on earlier benefit claiming is likely to be small, however, because most individuals in the 65–69 age group apply for benefits before reaching FRA. Also, the RET directly affects only about 5 percent of retired-worker beneficiaries each year because the majority of those individuals who are still working earn less than maximum specified by the RET (Olsen and Romig 2013). Research that examined how the elimination of the RET at FRA affected expected claiming age found that this effect was not significant among men aged 51–61 (Michaud 2008). For the reasons listed above and because of the limited covariates in the administrative data, this study does not control for the effect of the RET on benefit claiming.

Sample Size

The analysis looks at how Statement receipt affects claiming behavior for workers overall, by sex, by race (white, black), and by earnings level (low, high). The sample size for all workers aged 62 to 70 is 586,415. The sample sizes for male and female workers are 323,846 and 262,569, respectively. The sample size for non-Hispanic white workers is 510,197 and for non-Hispanic black workers is 52,797. The sample size for both low and high earners is 292,969.

Methodology

I begin by calculating the percentage of fully insured workers born 1913–1949 who claimed a retirement benefit on their own earnings record at age 62. Fully insured workers have at least 10 years of earnings and are therefore eligible to receive retirement benefits. Chart 1 plots the pattern of these age-62 claiming rates for 1975–2011. The rate increased for men until around the mid-1990s and afterward began to decline. Starting around 2000, the decline appears to become steeper. SSA began sending out the Statement to selected groups of older workers in 1995 and to all eligible workers aged 25 or older in 2000. Thus, there appears to be a correlation between Statement mailings and the decline in retirement benefit claiming at age 62.

I then estimate the effect of Statement receipt on claiming behavior and employment using a linear probability model (LPM). The LPM allows me to compare the probability of claiming (or being employed) at ages 62 through 70 for workers who received a Statement and those who did not. The dependent variable in this analysis is binary, equaling 1 if the worker claimed (or was employed) at a given age and equaling 0 otherwise. The advantage of using the LPM is that the coefficients are easy to interpret. For example, the coefficient on the dummy variable for receipt of the Statement can be interpreted as the change in the probability or likelihood of claiming at a given age—say, 62—if a Statement has been received.

I use three variations of the LPM:

• Model 1 estimates the effect of Statement receipt on the likelihood of claiming benefits at a given age, controlling for year effects.
• Model 2 estimates the effect of Statement receipt on the likelihood of claiming benefits at a given age, controlling for the varying FRAs.
• Model 3 estimates the effect of Statement receipt on the likelihood of employment, controlling for the varying FRAs.

For each model, I look at the effect of Statement receipt first for all workers and then separately by sex and race. For Model 2, controlling for the varying FRAs, I also look at the effect of Statement receipt by earnings level.

This project focuses on workers aged 62–70 in the years 1975–2007. This study period begins 20 years prior to the introduction of the Statement mailings and follows sample members who turned 62 in 1995, after receiving the first Statement mailing, for more than 10 years. I follow individuals through age 70 because monthly Social Security retirement benefits increase with each month the worker delays claiming, up to age 70. I use data only through 2007 to avoid possible complications arising from the Great Recession and the suspension of Statement mailings in 2011. Nonetheless, I am able to look at responses for the first 13 years of mailings.

Estimating the Effect of the Statement on Benefit Claiming

For Models 1 and 2, I use a separate equation for each year of age at which Social Security retirement benefits can be claimed, 62 through 70, using data for the calendar years 1975 through 2007. I set the binary dependent variable for each equation equal to 1 if retirement benefits were claimed at that age and 0 otherwise. I do not make distinctions by month within a claiming year. For example, I assume that claiming at age 62 and 0 months and at age 62 and 11 months are the same. Independent variables include the dummy variable for Statement receipt as well as dummy variables to capture the effects of demographic and economic factors that might influence when workers claim their retirement benefits.

In Model 1, I am able to compare the two periods of the Statement's implementation. In the first period, from 1995 to 1999, workers in selected age groups received just one Statement before age 62. In the second period, from 2000 to 2007, all workers aged 25 or older received annual Statements, and thus received multiple Statements before age 62. Therefore, I am able to compare the effects of receiving one and multiple Statements on claiming behavior. This LPM is written as:

where

• Claim^j_it = 1 if the individual claimed benefits at this age; else 0, for j = ages 62 through 70.
• SSS1_it = 1 if the individual received only one Statement before age 62 (birth years 1933 through 1937); else 0.
• SSS2_it = 1 if the individual received two or more Statements before age 62 (birth years 1938 through 1945); else 0.
• YR_it (dummy for year effects from 1975 through 2007) = 1 in the year the individual turns 62; else 0.

In this model, the dummies representing varying FRAs would be collinear with the dummies representing year effects, so I omit the FRA dummies.

In Model 2, I specifically control for the incremental variation in the FRAs from 65 to 66, which began to affect potential age-62 claimants in the sixth year of Statement mailings (2000). Including these dummy variables for the FRA in the estimations, along with the dummy variable for receipt of a Statement, allows me to separate the effect of the mailings from that of the varying FRAs. I am not able to look separately at receipt of one and of multiple Statements because the varying FRAs and the periods in which workers would have received multiple Statements coincide, so that the dummy variables would be collinear. Nonetheless, I think this specification provides useful estimates of the Statement's effect on benefit claiming as well as the responsiveness of claiming to changes in the FRA. The LPM is written as:

where

• Claim^j_it = 1 if the individual claimed benefits at this age; else 0, for j = ages 62 through 70.
• SSS_it = 1 if the individual received at least one Statement⁷ before age 62 (birth years 1933 through 1945); else 0.
• FRA65_2_it = 1 if the individual was born in 1938, for whom the FRA is 65 and 2 months; else 0.
• FRA65_4_it = 1 if the individual was born in 1939, for whom the FRA is 65 and 4 months; else 0.
• FRA65_6_it = 1 if the individual was born in 1940, for whom the FRA is 65 and 6 months; else 0.
• FRA65_8_it = 1 if the individual was born in 1941, for whom the FRA is 65 and 8 months; else 0.
• FRA65_10_it = 1 if the individual was born in 1942, for whom the FRA is 65 and 10 months; else 0.
• FRA66_it = 1 if the individual was born during 1943 through 1954, for whom the FRA is 66; else 0.

Estimating the Effect of the Statement on Employment

Claiming Social Security retirement benefits and deciding to work are separate and independent decisions. Individuals have four options with respect to claiming benefits and working.⁸ They can either:

• Claim benefits and continue to work,
• Claim benefits and stop working,
• Delay claiming benefits and continue to work, or
• Delay claiming benefits and stop working.

With Model 3, I examine whether individuals continued to work in response to the mailings or stopped working. For this portion of the analysis, I use an equation similar to equation (2) above, substituting a binary dependent variable that equals 1 for those who have earnings at each age 62 through 70 (and are therefore employed), and 0 if they do not, for EMP^j_it. I use the same independent variables: dummies controlling for Statement receipt and dummies controlling for the varying FRAs. The LPM is written as:

where

• EMP^j_it = 1 if the individual was employed at this age; else 0, for j = ages 62 through 70.
• SSS_it = 1 if the individual received at least one Statement before age 62 (birth years 1933 through 1945); else 0.
• FRA65_2_it = 1 if the individual was born in 1938, for whom the FRA is 65 and 2 months; else 0.
• FRA65_4_it = 1 if the individual was born in 1939, for whom the FRA is 65 and 4 months; else 0.
• FRA65_6_it = 1 if the individual was born in 1940, for whom the FRA is 65 and 6 months; else 0.
• FRA65_8_it = 1 if the individual was born in 1941, for whom the FRA is 65 and 8 months; else 0.
• FRA65_10_it = 1 if the individual was born in 1942, for whom the FRA is 65 and 10 months; else 0.
• FRA66_it = 1 if the individual was born during 1943 through 1954, for whom the FRA is 66; else 0.

I use FICA earnings data reported in the CWHS to create my employment variable. For each birth year from 1916 through 1945, I create an earnings variable for ages 62 through 70 that is equal to recorded FICA earnings. For younger birth cohorts (born from 1938 through 1945), I am not able to create an earnings variable for all ages from 62 through 70, given that members of these cohorts had not reached age 70 by my last year of analysis (2007). If my earnings variable is positive, I assume the individual was employed. For example, if the earnings variable for an individual aged 62 is positive, then I set EMP^j_it equal to 1 when j = 62.

Results

As noted earlier, the percentage of workers claiming benefits at age 62, the most popular age to claim Social Security retirement benefits, has been declining since the mid-1990s. The mid-1990s is also when the Statement was first mailed automatically to selected age groups of older workers. Beginning in 2000, the Statement was sent to all eligible workers aged 25 or older. The estimates I present here are intended to measure the effect that receiving the Statement, with its information about claiming age and benefit amounts, had on workers' claiming and employment decisions. Because delaying claiming and working longer would lead to larger monthly benefits for the rest of their lives, evidence that workers altered their decisions in this way would imply that the information they received informed their choices. Coefficient values for the independent variables range between 0 and 1 in value. For ease of interpretation, they are presented as percentages.

Effect of Statement Receipt on Benefit Claiming

Table 2 shows the estimated effects of the Statement mailings, controlling for year effects (Model 1) and for the varying FRAs (Model 2). In both models, Statement receipt is associated with a significantly greater likelihood of claiming at age 65: by 2.68 percentage points when controlling for the varying FRAs, and by 2.31 percentage points and 4.64 percentage points—depending on the number of Statements received—when controlling for year effects.⁹ Also in both models, receipt of the Statement was associated with significantly decreased claiming at younger ages. Model 2, controlling for the varying FRAs, showed a decrease in the likelihood of claiming at age 63 of 0.90 percentage point and a decrease in the likelihood of claiming at age 64 of 2.18 percentage points. Model 1, controlling for year effects, showed a decrease in the likelihood of claiming at age 62 of 3.36 percentage points for those receiving multiple Statements. Both models suggested that workers were less likely to claim at earlier ages following receipt of the Statement and more likely to claim at age 65, although the patterns of timing differ somewhat across the models. For ages 66 through 70, there was no clear pattern of effect of the mailings between the two models.

As a final note, in Model 2, at age 63, the magnitude of the Statement's negative effect was larger, at 0.90 of a percentage point, than that of any of the FRA dummies, whose absolute values ranged from 0.05 to 0.46. At ages 64, 65, and 67, the magnitude of the Statement effect was at the lower end of the range of magnitudes for the FRA dummies, with values of −2.18, 2.68, and 0.18, respectively. At ages 69 and 70, the magnitude of the Statement effect was, at −0.12 and 0.35 respectively, somewhere in the middle of the range of effects for the FRA dummies. I cite these comparisons simply to note that estimates of the effect of the mailings on behavior are similar to those of changing policy by raising FRAs for later birth cohorts.

Estimating the Effect of the Statement on Benefit Claiming by Sex and Race

Whereas Table 2 shows the Statement's effect on the likelihood of benefit claiming for all workers, Table 3 shows its effect on the likelihood of claiming benefits separately for men and women, again controlling for year effects and for the varying FRAs. The results for men and women in Model 1 were similar to those for all workers in magnitude and sign. However, receipt of the Statement was associated with a smaller change in the likelihood of claiming benefits for women than for men at claiming ages 62 and 65. For example, receipt of multiple Statements was associated with a reduction in the likelihood of claiming at age 62 by 3.81 percentage points for men and by 2.90 percentage points for women. Receipt of multiple Statements was associated with an increase in the likelihood of claiming at age 65 by 5.32 percentage points for men and 3.94 percentage points for women. However, receipt of one Statement was associated with a larger increase in the likelihood of claiming at age 64 for women (3.08 percentage points) than for men (2.60 percentage points).

The Model 2 results provide a possible explanation for the insignificant effect of Statement receipt on the likelihood of benefit claiming for all workers at age 62. For men, receipt of at least one Statement was associated with a significant and positive effect on the likelihood of claiming benefits at age 62, by 2.18 percentage points. For women, receipt of the Statement was associated with a reduction in the likelihood of claiming at age 62 by 3.99 percentage points. In this model, as in the model controlling for year effects, the Statement's effect on the likelihood of benefit claiming was smaller for women than for men for most ages. The effect of the Statement was also less likely to be significant for women than for men.

There are several possible explanations for why women's claiming may be less affected by Statement receipt than men's. For example, many of the women in my study are likely to be married. Studies show that married women tend to retire from the labor force when their husbands do, with the result that they might be less influenced by the Statement's estimates of their benefits (Johnson 2004). Other studies suggest that the real or perceived relative advantages of various retired-worker and spousal benefit claiming scenarios might encourage wives to retire early and thus be less influenced by Statement receipt to delay claiming (Munnell and Soto 2005). Finally, women are more likely than men to be a caregiver for an elderly or sick family member and thus be less likely to have the option to delay claiming.

Table 4 shows the effect of Statement receipt on the likelihood of claiming benefits for white and black workers, controlling for year effects and for the varying FRAs. Because white workers constitute a large majority of the study sample, their results were, not surprisingly, similar to those for all workers.¹⁰ In Model 1, the results for white workers are almost identical to those for all workers overall in sign, magnitude, and significance. For black workers, receipt of the Statement had a significant effect only at age 67, when it reduced the likelihood of claiming by 1.43 percentage points if one Statement was received and by 1.23 percentage points if multiple Statements were received.

In Model 2, the results for white workers are also very similar in sign, magnitude, and significance to those for all workers. The effect of Statement receipt on the likelihood of claiming for black workers is negative and significant at ages 63 and 64, and positive and significant at age 65.

Health status and socioeconomic factors might explain why Statement receipt has no significant effect on claiming for black workers at ages 62, 64, and 65 in Model 1 and why the negative effect on claiming at age 63 in Model 2 is so much smaller for black workers than for white ones. Older black workers are more likely than older white workers to have higher rates of diabetes, hypertension, and obesity—chronic conditions that can lead to disability. Black Americans also have lower life expectancy than white Americans (Martin and Murphy 2014). A higher prevalence of chronic health conditions and higher mortality is likely to result in early benefit claiming and a reduced likelihood of delaying claiming. On average, black workers also have lower educational attainment and lower earnings than white workers (Bureau of Labor Statistics [BLS] 2018). Individuals with lower earnings are less likely to have additional sources of retirement income to complement the benefits they receive from Social Security. Thus, black workers will be less financially likely to have the option to delay claiming of benefits—both because they are less likely to be able to continue to work and because they are less likely to have alternative sources of retirement income.

Estimating the Effect of the Statement on Benefit Claiming by Earnings Level

The effect of Statement receipt on benefit claiming varies across earnings levels. As a proxy for earnings, I use the primary insurance amount (PIA), which is the monthly retirement benefit workers receive if they claim at their FRA. The PIA is directly linked to earnings: As an individual's lifetime earnings increase, so does the PIA.¹¹ I calculated the median PIA and then used Model 2, controlling for the varying FRAs, to estimate the Statement's effect on claiming age for all workers, for those with a PIA less than the median, and for those with a PIA greater than the median.

Receipt of at least one Statement by age 62 had different effects on the two earnings groups (Table 5). For those with PIAs below the median, receipt of the Statement was associated with a decrease in the likelihood of claiming at age 62 by 4.07 percentage points. For those with PIAs above the median, receipt of the Statement was associated with an increase in the likelihood of claiming at age 62 by 3.36 percentage points.

These differing results might be explained by the relationship between earnings level and the likelihood of having other retirement savings. Higher earners have greater access to both defined benefit and defined contribution retirement plans than do lower earners (BLS 2017). Higher earners also participate in available employer-provided plans at higher rates than do lower earners. Further, higher earners are more likely to have personal retirement savings in addition to their employer-offered retirement plans. Given these other sources of retirement income, higher earners, who are also likely to receive higher Social Security benefits, might decide to claim early and let their 401(k) and personal saving investments appreciate. Lower earners, on the other hand, are less likely to have other sources of retirement income. Because they will be more dependent on Social Security benefits for their retirement income, lower earners may be more likely to continue to work and delay claiming to increase the amount of the Social Security benefits they will receive.

This offsetting difference by earnings level might explain why I find the overall effect of Statement receipt on claiming at age 62 to be insignificant. As noted above, the effect of Statement receipt on the likelihood of claiming at age 62 was positive for workers with a PIA above the median (3.36 percentage points) and negative for those with a PIA below the median (−4.07 percentage points).

The effect of Statement receipt on the likelihood of claiming at age 65 was positive for workers with a PIA above the median (3.00 percentage points) as well as for those with a PIA below the median (1.94 percentage points). Although the effect of Statement receipt on the likelihood of claiming at ages 63 and 64 was negative for workers with PIAs both above and below the median, it was significant only for those with PIAs above the median (−1.58 percentage points at age 63 and −4.09 percentage points at age 64). For those with PIAs below the median, receipt of the Statement had a small but significant positive effect on the likelihood of claiming at ages 68, 69, and 70 (0.17 percentage point, 0.19 percentage point, and 0.57 percentage point, respectively). For workers with PIAs above the median, receipt of the Statement had a small but significant negative effect on the likelihood of claiming at ages 67, 68, and 69, and an even smaller positive and significant effect on claiming at age 70.

Estimating the Effect of the Statement on Employment

Based on a broader CWHS sample, Table 6 shows the estimated effect of Statement receipt on the probability of being employed. I find that receipt of at least one Statement was associated with a significant increase in the percentage of individuals working. The estimates suggest that, at each age, the likelihood of working increased between 5 percentage points and 7 percentage points as a result of receiving a Statement. In other words, Statement receipt was associated with increased likelihood of being employed at ages 62 through 70. All of the FRA dummies were significant but the magnitude was less than half that of the Statement dummy. I also looked at the effect of Statement receipt on the likelihood of working by sex and race. What was true for all workers was also true for men and women and for black and white workers: Receipt of the Statement was associated with a 4–7 percentage point increase in the likelihood of working (not shown).

Robustness Results

The models I use contain a limited number of covariates, such as sex and race, and I used those covariates to separate the sample into different groups of interest. To determine if my results may have been driven by unobserved factors, I tested three different specifications to check the robustness of the effect of Statement receipt on claiming. The first specification included adding a trend term to my models to provide an additional control for exogenous changes in my dependent variable that are not explained by other variables. Second, to test whether the estimation framework itself might result in spurious results, I randomly assigned receipt of the Statement to individuals who never could have received one. The third specification added a variable to control for changes in the economy that might have affected when workers claimed their Social Security retirement benefits.

When I added a trend term to the model with year dummies, I found that the direction of the Statement's effect on the likelihood of claiming was the same as in the original model: a negative effect at ages 62 and 67 and a positive effect at ages 64 and 65 (Table 7). However, the magnitude of the effect of the Statement increased. For example, the effect of receipt of one Statement before age 62 on the likelihood of claiming at age 62, which was insignificant in the original model, was significant and negative in the model with the trend term. In addition, the effect of receiving multiple Statements on the likelihood of claiming at age 62 was −3.36 percentage points in the original model but −4.57 percentage points in the model with the trend term.

Adding a trend term to model 2 (controlling for varying FRAs) resulted in some changes in sign from the original specification of the Statement's effect on the likelihood of claiming at ages 62 through 70 (Table 8). However, the overall pattern of a reduction in early claiming and an increase in later claiming remained. For example, the effect of the Statement on the likelihood of claiming at age 65 was 2.68 percentage points in the original model and 2.94 percentage points in the model with the trend term. Without the time-trend term, controlling for the varying FRAs led to an insignificant effect on claiming at age 62; but controlling for the time trend led to a negative and significant reduction in claiming of 4.27 percentage points. Thus, including the trend term generally resulted in larger estimated reductions in early claiming.

The second robustness test involved randomizing receipt of the Statement. By randomly assigning mailings to individuals prior to the time when Statements were actually mailed, I tested whether the estimation approach employed in the analysis inappropriately found implementation effects.

As noted earlier, there was no variable in the administrative data I used that indicated whether a worker had received a Statement; but using workers' birth dates and the Statement implementation schedule, I was able to estimate which workers were likely to have received one and which were not. I assumed that workers who were the appropriate ages to receive a Statement according to SSA's implementation schedule actually received one.

To test the validity of my estimation approach, I randomly assigned and coded workers to have received a Statement in periods when they would not have, based on the implementation schedule and their birth year. Forty-four percent of all individuals in the sample potentially received a mailing; so, among those who could not, I randomly assigned 44 percent to the category of having falsely received a mailing. I then added a variable to my models that captured the effect of the random falsely received Statement. Tables 7 and 8 show the results of including a randomized variable representing the effect of receiving a falsely targeted Statement in the models controlling for year effects and the varying FRAs, respectively. The effect of the random falsely targeted Statement mailings was not significant in any of the models.

The third robustness test involved adding a variable to capture the effect of changes in the economy that occurred during the period when SSA was sending out the Statement. I tested the effect on benefit claiming of changes in four macroeconomic variables: the unemployment rate, the inflation rate, the interest rate on 3-month certificates of deposit, and the percentage change in per capita personal income. The values of these variables were for the year in which the worker turned 62. I derived the unemployment rate and inflation rate values using BLS data, the interest rate values using data from the Federal Reserve Bank of St. Louis, and the personal income values using data from the Bureau of Economic Analysis.¹²

All four variables had a similar effect on the relationship between receipt of the Statement and the likelihood of claiming at given ages. For simplicity, I limit the discussion to the effects of the change in per capita personal income.

Table 7 shows the effect of adding a macroeconomic variable to Model 1 (controlling for year effects). Receipt of the Statement still had a negative (but insignificant) effect on the likelihood of claiming benefits at age 62 and a significant positive effect on the likelihood of claiming benefits at age 65, but in both cases the magnitude of the effect was reduced. This suggests that the age at which benefits were claimed was also affected by what was happening in the larger economy. In fact, the macroeconomic variables had direct negative impacts on the likelihood of benefit claiming at both age 62 and age 65, with the magnitude varying between 1 and 2 percentage points in most cases.

Table 8 shows the effect of adding a macroeconomic variable to Model 2 (controlling for the varying FRAs). Receipt of the Statement still had a negative effect on the likelihood of claiming benefits at age 62 and a significant positive effect on the likelihood of claiming benefits at age 65. In this model, the magnitude of the Statement's effect on benefit claiming is similar to that of the original specification and the specification including the trend term. Also, in the FRA model, the magnitude of the direct effect of the macroeconomic variables on claiming at ages 62 and 65 was very small, less than 0.5 percentage point in most cases.

Discussion

This study is the first empirical analysis to demonstrate that Statement mailings had a statistically significant effect on the age at which workers claim their Social Security retirement benefits. Evidence herein suggests that workers who received a Statement were less likely on average to claim retirement benefits at age 62, the earliest claiming age, and more likely to claim at ages 64 or 65 than workers who did not receive a Statement. The effect of the Statement mailings on benefit claiming varied across demographic groups, being greater for men than for women, for white workers than for black workers, and for higher earners than for lower earners. Workers who received a Statement were also more likely to remain employed at older ages than workers who did not receive a Statement.

Mastrobuoni (2011) found that receipt of one Statement did not have a significant effect on benefit claiming behavior. I extended that research to look at the effect of receiving multiple Statements. I also found that receipt of one Statement did not have a significant effect on the likelihood of claiming benefits at 62, the age at which the largest number of people claim. However, I found that receipt of multiple Statements had a significant negative effect on the likelihood of claiming at age 62.

Liebman and Luttmer (2015) found that a one-time experimental intervention with several aspects—including informational mailings, online tutorials, specific information about Social Security, and vignettes—increased the likelihood that individuals were still working 1 year later by more than 4 percentage points. Yet the authors were unable to identify which of these aspects was the most important to that behavioral response. I extended their research by focusing on a single aspect of their intervention, an informational mailing: in this case, the Social Security Statement. I found that an informational mailing by itself can affect behavior. I also found statistically significant effects of the Statement on employment, similar in magnitude to those found by Liebman and Luttmer.

My results suggest that informational interventions could be a policy tool for promoting retirement security, along with complementary approaches such as tax incentives to encourage retirement saving and automatic enrollment in state-administered individual retirement accounts for workers whose employers do not offer retirement plans. I found that the Statement mailings increase the likelihood that recipients delay the claiming of retirement benefits, which results in a higher monthly benefit for the rest of the claimant's life. (Social Security benefits represent a significant proportion of the retirement income of all but the highest lifetime earners.) I also found that receipt of the Statement increased the likelihood of employment at ages 62 through 70. More time spent working results in more years of earnings and, consequently, increased Social Security retirement benefits. One implication of my findings is that low-cost informational interventions, in addition to direct policy levers, might be effective in increasing the retirement security of older Americans, by both raising the level of their retirement benefits through delayed claiming and increasing their employment at older ages.

Currently, SSA mails the Statement only to workers aged 60 or older who have not created a my Social Security account with which to access the Statement electronically. My findings on the effect of the Statement on benefit claiming suggest that the agency might consider outreach efforts to encourage more workers to sign up for a my Social Security account and, for those who have already created an online account, to check it more regularly. Whereas the agency sent Statements to more than 151 million workers in the last full year of mailings in 2010, only about 50 million workers had created a my Social Security account as of June 2020.

Limitations and Future Directions

A major limitation of the administrative data I use is the absence of information on health status, education, and family characteristics, and limited information on nonearnings income. Any of these variables might affect the age at which workers claim. Another limitation of my research is that I do not control for a major policy change: the elimination of the RET at FRA.

These limitations suggest several directions for future work. One obvious extension of this research would be to control for the change in the RET. Another extension would involve constructing a control variable for earnings. Possible examples of such a control variable might be lifetime earnings or average earnings for ages 45–55. Another extension could involve developing a way to control for compositional changes in the cohorts I study. These compositional changes might include education, race/ethnicity, the share of workers who are immigrants, and marital history.

Future work might include extending my analysis of how earnings levels are associated with the effect of Statement receipt on benefit claiming by looking at deciles or quintiles rather than the two broad categories of above and below the median PIA. This article examines how Statement receipt changes the claiming age from, for example, 62 to 63. Future work could focus on how Statement receipt affects the number of months by which claiming was delayed. In this article, if a worker delayed claiming from age 62 and 0 months to 62 and 11 months, we would not see any change in claiming age. Looking at ages in terms of months rather than years might reveal an even larger effect of the Statement on claiming.

Appendix A

2006 sample Social Security Statement