What is the effective Social Security tax on additional years of work?

INTRODUCTION

In the U.S., Social Security system retirement benefits are based on the average of the highest 35 years of earnings during an individual's working life. This "high-35" approach--especially when combined with wage-indexing of lifetime earnings and a progressive benefit formula--can lead to a situation where additional years of earnings at the end of a career do not improve benefit outcomes. In that situation, the Social Security payroll tax becomes a pure tax on labor for older workers, because there is no increase in future benefits that will (even partially) offset the statutory payroll tax. In this paper, administrative earnings records are used to assess the extent to which effective tax rates increase across and within groups for those around retirement age, how sensitive those conclusions are to various input assumptions and estimation procedures, and the implications of a budget-neutral change to benefit and tax formulas intended to mitigate rising effective tax rates for older workers.

The effective tax on an additional year of work is measured as the gap between the payroll tax paid and the change in the present value of benefits--that gap is then divided by the level of earnings to compute the effective tax rate. The change in the present value of benefits involves two sets of inputs: the first is the change in the individual's Primary Insurance Amount (PIA), which is used to calculate Social Security benefits at any given claim age, and the second is the present value discount factor (PVDF), which adjusts the stream of future benefits for survival probability and the time value of money. The change in PIA can be computed directly from the administrative data, but computing the value for the PVDF requires assumptions about mortality and discount rates.

In general, the analysis here shows that the high-35 effect is evident in the data--effective tax rates rise noticeably between ages 55 and 65, because more and more workers face benefit levels that do not rise as they continue earning. This pattern holds for both men and women, across lifetime earnings quintiles, and across both total and "career" worker populations. For men, the highest average effective tax rates are observed for earners in the middle lifetime earnings quintile, who are less likely to see big increases in their PIA from working another year. Males with low lifetime earnings in the data see the largest increases in PIA at each age, but that effect is offset to some extent by their higher mortality rates. For women, effective tax rates rise uniformly across lifetime earnings groups, and are generally negative for low earners.

The differences in means across age, income, and gender groups are modest relative to the tremendous heterogeneity within groups, however. If an individual in any group already has 35 years working at or near their lifetime (wage-indexed) average, their PIA will be little affected if they work another year, and the Social Security tax will approach a pure tax because there is no offsetting future benefit. That is indeed the case for a subset of workers in every group, and the change in effective tax rates by age for the highest taxed (for example, the 75th or 90th percentiles of effective tax rates) is somewhat larger than for the rest of the population within that age, earnings, and gender group.

The estimated levels of effective tax rates and the magnitude of the increase between ages 55 and 65 are somewhat sensitive to the specific population being studied and at least two key assumptions. Effective tax rates are generally lower for those estimated to still be in "career" jobs, where the end of a "career" is identified by a sudden and significant drop in earnings. This makes sense, because older workers who have left career jobs are even less likely to replace a year in their high-35 formula. In terms of assumptions, the lower is the discount rate used to compute the PVDFs, the lower are effective tax rates at any given age, but the steeper is the increase in effective tax rates across age groups. Also, adjusting for predictable differences in survival eliminates a noticeable share of the difference in average effective tax rates across lifetime earnings groups, because lifetime low earners have higher mortality rates. If low lifetime earners work another year, they are more likely to noticeably increase their PIA, but they are also less likely to actually receive the benefit.

The data used here are not well suited for a direct test of the hypothesis that differences in effective tax rates affect labor market exit behavior at older ages---one would need information about important individual control variables like health status, pension benefit formulas, or retiree health insurance provisions. However, the data do indicate that after controlling for lifetime earnings and gender, the effective tax rates for those who exit the (career) labor market in the next year are higher than the effective tax rates for those who do not. There are many possible competing explanations for that pattern, and the observation deserves more attention using data that is better suited to evaluating whether changes in effective tax rates actually do affect labor-force exits.

The data and methods developed for this paper are useful for getting a sense of how various proposed changes to the Social Security tax and benefit system would affect the distribution of effective tax rates across and within groups. One often-mentioned policy change is increasing the number of years in the benefit computation formula from 35 to 40. This has the effect of lowering the average effective tax rates at all ages by shifting the increase in effective tax rates to higher ages. People may not see the change in computation years as an advantage, however, because it means their benefits will generally be lower at any given claim age. That is particularly true for people who already have fewer than 35 years of earnings.

One suggested approach for offsetting the negative impact on benefit levels would be to pair the change in benefit computation years with a change in payroll taxes for people over a certain age (see, for example, Goda, Shoven, and Slavov (2006) or Butrica, Johnson, Smith, and Steuerle (2006)). The specific proposal considered here pairs an increase in benefit computation years from 35 to 40 with elimination of the employee share of the payroll tax for people age 62 and older (62 is the Early Eligibility Age (EEA) for retiree benefits). That would be an administratively feasible way to lower and flatten the age slope of effective tax rates, whereas introducing a direct link between years worked and payroll taxes would require new reporting and validation systems. From a budgetary perspective the two changes are basically offsetting in the long run because the budgetary savings from raising computation years would match the loss in payroll tax revenues. The data suggest that this fairly simple and budget-neutral change could have a significant impact on effective tax rates for workers nearing retirement.

DATA AND METHODS

The goal of this paper is to measure differences in effective tax rates across and within age, sex, and lifetime earnings groups. There are tradeoffs associated with choosing a data set for this type of analysis. Given the focus, the choice here is to use a large administrative data set with uncapped longitudinal earnings records from a fairly narrow birth cohort for which the actual data are complete through age 65. One drawback is that using the administrative data limits the demographic information somewhat, which makes it infeasible to directly assess how auxiliary benefits would affect the conclusions.

The calculation of effective tax rates is based on the earnings information for each person in the data set, combined with assumptions about mortality and discounting. One of the goals of this paper is to show how patterns of effective tax rates are sensitive to these assumptions, so the calculations are done using two approaches to mortality (with and without adjustments for differential mortality) and alternative values for the discount rate. Although the effective tax rate is a well-defined concept, the exact estimates are fairly sensitive to the details used in the computations.

Data

The data used in this paper are taken from the Continuous Work History Survey (CWHS) data system maintained by the Social Security Administration (SSA). The CWHS sampling frame is a one percent random sample of every Social Security Number (SSN) ever issued. The data sets are constructed by combining information from several SSA reporting systems that have been linked to the CWHS sample: the Summary and Detailed Earnings Record files (SER and DER), the Master Beneficiary File (MBF), and the core demographics file (Numident). The information from the various files is used to construct a longitudinal data record with each person's total earnings, Social Security beneficiary status, and basic demographics (birth, death, and sex).