Contributory Retirement Saving Plans: Differences Across Earnings Groups and Implications for Retirement Security

by
Social Security Bulletin, Vol. 77 No. 2, 2017

This study examines how earnings levels affect workers' access to, participation in, and contributions to defined contribution retirement plans. To what extent do these outcomes improve with higher earnings? Did the relationships change between 2006 and 2012? We match a nationally representative sample of Survey of Income and Program Participation respondents to data from their W-2 tax records. We find that access, participation, and contributions increase as earnings increase, even after controlling for key socioeconomic and labor-market covariates. Low earners are less likely to be offered a plan and to participate when one is offered, and they tend to contribute a smaller share of their earnings when participating. We also find that the earnings gradient changed little between 2006 and 2012, despite changing economic conditions.


Irena Dushi is an economist with the Office of Policy Evaluation and Modeling, Office of Research, Evaluation, and Statistics (ORES), Office of Retirement and Disability Policy (ORDP), Social Security Administration (SSA). When this article was written, Howard Iams was a senior research adviser to ORES, ORDP, SSA. Christopher Tamborini is a research analyst with the Office of Retirement Policy, ORDP, SSA.

The findings and conclusions presented in the Bulletin are those of the authors and do not necessarily represent the views of the Social Security Administration.

Introduction

Selected Abbreviations
DB defined benefit
DC defined contribution
OLS ordinary least squares
SIPP Survey of Income and Program Participation

In the United States, workplace pensions are a primary mode of retirement saving (Hardy and Shuey 2000; Herd 2009; O'Rand 2011; Poterba 2014; Shuey and O'Rand 2004; Warner, Hayward, and Hardy 2010). Because Social Security monthly benefits typically replace around 40 percent of monthly preretirement earnings, workers who wish to maintain their current standard of living after retiring must accumulate resources by other means; yet studies document low retirement saving levels (Fisher and others 2009; Knoll, Tamborini, and Whitman 2012). Estimates based on the 2013 Survey of Consumer Finances indicate that 41 percent of American households headed by individuals aged 55–64 have no savings in retirement accounts. Even more striking is the sharp variation by household income. The proportion of households headed by individuals aged 55–64 that have any retirement savings ranges from 9 percent in the lowest income quintile to 68 percent in the middle quintile and to 94 percent in the top quintile (Government Accountability Office 2015, Tables 1 and 3).

In recent decades, the dominant type of private pension offering shifted from traditional defined benefit (DB) plans to defined contribution (DC) plans such as the familiar 401(k). DC plan contributions today represent the primary means of private retirement saving among American workers. In this context, it has become increasingly important to understand who has access to DC retirement plans, who participates in them, and how much the participants contribute to them (Shuey and O'Rand 2004; Ekerdt 2010; Dushi and Iams 2015; Miller 2015; Tamborini and Purcell 2016).

In this article, we attempt to advance the understanding of how U.S. workers prepare for retirement by examining how DC pension savings vary across the earnings distribution and whether those patterns have changed in recent years. Specifically, we investigate the extent of an earnings gradient in access to, participation in, and levels of contribution to DC retirement plans. Do these three outcomes increase at the upper levels of the earnings distribution? This question is important because the connection between earnings and pension savings is likely to be a key factor influencing retirement resource accumulation during one's working life. Further, the increasing prevalence of DC-type plans is likely to have broadened the relationship between earnings and pension savings. In contrast with DB plans, which are generally mandatory and funded mainly by employers, DC plans are voluntary and require workers to decide what portion of their earnings to contribute; that is, to decide how much of today's consumption to give up for consumption in retirement. Consequently, a worker's earnings level is likely not only to be a major determinant of access to a DC plan but also to influence participation and contribution decisions.

Although a rather extensive literature examines why people save and how saving affects retirement wealth (for example, Poterba, Venti, and Wise 1998, 2000; Venti and Wise 1999), few studies have analyzed the extent of an earnings gradient in contributory retirement plans. Most research has addressed earnings as a control variable (for example, Butrica and Smith 2014; Knoll, Tamborini, and Whitman 2012) rather than as a central pathway to accumulation of retirement resources. In one of the few direct assessments of differential outcomes by earnings level, Dushi, Iams, and Tamborini (2011) find a positive earnings gradient in DC plan participation but pay little attention to the multiple pathways through which the gradient can arise, such as plan access and take-up.1 Other studies have found that earnings differentials contribute to variations in pension outcomes between women and men (Hardy and Shuey 2000), across educational-attainment groups (Tamborini and Kim 2017), and in retirement timing (Raymo and others 2011).

We also examine whether the relationship between earnings levels and DC pension outcomes changed between 2006 and 2012. The Pension Protection Act of 2006 may have led to increased DC plan participation and contributions because it incorporated automatic enrollment and default contribution rates. However, the Great Recession initiated a countervailing trend, with decreases in participation and contribution rates between 2007 and 2009 (Dushi and Iams 2015). The economy had largely recovered by 2012; did DC plan participation and contribution patterns change relative to prerecession years?

To explore these questions, we use a unique data set that matches a nationally representative sample of respondents from the 2004 and 2008 panels of the Survey of Income and Program Participation (SIPP) to individual-level earnings and contribution data from Internal Revenue Service Form W-2 tax records. The W-2 linkage reduces the measurement error that is common to self-reported data, especially in terms of DC plan contributions (Dushi and Iams 2011; Dushi and Honig 2015; Kim and Tamborini 2014). These matched data enable us to provide robust estimates of differentials in DC pension access and participation according to workers' position in the earnings distribution. We examine differences at the earnings-decile level of detail to provide a fine-grained analysis.

This study advances the understanding of variation in DC pension outcomes by workers' earnings levels. Our findings also provide insights into the possible significance of increasing earnings dispersion (Autor 2014; Kim and Sakamoto 2008) for the accumulation of retirement resources during working years. Earnings gradients in retirement savings, if sustained over the life course, can compound over time and help explain how and why retirement resources differ from one individual (or family) to another (Crystal, Shea, and Reyes 2016).

Background

U.S. retirement income is often described as a three-legged stool supported by Social Security benefits, employer-provided pensions, and personal savings and assets. After Social Security, employer-provided pensions are the most important source of retirement income (Shuey and O'Rand 2004).

Since the 1980s, employer-provided pensions have undergone a dramatic transition from consisting primarily of DB plans to consisting primarily of DC plans (Ekerdt 2010; O'Rand 2011). In DB plans, employees are enrolled automatically2 and the pension is funded mainly by employers. Covered employees do not have to decide whether and how much to contribute; consequently, they may not view the employer's contributions as a deduction from their paycheck. The DB pension benefit formula accounts for years of service and (usually) for earnings in only the last 3 to 5 years of work, which for many workers are the highest-earning years. Benefits are paid as a lifetime annuity.

By contrast, DC plans are generally voluntary and require workers to decide not only whether but also how much to contribute—and where to invest those funds.3 Because these contributions are deducted from gross pay, workers are likely aware of the direct link between retirement contributions and current consumption. In this context, a person's earnings level is likely to affect participation and contribution decisions.

Relating Earnings to DC Pension Outcomes

Our analysis explores three pathways likely to underlie an earnings gradient in DC pension savings. The first pathway is through a positive relationship between earnings level and plan access. To participate in a DC plan, an individual must be offered one by his or her employer. Studies of nonwage labor compensation have noted that employers face a competitive labor market for high-skill workers, making them more likely to offer pension plans—particularly at large firms (Dushi, Iams, and Lichtenstein 2015). Thus, if high earners are more likely to have access to a DC plan, then differential access by earnings level may be one source of the gradient in retirement savings.4

The second pathway is a positive relationship between earnings level and retirement plan participation (Butrica and Smith 2014; Copeland 2013). Low earners with tighter budget constraints may find it more difficult to divert current income toward retirement savings. Low earners may also have less incentive to save, in part because Social Security's progressive benefit formula will provide them with higher preretirement-income replacement rates. In addition, earnings level may shape a person's social networks and peer interactions (DiMaggio and Garip 2012), which in turn may influence saving decisions involving DC plans (Koposko and others 2015).

The third pathway in an earnings gradient involves the relationship between earnings level and DC plan contributions (Pattison and Waldron 2008). Among plan participants, both the dollar amounts and the percentages of gross earnings contributed to a DC account (up to the annual contribution limit) rise as earnings levels increase. This may reflect high earners' desire to replace a higher proportion of preretirement earnings, combined with greater budgetary latitude and greater tax benefits. Conversely, low earners may contribute less because of greater budgetary constraints, lower tax benefits, and a higher replacement rate from Social Security benefits. Hence, the marginal utility of reducing current consumption to fund future consumption is lower among low earners.

Prior research has assessed the determinants of DC retirement plans outcomes, but to date, studies either have examined earnings level as a control variable or have examined its influence on only one outcome. A recent study using representative data highlights the positive relationship between earnings level and DC-plan participation and contributions among U.S. workers in 2004, but it does not examine access or take-up rates (Dushi, Iams, and Tamborini 2011). Studies that examine factors besides earnings level have found evidence linking plan participation to a worker's family structure (Knoll, Tamborini, and Whitman 2012; Tamborini and Purcell 2016), education (Hardy and Shuey 2000; Tamborini and Kim 2017), and race/ethnicity (Kuan, Cullen, and Modrek 2015), as well as to employer firm size (Dushi, Iams, and Lichtenstein 2015). In addition, longitudinal research has shown that earnings and employment shocks influence participation and contribution decisions (Dushi and Iams 2015; Tamborini, Purcell, and Iams 2013).

In this article, we explore the earnings gradient not only in DC plan participation (the most studied dimension to date) but also in plan access, take-up, and contributions based on W-2 data. This attention to multiple outcomes, along with the use of matched administrative data, allows us to capture the earnings-level differentials in workplace retirement savings more fully than in prior research. In addition, our analysis examines the earnings gradient over a span of recent years reflecting different economic conditions and prevailing pension offerings.

Data

The data for this study come from the 2004 and 2008 panels of the SIPP, a nationally representative household survey administered by the Census Bureau, matched to W-2 tax records. In a given panel, two types of questionnaires are administered: Core and Topical Modules. The Core Module collects a common set of demographic and labor-market information, whereas each of a rotating set of Topical Modules covers a given topic in depth.

We focus on the survey waves that include the Topical Module on Retirement and Pension Plan Coverage, which provides information about respondents' employer-sponsored retirement plans. To test for changes in the relationship between earnings levels and DC plan outcomes in 2006 and 2012, we use data from wave 7 of the 2004 panel, with interviews conducted from February to May 2006; and from wave 11 of the 2008 panel, with interviews conducted from January to April 2012.

We link the SIPP data with respondents' tax records from employer-reported W-2 forms, which are available in SSA's Detailed Earnings Record file. These records provide information on annual wage-and-salary earnings and tax-deferred contributions to DC retirement plans for all jobs since 1990. We use W-2 information for calendar years 2006 and 2012.5

Our study population consists of full-time wage and salary workers aged 25–59 at the time of the interview.6 From this sample, we remove marginal earners by excluding workers with annual earnings totaling less than the equivalent of four quarters of Social Security coverage ($3,880 in 2006 and $4,520 in 2012). The study population is further restricted to SIPP respondents with linked W-2 tax records. For brevity, we refer to this sample as full-time workers (or, simply, workers) hereafter. The W-2 match rate is high: around 80 percent for the 2004 SIPP panel and around 90 percent for the 2008 panel. Potential bias because of nonmatched respondents is minimal (Czajka, Mabli, and Cody 2008; Davis and Mazumder 2011); nevertheless, we adjust the survey weights for nonmatches to preserve the national representativeness of our sample.7 We pool the data from both SIPP panels; our final matched sample contains 35,558 persons, of which 20,320 are from the 2004 panel (providing data for 2006) and 15,238 are from the 2008 panel (providing data for 2012).

Analysis

As noted above, we focus on three key indicators related to DC retirement savings plans: (a) access, (b) participation, and (c) contribution levels. A binary measure of access (zero/one) indicates whether the worker was offered a DC retirement plan by her or his employer. We define SIPP respondents who report that their employer offered a DC plan as having access to a plan. In addition, respondents with a positive contribution to a DC plan according to their W-2 tax records for the survey year are defined as having access to a plan, regardless of their SIPP response.

The second indicator is participation in a DC retirement plan. We define a worker as a plan participant if her or his W-2 record shows a tax-deferred contribution to a retirement account. We examine the participation rate among all full-time workers as well as that for the subset of workers who are offered a DC plan and elect to take it up.

The third indicator is annual contributions to a DC retirement plan among participants. Contribution amounts are from the W-2 records; we adjust the contribution amounts in 2006 to 2012 dollars. Based on this information, we calculate the contribution rate, defined as the percentage of total annual wages that a worker contributes to a DC account.8

We employ both descriptive and multivariate regression analysis to assess the earnings gradient for each outcome (access, participation, and contributions) while controlling for key covariates as described below. More specifically, we use a standard probit model to estimate the probability of having access to a DC plan and, separately, of participating in a plan.

We note that the estimates from the probit model of participation would not be accurate if the unobserved characteristics that affect the probability of being offered a plan were correlated with the unobserved characteristics that affect the probability of participating in the plan. In other words, the unobservable characteristics of workers whose employers do not offer a plan may differ from those of workers whose employers do offer a plan, and the latter workers may be more likely to participate for reasons unrelated to having received a plan offer (for example, because of their preference for saving). Probit estimates that do not control for that type of selection will likely be biased. To account for that possibility, we also estimate a bivariate probit (or Heckman selection) model. For identification purposes, in the bivariate model, we use two variables as exclusion restrictions in the plan offer equation; those variables measure the proportion of medium- and small-size firms in the respondent's state of residence. The exclusion restrictions are correlated with the probability of being offered a plan but not with the probability of participation. In addition, if the error terms (or unobservable characteristics) in the offer and participation equations are correlated—in technical terms, the rho coefficient—and the rho is statistically significant, then bivariate probit estimates are more appropriate than standard probit estimates. Finally, we use ordinary least squares (OLS) regression models to estimate contribution amounts and rates.

The main independent variable of interest, total annual earnings, is obtained from the W-2 records. To explore an earnings gradient, we sort workers by decile based on the earnings distribution in each study year. To test whether the earnings gradient changed from 2006 to 2012, the regression analyses use the pooled samples and include interaction variables between earnings deciles and year. Hence, the coefficients of earnings deciles give estimated effects for the reference year (2012), and the year-dummy and interaction terms give the additional effects for year 2006.

Our models also include controls for socioeconomic and labor market characteristics (based on SIPP data) that are expected to affect access to, participation in, and contributions to DC plans. These explanatory variables include sex, age (25–39, 40–49, 50–59), whether married, educational attainment (less than high school, high school graduate, some college, bachelor's degree), and race/ethnicity (non-Hispanic white, non-Hispanic black, non-Hispanic other, Hispanic). Dichotomous labor-market variables indicate the respondent's class of work (private versus public sector), firm size (fewer than 25, 25–99, and 100 or more employees), occupation (five broad categories), industry of employment (seven broad categories), and whether the employer matches DC plan contributions. We also control for household income and homeownership status. All reported estimates use sample weights adjusted for the probability of match to W-2 records.

Results

In this section, we discuss the results of our analysis. We address each of the three DC plan outcomes in turn.

Access to a DC Plan

Table 1 shows that the overall percentage of full-time workers who were offered a DC plan was 68.8 percent in 2006 and 73.1 percent in 2012. In both years, access differed significantly by earnings decile, increasing monotonically with higher earnings. For example, in 2006, 31.8 percent of workers in the lowest earnings decile were offered a DC plan by their employer, versus 91.5 percent of workers in the highest earnings decile—a statistically significant gap of nearly 60 percentage points. For those in the middle (5th) earnings decile, the offer rate was about twice that of workers in the lowest decile in both 2006 and 2012.

Table 1. DC retirement plan offer, participation, and take-up rates for full-time wage and salary workers aged 25–59 in 2006 and 2012, by earnings decile
Decile Offer rate a Participation rate b among—
All workers Workers offered a plan (take-up rate)
2006 2012 2006 2012 2006 2012
Total 68.8 73.1 49.4 51.9 71.8 71.0
1st (lowest) 31.8 37.5 13.7 17.2 43.0 45.8
2nd 48.8 55.1 26.4 30.9 54.1 56.1
3rd 58.6 64.9 36.2 38.6 61.8 59.5
4th 65.6 71.8 43.1 44.9 65.6 62.6
5th 68.7 74.9 45.8 48.2 66.6 64.4
6th 74.6 78.1 51.7 54.9 69.3 70.3
7th 76.8 82.5 56.9 60.3 74.1 73.1
8th 82.6 84.2 64.3 66.3 77.8 78.7
9th 89.2 89.3 74.1 75.4 83.1 84.4
10th (highest) 91.5 92.6 81.9 81.9 89.5 88.3
Sample size 20,320 15,238 20,320 15,238 14,259 11,253
SOURCE: Authors' calculations using data from SIPP 2004 Panel (wave 7) and 2008 Panel (wave 11) matched to Form W-2 tax records.
NOTES: Samples consist of respondents who were full-time wage and salary workers with matched W-2 records and who had earnings that qualified for four quarters of Social Security coverage (at least $3,880 in 2006; at least $4,520 in 2012) and who contributed to (or participated in) a DC plan.
Sample sizes are unweighted. Estimated offer and participation rates are weighted using SIPP complex survey weights, which are adjusted to account for respondents without a match to W-2 records.
a. Percentage of sample members who either reported in SIPP being offered or participating in a DC plan or whose W-2 records indicated contribution to a DC plan.
b. Percentage of sample members whose W-2 records indicated that they made any tax-deferred DC plan contributions in the given year. The total amount of tax-deferred contributions is recorded in Box 12 in the W-2 record.

Table 1 shows a slight improvement in the share of workers with access to a DC plan over the study period, particularly in the lower half of the earnings distribution. For example, the share of full-time workers in the 2nd decile with access to a DC plan increased from 48.8 percent in 2006 to 55.1 percent in 2012. However, substantial shares of workers still were not offered a DC plan in 2012, particularly in the first three deciles. In the lowest earnings decile, about 62 percent of workers did not have access to a DC plan.

Probit model estimates clearly reveal an earnings gradient in the probability of being offered a DC plan even after controlling for socioeconomic and labor-market characteristics (Table 2). Holding the covariates constant, workers in the highest earnings decile were 27.8 percentage points (or 37 percent relative to the mean) more likely to be offered a plan than were those in the lowest decile (the reference category). Those in middle of the distribution (the 5th decile) were 20.7 percentage points (or 28 percent relative to the mean) more likely to be offered a plan than were those in the lowest decile.

Table 2. Probit estimates of probability of being offered and of participating in a DC plan among full-time wage and salary workers aged 25–59, 2006 and 2012
Variable Offer Participation among—
All workers Workers offered a plan (take-up)
Standard probit model Bivariate probit model
Marginal effect Standard error Marginal effect Standard error Marginal effect Standard error Marginal effect Standard error
Earnings decile
1st (lowest) (omitted) . . . . . . . . . . . . . . . . . . . . . . . .
2nd .121** .013 .160** .022 .070** .022 .114** .021
3rd .167** .011 .230** .020 .105** .020 .158** .022
4th .196** .010 .277** .019 .127** .018 .207** .023
5th .207** .009 .297** .018 .143** .017 .223** .023
6th .220** .009 .345** .017 .180** .015 .281** .023
7th .236** .008 .374** .016 .194** .014 .308** .024
8th .240** .008 .409** .014 .229** .013 .370** .023
9th .263** .007 .457** .012 .258** .011 .428** .025
10th (highest) .278** .006 .481** .011 .270** .011 .490** .026
Year dummy (if 2006 = 1) .000 .017 -.019 .024 -.015 .027 -.047* .021
Interaction terms
Year × decile 1 (lowest) (omitted) . . . . . . . . . . . . . . . . . . . . . . . .
Year × decile 2 -.020 .024 -.005 .032 .001 .035 .012 .027
Year × decile 3 -.020 .024 .019 .032 .028 .032 .054 .027
Year × decile 4 -.018 .025 .031 .031 .031 .032 .051 .026
Year × decile 5 -.024 .025 .027 .031 .036 .031 .073 .026
Year × decile 6 -.011 .025 .011 .031 .004 .033 .051* .026
Year × decile 7 -.035 .026 .011 .031 .020 .032 .067* .026
Year × decile 8 .014 .025 .027 .031 .005 .033 .054* .026
Year × decile 9 .026 .026 .027 .032 -.001 .034 .059* .027
Year × decile 10 .007 .027 .051 .033 .041 .033 .068* .028
Mean estimated probability .744 .503 .737 .737
Pseudo R2 .192 .165 .105 . . .
Rho coefficient . . . . . . . . . .665** .113
χ2(1)= . . . . . . . . . 17.14
Probability greater than χ2= . . . . . . . . . .000
Sample size 35,558 25,512
SOURCE: Authors' calculations using data from SIPP 2004 Panel (wave 7) and 2008 Panel (wave 11) matched to Form W-2 tax records.
NOTES: Samples consist of respondents who were full-time wage and salary workers with matched W-2 records and who had earnings that qualified for four quarters of Social Security coverage (at least $3,880 in 2006; at least $4,520 in 2012) and who contributed to (or participated in) a DC plan.
Sample sizes are unweighted. Reported estimates are weighted using SIPP complex survey weights, which are adjusted to account for respondents without a match to W-2 records.
Model estimates control for demographic characteristics (sex, age, educational attainment, marital status, race/ethnicity); household characteristics (total income and homeownership); occupation, industry, firm size, and sector (public, private, nonprofit) of employment. The take-up probit model also controls for whether employer matches contributions.
. . . = not applicable.
* = statistically significant at the 5 percent level.
** = statistically significant at the 1 percent level.

When we account for the covariates, the coefficient of the study-year dummy shows that there was no difference between 2006 and 2012 in the probability of being offered a DC plan. Importantly, the interaction terms were not statistically significant, suggesting that the patterns of differential access by earnings decile were similar in both years.

Participation in a DC Plan

We examine DC plan participation—defined as contributing to a plan, according to W-2 records—by earnings decile, for the entire sample of full-time workers and separately for the subset of workers with access to a DC plan. Table 1 shows that a large fraction of all workers—around half—did not contribute to a DC retirement plan in either year. The gap in participation rates between low and high earners is large, ranging from about 14 percent (in 2006) and 17 percent (in 2012) for workers in the lowest earnings decile to about 82 percent (in both years) for those in the highest decile. This gap is a byproduct of the dual probabilities of whether a worker was offered a DC plan and whether, if receiving such an offer, the worker took up (that is, contributed to) the plan.

Over the study period, the participation rates for all workers increased slightly for every decile except the highest (which did not change). We also see a sharp positive earnings gradient in take-up rates, but the differences are somewhat attenuated relative to participation rates for all workers. Overall, about 71 percent of workers offered a plan contributed to that plan in each study year. Unsurprisingly, take-up rates increased as earnings levels rose. For instance, in the lowest decile, 43.0 percent of workers offered a plan elected to contribute to it in 2006, compared with 89.5 percent of those in the highest earnings decile. A similar pattern is evident for 2012.

Table 2 presents probit model estimates of the probability of DC plan participation among all workers and among those who were offered one. We find strong evidence of an earnings gradient in participation, even after adjusting for potentially confounding covariates. Among workers overall, those in the 2nd decile were 16.0 percentage points more likely to participate than were those in the 1st decile (the reference category), whereas those in the highest decile were 48.1 percentage points more likely to participate. Among workers with access to a plan, earners in the 2nd decile were only 7.0 percentage points more likely to take up the offer than were workers in the lowest earnings decile, whereas those in the highest earnings decile were 27.0 percentage points more likely to take up the offer. Regression results also reveal that, once we control for other explanatory variables, the earnings gradient in DC participation did not change substantively from 2006 to 2012, as indicated by the interaction terms that are not statistically significant. Furthermore, there were no significant increases between 2006 and 2012 in participation rates.

As noted in the analysis section, we use a Heckman bivariate probit model, which accounts for correlation in the unobserved characteristics, to jointly estimate offer and take-up probabilities. Our estimates indicate that the unobservable characteristics in the offer equation are correlated with the unobservables in the take-up decision and the rho coefficient is statistically significant. This finding suggests that workers whose employers offer a DC plan are more likely to participate and that the marginal effects by earnings decile from the standard probit model represent lower-bound estimates. Furthermore, although the overall pattern by earnings decile is the same for the standard and bivariate models, we find that the magnitude of marginal effects in the bivariate model's take-up equation is greater across all earnings deciles, and that the gap between the lowest and the highest deciles (11.4–49.0 percentage points) is larger. One difference worth noting is that workers in 2006 were less likely to participate (by 4.7 percentage points) than workers in 2012 were, as indicated by the year-dummy variable. Although the bivariate model's interaction terms in the top five earnings deciles suggest a positive additional effect on the probability of participation in 2006, the effect becomes small and insignificant when taking the negative year dummy (−4.7 percentage points) into account.

Contribution Rates and Levels

Among all DC plan participants, the median contribution rate was around 5 percent of annual salary in both 2006 and 2012 (Table 3). Contribution rates generally increased by earnings decile, ranging in 2006 from 3.9 percent in the lowest decile to 7.1 percent in the highest decile. In 2012, the pattern of median contribution rates across earnings deciles was largely similar to that for 2006.

Table 3. DC retirement plan median contribution rates and amounts among full-time wage and salary workers aged 25–59 who participated in DC plans in 2006 and 2012, by earnings decile
Decile Median contribution rate a Median contribution amount (2012 $)
2006 2012 2006 2012
Total 5.2 5.0 2,981 2,717
1st (lowest) 3.9 3.2 638 498
2nd 3.3 3.2 827 786
3rd 3.9 3.5 1,192 1,028
4th 3.7 3.9 1,376 1,371
5th 4.2 4.2 1,832 1,769
6th 4.9 5.0 2,385 2,427
7th 5.2 5.0 3,065 2,894
8th 5.8 5.2 4,112 3,781
9th 6.6 6.0 6,039 5,399
10th (highest) 7.1 6.7 12,368 11,902
Sample size 10,280 8,020 10,280 8,020
SOURCE: Authors' calculations using data from SIPP 2004 Panel (wave 7) and 2008 Panel (wave 11) matched to Form W-2 tax records.
NOTES: Samples consist of respondents who were full-time wage and salary workers with matched W-2 records and who had earnings that qualified for four quarters of Social Security coverage (at least $3,880 in 2006; at least $4,520 in 2012) and who contributed to (or participated in) a DC plan.
Sample sizes are unweighted. Estimated contribution rates and amounts are weighted using SIPP complex survey weights, which are adjusted to account for respondents without a match to W-2 records.
a. Percentage of total annual wages contributed to a DC plan.

In terms of amounts (in 2012 dollars), the median annual contribution dramatically increased with earnings level in both study years. In 2006, the median contribution was $2,981, but the differential across earnings deciles was substantial, ranging from $638 in the lowest decile to $12,368 in the highest one. Interestingly, median contribution amounts in 2012 were slightly lower than those in 2006 for all but the 6th decile. However, the earnings gradient was similar, with the median contribution amount increasing from $498 in the lowest decile to $11,902 in the highest one. In both years, for the majority (more than 70 percent) of workers contributing to a DC plan, the annual contribution level was lower than $3,100, an amount well below annual contribution limits ($15,000 in 2006 and $15,500 in 2012, not including additional “catch–up” contributions that older workers are eligible to make).

Table 4 presents the OLS regression estimates of DC plan contribution rates and amounts by earnings decile. The results confirm that a steep earnings gradient exists in both indicators, even after controlling for key socioeconomic and labor-market characteristics. Compared with workers in the lowest earnings decile (the reference group), median annual contribution amounts among workers in the 2nd through 4th earnings deciles were only slightly higher (from $341 to $776). Among workers in the 5th through 8th deciles, median contributions exceeded those of workers in the lowest decile by $1,074 to $3,039, whereas workers in the highest earnings decile contributed an average of about $9,864 more than did those in the lowest decile. The earnings gradients in each study year were largely similar, with only one interaction term revealing a significant difference: Compared with workers in the lowest earnings decile, those in the 8th decile contributed $630 more in 2006 than did those in the same decile in 2012. The contribution rates of workers in the 8th through 10th deciles were significantly higher than those of participants in the lowest earnings decile. There were no significant differences between 2006 and 2012 in contribution rates by earnings decile.9

Table 4. OLS estimates of DC plan contribution rates and amounts among full-time wage and salary workers aged 25–59 who participated in DC plans in 2006 and 2012
Variable Contribution rate Contribution amount (2012 $)
Coefficient Standard error Coefficient Standard error
Earnings decile
1st (lowest) (omitted) . . . . . . . . . . . .
2nd -.405 .422 341** 113
3rd -.369 .404 583** 117
4th -.200 .401 776** 117
5th -.051 .393 1,074** 122
6th .675 .413 1,721** 141
7th .672 .403 2,207** 147
8th .796* .396 3,039** 157
9th 1.646** .404 5,175** 196
10th (highest) 1.038** .395 9,864** 223
Year dummy (if 2006 = 1) .666 .680 156 119
Interaction terms
Year × decile 1 (lowest) (omitted) . . . . . . . . . . . .
Year × decile 2 -.315 .764 -7 154
Year × decile 3 -.141 .742 47 157
Year × decile 4 -.632 .722 -11 154
Year × decile 5 -.213 .724 146 161
Year × decile 6 -.709 .730 -70 179
Year × decile 7 -.147 .729 209 197
Year × decile 8 .391 .724 630** 212
Year × decile 9 .089 .722 404 250
Year × decile 10 -.206 .704 -125 280
Mean dependent variable 6.370 5,016
R2 0.124 0.498
Sample size 18,300
SOURCE: Authors' calculations using data from SIPP 2004 Panel (wave 7) and 2008 Panel (wave 11) matched to Form W-2 tax records.
NOTES: Sample consists of respondents who were full-time wage and salary workers with matched W-2 records and who had earnings that qualified for four quarters of Social Security coverage (at least $3,880 in 2006; at least $4,520 in 2012) and who contributed to (or participated in) a DC plan.
Sample size is unweighted. Reported estimates are weighted using SIPP complex survey weights, which are adjusted to account for respondents without a match to W-2 records.
Model estimates control for demographic characteristics (sex, age, educational attainment, marital status, race/ethnicity); household characteristics (total income and homeownership); occupation, industry, firm size, and sector (public, private, nonprofit) of employment; and whether employer matches contributions.
. . . = not applicable.
* = statistically significant at the 5 percent level.
** = statistically significant at the 1 percent level.

Discussion

Given the shift from DB plans to voluntary DC retirement saving plans as the dominant type of employer-provided pension, retirement income in the United States increasingly depends on several outcomes pertinent to DC pensions, and earnings level plays an important role in shaping those outcomes. Public policies that seek to improve retirement security should recognize how DC retirement plan outcomes vary across the earnings distribution. It is important to understand who has access to DC plans, who participates in them, and how much they contribute.

We find clear evidence of a steep earnings gradient in several DC plan outcomes, even after accounting for an array of socioeconomic and labor-market covariates. Access, participation, and contribution levels increase as earnings increase, in most cases monotonically. We find that earners in the bottom half of the earnings distribution are not only less likely to be offered a plan but are also less likely to participate when offered one. According to our estimates using W-2 records, less than 50 percent of all full-time wage and salary workers with earnings below the median participate in a DC plan and a substantial proportion of workers with access to a plan (about 29 percent) elect not to participate.

Our analysis suggests that low earners, and even workers in the middle of the earnings distribution who do participate, save lower dollar amounts and contribute smaller shares of their earnings. Because low earners receive relatively higher preretirement-income replacement rates from their Social Security benefits, one might argue that they have less need to save through DC-type plans. However, low earners are more likely to fall into poverty during retirement (Favreault 2009; Munnell 2004), and Social Security may be the only source of income for many of them.

A worker with low pension savings over long stretches of his or her working life would likely depend primarily on Social Security benefits in retirement and would thus be most sensitive to any future Social Security policy changes. Moreover, even among the majority of workers who save for retirement, the typically low annual contributions, even if sustained for many years, may not yield resource levels in later life that some may expect. For example, the median annual contribution for our sample was around $3,000. Assuming 30 years of contributions at that level and disregarding compound interest, inflation, and preretirement withdrawals, a worker would accumulate DC plan savings ranging from about $90,000 (assuming no employer contributions) to $135,000 (assuming 50 percent employer contributions).10 If those account balances were drawn down in monthly payments over 20 years, each monthly payment would be between $375 and $562.11

We also find consistency in the earnings gradient in DC plan outcomes for 2006 and 2012, despite changes in pension and economic conditions. Regression analysis reveals no statistically significant differences between study years in plan offer rates, participation rates, and contribution patterns.

Limitations and Future Directions

This study is not without limitations. For instance, our analysis shows the cross-sectional relationship between earnings and DC pension outcomes rather than longitudinal patterns. Workers with higher earnings prospects or those with changing earnings levels might alter their DC plan saving behaviors over time. We estimated a series of sensitivity tests to explore possible differences across age groups (using age-stratified models), which showed results similar to those presented here. In addition, whether the decision on how to invest DC plan savings varies by earnings level is an important issue not addressed in this article. Our analysis also does not explore possible interactions between pension design features (such as employer matching of employee contributions and account withdrawals) and earnings levels. Although we control for important covariates, other unmeasured variables could confound the correlations. The extent to which savings in contributory retirement plans vary by level of household resources (and the role of spousal earnings) is another factor not yet explored.

Notes

1 Plan take-up refers to participation among workers who have been offered a plan, as opposed to participation among workers overall.

2 However, employees generally must satisfy a years-of-service requirement (often as long as 5 years) to be vested in the plan.

3 The 2006 Pension Protection Act enables employers to enroll their employees automatically in a DC plan at a default contribution rate, from which employees can opt out or change the rate. In DC plans, employees bear the risks of investing and managing the account prior to and during retirement; they also bear the longevity risk. In DB plans, employers bear the investment and longevity risks.

4 Of course, employees who do not have access to a retirement plan through their employer can save through individual retirement accounts (IRAs). However, workers with lower disposable income are less likely to use a nonworkplace saving plan (Holden and Schrass 2017, Figure 5).

5 Access to these data is restricted and based on agreements between SSA and the Census Bureau (Davies and Fisher 2009; Olsen and Hudson 2009). The data are accessed at a secured site and undergo disclosure review before they are approved for release.

6 SIPP defines full-time work as 35 or more hours per week; we use the same definition here.

7 Drawing from previous work (Couch, Tamborini, and Reznik 2015), we use logistic regression to estimate the probability of a successful match, controlling for socioeconomic characteristics such as age, education, marital status, and race/ethnicity; we then multiply SIPP person-weights by the inverse of the match probability.

8 Wages in 2006 are also adjusted to 2012 dollars.

9 We also used a two-stage Heckman selection model to estimate the probabilities of plan take-up and contribution amount (or contribution rates) using the functional form for identification and exclusion restriction in the first stage and found that the rho coefficient (the correlation of the error terms in the two equations) was not statistically significant. Hence, Table 4 presents the standard OLS regression estimates rather than the Heckman selection model estimates.

10 Assuming 30 years of consistent contributions ignores how employment and earnings shocks affect a person's retirement savings (Dushi and Iams 2015). This example would be consistent with an individual investor whose investment returns keep up only with inflation, which might reflect such common mistakes as buying high and selling low (Malkiel and Ellis 2013, Chapter 4).

11 These amounts are consistent with estimates based on Survey of Consumer Finances data (Government Accountability Office 2015).

References

Autor, David H. 2014. “Skills, Education, and the Rise of Earnings Inequality Among the ‘Other 99 Percent.'” Science 344(6): 843–851.

Butrica, Barbara A., and Karen E. Smith. 2014. “401(k) Participant Behavior in a Volatile Economy.” Journal of Pension Economics and Finance 15(1): 1–29.

Copeland, Craig. 2013. “Retirement Plan Participation: Survey of Income and Program Participation (SIPP) Data, 2012.” EBRI Notes 34(8): 2–13.

Couch, Kenneth A., Christopher R. Tamborini, and Gayle L. Reznik. 2015. “The Long-Term Health Implications of Marital Disruption: Divorce, Work Limits, and Social Security Disability Benefits Among Men.” Demography 52(5): 1487–1512.

Crystal, Stephen, Dennis G. Shea, and Adriana M. Reyes. 2016. “Cumulative Advantage, Cumulative Disadvantage, and Evolving Patterns of Late-Life Inequality.” The Gerontologist. Advance publication. https://academic.oup.com/gerontologist/article/doi/10.1093/geront/gnw056/2632069/Cumulative-Advantage-Cumulative-Disadvantage-and.

Czajka, John L., James Mabli, and Scott Cody. 2008. Sample Loss and Survey Bias in Estimates of Social Security Beneficiaries: A Tale of Two Surveys. Final Report. Washington, DC: Mathematica Policy Research, Inc.

Davies, Paul S., and T. Lynn Fisher. 2009. “Measurement Issues Associated with Using Survey Data Matched with Administrative Data from the Social Security Administration.” Social Security Bulletin 69(2): 1–12.

Davis, Jonathan M. V., and Bhashkar Mazumder. 2011. “An Analysis of Sample Selection and the Reliability of Using Short-Term Earnings Averages in SIPP-SSA Matched Data.” Center for Economic Studies Paper No. CES-WP-11-39. Washington, DC: Census Bureau.

DiMaggio, Paul, and Filiz Garip. 2012. “Network Effects and Social Inequality.” Annual Review of Sociology 38: 93–118.

Dushi, Irena, and Marjorie Honig. 2015. “How Much Do Respondents in the Health and Retirement Study Know About Their Contributions to Tax Deferred Contribution Plans? A Cross-Cohort Comparison.” Journal of Pension Economics and Finance 14(3): 203–239.

Dushi, Irena, and Howard M. Iams. 2011. “The Impact of Response Error on Participation Rates and Contributions to Defined Contribution Pension Plans.” Social Security Bulletin 70(1): 45–60.

———. 2015. “The Impact of Employment and Earnings Shocks on Contribution Behavior in Defined Contribution Plans: 2005–2009.The Journal of Retirement 2(4): 86–104.

Dushi, Irena, Howard M. Iams, and Jules Lichtenstein. 2015. “Retirement Plan Coverage by Firm Size: An Update.” Social Security Bulletin 75(2): 41–55.

Dushi, Irena, Howard M. Iams, and Christopher R. Tamborini. 2011. “Defined Contribution Pension Participation and Contributions by Earnings Levels Using Administrative Data.” Social Security Bulletin 71(2): 67–76.

Ekerdt, David J. 2010. “Frontiers of Research on Work and Retirement.” The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences 65B(1): 69–80.

Favreault, Melissa M. 2009. A New Minimum Benefit for Low Lifetime Earners. Washington, DC: Urban Institute. http://webarchive.urban.org/uploadedpdf/411853_anewminimumbenefit.pdf.

Fisher, Jonathan, David S. Johnson, Joseph T. Marchand, Timothy M. Smeeding, and Barbara B. Torrey. 2009. “Identifying the Poorest Older Americans.” The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences 64B(6): 758–766.

Government Accountability Office. 2015. Retirement Security: Most Households Approaching Retirement Have Low Savings. GAO-15-419. Washington, DC: GAO. http://www.gao.gov/products/GAO-15-419.

Hardy, Melissa A., and Kim Shuey. 2000. “Pension Decisions in a Changing Economy: Gender, Structure, and Choice.” The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences 55(5): S271–S277.

Herd, Pamela. 2009. “The Two-Legged Stool: The Reconfiguration of Risk in Retirement Income Security.” Generations 33(3): 12–18.

Holden, Sarah, and Daniel Schrass. 2017. “The Role of IRAs in US Households' Saving for Retirement, 2016.” ICI Research Perspective 23(1). https://www.ICI.org/pdf/per23-01.pdf.

Kim, ChangHwan, and Arthur Sakamoto. 2008. “The Rise of Intra-Occupational Wage Inequality in the United States, 1983 to 2002.” American Sociological Review 73(1): 129–157.

Kim, ChangHwan, and Christopher R. Tamborini. 2014. “Response Error in Earnings: An Analysis of the Survey of Income and Program Participation Matched With Administrative Data.” Sociological Methods & Research 43(1): 39–72.

Knoll, Melissa A. Z., Christopher R. Tamborini, and Kevin Whitman. 2012. “I Do…Want to Save: Marriage and Retirement Savings in Young Households.” Journal of Marriage and Family 74(1): 86–100.

Koposko, Janet L., Helen Kiso, Douglas A. Hershey, and Paul Gerrans. 2015. “Perceptions of Retirement Savings Relative to Peers.” Work, Aging and Retirement 2(1): 65–72.

Kuan, Kai Yuan, Mark R. Cullen, and Sepideh Modrek. 2015. “Racial Disparities in Savings Behavior for a Continuously Employed Cohort.” NBER Working Paper No. 20937. Cambridge, MA: National Bureau of Economic Research.

Malkiel, Burton G., and Charles D. Ellis. 2013. The Elements of Investing: Easy Lessons for Every Investor. New York, NY: John Wiley & Sons.

Miller, Mark. 2015. “Pushing Aside 401(k)'s for Mandatory Savings Plans.” New York Times, December 11.

Munnell, Alicia H. 2004. “Why Are So Many Older Women Poor?” Just the Facts on Retirement Issues No. 10. Chestnut Hill, MA: Center for Retirement Research at Boston College.

Olsen, Anya, and Russell Hudson. 2009. “Social Security Administration's Master Earnings File: Background Information.” Social Security Bulletin 69(3): 29–45.

O'Rand, Angela M. 2011. “The Devolution of Risk and the Changing Life Course in the United States.” Social Forces 90(1): 1–16.

Pattison, David, and Hilary Waldron. 2008. “Trends in Elective Deferrals of Earnings from 1990–2001 in Social Security Administrative Data.” Research and Statistics Note No. 2008-03. Washington, DC: SSA.

Poterba, James M. 2014. “Retirement Security in an Aging Population.” American Economic Review 104(5): 1–30.

Poterba, James M., Steven F. Venti, and David A. Wise. 1998. “Implications of Rising Personal Retirement Saving.” In Frontiers in the Economics of Aging, edited by David A. Wise, 125–167. Chicago, IL: University of Chicago Press.

———. 2000. “Saver Behavior and 401(k) Retirement Wealth.” American Economic Review 90(2): 297–302.

Raymo, James M., John R. Warren, Megan M. Sweeney, Robert M. Hauser, and Jeong-Hwa Ho. 2011. “Precarious Employment, Bad Jobs, Labor Unions, and Early Retirement.” The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences 66B(2): 249–259.

Shuey, Kim M., and Angela M. O'Rand. 2004. “New Risks for Workers: Pensions, Labor Markets, and Gender.” Annual Review of Sociology 30: 453–477.

Tamborini, Christopher R., and ChangHwan Kim. 2017. “Education and Contributory Pensions at Work: Disadvantages of the Less Educated” Social Forces. Advance publication.

Tamborini, Christopher R., and Patrick Purcell. 2016. “Women's Household Preparation for Retirement at Young and Mid-Adulthood: Differences by Children and Marital Status.” Journal of Family and Economic Issues 37(2): 226–241.

Tamborini, Christopher R., Patrick Purcell, and Howard M. Iams. 2013. “The Relationship between Job Characteristics and Retirement Savings in Defined Contribution Plans During the 2007–2009 Recession.” Monthly Labor Review 136(5): 3–16.

Venti, Steven F., and David A. Wise. 1999. “Lifetime Earnings, Saving Choices, and Wealth at Retirement.” In Wealth, Work, and Health: Innovations in Measurement in the Social Sciences, edited by James P. Smith and Robert J. Willis, 87–120. Ann Arbor, MI: University of Michigan Press.

Warner, David F., Mark D. Hayward, and Melissa A. Hardy. 2010. “The Retirement Life Course in America at the Dawn of the Twenty-First Century.” Population Research and Policy Review 29(6): 893–919.