Public Comments | Retirement

EPI comment regarding proposed information collection on ACS Methods Panel Tests

Download PDF

U.S. Census Bureau, Department of Commerce
Response to Notice of Proposed Information Collection on American Community Survey Methods Panel Tests
Comment submitted via Federal e-Rulemaking Portal (, Docket number USBC– 2017–0006.

To whom it may concern:

I appreciate the opportunity to respond to your request for public comments regarding proposed information collection on American Community Survey (ACS) Methods Panel Tests. The Economic Policy Institute (EPI) is a nonprofit, nonpartisan think tank created in 1986 to include the needs of low- and middle-income workers in economic policy discussions. EPI and organizations affiliated with EPI through the Economic Analysis and Research Network (EARN) of state and local organizations rely on ACS data for research on a range of topics related to the financial wellbeing of American workers and their families.

The ACS Methods Panel has been tasked with increasing survey efficiencies and improving data quality while reducing respondent burden and improving response rates. With these ends in mind, I encourage the panel to study ways to improve the quality of data on retirement income and coverage. I also encourage the panel to consider adding questions on other topics, such as contingent work, that would allow researchers to explore new dimensions in labor economics and related fields.

Census study finds significant underreporting of retirement income. In a widely-discussed working paper, Census researchers Adam Bee and Joshua Mitchell (2017) documented that Census household surveys significantly underreport retirement income, including income from 401(k)-style defined contribution (DC) plans, Individual Retirement Accounts (IRAs), and traditional defined benefit (DB) pensions.1 Comparing survey data to data from tax and Social Security records, Bee and Mitchell found that the median income of seniors was 30 percent higher based on administrative records.

Though the Bee and Mitchell study focused on the Current Population Survey’s Annual Social and Economic supplement (CPS ASEC), the authors noted that the ACS and the Survey of Income and Program Participation (SIPP) also underreport retirement income. Bee and Mitchell found that the underreporting was mostly due to respondents not reporting certain types of income at all rather than reporting low positive values. However, another Census study noted that some ACS respondents mistakenly reported monthly instead of annual income, so the problem can occur at both the extensive and intensive margins (O’Hara and Bee 2016).2

The underlying reasons for this underreporting remain poorly understood, though it likely stems from confusion or fatigue. Respondents have no incentive to minimize income in Census surveys. Therefore, Census may be able to mitigate the problem by rephrasing, reordering, or streamlining questions, and perhaps supplementing survey data with tax and Social Security records in ways Census has explored for the SIPP.

The shift from DB to DC plans has muddied definitions of “employer benefit,” “employer pension,” and “retirement income.” Private-sector employers’ shift from DB to DC retirement plans has complicated efforts to measure income at different stages of the lifecycle. Traditional DB pensions in the private sector are funded in advance by employers and provide income to retired or disabled workers and their spouses for the rest of their lives. Thus, these pensions are easily understood as a type of employer benefit providing income to mostly older households. Though “pension entitlements” can also be treated as financial flows to working-age households (which is how they appear in national income accounts),3 Census surveys consistently treat DB pension benefits as retirement income, avoiding double counting. In contrast, some DC and IRA funds are counted twice in survey income measures—first as earned income and later as retirement income–while some distributions from these accounts are never counted at all, as will be explained below.

Census surveys refer to both DB and DC plans as “employer pensions,” even though the latter shift much of the responsibility for saving and investing onto workers and are rarely annuitized like traditional DB pensions. Substituting the phrase “employer-sponsored retirement plan” could better describe DC and hybrid cash balance plans and would be broad enough to encompass traditional DB pensions as well. Listing the most common types of plans in instructions provided to respondents and interviewers would minimize confusion. It is also important to distinguish between employer-sponsored plans and IRAs, especially payroll deduction IRAs being implemented in California and other states. In these state-sponsored initiatives, employers facilitate payments to a “Secure Choice” IRA but do not sponsor the plan.

In contrast to the fairly consistent treatment of DB pension benefits, Census survey questions about DC plan distributions are inconsistent and poorly worded, increasing the potential for confusion. This ambiguity extends to IRAs, which are mostly composed of funds rolled over from employer plans (though this could change if the state-sponsored IRAs are successful).

Employee contributions to retirement accounts may be counted twice as income. Most DC plans are funded through a combination of employee contributions, employer matching contributions, and investment income. Employee contributions to DC plans and IRAs are not excluded from measures of income received by working-age households in Census surveys, even if these funds are later included in measures of income received by retiree households. This leads to double counting employee contributions over the life cycle, first as earned income and later as retirement income. Census should therefore collect data on employee contributions to retirement plans and perhaps exclude these from summary income measures, or at least provide researchers with the information needed to do this themselves.

Meanwhile, other retirement account funds are never included in income measures. While some income is counted twice, lump sum distributions from retirement accounts are omitted from some measures of retirement income. As a result, employer contributions and investment returns accruing to these accounts may never be included in survey income measures.

Census surveys are inconsistent in their treatment of lump sum, as opposed to “regular,” distributions:

  • Instructions accompanying the ACS questionnaire specify that respondents should include “regular income from annuities and IRA and KEOGH retirement plans,” suggesting that lump-sum payments should be omitted.4
  • Before 2014, CPS ASEC asked about “regular payments from IRA, KEOGH, or 401(k) accounts.”5 The wording was changed in 2014, as will be discussed below.
  • SIPP collects information on both “regular” and “lump sum” distributions, but does not include the latter in summary income measures.6

The distinction between “regular” and “lump sum” distributions is based on a definition of income that focuses on regularity of payments rather than on accounting for financial flows between actors while ignoring internal transactions, such as withdrawals from savings. The decision to differentiate between “regular” distributions and lump sums and include only the former in some measures of income has made it difficult to assess the effects of the shift from traditional pensions to savings plans on retiree living standards.

Poorly-worded survey questions exacerbate the problem. In the ACS, the question about sources of retirement income aside from Social Security refers to “retirement, survivor, or disability pensions.” The accompanying guide elaborates: “Include retirement, survivor or disability benefits received from companies and unions, federal, state, and local governments, and the U.S. military. Include regular income from annuities and IRA or KEOGH retirement plans.” The most common type of retirement plan, a 401(k) plan, is never mentioned by name, even though some respondents may not consider 401(k) plans to be a “pensions” or “benefits received from companies.”7 The guide also specifies “regular income from annuities and IRA or KEOGH plans” without clarifying whether to exclude lump sum distributions from other (employer) plans.

Census took steps to alleviate the problem with the CPS ASEC redesign. In response to concerns about measures of unearned income, Census reworded, reordered and added questions to the CPS ASEC in 2014. Among other things, Census separated questions about pensions from those about retirement accounts. Census also dropped questions about “regular payments” from retirement accounts in favor of asking about any withdrawals or distributions from “accounts such as a 401(k), 403(b), IRA, or other account designed specifically for retirement savings.” Respondents can now choose to provide amounts withdrawn or distributed monthly, quarterly, every six months, or annually. They are also asked if any of the money withdrawn was rolled over or reinvested.8

The CPS ASEC redesign led to a modest increase in retiree income. Based on a split-sample test, the CPS ASEC redesign more than quadrupled the measure of IRA, Keogh, and 401(k) income. However, because the amounts were still small, these and other changes to the survey resulted in a fairly modest (4.6-4.7 percent) increase in median income for householders 65 and older, according to Census and Mathematica researchers.9

This suggests that the CPS ASEC redesign, though a step in the right direction, was only partly successful in capturing underreported income documented by Mitchell and Bee. 10 One explanation may be that imputed data used in the split-sample tests did not make use of some of the new questions and methods being tested. Therefore, the redesigned CPS ASEC may do a better job of capturing retirement income once the new questions and methods are fully incorporated.11

Surveys also miss income from traditional DB pensions. While Census surveys’ failure to capture lump sum distributions from account-style plans was widely recognized, an unexpected finding of Bee and Mitchell’s study was that respondents also failed to report significant income from traditional DB pensions. There should be little ambiguity about whether DB pensions constitute “income,” except to the extent that people are taking lump sum distributions from these plans. Though Bee and Mitchell do not explore this angle, it is possible that some, if not most, of unreported DB income takes the form of lump sum distributions.

Bee and Mitchell note that lump sum distributions are not the norm for employer-sponsored plans—at least among seniors (younger recipients and IRA recipients do tend to take lump sums). Tax records analyzed by Bee and Mitchell show that 96 percent of seniors who received income from an employer-sponsored plan in 2012 also received such income in 2013, with 81 percent receiving an amount in 2013 that was within 10 percent of what they received the previous year (Bee and Mitchell, Appendix Table 10). Bee and Mitchell also estimated, based on tax records, that 70 percent of retirement income received by seniors in 2012 came from DB pensions, with most of the remainder coming from IRAs rather than directly from DC plans (Bee and Mitchell, Table 9).

The fact that the largest source of senior income after Social Security is DB pensions and the fact that most seniors receive regular payments from employer plans do not rule out the possibility that much unreported income, even from DB pensions, takes the form of lump sum distributions. In fact, Bee and Mitchell note that IRA distributions are both more likely to go unreported and more likely to vary significantly from year to year, suggesting that lump sum distributions from other sources may also be missing from survey data.

A study by the Employee Benefit Research Institute, using plan data provided by Aon Hewitt, found that a third of participants age 55-69 who received payments from defined benefit plans, including cash balance plans, opted not to receive annuitized benefits. The annuitization rate was even lower for participants aged 70-75.12 Because the focus of the study was on participant choice, the author excluded participants with short tenures and very small balances who might have been forced to cash out. Therefore, if the plans in the study are representative of all DB plans, the overall share of participants who cash out may even be higher than a third.

Bee and Mitchell found that 44 percent of seniors in the CPS ASEC reported employer retirement plan income in 2012, with recipients reporting an average of $20,138 (Bee and Mitchell, Table 8). Based on tax data, however, Bee and Mitchell found that 68 percent of seniors received such income, with the missing amounts averaging around $27,000.13 It seems likely that these are mostly lump sum distributions, since it is difficult to imagine why seniors with larger-than-average annual benefits would be more likely to omit this income in a survey.

Clearly, more work needs to be done to assess problems survey respondents have with questions about retirement plans and retirement income and to identify ways to minimize these problems. It seems likely that the problems identified by Bee and Mitchell are not unique to the CPS ASEC and were not fully addressed by changes to that survey. I therefore urge Census to take steps to explore ways to improve information gathering on retirement plans and retirement income across all surveys, including the ACS.

In particular:

  • Census should use easier-to-understand and consistent language when asking respondents about retirement plan participation and income, including listing the most common types of plans.
  • Census should eliminate arbitrary distinctions between lump sum and regular distributions from retirement plans in all surveys and make clear that respondents should report irregular or lump-sum payments from these plans in measures of retirement income.
  • Census should ask survey respondents about retirement plan contributions in order to allow researchers, if they wish, to calculate earnings net of these contributions and avoid counting the same income at two stages of the lifecycle. Tracking employee contributions (and, ideally, employer contributions) would also be useful for assessing the use of tax-qualified plans.

I encourage Census researchers to compare notes on retirement income measures with counterparts working on Federal Reserve’s Survey of Consumer Finances and at the University of Michigan’s Health and Retirement Study, as well as researchers at other government and academic research centers. I also encourage Census to continue to look for ways to use administrative data to supplement or benchmark survey data.

Finally, if Census is considering adding questions to the ACS or other surveys, my colleagues and I at the Economic Policy Institute are very interested in the following topic areas:

  • Retirement plan participation, including plan type
  • Hourly wages
  • Contingent work
  • Union membership
  • Employer size
  • Sexual orientation and gender identity
  • Prior incarceration or felony conviction

We would be happy to provide more specifics if there is interest in exploring these additional topic areas.

Monique Morrissey
Economic Policy Institute


1. “Do Older Americans Have More Income Than We Think?” SESHD Working Paper #2017-39, July 2017.

2. Amy O’Hara and Adam Bee, “Preliminary Research for Replacing or Supplementing the Income Question on the American Community Survey with Administrative Records,” March 1, 2016 memo for the ACS Research and Evaluation Advisory Group.

3. Irina Stefanescu and Ivan Vidangos, “Introducing Actuarial Liabilities and Funding Status of Defined Benefit Pensions in the U.S. Financial Accounts,” Board of Governors of the Federal Reserve System, FEDS Notes, October 31, 2014.

4. U.S. Census Bureau, “Your Guide for the American Community Survey,” (2017 edition).

5. U.S. Census Bureau, “ASEC Technical Documentation 2013.”

6. Social Security Administration researchers found that including lump sum distributions would increase measures of mean income for families headed by persons 65 or older with such distributions by 15 percent. However, since only 19 percent of 65-or-older families received lump sum distributions, mean income for families in this age group increased by less than 3 percent after including lump sums (Howard M. Iams and Patrick J. Purcell, “The Impact of Retirement Account Distributions on Measures of Family Income,” Social Security Bulletin, Vol. 73, No. 2, 2013).

7. It is worth noting that in the post-redesign CPS ASEC, questions about pension income from a previous employer or union explicitly exclude “distributions or withdrawals from IRAs, 401(k)s, or similar accounts.” Questions about such distributions are in a separate section (U.S. Census Bureau, “CPS ASEC 2017 Guide and Questionnaire”).

8. U.S. Census Bureau, “CPS ASEC 2017 Guide and Questionnaire.”

9. John L. Czajka and Randy Rosso, “Redesign of the Income Questions in the Current Population Survey Annual Social and Economic Supplement: Further Analysis of the 2014 Split-Sample Test,” Mathematica Policy Research Report, September 27, 2015; Jessica L. Semega and Edward Welniak, Jr., “The Effects of the Changes to the Current Population Survey Annual Social and Economic Supplement on Estimates of Income.” Proceedings of the 2015 Allied Social Science Association Research Conference, 2015.

10. Though the Bee and Mitchell working paper was released in 2017, it examines income measures before the CPS ASEC redesign because post-redesign administrative data was not yet available.

11. Czaika and Rosso (2015).

12. Sudipto Banerjee, “Annuity and Lump-Sum Decisions in Defined Benefit Plans: The Role of Plan Rules,” EBRI Issue Brief No. 381, January 2013.

13. The comment submitted to the panel mistakenly estimated that the missing amounts averaged around $40,000. This is the corrected version.

See related work on Retirement

See more work by Monique Morrissey