C. Adam Bee and Joshua Mitchell, two economists at the Census Bureau, recently released an analysis of retirement income that qualitatively changes our understanding of the well-being of retirees. The analysis matched administrative data (essentially tax filings) with the reported income in the Current Population Survey (CPS), which has been the standard basis for the measurement of household income, including retirement income.

Bee and Mitchell found that the income of households in the administrative data was substantially higher than what was reported in the CPS. The overall median for households over age 65 in the administrative data was $44,371 in 2012 (the year that was the basis of their analysis), 30.4 percent higher than the $34,037 reported in the CPS for the same households. Their analysis found sharply higher incomes at all points along the income distribution than what was reported in the CPS. They also show a corresponding reduction in poverty rates among older households.

This is good and important news. While there is much room for additional analysis based on the Bee and Mitchell findings, there are two points that jump out. First, defined benefit (DB) pensions have been more effective in supporting retirement incomes than we had realized. This is good news. The second point is not good. There is nothing in the Bee and Mitchell analysis that suggests we had been overly pessimistic about the extent to which defined contribution (DC) pensions will provide adequate income to future retirees.

On the first point, by far the largest single source of the gap between the income as measured in the administrative data and income as measured in the CPS is uncounted income from DB pensions. This is due to both the fact that many people who receive a DB benefit do not report it on the CPS and also that many people who do receive a DB benefit under-report the amount.[1] Bee and Mitchell find that DB pensions make a large contribution to the income of older households for the 3rd decile and above in the income distribution. Their results are shown in the table below.



Administrative Data


CPS Data


Income Decile































































































Source: Bee and Mitchell 2017, Table 9.

Their analysis shows that in 2012, on average the third decile of households got 11 percent of their income, or $2073, from a DB pension.[2] For the fourth decile the average is 18 percent for an annual payment of $4,531, for the fifth decile the share is 22 percent which comes to $7,151 and for the sixth decile the share was 27 percent for an annual pension income of $11,291. (Since these numbers are in 2012 dollars, add roughly 10 percent to put them in 2017 dollars.) These data indicate that defined benefit pensions have provided a substantial part of the retirement income, not just for the upper middle class, but for retirees who are at the middle- and even lower-middle-end of the income distribution.

By comparison, the CPS put all annual retirement income (both DB and DC) at $923 for the third decile, $1960 for the fourth decile, $4012 for the fifth decile, and $6987 for the sixth decile. It is worth noting that even these numbers likely understate actual income from DB plans since some participants cash out their pension and take a lump sum. The income from this would then show up as DC income (if placed in an IRA) or as interest earnings.

On the other side, there is nothing in the Bee and Mitchell analysis to suggest that DC pensions will provide substantially higher benefits that would be expected based on asset holdings reported in the Survey of Consumer Finance (SCF) and other sources. Their analysis shows DC accounts adding $377, $755, $1950, and $2927 to the retirement incomes of the third, fourth, fifth, and sixth deciles of the income distribution. These numbers are low in part because it includes many older households who likely would never have had DC pensions during their working lifetime, but the point is that the upward revision of income is not due to the finding of large amounts of previously uncounted DC income.

By comparison, the SCF showed average net wealth outside of primary residences for the middle quintile (the fifth and sixth deciles of the income distribution) for near retirees (ages 55–64) at $136,100, $100,600, and $89,300 in 2007, 2010, and 2013, respectively (all numbers are in 2013 dollars).[3] This net wealth figure would include all assets in DC retirement accounts, as well as any non-retirement assets held by these households. If we assume a roughly 6 percent annual payout can be sustained over a retirement expected to range between 20–25 years, this implies a range of annual income from $5,358 based on the 2013 figure to $8,166 based on the 2007 figure.

Even the 2007 figure is considerably less than DB pension income found for the middle quintile of households by Bee and Mitchell. Furthermore, older households are much less likely now than in prior decades to have paid off the mortgages on their houses, if they are homeowners, which is also less likely than in prior decades. In 2013, the SCF showed that the middle quintile of households in this 55 to 64 age category had an equity stake equal to 54.6 percent of the value of their home. This compares to a stake of 81.0 percent in 1989. The net wealth for the middle quintile in the 2013 SCF was roughly equal to the outstanding debt on their mortgage, meaning that if they used to assets to fully pay off their mortgage they would have no retirement income.

Unfortunately, the Bee and Mitchell analysis provides us no reason to think this assessment is wrong. It does show that other forms of income are also somewhat undercounted in the CPS, which means that retirees may have somewhat more income from Social Security, Supplemental Security, and other sources than the CPS has led us to believe. However, their analysis gives us little reason to believe that retirement income from 401(k)s will be larger than survey data suggests, which means that the gap with current DB income is even larger than we had previously believed.  

[1] A redesign of the survey in 2014 likely reduced under-reporting, but the more recent administrative data is not yet available, so Bee and Mitchell have not been able to assess the extent to which the revised CPS may still be missing retirement income.

[2] These numbers are taken from Table 9 in their analysis.

[3] These data are taken from David Rosnick and Dean Baker, 2014, β€œThe Wealth of Households,” Center for Economic and Policy Research, available at http://cepr.net/documents/wealth-scf-2014-10.pdf.


Link corrected, thanks AlaninAz.

Thanks to Harlan for pointing out the duplication, not sure what happened.