The problem with the figures, it said, "is not subtle." 

Also read: Drop in families with retirement plans

According to the Census data, defined contribution income in all retired households stood at $18 billion for U.S. households while defined benefit income totaled $394 billion.

By comparison, the Federal Reserve data – considered the gold standard for data on household wealth – put the DC total at $220 billion and DB total at $535 billion, while IRS data came up with $229 billion for DC and $623 billion for DB. 

So did people simply not report available retirement income for the Census survey? In a way, yes.

The Center for Retirement Research brief laid at least part of the blame on the framing of the survey questions, which it said were phrased in such a way as to exclude "as-needed withdrawals from retirement accounts," and also limited respondents to reporting only two sources of retirement income. 

The Federal Reserve, on the other hand, did not limit responses to include only "regular" retirement income, and also "did not take a bundled approach … but rather asks about each source of retirement income separately."

When the center broke down and reviewed the data across five segments of income, it found that the biggest discrepancies between Census and Fed figures occurred in the top two quintiles – althought it found discrepencies elsewhere, too.

Looking at average annual 401(k)/IRA income for retired households age 65-84, Census found that bottom quintile figure of $18 billion differed from the Federal Reserve figure of $215 billion.

Census had a second-quintile figure of $71 billion compared to the Federal Reserve's $1.3 trillion.
In the third quintile, Census came up with $313 billion, compared to the Fed's $2.8 trillion.

The Census survey's fourth-quintile figure was $623 billion vs. $5.3 trillion for the Fed, while the fifth's came in at $1.8 trillion for Census vs. $16.1 trillion for the Fed. 

The center also took issue with another aspect of the Census survey.

"The general rule-of-thumb for describing the income or wealth of a typical respondent from a population sample is to report the median. Since populations are rarely symmetrically distributed, the mean value can easily be skewed by the holdings of those at the top of the income distribution," it said. 

On the other hand, "when looking at the distribution of various components of income or wealth by quintile, the likelihood of a few extreme values distorting the results is substantially reduced. Thus, one can select either the mean or the median within the quintile.

Also read: Hindsight is 20/20 in retirement savings

That choice produces very different results, because income components and wealth are asymmetrically distributed within income brackets as well as across income brackets." 

And indeed, the center found that median holdings for the bottom three quintiles were zero, while the fourth quintile was $14,000 and the fifth (highest) quintile was $120,000.

Looking at the mean instead, the first quintile had a mean of $206; the second, $11,376; the third, $17,392; the fourth, $69,225; and the fifth, $400,647.

The center's brief also said that the importance of Social Security as a source of retirement income is "significantly less than previously thought" for the top two quintiles, while it's the main source of retirement income for the three lowest quintiles. 

Moreover, the brief also said that because of younger retirees' dependence on DC income rather than DB income, the problem of retirees in middle-income households coming up short on retirement income will only grow. 

The good news, the center said, is that the government is considering changing some questions and adding others to make responses to its population survey more accurate.

Also read: Think tank questions feasibility of Hatch plan