Income data from the Census may not tell full story on middle-class trends

The Census Bureau published its annual report on household income and poverty on Wednesday. The report uses the latest data from the Current Population Survey (CPS), a nationally representative survey that has tracked household incomes since shortly after World War II. This year’s report, which covers incomes received in 2017, contains plenty of good news. Median income increased by 1.8 percent in 2017, after accounting for inflation. The Gini index and other measures of income inequality did not show a meaningful change this year. The official poverty rate went down by 0.4 percentage points, marking the third consecutive year of declining poverty.

The Census household income report always attracts wide coverage in the press. Indeed, it made the front page of The Wall Street Journal. Many reporters and analysts will pore over the new data and draw conclusions about trends in the economic well-being of Americans. The resulting news stories deserve our attention, but it is important to keep a vital question in mind: Does the CPS give us an accurate picture of household incomes?

In many recent years, the answer has been “No.” Compared to the national income and product accounts (NIPA) produced by the Bureau of Economic Analysis (BEA), the CPS often gives us a strikingly different picture of the recent trend in household income. For example, we saw large discrepancies in real per capita income growth in both 2014 and 2015.

Yearly discrepancies can add up over a business cycle. Chart 1 below shows trends in real per capita income estimated from the NIPA statistics and in the Census Bureau’s CPS in two business-cycle recoveries. The first of these recoveries followed the 2001 recession; the more recent one followed the Great Recession. According to the NIPA estimates, real per capita money income increased a total of 7.5% during the 2002-2007 recovery. It increased 14.7% in the 2010-2017 recovery. In contrast, real per capita income reported in the CPS increased just 2.0% in the earlier recovery and 10.7% in the recovery following the Great Recession. (In calculating real income growth, we adjust nominal incomes using the same measure of price inflation used by the Census Bureau.) Compared with the real income growth seen in the national income and product accounts, per capita income growth was substantially slower as reported by the families interviewed in the CPS.

In contrast to the period between 2002-2014, in the last three years the income gains in the CPS have equaled or exceeded those seen in the national income accounts. In fact, in 2017 the two data sources show an identical percentage gain in average real income (1.7%).

Is this the start of a new era of economic data harmony? Probably not, but it wouldn’t be unprecedented. As shown in an earlier analysis, the gap between the CPS and BEA data hasn’t always existed. From 1973 to 2003, the two datasets showed remarkably similar income trends. Since then, they have diverged significantly (see Chart 2).

Why the gap after 2001? Bruce Meyer of the University of Chicago and his colleagues show that data from nationally representative household surveys, such as the CPS, have deteriorated in quality over time. Americans are less likely to respond to surveys. Among those who do agree to be interviewed, a smaller fraction gives complete or accurate answers. For some income items, we can compare respondents’ answers to income reports in administrative records, such as the W-2 files. Survey responses to questions about public assistance, government transfers other than Social Security, self-employment earnings, interest, and dividends are especially problematic.

While the CPS interviews give us valuable information about differences in the economic well-being of Americans, it is important to check the validity of the responses against other, usually more accurate sources, such as tax return data and Social Security records. When judging recent trends in overall well-being, it makes sense to verify that the basic trends revealed in the household survey correspond with those we can also see in an independent and more reliable data source.