Anyone with detailed knowledge of the annual Census Bureau data on household income may be surprised by the new study Robert Pear describes in today’s Times. From the article:

In a grim sign of the enduring nature of the economic slump, household income declined more in the two years after the recession ended than it did during the recession itself, new research has found.

Between June 2009, when the recession officially ended, and June 2011, inflation-adjusted median household income fell 6.7 percent, to $49,909, according to a study by two former Census Bureau officials. During the recession — from December 2007 to June 2009 — household income fell 3.2 percent.

DAVID LEONHARDT

Thoughts on the economic scene.

The annual Census data, which is better known than the monthly data that forms the basis of the study, presents a somewhat different picture. It shows that inflation-adjusted median income fell 3.6 percent from 2007 to 2008, more sharply than in any year since. From 2008 to 2010 — a two-year period — income fell 2.9 percent.

What explains the difference between the two surveys?

For one thing, the monthly data (which goes through June of this year) is more current than the annual data (which ends in 2010) and reflects the economy’s continued weakness.

But there is also a second, more surprising reason for the difference: The annual data may be less reliable in some ways than the monthly data.

The authors of the study, Gordon W. Green Jr. and John F. Coder, point out that people are asked about their prior year’s income a few months into the next year. If the economy has weakened during the window between the end of the year and the survey date, some respondents may incorrectly describe their income from their previous year. Their answers will be affected by the economy’s deterioration, and they will tell the surveyor that they made less the previous year than they actually did.

A similar dynamic happens in surveys about health insurance. When people are asked about their insurance status from the previous year, they sometimes misunderstand and instead describe the coverage they have at the time of the survey.

You can see how this would affect the Census income survey, given that the economy was weakening so much in 2009. When asked about their 2008 income, some instead may have described their 2009 income.

In an e-mail to Mr. Pear, the two researchers offered more detail:

In household surveys, respondents are often asked to report events retrospectively for some given time period. When a respondent forgets the exact dates for a sequence of events, this often results in a known survey bias called “telescoping” in which the reporting of the events is telescoped either forward or backward.

The reference periods between the Census Bureau’s annual estimates of income and our monthly estimates of income are quite different. The Census Bureau’s survey is conducted in February, March and April of a given year and respondents are asked to report their income during the previous calendar year. The reference period for the monthly estimates is the 12-month period immediately prior to when the question is asked. Thus, the Census Bureau’s approach requires people to recollect events over a longer reference period.

We think that a telescoping problem may explain the discrepancy between the Census Bureau’s reporting of annual changes in household income between 2007 and 2008 and our monthly estimates. The Census Bureau reports a sharp drop (3.6%) in the annual amount of median household income over this time period whereas the monthly estimates are relatively flat.

Between January 2007 and the Spring of 2009 (when the Census Bureau collected its annual income estimates for 2008), the unemployment rate (seasonally adjusted) increased dramatically. The unemployment rate was in Jan. ’07 was 4.6%, in Jan. ’08 was 5.0%, in Dec ’08 (the end of calendar year reference period) was 7.3%, and continued to rise into the Spring of 2009 when the Census Bureau conducted its survey: Feb ’09 (8.2%), Mar. ’09 (8.6%), and Apr. ’09 (8.9%)

Now, if some of the respondents to the Census Bureau’s survey in the Spring of 2009 erroneously “telescoped” their current unemployment experience back to the previous year’s calendar income, this could have resulted in a calendar year 2008 estimate for income that was too low, hence producing the 3.6% decline in household income between 2007 and 2008 in the Census Bureau’s estimates. The monthly estimates do not have this telescoping problem to the same degree because the reference period is the 12 months immediately prior to interview and the data are collected in real time. This could explain why the Census Bureau estimates are different than the monthly estimates over this time period.

Article source: http://feeds.nytimes.com/click.phdo?i=4921e753c9558c8ddc3cc0c1d453c7ad