September 20, 2012

Revising the Past, Using 2010 Census Data

Each decade’s census not only provides fresh statistics on the U.S. population, but also is the basis for updating a broad range of previously published federal demographic estimates and survey benchmarks. Though necessary, the changes can produce hiccups in trend data. One example was in last week’s Census Bureau release of income, poverty and health insurance statistics for 2011.

In its report, the Census Bureau not only published new income, poverty and health insurance statistics for 2011, but also updated previously published 2010 data based on new information from the 2010 Census. The changes were relatively small, but data users need to account for them in order not to be analyzing outdated estimates.

The data on median household income, poverty rates and health insurance coverage come from the Current Population Survey (CPS) Annual Social and Economic Supplement (ASEC). The ASEC collects information each March from 100,000 households about finances and insurance for the previous year, so the March 2012 supplement collected data for 2011.

In surveys such as the CPS, responses from a sample of Americans are generalized to the entire population. To do that, individual survey respondents are weighted so the CPS totals match official population shares by sex, age, race and Hispanic origin. Once a decade, after each census, those sample weights are revised so the reported CPS survey demographics are in sync with new information about the composition of the nation’s population. For the rest of the decade, the weights are based on the Census Bureau’s annual population estimates that are benchmarked to the previous census and updated each year using government records on births, deaths, migration and immigration.

As the report’s authors explained, the newly published 2011 income, poverty and health insurance statistics were benchmarked to the 2010 Census, but the 2010 data published a year earlier used “population controls” based on estimates updated from the 2000 Census. (The process of benchmarking data to each new census is quite complex and lengthy.) In order to have an accurate comparison of year-to-year change, bureau analysts recalculated the 2010 statistics so that they too were benchmarked to the newest census. This ensures that any measured change from 2010 to 2011 is real, and not a byproduct of revised population weights.

Differing Impacts, Mainly Small

Most changes to 2010 income, poverty and health insurance statistics resulting from the updates were small, as the authors noted [see page 77]. Median household income for 2010, originally reported as $49,445, was slightly lower ($49,276) once the figures were updated with 2010 Census benchmarks. This 0.3% change was statistically significant, according to the Census Bureau. In general, the revised 2010 income estimates were lower than the original numbers published last year, although the declines were less than 1%, according to the report’s authors.

When the figures were updated with 2010 Census data, median household income was slightly lower in 2010 in all regions and for most households headed by someone younger than 65, for most types of family and non-family households, and for households headed by a U.S.-born householder. Median household income was higher for older householders and foreign-born householders with the revised numbers.

Poverty rates were affected “only minimally” once the 2010 Census population benchmarks were applied to the 2010 rates. The nation’s overall poverty rate for 2010, reported last year as 15.11%, was barely higher, 15.14%. The number in poverty, originally 46.18 million, was 46.34 million.

The statistics on Americans who do not have health insurance showed a more varied pattern, though any changes were relatively small. The overall number and share of those uninsured in 2010 did not show a statistically significant difference once the 2010 Census benchmarks were used. The revised uninsured rate for 2010 was higher for households headed by people from some younger age groups and for households headed by Asians when 2010 Census benchmarks were factored in. It was lower for households headed by non-Hispanic whites, and for households in the Midwest and South.

Effect on Other Data

The revised population controls from the 2010 Census also are being implemented in other government (and non-government) surveys, and throughout other government datasets.  But when it comes to adjusting previous years of data to account for new information from the census, practice varies.

The Census Bureau itself produces a set of “intercensal population estimates” by age, sex, race and Hispanic origin for the years between 2000 and 2010 that account for results of both censuses and smooth any discrepancies over the entire time period to avoid a big discontinuity in the data. When compared with the bureau’s previously released annual population estimates (called “postcensal estimates”), there was only a small difference for the national population, but larger differences in some locations and population sub-groups.

Updated population controls were incorporated into the monthly CPS labor force survey beginning with data released for January 2012. The Bureau of Labor Statistics, which uses the CPS to produce unemployment estimates and other labor-force statistics, has issued guidance to users about updated population controls, and produced a research series of estimates that smooth the discontinuities due to the new 2010 data.

The Census Bureau’s American Community Survey adopted new 2010 population controls beginning with 2010 data released last year. As the bureau’s analysis shows, the revisions had a modest impact (the change was less than 1%) on the national and state totals, but had larger effects on the size of some groups, such as men ages 25 to 34 and adults ages 75 and older.

Data from earlier years of the ACS were not revised to account for new information from the 2010 Census, although the Census Bureau alerted users to be cautious about comparing 2009 and 2010 ACS data. If earlier years of the ACS had been revised, according to a Pew Hispanic Center analysis, real growth in the foreign-born population from 2009 to 2010 would have been “substantially smaller” than reported in the ACS.

The 2009 ACS reported estimate of foreign-born Americans was 38.5 million, but would have been 39.3 million if new population assumptions were applied, the analysis said. The 2010 ACS foreign-born estimate was 39.9 million. Thus, the reported increase of 1.5 million would have been only about 616,000 had the 2009 figures been revised, according to the Pew Hispanic analysis.

The updated census information has a major impact at the National Center for Health Statistics, where population totals are the denominator for birth rates. When the center published final birth data for 2010 in August, it revised previously published birth and fertility rates for 2001-2009 in light of new information from the 2010 Census.

The center sponsored a session at a conference this summer on the impact of the 2010 Census on trends and variations in health and population statistics (slides can be found on this web page.).

As Brady E. Hamilton’s presentation showed, using revised population controls based on the 2010 Census resulted in varying changes by race and ethnicity in birth and fertility rates. A presentation by Gretchen Livingston and Jeffrey Passel of the Pew Hispanic Center added depth to the center’s past work on how the 2009 ACS probably understated the size of the foreign-born population, and included details on the impact on subgroup birth and fertility rates.