April 5, 2007
CENSUS REVISES ESTIMATES OF THE NUMBER OF UNINSURED PEOPLE
By Leighton Ku
On March 23, the Census Bureau announced that it has revised estimates of the number of uninsured people for 2004 and 2005, after it detected an error in how data from its Current Population Survey have been tabulated. The number and percentage of people who were uninsured in 2005 changed from 46.6 million (15.9 percent of the population) to 44.8 million (15.3 percent).
The Bureau also stated that the same error occurred in previous years and “the effect of the revision is relatively constant from one year to the next,” so that there is little impact on trends in insurance coverage. For example, the revised estimates show that the number of uninsured grew by 1.3 million from 43.5 million in 2004 to 44.8 million in 2005. The earlier, unrevised estimates also showed an increase of 1.3 million (from 45.3 million uninsured in 2004 to 46.6 million in 2005). While the estimates for each year changed, the trend across years was the same.
What Changed in the Census Data?
Every March, the Current Population Survey asks about health insurance coverage in the previous year. The technical problem the Census Bureau found is that sometimes its computer programs failed to account properly for situations in which a respondent said that all household members had the same type of private health insurance plan (whether employer-sponsored or individually purchased) as the policy holder. In such cases, the Census Bureau’s computer programs sometimes correctly counted dependents as having private coverage but sometimes incorrectly counted them as being uninsured or having another kind of insurance. This led to a slight underestimate of the number of people with private coverage and a slight overestimate of the number of uninsured. The Census Bureau has now fixed this problem and correctly assigned insurance coverage for these individuals.
It is worth remembering that although the Current Population Survey asks whether a person lacked health insurance over all of the prior year, comparisons with other surveys indicate that its estimates correspond more closely with the number of people uninsured at a given point in time, rather than with the number who are uninsured for a full year.
The Census Bureau plans later this year to revise its estimates for years from 1995 through 2003, but has suggested that the general trends will remain about the same as before. Thus, earlier Census findings about substantial growth in the number of uninsured Americans from 2000 to 2005 are likely to be essentially unchanged.
The new Census findings represent appropriate technical revisions. People using these data, however, need to be careful not to mix revised and unrevised estimates. For example, the Galen Institute recently published estimates of the number of uninsured from 2001 through 2005 (which it attributed to former White House health policy adviser Doug Badger) which inappropriately mix the unrevised estimates for 2001 to 2003 and the revised estimates for 2004 to 2005. Mixing revised and unrevised data could lead to the misleading and incorrect interpretation that the number of uninsured fell from 2003 to 2004. Until the Census Bureau releases revised estimates for earlier years, the multi-year time trends it reported in August 2006 remain the best estimates to use of the direction of health insurance coverage in the United States in recent years.
 Census Bureau press release, “Census Bureau Revises 2004 and 2005 Health Insurance Coverage Estimates,” March 23, 2007. The press release, a user note and several tables of selected estimates for 2004 and 2005, including new state-by-state estimates of the uninsured, can be found at www.census.gov/hhes/www/hlthins/usernote/schedule.html.
 Galen Institute, Health Policy Matters, March 30, 2007.
 Carmen DeNavas-Walt, et al. Income, Poverty, and Health Insurance Coverage in the United States: 2005, Report P60-231, Census Bureau, August 2006. As it reports, the Census Bureau has occasionally made other methodological changes from one year to another that make it difficult to compare trends over many years accurately.