September 21, 2001

Counting the Uninsured: A Guide for the Perplexed

PDF of report

If you cannot access the files through the links, right-click on the underlined text, click "Save Link As," download to your directory, and open the document in Adobe Acrobat Reader.

On September 28, the Census Bureau will release new estimates of the number of Americans who lacked health insurance in 2000, based on data from the March 2001 Current Population Survey (CPS), a nationally representative survey of the U.S. population that is the main source of national data on poverty and insurance status. This year, the health insurance estimates will differ from those released in prior years because the Census Bureau has changed the way that it determines insurance status. The number of uninsured people will be lower than what was reported in earlier years, but most of the difference will be due to changes in survey methods, not because of actual changes in the number of people who lack health insurance.


The New Method Finds More with Insurance and Fewer Uninsured

Each March, the Census Bureau asks respondents in the Current Population Survey whether they had any health insurance coverage in the prior calendar year from an employer or union, from Medicaid, from Medicare or from other sources.(1) If a person answers 'No' to all of these questions, the Census Bureau has traditionally determined that the person is uninsured. Last year, the Census Bureau pilot-tested a new method of checking insurance status and determined it to be more accurate. Based on those findings, the new method has become the standard method this year.

The new approach identifies those who did not report having any type of health insurance and asks them a "verification" question. Basically, the survey asks, "It looks like you did not have any health insurance last year. Is that correct? If that is not right, what type of insurance did you have?" In the March 2000 pilot test, 7.7 percent of those who did not initially report having any health insurance changed answers when asked the verification question.(2) The estimate of the number of people uninsured in 1999 fell from 42.6 million with the old method to 39.3 million under the new method. This change in methodology yields a better estimate of the number of uninsured people. The Census Bureau acted appropriately in testing and accepting the new method.

The change may, however, cause some initial confusion. Last year at this time, the Census Bureau announced there were about 43 million uninsured people; this is the number with which the public is most familiar. Although the Census Bureau has not announced the new estimate from the March 2001 CPS (the estimate that will be released September 28), it will probably be in the range of 37 to 39 million uninsured. This does not mean the ranks of the uninsured fell by 4 to 6 million people between 1999 and 2000.

It will be important to make apples-to-apples comparisons and to compare the verified tallies of the uninsured in 1999 and 2000. If, for example, the new "verified" data show there are 38.0 million uninsured in 2000, then the change in the number of uninsured would be 39.3 million (based on the verified estimates for 1999) minus 38.0 million in 2000, for a reduction of 1.3 million. The Census Bureau also will need to determine if any such change is statistically significant or whether the apparent difference might have been due to sampling error.


State-level Estimates of the Uninsured

Because the number of Current Population Survey respondents in any given state is limited, the Census Bureau provides state-level estimates of the number and percentage of people who lack insurance by averaging numbers for two or three years. Because there are verified data for the number of uninsured people in 1999 and 2000, Census can readily compute a 1999-2000 two-year average for each state. To measure whether the number of uninsured is rising or falling in each state, the Census Bureau also has made adjustments to the 1998 data that lower the state-by-state estimates of the number of uninsured in 1998 "as if" the verification question had been asked. Census has used these adjusted numbers to compute an adjusted 1998-99 two-year average of the number of uninsured people in each state. The Census Bureau will assess whether the number of uninsured rose or fell in each state by comparing the 1999-2000 and 1998-1999 averages. These assessments will be included in the data to be released on September 28. It will be more difficult to measure changes from earlier periods because the verification questions were not asked in earlier years.(3)


Changes in Methods of Asking About Children's Use of Public Insurance

Another methodological change may affect the reporting of children's insurance coverage. For the first time, the Census Bureau explicitly asked families this March whether their children participated in the State Children's Health Insurance Program (SCHIP) and specifically asked this question using the local name for the program (e.g., Child Health Plus in New York or Healthy Families in California). In the past, the Census Bureau did not explicitly ask about SCHIP participation, although respondents could report it when asked an open-ended question about other insurance coverage. This change will lead to some increase in the number of children who are reported to have insurance through SCHIP or Medicaid and, therefore, some reduction in the reported number of uninsured children. A useful benchmark for comparison is that administrative data from the states indicate that the number of children covered by SCHIP grew by about 1.4 million between 1999 and 2000. (This does not necessarily mean the number of these children with insurance increased by the full 1.4 million, as some may have shifted from other forms of insurance to an SCHIP-funded program.)


Uninsurance Rates May Climb in 2001

The Census data being released on September 28 describe insurance status for calendar year 2000, which was one of the strongest economic periods in recent memory. The number of uninsured people may be on the rise again in 2001, however, because of the economic slowdown and recent increases in health insurance costs. For example, the national unemployment rate in August 2000 was 4.1 percent, close to the lowest rate in the past 30 years, but the rate rose to 4.9 percent by August 2001. In addition, recent data show that private health insurance premiums rose 11 percent this year, which makes it harder for employers and employees to afford insurance.(4) If the new Census data show that the number of uninsured people fell in 2000, policymakers should be pleased but cautious. Improvements in 2000 may prove fleeting.

Still More Data Are Coming

The Census Bureau will release another set of health insurance data a few months from now. In 2001, the number of people surveyed for the March CPS was augmented to increase the statistical accuracy of the CPS, particularly for measurements at the state level. The sample was increased to improve estimates of the number of uninsured and low-income children, which are used in determining state grants in the SCHIP program. The data the Census Bureau will release September 28 will be based on the traditional core sample of about 50,000 households. After reviewing the data from the expanded sample, which includes about 29,000 additional households, further data will be released this coming winter. The estimates from the larger survey will probably be similar to those released on September 28 for statistics at the national level, but state-level estimates may change to a greater degree because of the increase in sample sizes.

End notes:

1. Thus, the March 2001 CPS measured insurance status in 2000, while the March 2000 survey measured it for 1999.

2. Charles Nelson and Robert Mills, "The March CPS Health Insurance Question and Its Effect on Estimates of the Uninsured," U.S. Bureau of the Census, August 2001. Available at

3. The Census Bureau will continue to release data files that include estimates of the number of uninsured for 1999 and 2000 using the old method. One way to measure longer time trends is to compare estimates from an earlier year to the estimates for 2000 as determined under the old method. Since the measures would be comparable, this should provide a relatively accurate estimate of the change in insurance coverage over the years.

4. Kaiser Family Foundation-Health Research and Educational Trust, "Employer Health Benefits 2001: Summary of Findings," September 2001.