NHIS

Using and Comparing Data on Health Insurance from Multiple Sources

Why does the Government collect health insurance data in multiple surveys?

Multiple surveys are needed to address the full range of data needs, and each of these surveys addresses multiple topics and have different strengths. The Census Bureau’s Current Population Survey (CPS), primarily a survey related to employment, is large enough to yield State data and relates health insurance coverage to employment and other benefits. The HHS National Health Interview Survey (NHIS), the broadest survey of health, is useful in analysis of relationships between coverage and access to care and to health status and is large enough to examine these factors in small population subgroups and selected States. The HHS Medical Expenditure Panel Survey (MEPS) provides the depth we need to look at expenditures and utilization. Because health insurance is so important, it is collected on each of these and other surveys so that it can be related to other content or detail that is unique to those surveys.

Additional information on these three surveys (including their design, content and other results) can be found at the following sites:

  1. Current Population Survey
    external icon
    Conducted by the Bureau of the Census for the Bureau of Labor Statistics
  2. National Health Interview Survey
    Conducted by the National Center for Health Statistics
    Centers for Disease Control and Prevention
    Department of Health and Human Services
  3. Medical Expenditure Panel Survey
    external icon
    Conducted by the Center for Cost and Financing Studies
    Agency for Healthcare Research and Quality
    Department of Health and Human Services

Why do different surveys come up with different numbers of people without health insurance coverage?

The number of uninsured often varies from survey to survey. Because each survey is designed to address different topics and issues, there are differences in the methods used. Some of these differences are in the statistical design of the survey and the way in which the data are processed and analyzed. Other differences are in the basic concepts themselves – the definition of insurance, the reference period (the period of time the individual is asked about), and the length of time an individual would need to be without insurance coverage to be counted as “uninsured.”

These differences in design – some subtle, some more fundamental – all contribute to different estimates of health insurance coverage. A more detailed discussion can be found in a paper prepared by the HHS Office of Health Policyexternal icon.

For 2000, the three major surveys yielded the following estimates of the uninsured:

Percent uninsured, 2000 When persons were surveyed Measure
CPS 14.0 March 2001 Uninsured for the full year 2000
NHIS 14.7 Throughout the year 2000 Uninsured at the time of the interview
MEPS 16.1 Early 2000 Uninsured throughout the first half of 2000

Why do different surveys show different trends?

The three major government surveys diverge somewhat in their findings of what is going on with health insurance coverage between 1999 and 2000. The CPS shows a decrease from 14.3 percent uninsured in 1999 to 14.0 percent in 2000. The two other surveys do not show a significant change. In the NHIS, there is a slight increase from 14.2 percent in 1999 to 14.7 percent in 2000, but this change is not statistically significant. Similarly, the MEPS shows a slight increase from 15.8 to 16.1 percent, a change that is also not statistically significant.

In interpreting a complex picture on health insurance coverage, there are several important cautions:

  • Most of the differences shown between 1999 and 2000 are relatively small, and two of the surveys show no statistically significant change.
  • For the most part, there have not been large changes in insurance coverage in the last several years.
  • While the comparison of any two years of data may vary across surveys, the longer term trends are consistent from these different sources. Since differences between any two years are more subject to random or sampling variation than longer term trends, it is important to view these yearly differences in context. Generally, the overall trend is for lower observed levels of the uninsured. The first quarter 2001 data from the NHIS suggests that this longer term trend may be consistent across the surveys. Data from the NHIS will be released on a quarterly basis throughout 2001 to monitor these trends.
  • Further analysis will be required to better understand the trends and the differences among surveys. For example, as more data are available for the NHIS later in the year, it will be possible to compare data on the types of coverage (e.g., private, Medicaid, etc.) to see how comparable these trends are.

Findings on health insurance coverage for children (under 18 years) also vary across the surveys. For 2000, the CPS finds that 11.6 percent of children were uninsured; the NHIS reports 12.2 percent uninsured; and the MEPS reported that 13.9 percent of children were uninsured.

Changes between 1999 and 2000 are also somewhat divergent. From 1999 to 2000, the CPS shows a statistically significant decline in the percent of uninsured children, while the NHIS and MEPS show small increases that are not statistically significant. Despite this small increase in 2000, the trend shown in the NHIS for children is one of decline, reinforced by the first quarter data from the 2001 NHIS, which shows a decrease to 11.5% uninsured.

What are the right numbers to use in presenting data on health insurance coverage?

There is no one “best” estimate that serves all purposes. In fact, given that the measurement of health insurance coverage is difficult, it is useful to have estimates derived from different methods and sources. This multiplicity of data sources gives us a way of cross-checking results and gaining insight from different perspectives. As noted above, these estimates come from surveys that measure insurance using different approaches (e.g., point in time, full year uninsured, etc.). Analysts should use the measure most relevant to the analysis or statement being made.

Page last reviewed: November 6, 2015