Methodology for Report “Number of Uninsured Children Stabilized and Improved Slightly During the Pandemic”

Data Sources and Changes to Race/Ethnicity Questions and Age Categories for Children

This report from the Georgetown University Center for Children and Families (CCF) uses data from the U.S. Census Bureau American Community Survey (ACS). The ACS randomly selects about 3.5 million households each year to be surveyed and conducts the survey year-round, publishing results annually.

This report uses two ACS data products: 1) Health Insurance Historical Table HIC-5. Health Insurance Coverage Status and Type of Coverage by State Children Under 19: 2008 to 2021, and 2) the 2021 1-Year ACS Estimates Detailed Tables published by the Census Bureau on Please note that, because of differences in sample size and data processing, the estimates published in this report may differ from other estimates produced using either the 5-year ACS estimates or ACS microdata (including the Census Bureau’s Public Use Microdata Sample (PUMS) or the University of Minnesota’s Integrated Public Use Microdata Series (IPUMS)), despite the fact that all of these datasets are based on the same American Community Survey.

Because of data quality issues related to the pandemic, the Census Bureau did not publish standard 1-year estimates for 2020 but instead only released a set of 1-year experimental estimates. The experimental estimates are not available through and the Census Bureau notes that these experimental estimates should not be compared to other ACS 1-year estimates, so CCF excludes 2020 ACS data from all of its analyses.

In 2017, the Census Bureau updated the age categories used in the detailed tables to define children as individuals under age 19 (in previous years, children were classified as individuals under age 18). Because of this change, detailed tables from before and after 2017 should not be compared. When using the detailed tables, CCF therefore limits trend analysis to years 2017 and after. When examining longer-term trends, CCF instead uses the HIC-5 table, which standardizes age groups to children under age 19.

Margin of Error, Data Reliability and Suppression, and Statistical Significance

The Census Bureau provides a margin of error (MOE) at a 90 percent confidence level for each estimate it publishes. Because ACS data are based on a sample of the population (as opposed to the full population), there is a level of uncertainty associated with each estimate. This uncertainty is captured by the margin of error. When CCF calculates a new estimate based on ACS data (such as when we combine racial/ethnic groups or calculate percentages/rates), we also calculate its margin of error, using formulas provided by the Census Bureau in their handbook: “Understanding and Using American Community Survey Data: What All Data Users Need to Know” (September 2020). CCF does not take the margin of error into account when ranking states by the number and percent of uninsured children by state. Although we do not publish margins of error in this report, they are available upon request.

CCF calculates coefficients of variation (CVs, also known as relative standard errors) to measure data reliability for each estimate. The CV measures the relative amount of error in an estimate by comparing how large its standard error is to the estimate itself. The lower the CV, the more reliable the estimate. CCF suppresses any estimate with a CV larger than 25 percent.

CCF uses the Census Bureau’s Statistical Testing Tool to determine statistical significance between estimates at a 90 percent confidence level. Statistical significance indicates that differences between estimates are due to something besides chance. For example, an estimate marked as significant at a 90 percent confidence level indicates that there is a 90 percent likelihood that the difference is not due to chance or sampling error. Margins of error are a critical part of determining statistical significance; two estimates with high levels of uncertainty (e.g., high margins of error) indicate that the difference could be due to sampling error and are less likely to “pass” the significance test. Differences between estimates should not be assumed to be statistically significant unless specifically discussed or marked as such.

Geographic Levels

The Census Bureau publishes 1-year ACS estimates for all geographic areas with a population of 65,000 or more, which includes all regions, states (including the District of Columbia), and some counties. Please note that 1-year estimates will differ from 5-year estimates, which CCF may use elsewhere for analyses of smaller geographic levels like counties or school districts. CCF uses Census Bureau designations to report regional data.

Poverty Status

Data on poverty thresholds only include individuals for whom the Census Bureau could determine poverty status for the past year. This population is slightly smaller than the total non-institutionalized population of the U.S. (the universe for all other data used in this report). The Census Bureau determines an individual’s poverty status by comparing that person’s income in the past 12 months to Census Poverty Thresholds (CPTs). Notably, Census Poverty Thresholds differ from the poverty guidelines (commonly known as the Federal Poverty Level or FPL) determined by the Department of Health and Human Services (HHS), and may differ considerably for the separate FPLs that HHS determines for Alaska and Hawaii. Additionally, Census Poverty Thresholds may include some income sources that state Medicaid and CHIP agencies would not count purposes of determining income eligibility using Modified Adjusted Gross Income (MAGI).

Health Insurance Coverage and Medicaid Undercount

ACS data represents a “point-in-time” estimate of an individual’s insurance coverage, meaning that the survey collects information on the respondent’s coverage only at the moment they complete the form, not at another point during the year. (The ACS is conducted over the course of the year.)  The ACS groups coverage into the following categories: employer-based health insurance only, direct purchase health insurance only, Medicare coverage only, Medicaid/means-tested public coverage only (including CHIP), TRICARE/military health coverage only, VA health coverage only, two or more types of health insurance coverage, and no health insurance coverage. CCF combines Medicare, TRICARE/military coverage, VA coverage, and two or more types of coverage and reports the calculations as “Other.” Except for Other, all coverage categories refer to individuals who reported having only one source of coverage. The Census Bureau does not consider access to Indian Health Service (IHS) services alone as a comprehensive form of health insurance coverage. Consequently, individuals who indicate that IHS is their only source of coverage are designated as uninsured. Individuals can report more than one source of coverage, so totals may sum to more than 100 percent.

Please note that ACS estimates are not adjusted by the Census Bureau (or by CCF) to address the “Medicaid undercount” often observed when comparing surveys to the reported numbers of individuals enrolled in Medicaid and CHIP using federal and state administrative data. For example, ACS data show that 30.5 million children had Medicaid coverage (either alone or in combination with another type of coverage) in 2021, while administrative data show average enrollment over the same period equal to 40.2 million, a difference of nearly 10 million children. This undercount occurs in the majority of surveys and is not unique to the ACS, though the extent of the undercount varies among federal surveys. Researchers attribute the Medicaid undercount to a combination of factors like sampling error and differences in demographic characteristics (for example, adults and individuals with higher incomes are less likely to report Medicaid coverage); reporting error, where individuals may respond that they do not have coverage or have a type of coverage other than Medicaid; and differences in how surveys and administrative data define coverage, such as estimates taken at one point in time versus over the course of a year. Additionally, recent research on the decennial Census shows that young children are consistently and significantly undercounted, likely worsening the Medicaid undercount among children. In 2021, the Medicaid continuous coverage provision may affect children’s reported coverage source—including uninsurance—if families are unaware that they still have Medicaid coverage.

 Demographic Characteristics

“Children”, as noted above, are defined as individuals under age 19 (ages 0-18).

The American Community Survey allows respondents to self-identify as the following races: White alone, Black/African-American, American Indian/Alaska Native, Asian, Native Hawaiian/Pacific Islander, “Some other race,” and “Two or more races.” To improve sample sizes and data reliability, CCF combines estimates for Asian alone and Native Hawaiian or Other Pacific Islander alone and reports the calculations as “Asian, Native Hawaiian, or Other Pacific Islander” and combines “Some other race” alone and “Two or more races” and reports the calculations as “Other or Multiracial.” Except for Other or Multiracial, all racial categories refer to individuals who reported belonging only to one race.

The Census Bureau recognizes and reports race and Hispanic origin (i.e., ethnicity) as separate and distinct concepts and variables. “Hispanic or Latino” refers to individuals who self-identified as being Hispanic or Latino, while “non-Hispanic/Latino” refers to individuals who indicated that they were not of Hispanic or Latino origin. CCF calculates estimates for non-Hispanic or Latino populations by subtracting estimates for Hispanic or Latino individuals from the total population estimate for children. As “Hispanic or Latino” refers to a person’s ethnicity, Hispanic and non-Hispanic individuals may be of any race.

Beginning in 2020, the Census Bureau made changes to the race and ethnicity questions on the ACS which may affect health coverage comparisons related to race and ethnicity between the 2021 and 2019 1-year estimates. These included changes to the instructions and examples listed with some race and Hispanic origin response options, additional write-in response options for “White” and “Black or African American” categories, and changes to the way the Census Bureau processes write-in responses. These changes may affect the “Some Other Race” or multiple race categories in particular. For example, 10.2 million children identified as another race or multiracial in 2019, representing 13 percent of the total child population. In 2021, this number increased to 21.6 million or 28 percent of the child population. This increase, which may be related to changes in question design, may also affect the distribution of children across other racial groups. As a result, the Census Bureau recommends caution in comparing 2021 and 2019 ACS estimates related to race/ethnicity.