Methodology
Data Sources
Georgetown University Center for Children and Families (CCF) uses the U.S. Census Bureau American Community Survey (ACS), an annual survey of approximately 3.5 million individuals, to analyze national, state, and local trends in health insurance coverage. The data in this report come from three sources:
- 2016-2019 Public Use Microdata Sample (PUMS), a two-thirds sample of the full ACS data file. This sample allows for disaggregation by detailed Latino ethnicity. Files are downloaded from census.gov FTP platform.
- 2019 Integrated Public Use Microdata Sample (IPUMS), a recoded and enhanced version of PUMS which enables the analysis of parental characteristics (such as comfort level with English).
- 2019 Puerto Rico Community Survey Public Use Microdata Sample (PUMS), a sample of the full PRCS data file.
Other analyses and reports from CCF, including kidshealthcarereport.ccf.georgetown.edu and “Children’s Uninsured Rate Rises by Largest Annual Jump in More than a Decade” (October 2020) use the American Community Survey detailed tables, published on data.census.gov. The detailed tables are based on the full sample of ACS results, but do not allow for detailed disaggregation. Consequently, estimates may vary between CCF’s analyses.
Margin of Error, Statistical Significance, and Data Suppression
Following the instructions given in the “Calculating Margins of Error the ACS Way Using Replicate Methodology to Calculate Uncertainty” webinar (February 2020), standard error and coefficients of variation are computed using successive differences replication (SDR) in STATA statistical software. Margin of error calculations are not published in this report but are available upon request.
Statistical significance is determined using the U.S. Census Bureau “Statistical Testing Tool” with a confidence interval of 90 percent. In other words, when the difference between two values is marked as significant, 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, or high margins of error, indicate that the difference could be due to chance or sampling error. Consequently, they are less likely to “pass” the significance test.
To ensure accuracy and consistency, CCF calculates the coefficient of variation (CV; also known as the relative standard error) for each estimate. CCF follows the instructions included in the Census Bureau’s publication, “Understanding and Using American Community Survey Data: What All Data Users Need to Know” (September 2020). CVs produce a comparable indicator of how large the error is by dividing the standard error of an estimate by the estimate itself. The lower the CV, the more reliable the estimate. Estimates with CVs greater than 25 percent are not presented in this analysis. Applying this rule results in the suppression of several estimates for Latino subgroups (for example, the uninsured rate for Puerto Rican children in Arizona) and of several states in figure 7 and the appendix tables (Alaska, Delaware, District of Columbia, Hawaii, Iowa, Maine, Montana, New Hampshire, North Dakota, Rhode Island, South Dakota, West Virginia, and Wyoming).
Demographic Characteristics
Children refers to individuals under age 19 (0 to 18 years of age). In 2017, the Census Bureau changed the upper bound for children from 18 (0 to 17 years of age) to 19 (0 to 18 years of age) on the detailed health insurance tables published on data.census.gov, thus making comparisons between earlier years difficult. Because this report uses microdata from PUMS and IPUMS, it is possible to compare trends from 2016 to 2019.
As noted in Appendix A, the Census Bureau distinguishes between race and Hispanic origin/Latino ethnicity. For the purposes of this analysis, “Latino” refers to all those who indicated that they were of Hispanic or Latino origin on question five of the ACS. “Latinx” can also be used to respect various gender identities and expressions. “Non-Latino” refers to all those who indicated that they were not of Hispanic or Latino origin on question five of the ACS. Latino and Non-Latino individuals may be of any race.
English Proficiency
Question 14c of the ACS asks respondents “How well does this person speak English?” and provides them with the following options: Very well, Well, Not well, Not at all. For the purposes of this paper, individuals who indicate “Not well” or “Not at all” are categorized as “Language Other than English (LOE).” CCF linked parental language proficiency to children using the Integrated Public Use Microdata Sample’s (IPUMS’) “attach characteristic” feature.
Health Coverage
ACS data is collected over the course of a year and represents a “point-in-time” estimate of a person’s insurance status. That is, the survey collects information on whether the respondent is insured at the moment they complete the form, not if they have been insured/uninsured at any point during the year. The US Census Bureau does not consider Indian Health Service (IHS) access a comprehensive form of coverage. Consequently, those who indicate that IHS is their only source of coverage are designated as uninsured.
Citizenship Status of Child and Parent
Unlike the decennial Census, the ACS collects data on respondents’ citizenship status. Citizenship status is not the same as immigration status; a respondent classified as non-citizen can be lawfully-residing or undocumented. For the purposes of this analysis, “citizen” includes any person born in the U.S., any person born abroad to American parents, and any naturalized citizen. Citizenship status of children’s parents was computed using the Integrated Public Use Microdata Sample (IPUMs) “attach characteristic” feature. More information on how the University of Minnesota codes the survey responses to create these enhanced variables, please see “Frequently Asked Questions (FAQ) Extract Option: Attach Characteristics.”
Poverty Status
The Census Bureau determines an individual’s poverty status by comparing their estimated income to the Census Poverty Thresholds. Though the overall Census Poverty Thresholds are similar to the Department of Health and Human Services Federal Poverty Level Guidelines, there are significant differences for Alaska and Hawaii. Further, the Census does not adhere to the same MAGI formula for computing income that state Medicaid and CHIP programs use when determining income-based eligibility.
Medicaid Expansion Analysis
This report relies on ACS data collected in 2019. For this reason, expansion status is determined based on if a state had implemented expansion for the majority of 2019. The expansion states include: Alaska, Arizona, Arkansas, California, Colorado, Connecticut, Delaware, District of Columbia, Hawaii, Illinois, Indiana, Iowa, Kentucky, Louisiana, Maryland, Massachusetts, Michigan, Minnesota, Montana, Nevada, New Hampshire, New Jersey, New Mexico, New York, North Dakota, Ohio, Oregon, Pennsylvania, Rhode Island, Washington, West Virginia, and Vermont. The following states are categorized as non-expansion: Alabama, Florida, Georgia, Idaho (implemented in November 2019), Kansas, Mississippi, Missouri (plans to implement in 2021), Nebraska (implemented in 2020), North Carolina, Oklahoma (plans to implement in 2021), South Carolina, South Dakota, Tennessee, Texas, Utah (implemented in 2020), Wisconsin, and Wyoming. Maine and Virginia were excluded from this analysis given that these estimates were also used to calculate change over time.
