Natalie Mac Arthur
Senior Research Associate

e shadac@umn.edu

2020 Public Use Microdata Area (PUMA) Updates in the 2022 American Community Survey

A Public Use Microdata Area (PUMA) is a type of geographic unit created for statistical purposes. PUMAs represent geographic areas with a population size of 100,000–200,000 within a state (PUMAs cannot cross state lines). PUMAs are the smallest level of geography available in American Community Survey (ACS) microdata. They are designed to protect respondent confidentiality while simultaneously allowing analysts to produce estimates for small geographic areas. 

Every ten years, the decennial census results are used to redefine ACS PUMA boundaries to account for shifts in population and continue to maintain respondent confidentiality. This process is intended to yield geographic definitions that are meaningful to many stakeholders. 

Most recently, new PUMAs were created based on the 2020 Census; these 2020 PUMAs were implemented in the ACS starting in the 2022 data year. Although Public Use Microdata Area components remain consistent to the extent possible, they are updated based on census results and revised criteria. Therefore, they are not directly comparable with PUMAs from any previous ACS data years. For example, the 2020 PUMAs used in the 2022 data year are distinct from the 2010 PUMAs, which were used in the 2012–2021 ACS data years. 

The 2020 PUMAs were created based on definitions that include two substantive changes relative to the 2010 PUMAs: 

1) An increase in the minimum population threshold for the minimum size of partial counties from 2,400 to 10,000. Increasing the population minimum for a PUMA-county part aims to further protect the confidentiality of respondents. However, exceptions are allowed on a case-by-case basis in order to maintain the stability of PUMA definitions (that were based on the previous minimum of 2,400) and due to unique geography.

2) Allowing noncontiguous geographic areas. Allowing PUMAs to include noncontiguous geographic areas aims to avoid unnecessarily splitting up demographic groups in order to provide more meaningful data. This change is not intended to create highly fragmented PUMAs. 

Other than the two changes listed above, PUMA criteria remained consistent, such as treating 100,000 as a strict minimum population size for PUMAs. The maximum population size for PUMAs can exceed a population of 200,000 in certain instances due to expected population declines or geographic constraints. 

Generally, counties and census tracts are the building block geographies for PUMAs. This maximizes the stability of PUMA boundaries and therefore reduces disclosure risks. Counties may be combined or split as needed, depending on population size. Additionally, PUMAs are designed to avoid unnecessarily splitting metropolitan areas, minor civil divisions (MCDs) with a functional government, and areas of American Indian/Alaska Native (AIAN) and Native Hawaiian people. Carlton County in Minnesota, for example, illustrates that splitting a county was necessary to avoid splitting Fond du Lac reservation into multiple PUMAs (U.S. Census Bureau, 2022).

Figure 1: Minnesota PUMA 00400 Includes Fond du Lac Reservation in Its Entirety

Note: Minnesota PUMA 00400 featuring Fond Du Lac Reservation in its entirety. Reprinted from Final Criteria for Public Use Microdata Areas for the 2020 Census and the American Community Survey from https://www2.census.gov/geo/pdfs/reference/puma2020/2020PUMA_FinalCriteria.pdf.

Due to periodic changes in Public Use Microdata Area boundaries, states as a whole are the most granular geographic variable in the ACS microdata that can be compared consistently across all years. However, IPUMS USA offers harmonized geographic variables that greatly expand the options for analyzing ACS data across both time and geography. These variables are less granular than PUMAs, but more granular than states. For example, CONSPUMA and CPUMA0010 combine ACS PUMAs to create consistent geographies that can be compared across multiple decades of PUMA definitions; these have not yet been updated to include the new 2020 PUMAs. Currently, CITY, COUNTYFIP, and MET2013 allows users to compare cities, counties, and metropolitan areas, respectively, over years 2005-2022.

IPUMS USA also provides in-depth geographic resources for ACS data users, including interactive maps of the 2010 and 2020 PUMAs. For example, below is a map of the Twin Cities metropolitan area, Minnesota, showing 2020 PUMAs in blue and 2010 PUMAs in orange (IPUMS USA, 2020). This map illustrates that the updated PUMA criteria resulted in changes to PUMA boundaries in urban areas and some aggregation of PUMAs in more rural areas.

Figure 2: IPUMS USA Map of Twin Cities Metropolitan Area, Minnesota 2010 and 2020 PUMAs

Note: IPUMS USA Map of Twin Cities Metropolitan Area, Minnesota 2010 and 2020 PUMAs. Reprinted from 2020 PUMA DEFINITIONS: WEB MAP OF 2010 AND 2020 PUMAS from IPUMS USA, https://usa.ipums.org/usa/volii/pumas20.shtml.

Looking for other data and information from the American Community Survey? Learn more with the following SHADAC products:  

2022 ACS Tables: State and County Uninsured Rates

SHADAC Comments on Proposed 2025 American Community Survey Health Insurance Coverage Instrument Changes

American Indian/Alaska Native (AIAN) and Native Hawaiian and Other Pacific Islander (NHPI) data added to State Health Compare ACS estimates

Explore the Data on State Health Compare

Sources: 

U.S. Census Bureau. (2022, July 25). Final Criteria for Public Use Microdata Areas for the 2020 Census and the American Community Survey. https://www2.census.gov/geo/pdfs/reference/puma2020/2020PUMA_FinalCriteria.pdf

IPUMS USA (2020). 2020 PUMA Definitions: Web Map of 2010 and 2020 PUMAs. https://usa.ipums.org/usa/volii/pumas20.shtml.

ACS 5-year Estimates: State and County Uninsured

January 2024:

Following the release of 2022 single-year estimates of health insurance coverage, household income, and poverty levels, the U.S. Census Bureau has now made available 5-year American Community Survey data files. These ACS 5 Year estimates are generated by pooling together five years of American Community Survey (ACS) data to produce estimates for areas and subgroups with smaller populations.

The interactive map provided below offers users the opportunity to explore health insurance coverage estimates, specifically the percentage of uninsured individuals for each state and all counties for the pooled years 2018-2022.* These data can be accessed via the Census Bureau’s data.census.gov tool. Click on a county in the map below to view state/county data tables. Counties are easily searchable through bookmarks in each state file.

2018-2022 American Community Survey (ACS) 5-Year Estimates: Percent Uninsured, Total Civilian Noninstitutionalized Population by County

Click here to view estimates for Puerto Rico by municipio.
Click here to view a 50-state table of estimates.

---

About the American Community Survey and ACS Data

The ACS is a household survey that began in 2005 and produces annually updated data on a variety of population characteristics, including health insurance coverage. In total, the ACS surveys approximately three million US households each year. An important feature of the ACS is that it includes a large enough sample for state‐level and sub‐state estimates. The Census Bureau provides ACS 1 year estimates and ACS 5 year estimates.

The Census Bureau publishes 1-year estimates for areas with populations of 65,000 or more and 5-year estimates (covering 60 months) for all statistical, legal, and administrative entities.

The ACS began asking survey respondents about health insurance coverage during the 2008 calendar year. Specifically, the survey asks respondents about current coverage for each person in the respondent’s household. A person is categorized as “insured” if he or she has coverage at the point in time at which the survey is administered.

*The U.S. Census Bureau has extensively cautioned against the use of single-year 2020 ACS estimates due to disruptions caused by the COVID-19 pandemic (e.g., limited means of data collection, such as shutdowns of mail operations, switches to telephone-first methodologies, etc., leading to low response rates and nonresponse bias). However, the Census Bureau believes that course corrections to address nonresponse bias, and the larger sample resulting from pooled data mean that the “data are fit for public release, government and business uses, and understanding the social and economic characteristics of the U.S. population and economy.”[1]

How Are These Estimates Different from the Estimates that SHADAC Publishes Using Census Bureau Micro-Data Files?
Two definitions used by the Census Bureau to generate the tabulations above differ from those that SHADAC uses to generate tabulations for State Health Compare. The definitional differences are as follows:

Family
The Census Bureau defines a family as “all related people in a household.”

SHADAC defines a family using a measure called the “Health Insurance Unit” (HIU), which includes all individuals who would likely be considered a family unit in determining eligibility for either private or public coverage.

To learn more about the 2020 update of SHADAC's Health Insurance Unit (HIU) see our HIU resource page, which houses two issue briefs: The first describes the SHADAC HIU, its purpose, the most recent update, and improvements to HIU data inputs; and the second outlines the impacts of using the SHADAC HIU in analysis so that researchers can assess whether the SHADAC HIU is suitable for their research and what the potential impacts of its use might be. The page also provides a link to STATA and SAS codes to aid in the use of the HIU variable.

Family Income
The Census Bureau determines family income as a percentage of the Federal Poverty Level (FPL), which is a definition of poverty used primarily for statistical purposes. For example, FPL is used to estimate the number of Americans living in poverty each year.

SHADAC determines family income as a percentage of the U.S. Department of Health and Human Services’ Federal Poverty Guideline (FPG), which is a measure used for administrative purposes. For example, FPG is used to determine eligibility for federal programs such as Medicaid and CHIP, as well as the Supplemental Nutrition Assistance Program (SNAP).

Check out our blog post from April 2023 to learn more about the difference between FPL and FPG.

---

Related ACS Materials:

Blog: 2022 ACS: Declining Uninsured Rates for the U.S. and States are Supported by Private and Public Coverage Increases

Blog: 2022 ACS Tables: State and County Uninsured Rates, with Comparison Year 2021

Webinar: An Annual Conversation with the U.S. Census Bureau: 2022 Health Insurance Coverage Estimates from the ACS and CPS

---

[1] U.S. Census Bureau. (2022, February 7). Census Bureau Update on 2016–2020 American Community Survey (ACS) 5-Year Estimates [Press Release CB22-RTQ.01]. https://www.census.gov/newsroom/press-releases/2022/acs-5-year-estimates-update.html

SHADAC Staff

p 612.624.4802
e shadac@umn.edu

Race/Ethnicity Data in CMS Medicaid (T-MSIS) Analytic Files: 2021 Data Assessment

The Transformed Medicaid Statistical Information System (T-MSIS) is the largest national database of current Medicaid and Children’s Health Insurance Program (CHIP) beneficiary information collected from U.S. states, territories, and the District of Columbia (DC).¹ T-MSIS data are critical for monitoring and evaluating the utilization of Medicaid and CHIP, which together provide health insurance coverage to almost 90 million people.²

Due to their size and complexity, T-MSIS data files are challenging to use directly for research and analytic purposes. To optimize these files for health services research, Centers for Medicare and Medicaid Services (CMS) repackages them into a user-friendly, research-ready format called T-MSIS Analytic Files (TAF) Research Identifiable Files (RIF). One such file, the Annual Demographic and Eligibility (DE) file, contains race and ethnicity information for Medicaid and CHIP beneficiaries. This information is vital for assessing enrollment, access to services, and quality of care across racial and ethnic groups in the Medicaid/CHIP population, whose members are particularly vulnerable due to limited income, physical and cognitive disabilities, old age, complex medical conditions, housing insecurity, and other social, economic, behavioral, and health needs.    

To guide researchers and other consumers in their use of T-MSIS data, CMS produces data quality assessments of the completeness of race and ethnicity data along with other data such as enrollment, claims, expenditures, and service use. The Data Quality (DQ) assessments for race and ethnicity data have been posted for data years 2014 through 2021 and indicate varying levels of “concern” regarding race and ethnicity data completeness. Some data years have multiple data versions (e.g., Preliminary, Release 1, Release 2), each with their own DQ assessment.

While completeness of race and ethnicity data reported to CMS has historically remained inconsistent among the states, territories, and DC, SHADAC has been monitoring the quality of these data over time. We are excited to discuss a noticeable improvement in quality as discussed below. This blog explores not only the 2021 Data Release 1, the most recent T-MSIS race and ethnicity data for which a DQ assessment is available, but also a brief analysis of data quality trends over time that we plan to follow in future T-MSIS file releases.

Evaluation of T-MSIS Race and Ethnicity Data

DQ assessments for each year and data version of T-MSIS data are housed in the Data Quality Atlas (DQ Atlas), an online evaluation tool developed as a companion to T-MSIS data.³ The DQ Atlas assesses T-MSIS race and ethnicity data using two criteria: the percentage of beneficiaries with missing race and/or ethnicity values in the TAF; and the number of race/ethnicity categories (out of five) that differ by more than ten percentage points between the TAF and American Community Survey (ACS) data. Taken together, these two criteria indicate the level of “concern” (i.e., reliability) for states’ T-MSIS race/ethnicity data. To construct the external ACS benchmark for evaluating T-MSIS data, creators of the DQ Atlas combine race and ethnicity categories in the ACS to mirror race and ethnicity categories reported in the TAF (see Table 1). More information about the evaluation of T-MSIS race and ethnicity data is available in the DQ Atlas’ Background and Methods Resource.

Five “concern” categories appear in the DQ Atlas: Low Concern, Medium Concern, High Concern, Unusable, and Unclassified. States with substantial missing race/ethnicity data or race/ethnicity data that are inconsistent with the ACS – a premier source of demographic data – are grouped into either the High Concern or Unusable categories, whereas states with relatively complete race/ethnicity data or race/ethnicity data that align with ACS estimates are grouped into either the Low Concern or Medium Concern categories. The Unclassified category includes states for which benchmark data are incomplete or unavailable for a given data year and version.  

Table 1. Crosswalk of Race and Ethnicity Variables between the TAF and ACS

Federal Survey Sample Size Analysis: Disability, Language, and Sexual Orientation and Gender Identity

October 2023:

The Medicaid and CHIP Payment and Access Commission contracted with the State Health Access Data Assistance Center (SHADAC) to conduct a review and sample size analysis of 13 federal population health surveys to assess the quality and availability of data for certain populations facing health inequities, including people with disabilities, language access needs, and people who indicate their sexual orientation or gender identity (SOGI). This brief provides an overview of each federal survey and identifies which surveys included questions about disability, primary language, limited English proficiency (LEP), and sexual orientation and gender identity (SOGI). Below is an overview table of the Federal surveys included in this analysis. 

Federal Survey	Periodicity	Target Population	States, DC, and Territories Represented	Most Recent Data Year	Total Population Unweighted Sample Size	Total Unweighted Sample Size for Medicaid 1
ACS	Annual	Civilian, non-institutionalized U.S. population	50 states, D.C., Puerto Rico	2021	3,252,599	434,900
BRFSS	Annual	U.S. adults > 18 years	49 states (no Florida in 2021), D.C., Guam, Puerto Rico, Virgin Islands	2021	438,693	50,306
CPS	Annual	Civilian, non-institutionalized U.S. population	50 states, D.C.	2022	152,732	24,155
HPS	Panel	U.S. adults > 18 years	50 states, D.C.	2023	59,290	2,764
MEPS-HC2	Panel	Civilian, non-institutionalized U.S. population	No state estimates	2020	27,805	6,504
MCBS3	Annual	All individuals enrolled in Medicare	No state estimates	2020	13,916	2,593
NAMCAHPS4	Fielded once	Adults enrolled in Medicaid	50 states, D.C.	2014	272,679	272,679
NHANES	Periodic	Civilian, non-institutionalized U.S. population	No state estimates	20205	9,254	2,841
NHIS	Annual	Civilian, non-institutionalized U.S. population	50 states, D.C. when two years combined in the RDC	2022	35,115	4,918
NSCH6	Annual	U.S. children < 18 years	50 states, D.C. when two years combined	2021	50,892	11,259
NSDUH	Annual	Civilian, non-institutionalized U.S. population, age 12 +	No state estimates	2021	58,034	11,008
PRAMS	Annual	Pregnant women in 47 states, D.C., and Puerto Rico	See target population	2021	NA	NA
SIPP7	Panel	Civilian, non-institutionalized U.S. population	50 states, D.C.	2022	40,510	5,835

Notes: ACS is American Community Survey. BRFSS is Behavioral Risk Factor Surveillance System. CPS is Current Population Survey. HPS is Household Pulse Survey. MEPS-HC is Medical Expenditure Panel Survey-Household Component. MCBS is Medicare Current Beneficiary Survey. NAMCAHPS is Nationwide Adult Medicaid Consumer Assessment of Healthcare Providers & Systems. NHANES is National Health and Nutrition Examination Survey. NHIS is National Health Interview Survey. NSCH is National Survey of Children’s Health. NSDUH is National Survey on Drug Use and Health. PRAMS is Pregnancy Risk Assessment Monitoring System. SIPP is Survey of Income and Program Participation. DC is District of Columbia.  RDC is Research Data Center.  NA is not available.
1 Unless otherwise noted, where relevant, Medicaid was defined using the typical MACPAC hierarchy:
                (1) Medicare; (2) Private with no Medicare; (3) Medicaid with no Medicare or private; (4) Other type of insurance with no Medicare or private; (5) Uninsured all year
2 Medicaid coverage was defined based on the most recent date of data collection (e.g., December).
3 MCBS data uses “full year dual” to define Medicaid, meaning people who were enrolled in Medicare and Medicaid all year.
4 NAMCAHPS limited to adults enrolled in Medicaid.
5 Sample size analysis conducted using 2017-2018 file, because of pandemic impacts on data collection in 2020.
6 The NSCH includes three categories of children’s coverage (public, private, uninsured); we code children in public coverage as having Medicaid, because Medicare is much less common among children.
7 SIPP data uses the annual coverage variables to define Medicaid.
Source: SHADAC analysis of federal survey data.

The findings show that the majority of surveys include questions to identify individuals with a functional disability. Still, fewer than half of these surveys include questions about intellectual disability, serious mental illness, primary language, LEP, and SOGI. Of the surveys that include these questions, the majority have a sufficient sample for reporting about Medicaid-covered individuals for each demographic group.

SHADAC Staff

p 612.624.4802
e shadac@umn.edu

Considerations from SHADAC: Proposed Revisions to Federal Standards for Collecting Race/Ethnicity Data

Response from SHADAC

The State Health Access Data Assistance Center (SHADAC) is a health policy research center within the University of Minnesota School of Public Health. We provide technical assistance (TA) to states in the areas of data use and analysis to support policy. The below comments are based on SHADAC’s specific experience supporting states as they address challenges related to the collection of data on race, ethnicity, language, disability, sexual orientation, and gender identity via the Medicaid application.

SHADAC was thrilled to see the revised OMB standards related to the collection of race and ethnicity. The proposed revisions align with available evidence, are consistent with the changes made by leading states, and most importantly, explicitly state that these standards should serve as a minimum baseline with a call to collect and provide more granular data. However, while these standards are specifically named as minimum reporting categories for data collection throughout the Federal Government, if adopted, they are likely to shape data collection and reporting across all sectors, including the states with whom we work that collect race/ethnicity data through the Medicaid application process.

State Medicaid programs have the flexibility to develop their own applications and race and ethnicity questions as long as they meet CMS-established reporting requirements (which follow current OMB guidance). Although many states follow the application model that CMS developed in 2013, our research has found 65 different variations in how these data are collected across all states and modes (e.g., paper, online). Once collected, this information is used in a wide variety of ways: to support state analysis and operation; in reporting to CMS; and to feed into the T-MSIS, which in turn is released as the research-ready T-MSIS data files used by researchers studying Medicaid program health outcomes, health equity, and administrative accountability.

Given the role OMB standards play in shaping state data collection approaches and the obvious impact and reach of these data, we respectfully submit our comments aimed at supporting the adoption of a revised set of standards.

Cross Program and Agency Alignment

Guidance needs to include detail on how any revisions will filter through data collection and reporting across all sectors that impact states/federal interactions. Will all federal agencies adopt the proposed new standards on the same timeline?

For example, all Food and Nutrition Services (FNS) state agencies are currently required to collect race and ethnicity according to prescribed specifications: race and ethnicity must be collected using a two-question format (ethnicity must be collected before race), and the minimum categories must align with OMB standards. Therefore, states with a combined Medicaid/SNAP application currently cannot offer a combined race/ethnicity question even though the most current evidence shows that is the preferred approach. States will want to know whether the adoption of the new proposed OMB standard allowing for a combined race/ethnicity question will supersede the existing FNS guidance on how to collect this information. Or will states that want to pursue a combined question on their combined Medicaid/SNAP application have to wait for updated FNS guidance to be released?

Similarly, existing guidance requires states to share data (including demographic data like race and ethnicity) with health plan carriers via a federally standardized 834 template. States cannot make changes to existing data fields on this form—any changes made currently result in enrollment transaction failures. If OMB standards are updated, how and when will changes be made to the 834 file form to align with these new changes? 

Finally, the revisions have implications for state data reporting of race and ethnicity into the Transformed Medicaid Statistical Information System (T-MSIS). If approved, states will need guidance on how to transition to a single data element (combined race and ethnicity). They will need to understand how to streamline or roll-up more granular data and more write-in data and how to manage the likely increase in people who report more than one race.

Data Granularity

The revised guidance rightfully acknowledges the challenges associated with collecting more granular data. If the final guidance recommends the inclusion of an open text response option for each detailed category, states will need detailed, individual guidance on how best to manage those data. While most states prioritize granularity, they simply don’t have the staff time or, in some cases, the expertise to develop the thoughtful and detailed cleaning and coding procedures required to process a large volume of write-in data. In addition, many current IT systems don’t have the functionality to collect and store a large volume of write-in data.

One way to support this would be to develop or enhance an existing hierarchy that includes a long and diverse list of racial and ethnic categories, such as the Centers for Disease Control Race Category and Ethnicity Groupings. Another way to support the use and reporting of disaggregated data is to provide technical support and guidance around data use standards. This could include standards around recency, maximum “refusal to answer” thresholds, data missingness thresholds, and sample size thresholds that guide suppression of publication.

Technical Assistance

States need direct technical assistance (TA) to support the successful implementation of revised standards. This will need to go beyond documentation and guidance. To do this well, states should have access to experts who can provide practical, hands-on advice that is responsive to their specific needs. The type of assistance needed will vary. States with sophisticated analytic teams may seek advice on how to bridge race and ethnicity data collected under one method with data collected under a revised method. Other states may need more fundamental support to implement the expanded race/ethnicity option in a way that is the least disruptive to their current data collection structure.

We would also recommend that any TA effort include a forum for states to discuss implementation challenges with their peers and serve as a repository for promising practices. States will face different challenges but always appreciate having a forum to discuss problems and solutions with other state officials. In addition, there is limited research on best ways to collect data on race and ethnicity and even less on operational considerations for collecting this information in an administrative context like a Medicaid application. Technical assistance should encourage and support the evaluation of systems’ improvements and serve as a clearinghouse for sharing key findings.

To build trust with the enrollees from whom this data is being collected, states would also benefit from technical assistance or model training materials on the best ways to communicate why this data is being collected and how it is being used (e.g., TA on effective forms of community engagement, Navigator or assister training, or instructional language scripts for frontline workers.)  

Funding

Questions that collect race and ethnicity data within the Medicaid application are embedded in many other complex and multifaceted questions about topics such as employment, income, household composition, etc. The systems that collect and house this information vary in their age, technical capacity, and purpose. Almost uniformly, they are complex and expensive to change, both in terms of data collection but also data transfer, reporting, and sharing. Ideally, states can access new dedicated funding or receive clear guidance on how existing federal money can be used to make necessary IT and systems changes.

The funding associated with this effort should also be flexible. It should support direct costs like system modifications but also priorities such as stakeholder engagement, instructional language development, and training—all of which are critical to successfully implementing the revised data collection guidance.

Urgency of Additional Guidance

While we are encouraged to see revised standards related to race and ethnicity data collection, we want to stress the urgency of revised guidance related to data collection on sexual orientation, gender identity, primary language, and disability. The current standards on sex are woefully inadequate, to the point of causing harm to people who identify as belonging to a sexual or gender minority population. Standards around disability and language are similarly lacking. Given the intersectionality of these concepts as they relate to identity, it’s important to make progress on updating standards for all areas, not just one. In addition, if states are to make significant investments to modify application wording as well as data systems and reporting structures to improve the collation of race and ethnicity, it makes sense to do this with the knowledge and understanding of timing for future changes regarding sexual orientation, gender identity, primary language, and disability. At the best, these timelines could be aligned, and at the least, states should have visibility to when those revisions are likely to be released.