The Demographic Research area of the Center for Economic Studies (CES) within the U.S. Census Bureau is responsible for researching and developing innovative ways to use administrative records in decennial census and survey operations. Our team of demographers, economists, geographers, and sociologists evaluate a wide array of administrative data from other federal agencies, state governments, and third party organizations. We assess the quality and coverage of these datasets and investigate how they may be useful for the Census Bureau’s data collection and processing efforts.  In addition, we use linked census, survey, and administrative records data to conduct scholarly research and to create estimates that could not be created without linked data to better inform the American Public.

Much of the work we do hinges on the ability to link records for people across different data sources. We are able to do this because another area at the Census Bureau first uses personally identifiable information (PII), such as name, date of birth, etc., to assign anonymous unique identifiers to individuals in our census, survey, and administrative data sets. They then strip off all PII and provide an anonymized file that includes these unique identifiers to researchers like myself to investigate important research questions.

One type of analysis we often perform involves linking survey data to administrative records to see if responses to survey questions match what we find in administrative records for people found in both data sources. For example, my colleagues and I linked responses from the Current Population Survey (CPS) on Medicare coverage to Medicare enrollment data and measured the extent of survey misreporting of Medicare coverage.  In this study, we found that survey responses were mostly consistent with enrollment data but we did note a small undercount of Medicare coverage in the CPS. In another case, we linked responses by American Indians and Alaska Natives regarding Indian Health Service (IHS) coverage in the American Community Survey (ACS) to IHS Patient Registration data. With this study, we found much higher levels of discordance between survey responses and administrative records. While some of the differences we found were likely due in part to definitional differences between the data sources, our analysis also suggested true inconsistencies in reporting of Indian Health Service coverage.

We also use linked data to understand how people’s responses to decennial census and survey questions change over time. For example, we have examined responses to census and survey questions on race and Hispanic origin. In one study on American Indians and Alaska Natives, we found considerable changes in racial responses between the 2000 and 2010 censuses, and by linking individuals to their responses in ACS data we were able to evaluate the characteristics of those who changed their race and those who did not.  In another project we evaluated how people reported their Hispanic origin in the 2000 and 2010 censuses and the ACS and examined the characteristics associated with a change in response, including the impact of changes in question wording and other data collection aspects.

My current work uses linked survey and administrative records data to increase our understanding of participation in social safety net programs.  This work is part of a joint project between the Census Bureau and the U.S. Department of Agriculture’s Economic Research Service and Food and Nutrition Service, as well as multiple state partners.  States that participate in the project send us data on people that receive Supplemental Nutrition Assistance Program (SNAP), Women, Infants, and Children (WIC), as well as data on Temporary Assistance for Needy Families (TANF) benefits. We link these records to ACS data to estimate eligibility and participation in each of these programs by various demographic, socioeconomic, and household characteristics.  We send our estimates of eligibility and participation back to the states with the aim of providing data that can inform program administration. For example, if we find that a particular characteristic or geographic area is associated with high rates of eligibility for a particular program but low rates of participation, it may indicate the need for further outreach.

The team I work with recently developed a visualization displaying these estimates for a few states. The visualization allows users to examine WIC eligibility and participation rates among infants and children at the county level by various characteristics. We are currently developing a similar visualization for SNAP recipients, which will include both children and adults. This is an example of the estimates we can produce with blended data that provide the public with additional information.

Renuka Bhaskar is an OSU alumna and a senior researcher in the Center for Economic Studies at the U.S. Census Bureau. Any opinions and conclusions expressed herein are those of the author and do not represent the views of the U.S. Census Bureau.