Imputation research for the 2020 Census1

Abstract

Abstract. For the 2010 Census, the count imputation procedure filled in housing unit status and size for the small proportion of addresses (less than one-half percent) where this information was unknown. The small proportion was due in part to an extensive nonresponse followup (NRFU) field operation geared towards resolving addresses so that a status and count were known. For 2020, the Census Bureau is researching two changes to the NRFU field operation to reduce cost. The first is the possible use of administrative records (AR) to provide a status and count for some nonresponding addresses. The second is potentially reducing the number of visits made to nonresponding addresses. Although using AR will help resolve some of the remaining unresolved cases, the proportion of addresses in need of count imputation may be higher in 2020 due to the reduction in NRFU fieldwork. The 2010 count imputation model was developed assuming a small amount of missing data. This research looks at potential count imputation models to handle increased missingness. The paper also articulates the downstream characteristic imputation ramifications from the same missing data challenge.

A Modeling Approach for Administrative Record Enumeration in the Decennial Census

“…. SeeKeller (2016) for a comparison of direct use of AR and traditional imputation modeling for the cases that are unresolved at the conclusion of the NRFU operation.…”

A Distance Metric for Modeling the Quality of Administrative Records for Use in the 2020 U.S. Census

“…However, the approach can be implemented adaptively as new administrative record information is obtained during the NRFU operation. In the context of the 2015 Census Test, Keller (2016) describes multi-phased integration of administrative record modeling as an adaptive component throughout NRFU. For this test, after initial AR Occupied and AR Vacant cases were identified, the Census Bureau received additional IRS 1040 and IRS 1099 information.…”

“…The distance function methodology can also be used after data collection is complete, as an alternative to unit imputation of status and population size for unresolved housing units. In the context of the 2015 Census Test, Keller (2016) documents a modification of the optimization approach: refitting and determining AR Vacant and AR Occupied cases by lowering the average constraint values in the optimization approach -thus identifying more AR Vacant and AR Occupied cases. The new cases that remained unresolved addresses after the full visit strategy are assigned occupancy status and enumerated using administrative records rather that via an imputation.…”

Census 2020 Why Increasing Self-Response is Key to a Good Count

“…Unfortunately, though, there are serious issues regarding the completeness and accuracy of administrative records. For example, the Census Bureau found that about 20% of housing units determined to be vacant using administrative records in census tests conducted in 2016 in Los Angeles County, California, and Harris County, Texas, were actually occupied 3 …”