Triple System Estimation of National Population Counts Through Log-linear and Latent Class Modeling

Oregon Convention Center 

The Continuous Count Study (CCS) is an important research effort within Census to generate lower level geographic and demographic estimates throughout the decade. One part of the study involves the use of administrative records to generate these estimates. The Demographic Frame is a comprehensive database of person-level data containing demographic characteristics and addresses associated with each person. Its information is derived from administrative, third-party, decennial census, and survey data sources. Prior modelling work has focused on predicting the probability that a given individual is found at a particular address on Census Day.

This research will examine a triple-system estimation of person counts where the three systems are: 2020 Census, 2020 American Community Survey (ACS), and a particular vintage of the Demographic Frame. We follow the ideas of Van der Heijden et al (2018, 2021), who estimated populations in New Zealand where individuals may be found on any number of lists, including being missing from all lists. We will be making use of the R package cvam (Schafer 2021) to fit the models, including a latent class model to assign demographic information.

Multiple Systems Estimation

Latent Class Modeling

Administrative Records 

Government Statistics Section