|
Abstract:
|
In 2020, the U.S. Census Bureau conducted its decennial Census using new methodology for initial contact with the public. This new design provided two distinct contact strategies, Internet First and Internet Choice, based on data taken from the U.S. Census Bureau’s Planning Database and the Federal Communications Commission (FCC) semiannual Internet Access Services Reports. This evaluation of the contact-strategy definition focused on determining which internet access dataset is more appropriate for future Census tests and the 2030 Census, FCC or ACS 5-year internet access data. A cross-sectional causal inference approach was applied to retrospectively determine how the ACS 5-year internet access data would have performed in place of the FCC internet access data during the 2020 Census while also exploring alternate forms of the current contact-strategy definition. This required advanced statistical methodology for predicting 2020 Census internal response rates via ensemble modeling using a weighted combination of machine learning algorithms to maximize predictive power and minimize reducible error to investigate various formulations of the contact-strategy definition.
|
