In 2022, the U.S. Census Bureau faced potential budget overruns for the American Community Survey (ACS) nonresponse follow-up operation, primarily due to rising wages and declining survey responses. To manage these costs, the bureau capped the monthly workload for Computer-Assisted Personal Interviewing (CAPI) at 60,000 cases and sometimes closed cases early. While these measures stabilized the budget, they were short-term solutions. The bureau sought a more data-driven method to optimize the CAPI workload without compromising data quality.

One key strategy involved adapting a vacancy prediction model used in the 2020 Census to identify likely vacant housing units during the ACS CAPI data collection. By limiting contact attempts on these units, interviewers could allocate more time to households more likely to respond, thereby improving the chances of obtaining complete interviews. This report summarizes the modeling approach from the 2020 Census and the subsequent research conducted to tailor it for ACS CAPI data collection.