Initial American Housing Survey (AHS) estimates in 2017 and 2019 showed a sharp increase in single attached units compared to 2015 and earlier, and a decrease in housing units in multi-unit structures having fewer than 20 apartments. Census determined that the large differences may have been due to questionnaire changes and respondents’ difficulty distinguishing single attached units from multi-units. As a result, structure type responses were edited in the 2017 and 2019 data for units reporting as single attached but having characteristics of a multi-unit, and vice versa.  

In 2017, these changes were made based either on a set of criteria (rule-based) using qualitative variables or on the output of a model that assigned structure type. In 2019, a logistic regression model was used. These edits brought the national estimates and most metro-level estimates of single attached units and of multi-units closer to American Community Survey (ACS) estimates.