Citizenship question effects on household survey response

Differential coverage across demographic groups in a census or survey can reduce the accuracy and representativeness of the resulting statistics. Researchers traditionally have used community-level measures to study response behavior and coverage, which can obscure patterns for small population groups. We illustrate this using household-level citizenship and immigration status. We construct household-level characteristics using administrative records for each address in a randomized control trial (RCT) survey that measured the effects of including a citizenship question on a decennial census questionnaire. Our results show that the self-response rate to the questionnaire without the citizenship question ranges from 70.4% in households with only U.S.-born non-Hispanic Whites to 27.5% in those with at least one likely undocumented person (a 42.9 percentage point gap). Including the citizenship question widens the gap by a statistically significant 2.4 percentage points. Compared to households with all U.S.-born non-Hispanic Whites, the household roster omission rate in households with at least one likely undocumented member is 6.0 times higher without the citizenship question and 10.4 times higher with the question. These patterns help explain why administrative record-based population data include more non-citizens than survey-based official statistics.