The CPS citizenship question and survey refusals : causal and semi-causal evidence featuring a two-stage regression discontinuity design

The unsuccessful attempt to add a citizenship question to the Census has drawn attention to citizenship questions on other surveys. Simultaneously, researchers have noted a recent increase in Current Population Survey non-response. We combine these topics, studying the effect of the CPS citizenship question on refusals. We use the question's sudden introduction in 1994 as a natural experiment and obtain causal estimates via a regression discontinuity design (RDD). In January 1994, we find an immediate and sustained 20-50% jump in refusals. However, this cannot be attributed to the question alone, as numerous other survey characteristics were revised. We employ a two-stage RDD to relate state-specific refusal discontinuities to state characteristics. Discontinuity size is positively related to non-citizen and Hispanic populations, and a proxy for citizenship question item non-response. An 8% increase in refusals is potentially attributable to the question. Moreover, at the threshold, there is weak evidence of a discrete decrease in states' reported Hispanic populations. When non-citizenship is observable, state non-citizen population is positively related with refusals. These results imply the question makes non-citizens and Hispanics reluctant to take the survey. We recommend there be a trial to precisely estimate the question's effects, and decide if it merits continuation.