Combining the list-experiment and direct question to improve estimation of abortion incidence

Abortion has been found to be severely underreported overall, and underreported differentially across groups, when using a direct question. The list-experiment method attempts to overcome these reporting biases indirectly by asking how many items an individual has experienced, but not which, where abortion is one of the items asked to a randomly assigned "treatment" group but not to a control group. Abortion incidence is estimated as the difference in the mean number of items reported between the treatment and control groups. If list-experiment respondents are also asked a direct abortion question, a combined-data estimator can be constructed from respondents with and without affirmative responses to the direct question. We assess for four US states how this combined estimator may improve estimation of cumulative lifetime abortion incidence relative to the direct question or the list experiment alone. Our combined-data estimate across the four states is 12.9% (95% CI, 10.5, 15.4), which is substantively and statistically higher than both the list-experiment estimate (11.0%, CI, 8.9, 13.2) and the direct-question estimate (9.6%, CI, 8.6, 10.5). Bias by state is much more variable for the direct question than for the list experiment. We conclude that the combined-data estimator improves estimation especially over the direct question.