Addressing Abortion Underreporting in Surveys with the List Experiment: Lifetime and Five-Year Abortion Incidence with Multivariate Estimation of Socio-demographic Associations in two U.S. States.

Abstract

Limited data are available on the characteristics and outcomes both of people who have and who have not had an abortion. Administrative data sources contain information on aggregate abortion counts and some demographic characteristics describing individuals who had an abortion but not on those who did not have an abortion. They are therefore of limited use for analyzing the characteristics, reproductive behaviors, and attitudes associated with abortion risk. Direct questions in population representative surveys yield downwardly biased estimates of abortion and likely differential underreporting of abortion by socio-demographic characteristics. In the present study, we evaluate the effectiveness of an indirect survey method, the list experiment, for improving estimates of abortion risk and differentials in population-representative surveys. We estimate cumulative-lifetime abortion incidence in 2017 and five-year incidence in 2021 using two cross-sectional surveys administered in Delaware and Maryland and evaluate the five-year estimates against external benchmarks from administrative data. We use multivariate regression with the list-experiment data to examine abortion incidence by socio-demographic predictors. We find that list-experiment estimates of five-year abortion incidence are similar to estimates calculated from external data: that cumulative lifetime abortion incidence increases monotonically with age, and that five-year incidence is inverse U-shaped. Black adults are found to be much more likely to have had an abortion both in the past five-years and over the reproductive lifetime, before and after controlling for age, parity, relationship status, education, and household income. We conclude positively about the validity and utility of the list experiment method.