Abortion Reporting and Underreporting: Can Better Design Yield Better Data?

Abstract

Background Induced abortion is a ubiquitous yet elusive phenomenon. It is among the most common health experiences, yet despite investigators’ best efforts, we know relatively little about the specifics of its occurrence in most low-resource settings. In India, abortion is broadly legal but poorly measured. In this dissertation, we examined aspects of survey design and question methodology that could impact women’s willingness to report abortion on face-to-face surveys.

Data and Methods We used PMA2020 data from Rajasthan, India, which is a cross-sectional survey that is representative of women age 15 to 49. In Aim 1, we estimated the prevalence of induced abortion overall and for subgroups using list experiment and direct questions. In Aim 2, we assessed failures in abortion reporting using response time paradata from direct and list experiment questions. And in Aim 3 we sought to determine whether interviewer-respondent familiarity and respondent’s prior survey experience were associated with improved reporting of abortion via the direct abortion questions.

Results Despite having a large sample size of reproductive age women, the list experiment estimate of lifetime experience of abortion was actually significantly lower than the direct abortion estimate (1.8% versus 3.5%). Further investigation into the list experiment assumptions revealed evidence of violations. We did not find evidence that poor numeracy or poor cognitive ability (as measured by schooling) explains the list experiment’s poor performance. However, we did identify a significant editing effect whereby women who reported an abortion on the direct questions took 11.6 (95% CI 7.2-16.0) seconds longer to respond to the treatment list compared to women who reported no abortion on the direct questions. Regarding respondent acquaintance with the interviewer or the respondent’s prior participation in a PMA2020 survey, we find these aspects of familiarity were not statistically significantly associated with abortion reporting, adjusting for respondent, interviewer, and community characteristics.

Conclusions This dissertation provides a thorough investigation of abortion reporting and underreporting on a face-to-face survey in Rajasthan, India. It is also among the most in-depth studies of this phenomenon and the social and cognitive processes involved. Many of the specific analyses constitute the first investigations exploring these relationships with regard to abortion. As such, this dissertation contributes significantly to the literature on survey based abortion reporting. Many challenges remain in terms of our interest in identifying survey design features that would increase the validity of abortion reporting on face-to-face surveys.