Using Multilevel Regression with Post-Stratification to Improve Abortion Incidence Estimation

Marija Pejcinovska , University of Toronto
Monica Alexander, University of Toronto

Population-level estimates of the incidence of induced abortion are important for the improvement of maternal health and access to family planning. Yet reliable estimates are lacking in many countries, especially in ones with restrictive abortion laws or where abortions are highly stigmatized. Innovative direct and indirect methods based on survey responses, such as the List Experiment, the Confidante Method, the Network Scale-Up Method and so on, have been proposed in attempts to improve the estimation of the incidence of induced abortion. While many of these approaches have been discussed and contrasted in the literature, a little less attention has been paid to formally assessing the accuracy and validity of the estimates they produce. In this work we propose using a survey technique known as multilevel modelling with poststratification (MRP) to assess and adjust for various survey biases, including non-representativeness in both survey respondents and their reported networks. We evaluate our approach on abortion-related data from the 2018 Performance Monitoring for Action (PMA) survey in Uganda.

See extended abstract

 Presented in Session P1. Postercafe