Dealing with Incomplete Data and Attrition in Longitudinal Studies

Session Number: 2503
Track: Quantitative Methods: Theory and Design
Session Type: Skill-Building Workshop
Tags: attrition, Impact evaluation, incomplete data, inverse probability weights, longitudinal evaluation, Multiple Imputation, power, quantitative methods
Session Facilitator: Geetha Nagarajan [Senior Evaluation Advisor - Social Impact]
First Author or Discussion Group Leader: Olga Rostapshova [Social Impact]
Second Author or Discussion Group Leader: Geetha Nagarajan [Senior Evaluation Advisor - Social Impact]
Third Author or Discussion Group Leader: Lana Basneen Zaman [Program Manager - Social Impact]
Time: Nov 10, 2017 (01:45 PM - 03:15 PM)
Room: PARK TWR STE 8206

Audience Level: Intermediate

Session Abstract (150 words): 

In quantitative surveys, especially longitudinal studies, incomplete data due to missing responses to survey questions and sample attrition presents a substantial challenge. Missing data can create problems including decreased power, bias in findings, and uncertainty on how to make the most of the data available. In this session, we explore 3 techniques evaluators may use during and post data collection to maximize evaluation power, even when data is incomplete: 1) replacements of attrited sample (during data collection), 2) inverse probability weighting, and 3) multiple imputation for missing data (post data collection). Presenters will provide an overview of each methodology with brief demonstrations. Drawing on examples from impact evaluations of development programs, they will describe how best to employ these methods the suit the context. Participants will then be invited to share examples of studies they would like to work on and discuss in groups what methodologies may be most effective.