Bayesian Additive Regression Trees for Non-Ignorable Missing Data
Date: Friday, April 19th 2024
Time: 10:10am WET (11:10am CET)
Join the Webinar
- Webinar is free and open to everyone.
- To register for the event please click here.
Speaker
Yong Chen Goh, PhD student in Statistics, Maynooth University, Ireland
Yong Chen is a 4th year PhD candidate at Maynooth University. Her research focuses on tackling non-ignorable missing data problems in the context of Bayesian tree-based methods. She completed a BSc in Actuarial Science from City, University of London, followed by an MSc in Mathematics from Queen Mary, University of London.
Abstract
Dealing with missing data has been a pervasive challenge in data analysis. Despite numerous methods available to address this issue, many still result in biased or erroneous conclusions. This is particularly true when assumptions about the underlying process that generates missing data are oversimplified or unrealistic. In cases where the data are Missing Not at Random (MNAR), it is crucial to consider the process that causes the missingness by jointly modelling the data and missing data indicator. Through a selection model, we propose two methods for handling data with non-ignorable missingness in the response variables. The first approach uses Bayesian Additive Regression Trees (BART) to model the outcomes while using a probit regression model to account for missingness. In scenarios where the relationship between the missingness and the data is more complex or nonlinear, we propose a probit BART model to characterise the missing data generating process. Both models are presented in the univariate and multivariate framework.
Category: Events