Bayesian models & missing data
Overview:
The goal of this workshop is to serve as an introduction to Bayesian models and tools for analyzing missing or partially observed data. Specifically, we will cover the different types of missing data that one can encounter when working on real problems and various approaches for analyzing the incomplete data under different assumptions.Ìý We will begin by considering problems where observations for some characteristics are completely missing in the original dataset. Then we will address Bayesian models for partially observed values, e.g. for censored or measurement-error contaminated values.
Participants will get access to several worked examples written in STAN, NIMBLE, and other R packages (e.g. mitools) that are often used in the analysis of data with missing values.Ìý
At the end of this workshop, you will be able to:
ÌýÌý - Understand the different kinds of assumptions one can choose from for missing data models with completely missing observations;
ÌýÌý - Understand the importance of incorporating appropriate uncertainty in any analysis where there are missing values;
ÌýÌýÌý- Identify different patterns of missing values for multivariate datasets and how they affect analyses;
ÌýÌýÌý- Identify different ways that data can be partially observed and the choices of assumptions for how that occurs;
ÌýÌýÌý- Fit Bayesian models in STAN and NIMBLE to data with missing values or Bayesian inference in commonly used models.
Pre-requisites:
ÌýÌý - An undergraduate/graduate introduction to probability;
ÌýÌý - Knowledge of R;
ÌýÌý - An introduction to Bayesian statistics and methods.
Date: Friday, 5 May 2023.
Time: 10 a.m. to 12 p.m.
Location: hybrid (in-person at Burnside Hall 1104, and online via Zoom).
Instructor: Prof. Russell Steele, Dept. of Mathematics and Statistics, McGill.
Ìý
