Methods for Missing Data
Presenter: Dr Emma Eastoe
This course deals with the problem of missing data common in many social surveys; problems of bias and inefficiency of naive statistical methods; alternative procedures: basics and complications; MCAR, MAR and non-ignorable missing data; selection bias and the problem of dropout in panel studies. The course will also cover appropriate statistical analysis in appropriate software. The methods will be illustrated by case study analyses.
Cost
- Lancaster University staff and postgraduates - £50
- External staff member from an academic institution on a NODE course - £120
- External Postgraduate research student from an academic institution on a NODE course - £60
- External participant from a non academic institution on a NODE course - £440
The course fees include all supporting documentation, refreshments and lunches.
Topics
Particular topics will be:
- Assumptions for missing data methods;
- problems with conventional methods;
- Maximum Likelihood (ML) with missing data;
- ML with the EM algorithm; ML for contingency tables;
- multiple imputation (MI) for missing data;
- data augmentation;
- MI for the multivariate normal model;
- Markov Chain Monte Carlo (MCMC) approach;
- MI with SAS;
- MI with categorical and non-normal data;
- combining MI results;
- likelihood ratio tests;
- nonparametric methods;
- Bayesian statistics;
- bootstrap methods.
Learning
Students will learn through the application of concepts and techniques covered in the course to real data sets. Students will be encouraged to examine issues of substantive interest in these studies.
Successful students will be able to:
- understand the problems of missing data in social studies
- perform advanced statistical procedures
- apply theoretical concepts
- identify and solve problems
- analyse data and interpret statistical output
