Researchers
Professor Michael
G Kenward, Dr James Carpenter, London
School of Hygine and Tropical Medicine
Professor H. Goldstein,
Institute of Education
Professor G. Molenberghs,
Limburgs University
December 2002 - December 2004
Context
- Missing data is ubiquitous in social science research, especially
if data are longitudinal;
- This causes uncertainty in scientific inference, that goes beyond
familiar sampling variation;
- When data are missing, the sensitivity
of conclusions to assumptions about
the missing data should be examined.
Aims and Objectives
- Explore how analyses with missing data can be formalised and
used by social scientists using multi-level modelling;
- Compare and where necessary substantially develop the most appropriate
methods;
- Provide software for use by social scientists;
- Provide a web site for dissemination, discussion and training;
- Organise workshops.
Methodological aspects
- The reasons for missing data are the key to choosing an appropriate
analysis. However, these are invariably unknown;
- Therefore, methodology which transparently conveys the sensitivity
of analysis to the assumptions about the missing data is required;
- There is a particular need for this in multi-level
models which are widely used in social
science.
|
Sensitivity of treatment estimate (blue plain) to effect of rescue medication.
95% confidence plains in red. Null plain in black.
Research Design
- Review the three broad classes of approach to missing data emerging
in the recent literature;
- Understand how they relate to each other;
- Develop them to address research questions arising from Belgian
Health Interview Survey and Institute of Education's class size
project;
- Foster accessible discussion of recent methodological developments
using forthcoming web site;
- Promote good practice by making example
analyses available on the web.
Outputs
- Coming soon: www.missingdata.org.uk
, containing teaching, discussion and research;
- Articles, software and reviews posted on web site;
- Four workshops in 2004.
|
