LEMMA 3: Longitudinal Effects, Multilevel Modelling and Applications
Principal Investigator: Professor Fiona Steele
Social science is all about understanding complex social processes. For example, the processes by which people from families with differing socio-economic status obtain markedly different life outcomes are complex and develop over time. It has long been recognised that understanding such processes requires longitudinal data comprising repeated measurements of the key factors over time, and there has been substantial investment in the collection of such data.
The overarching objective of LEMMA 3 was to build capacity in the analysis of longitudinal data. LEMMA 3 aimed to
- review and synthesise important developments in longitudinal data analysis;
- develop and adapt new methodology that addresses important problems in social research today;
- apply the newly developed methods to substantive research projects in collaboration with experts from medical sociology, health psychology, economics, education and developmental psychology; and
- implement the methodological research in the user-friendly STAT-JR software environment. STAT-JR has been developed to overcome one of the biggest barriers facing social researchers, namely, learning to use statistical software packages.
LEMMA 3 ran an extensive programme of training and capacity building activities for academic and non-academic users. LEMMA 3 also provided support to other trainers who wish to develop their own courses in longitudinal data analysis. These activities are augmented by further developing online training materials and continuing to make available free software to UK-based academic users.
Impacts (March 2013)
- Multiple imputation methods for missing data in multilevel data structures have been extended to situations where the model of scientific interest is for multivariate responses of mixed type and where missing data may be on the responses and/or covariates. These methods have been implemented in the REALCOM-Impute software (which interoperates with MLwiN and Stata) and, with substantial gains in computational efficiency, in the new Stat-JR software system.
- Multiple imputation methods have also been applied for the first time in the area of administrative data linkage, which can be viewed as a case of partially known (missing) data where a set of candidate records from a secondary file is to be linked to a single record in a basic file of interest.
- The LEMMA online training course on multilevel modelling currently comprises over 1400 registered users from outside academia. There are almost 10000 registered users of this online resource.
- Since January 2012 the MLwiN software package which has been maintained and developed under LEMMA 3 has been purchased by over 70 users from outside academia.
LEMMA 3 was based at the Centre for Multilevel Modelling at the University of Bristol. For further information please see the LEMMA 3 website.