Training In Latent Variable Modelling Based on Archived Social Science Datasets in Northern Ireland
This project is supported by the Economic & Social Research Council (ESRC) through the Research Development Initiative (RDI). It aims to provide advanced quantitative training for the region of Northern Ireland through collaboration between the University of Ulster and the Queens University of Belfast. The main objectives are to:
1. Provide high quality training and support in advanced multivariate statistical modelling in a latent variable framework.
2. Provide a series of Master Classes addressing state of the art analytical techniques.
3. Develop and maintain a range of online learning and support resources.
4. Inform researchers and research students on the availability, access, and use of archived datasets.
5. Increase awareness of current developments in quantitative statistical analysis of social science data.
Traditionally social science disciplines have taught statistics in terms of a particular statistical models such as correlation, multiple regression, factor analysis or, analysis of variance (ANOVA). However, most of the commonly used statistical models can be formulated within a single, more general, linear framework. This 'latent variable modelling' framework provides a flexible approach to statistical analysis where models can be specifically tailored to meet the researcher's needs. The adoption of latent variable modelling has been rapid over the last 30 years and is now considered the method of choice in most social science disciplines. Despite the widespread adoption of such techniques there is unmet demand for relevant training from junior and senior researchers and research students.
Tremblay and Gardner (1996) documented the increasing use of latent variable modelling in psychological research, finding the percentage of articles employing such techniques had doubled between 1987 and 1994. Subsequent studies have demonstrated the continued increase in latent variable modelling in psychology (Nachtigall et al. 2003). This increased use in psychology has been mirrored in many other social science disciplines. Indeed, a journal devoted to the application of latent variable models, Structural Equation Modelling: A Multidisciplinary Journal, was launched in 1994 and is now the highest ranked journal in the Social Sciences/ Mathematical Methods category. In the light of this, it is increasingly likely that social scientists, who are consumers of research in both an academic research and a professional context, will encounter latent variable modelling in the relevant research literature.
In 2006 an ESRC commissioned report, Assessment of Needs for Training in Research Methods in the UK Social Science Community, identified an unmet demand for training in research methods and statistics: “The surveys of researchers indicated an increasing demand for training in quantitative methods, relative to qualitative methods, with increasing seniority” (p.75). This project aims to provide training in latent variable modelling based on a summer school and follow-up Master Class workshops for research students and researchers in Northern Ireland. The training programme is designed to provide all research students and researcher in Northern Ireland with the opportunity to develop their general skills in advanced quantitative analysis and attend workshops covering specialized analyses such as models for the analysis of change and latent class analysis.
The summer school will be a five day course. The first two days will provide an introduction to the general linear model and demonstrate how many traditional statistical tests can be formulated in this framework. Days 3 and 4 will extend this theme by showing the additional flexibility that can be achieved by working within a framework that can accommodate (1) observed and latent variables, (2) multiple group analysis, (3) formal tests of model fit, (4) appropriate estimators, and (5) ordinal and categorical variables. The final day of the summer school will cover how categorical latent variables can be modelled and how longitudinal data can be handled. The summer school will be delivered by means of lectures and practical sessions. Each day will conclude with a questions and answers session and there will be opportunities for the participants to discuss their own studies with the tutors.
A novel aspect of this programme is that all practical activities will be based on archived social science datasets. Data from both cross-sectional (e.g. Offending, Crime and Justice Survey, British Psychiatric Morbidity Survey, General Household Survey) and longitudinal studies (e.g. Longitudinal Study of Young People in England, Millennium Cohort Study) will be used. The use of archived social science datasets allows participants to formulate and test complex hypotheses based on real-life data. This provides the opportunity to deal with common problems in social science research such as non-normality and missing observations.