Short Biography

Irini Moustaki Irini Moustaki is a professor of Social Statistics at the London School of Economics and Political Science and was Head of the Statistics Department from January 2010 to December 2012. She has developed methodology for the analysis of categorical and mixed type data collected in Social and Health related Surveys. She is in expert in latent variable models and structural equation models that are widely used in Social Sciences and Educational Testing for measuring and connected unobserved constructs such as attitudes, health status, behaviours, intelligence, performance, etc.  Her methodological work includes treatment of missing data, longitudinal data, detection of outliers, goodness‐of‐fit tests and advanced estimation methods. Furthermore, she has made methodological and applied contributions in the areas of comparative cross‐national studies and epidemiological studies on rare diseases. 

In 2008 she co‐authored the book, Analysis of Multivariate Social Science Data with Bartholomew, Steele and Galbraith that provides a non‐mathematical treatment of advanced statistical techniques for social scientists and in 2011 she joined as a co‐author and published the third edition of the seminal book on Latent Variable Models and Factor Analysis: a unified Approach with Bartholomew and Knott. She has been the Editor in Chief of Psychometrika since November 2014. She was awarded an honorary doctorate in Statistics from the Faculty of Social Sciences, University of Uppsala in 2013.   

 

Course description

Latent variable models are a broad family of models that can be used to capture abstract concepts by means of multiple indicators. Social scientists know them best in the form of factor analysis and structural equation models, in which continuous latent variables are captured by means of continuous observed variables. However, social surveys and many other applications often yield observed variables that are categorical instead of continuous. In this case, the appropriate latent variable models are latent trait (or item response theory) models for continuous latent variables and latent class models for categorical latent variables.  

These methods can also be used to compare the distributions of latent variables between different groups. A common example is comparison of countries using data from cross‐national surveys. Before doing so, we should also assess the extent to which we have measured the same concept in the same way across groups. Ignoring this question means that we cannot be confident about making valid comparisons of like with like. This question of “measurement equivalence” or “differential item functioning” can also be examined within the models.   

This workshop aims to introduce participants to latent trait and latent class models (Days 1‐2), and to multiple group latent trait analysis (Day 3). It provides training in the use of the Mplus programme to carry out the analyses. 

 

Software

Mplus and R. Prior knowledge of either Mplus or R is not required.  

 

Prerequisites

Participants should be familiar with logistic regression modelling. Familiarity with factor analysis and structural equation modelling would be an advantage, but is not essential.  

 

References

Bartholomew, D.J, Steele, F., Moustaki, I. and Galbraith, J. (2008) Analysis of Multivariate Social Science Data (2nd ed.) Chapman and Hall/CRC.