The tutorials will take place at the London School of Economics campus on Thursday the 9th of December 2010. The number of participants to the tutorials is limited and restricted only to those who attend the conference. The host is the Department of Statistics, London School of Economics, Houghton Street, London WC2A 2AE. View the location for the Tutorial Venue by downloading the map of the LSE area and campus. For further information please contact Dr Irini Moustaki.
The one-day tutorial will cover recent developments in latent variable modelling and item response theory. Latent variable models are widely used in Social Sciences and Medical Statistics for measuring unobserved constructs such as abilities, attitudes, stages of a disease etc. Various models for modelling categorical and continuous observed variables will be covered such as the multidimensional latent trait model, graded response model, model for nominal responses, the partial credit model as well as latent class models. Multi-group analysis and structural relations among latent variables will also be covered. For each model, issues of model specification, estimation and goodness-of-fit will be discussed. Specialized commercial software and R-routines will be used for illustrating the different models using real applications.
Prerequisites: linear regression model and some knowledge on categorical data analysis.
These tutorial lectures will review the theory and the practice
of some of the modern methods of statistical signal extraction, and
they will encompass some of the most recent developments. The
lectures will begin with a review of the well-established
Wiener--Kolmogorov theory of linear filtering, together with its
interpretations within the time domain and the frequency domain. The
theory will be extended to encompass short nonstationary
sequences. Various alternative algorithms for realising the filters
will be presented. The lectures will proceed to describe filtering
methods that exploit the concepts of the Fourier analysis. Such
methods presuppose that the de-trended data can be expressed as a
combination of trigonometrical or complex exponential functions. The
techniques that enable these methods to be applied to trended and
nonstationary data will be described. A variety of recently developed
filtering methods that continue to evade a full theoretical analysis
will also be described. These will include the methods of singular
spectral analysis and of empirical mode decompositions.
