Monday 13 December 1999 - 4pm, Room 122, Pearson Building
Department of Statistical Science - University College London
Speaker: Panagiota Tzamourani (National Centre for Social Research)
joint work with Martin Knott (London School of Economics)
Semiparametric estimation of the one-factor latent trait model for binary data
In our talk we will discuss the one-factor latent trait model for binary
items.
The estimation of latent trait models requires an assumption about the
distribution of the latent factor. The usual assumption made is that the
latent factor is normally distributed. The estimation of the model is
done using an EM algorithm, and the latent distribution is approximated
by a set of quadrature points and weights. To avoid assuming a rigid form
for the latent distribution the EM algorithm is modified so that either the
weights are estimated over a fixed grid of points, or that both the points
and weights of the prior are estimated together with the item parameters.
To assess the information that is available in the data about the latent
distribution we measure its variability using bootstrap procedures. Results
suggest that there is information about the prior in the data and the form
of the prior may differ from the N(0,1).
Useful Web sites:
The programme of our statistical seminars is available at:
http://www.ucl.ac.uk/Stats/research/journals.html
Map with instructions on how to get to our department:
http://www.ucl.ac.uk/Stats/map.html
----------------------------------------------------------------------------
---------------
K. Skouras, Lecturer in Statistics, Department of Statistical Science,
University College London, Gower Street, London WC1E 6BT, United Kingdom.
Tel: 0171-4193652 Fax: 0171-7383 4703
e-mail: [log in to unmask]
----------------------------------------------------------------------------
---------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|