[Forum SIS] Seminario prof. Fox, Univ. Twente, all'Università di Milano-Bicocca

[Franca Crippa]

Seminario internazionale:

Random Item Effects Modeling, prof. Jean-Paul Fox, University of Twente
28 e 29 aprile 2011, Universit¨¤ di Milano-Bicocca, Dottorati in Statistica
e in Psicologia

per informazioni: prof.ssa Franca Crippa, Universit¨¤ di Milano-Bicocca;
franca.crippa a unimib.it

Theoretical frame: 28/4/2011, 10,30-12,30, Seminario del Dottorato in
Statistica, Sala Lauree della Facolt¨¤ di Statistica, U7-2019, 2¡ã piano Ed.
U7.
Applications: 29 aprile 2011, 10-13, Seminario della Scuola di Dottorato in
Psicologia, Sala Lauree della Facolt¨¤ di Psicologia, 2¡ã piano Ed. U6.
         
Abstract: Random Item Effects Modeling   
The multilevel item response modeling framework can be extended with random
item effects parameters. The random item parameters can be assumed to be
clustered within groups. This way, a hierarchical item structure can be
explicitly modeled, which includes an error component at the item level.   
A typical application of random item effects modeling is in cross©\national
survey research. In cross©\national survey research, measurement invariance
is required to make meaningful comparisons. The random item effects
structure can be used to allow random item variation across countries and to
obtain achievement estimates on a common scale. This makes it unnecessary to
define anchor items or to equate scores.   
The random item effects model is defined for the multilevel binary and/or
polytomous response data. An MCMC estimation method is used to estimate
simultaneously all parameters. A Bayesian testing procedure is proposed for
evaluating simultaneously multiple invariance hypotheses. Measurement
invariance hypotheses can be tested simultaneously through direct evaluation
of the variance components using the Bayes factor, highest posterior density
region test, or the deviance information criterion. Response data from
European Social Survey (ESS) are used to illustrate the modeling framework.
It will be shown that the MLIRT R routine can be used to estimate the model.


 

-------------- parte successiva --------------
Un allegato HTML ¨¨ stato rimosso...
URL: <http://www.stat.unipg.it/pipermail/sis/attachments/20110419/1e685636/attachment-0001.html>