Tuesday, November 23, 2004

MCAT and Factor Analysis

Multidimensional CAT is based on factor analysis theory, so it worth our talking here about them.

posted by Shunkai Fu @ 4:44 AM

5 Comments:

At 4:49 AM, Blogger Shunkai Fu said...

Factor analysis (continous latent and indicator variable) sis one kind of SEM, and there are some related or similar techniques, such as Latent Trait Analysis (continuous latent variables, discrete indicators), Latent Class Analysis (discrete latent variables, discrete indicators), and Latent Profile Analysis (discrete latent variables, continous indicators). For IRT, LTA is appropriate. However, I use LCA in my work about using Bayesian decision theory to measure examinees in a discrete categories (master, non-master).

 
At 4:50 AM, Blogger Shunkai Fu said...

Factor analysis is called Q-factor analysis, and we need distinguish it from Cluster Analysis. The following link may be useful: http://www.qmethod.org/Issues/cluster_vs_Q.htm

 
At 8:57 AM, Blogger Shunkai Fu said...

Cluster analysis is an exploratory approach to find potential clusters exist in the sample data. However, discriminant analysis is one kind of supervised way. From my point view, we can use any one of them to check the result of another one. For example, clusters gotten by using CA can be proved or evaluated by DA.

 
At 3:49 PM, Blogger Shunkai Fu said...

I have written a paper with my supervisor Prof. Michel Desmarais on CAT Based on Bayesian Decision Theory. We compared the performance with IRT and Bayesian Networks. The result shows that they are comparable, but Bayesian approach requires much less computation comparing with IRT.

 
At 5:10 PM, Blogger Shunkai Fu said...

The following site provide important information on alternatives to SEM/CFA. We can find that Latent Class Analysis is a potential tool for IRT.

http://www.nc.gsu.edu/~mkteer/relmeth.html

 

Post a Comment

<< Home