[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: St: How to handle missing observations in the factor-principal component analysis

From   "Verkuilen, Jay" <[email protected]>
To   <[email protected]>
Subject   RE: st: St: How to handle missing observations in the factor-principal component analysis
Date   Wed, 19 Dec 2007 10:23:36 -0500

Maarten Buis wrote:

>>-ice- is explicitly designed to deal with exactly that problem, so yes
you can and should use -ice- in that case.<<

Several points:

You could also get Amelia II, which is free and generates multiply imputed Stata datasets. 

I want to echo the concern about mean imputation. This will do BAD things to factor analysis. Yet another reason for SPSS to move off to the dustbin of history....  

As to the ordinal factor analysis issue, I wasn't clear that the data weren't binary. If so, a simple method is to compute the tetrachoric correlations and factor-analyze those. This is not ideal, but given some of the limiations of Stata's exploratory factor analysis program (no standard errors for loadings, ability to work with the correlation matrix only), I suggest getting the free program CEFA, downloadable from Michael Browne is, unarguably, the expert on exploratory factor analysis. If CEFA doesn't do it, look long and hard.... 

If not, has at least one worked examples of ordinal factor analysis, as does Rabe-Hesketh and Skrondal's Stata Press book. You can also work out confirmatory linear factor analysis in gllamm. 

It's not difficult to fit the Rasch model using xtlogit, or to use raschtest. This model is a viable alternative for relatively short scales and probably what I'd recommend the original poster try. 

Alternatively, look up Mokken scaling. This is a non-parametric IRT model for binary and polytomous data. There are Stata commands that can be downloaded. 



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index