Holzinger (Psychometrika, vol. 9, no. 4, Dec. 1944) and Thurstone (Psychometrika, vol. 10, no. 2, June 1945; vol. 14, no. 1, March, 1949) discussed an alternative method for factoring a correlation matrix. The idea was to enter several clusters of items (tests) in the computer program beforehand, and then test them, optimize them and produce the residual matrix (which may show the necessity of further factoring). These clusters could stem from theoretical and substantive considerations, or from an inspection of the correlation matrix. It was an alternative to producing one factor at a time until the residual matrix becomes negligible, and was attractive because it spared much calculation time for the computers in that era. That reason soon lapsed but the method is still interesting as an alternative kind of confirmatory factor analysis.
My problem is: I would like to know the exact procedure (especially the one by Holzinger) but I cannot get hold of these three original publications (except the first two pages), unless against big expenses, nor can I find a thorough discussion of it in another publication, except perhaps in H.H. Harman (1976): Modern factor analysis, Section 11.5, but that book has disappeared from the university library, while on Google-books it is incomplete. Has anyone a copy of these publications, or is he/she familiar with this type of factor analysis?
I have got them! But if anyone is familiar with this method, I would still like to know his/her thoughts about it.
An old question (I'm now searching for the same thing, that's why I came across it), but if you're still interested in this topic, this might help: http://www.rug.nl/research/portal/publications/a-comparison-of-confirmatory-factor-analysis-methods(81618545-e619-4573-a903-7b5b6b63af4b).html
Thank you Thijs, but I know that dissertation very well. My interest in the topic started after reading it in 2008. However, the so-called OMG method that Ilse applied does not correspond to the multiple group method of factoring, as I learned when I studied the articles mentioned in the comment accompanying my question. That method still works with a residual correlation matrix. Ilse's OMG is, in fact, based on item analysis in line with classical test theory (as opposed to IRT). In my article about optimization of predicted clusters (PCO), as an alternative to confirmatory factor analysis, I explained all this. PCO is a recommendable alternative to CFA if one's correlation matrix does not obey the common factor model and one's investiation is still in a stage too exploratory to allow for precise prediction of the many residual correlations. This is often the case in studies of psychopathological phenomena (compare the arguments and findings network approach of Borsboom, Cramer, Fried, Epskamp et al).
The article on PCO: Article Testing Predicted Clusters: A New Approach to This Powerful ...
- Badges
- Science topic