After running Exploratory Factor Analysis in spss v.17 I deleted 4 measurement items which yielded low factor loadings. Now I want to proceed to Confirmatory Factor Analysis to calculate AVE and CR. My Question is should I repeat the EFA to obtain the new factor loadings after deletion of those 4 variables(items) and conduct the CFA based on them or i should do it with the original factor loadings? Thnx 4 your help.
$\begingroup$
$\endgroup$
2
-
$\begingroup$ Of course you should redo factor analysis $\endgroup$– ttnphnsJul 9, 2012 at 9:21
-
1$\begingroup$ Any method of EFA (except the so-called alpha FA) treats the input variables as the population, not sample, of items on the research "field". Everything (loadings, communalities, restoring of correlations) change when you drop some variables. So, you should rerun and rerun EFA until you stop dropping variables and are satisfied with the factors you've got. Then you can proceed to CFA or cross-validation. $\endgroup$– ttnphnsJul 9, 2012 at 14:22
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
0
In my view, exploratory factor analysis is mainly use to identify the number of dimensions and describe a simple structure of interrelated measures. In addition, it is used to verify items cross-loadings (items loading on two or more factors), or unreliable indicators (low communality or a very low unique factor-specific loading, <.3 means less than 9% of explained variance). As the name suggests, it is exploratory in nature; you can repeat it with different subset of variables, or with some items singled out, to isolate a reliable factor structure.
Based on a pre-defined pattern matrix of item/factor loadings, a confirmatory factor model is mainly used to test hypotheses about your underlying construct (which also means you have to check assumptions for associated tests). This would obviously be a biased approach to test your factor structure on the same sample of individuals, and you will get over-optimistic measures of goodness of fit or Average Variance Extracted. In case you have a large sample, you can use some kind of internal cross-validation (e.g., split your sample using 50:50 and test factor structure on the holdout sample), but that does not guarantee the generalizability of your results beyond the sample at hand.
