Message/Author 

CMW posted on Monday, February 03, 2003  9:03 pm



Is it possible to compute standard errors for the factor loadings from an EFA with categorical indicators using one of the WLS estimators? Is, for example, Cudeck and O'Dell's 1994 work for ML extendable to this case? 

bmuthen posted on Monday, February 03, 2003  9:18 pm



You can do this by putting the EFA model into a CFA. That is, imposing the same number of restictions. I am not familiar with the article you mention. In principle one could derive s.e.'s for rotated solutions in line with the work by Jennrich in Psychometrika  I have not seen this attempted for the Mplus WLS. 


Hello: Is the test statistic produced from the Geomin rotated loadings a zstatistic? Or how large of a sample do you need for this test statistic to be normally distributed? Best, Tom 


Yes, it is a zstatistic. I would think it is approximately normal at n=30. You can do a Monte Carlo study to verify this. 


Thank you Linda! 

ywang posted on Monday, September 21, 2009  6:44 pm



Dear Drs. Muthen: I am working on an EFA model with one factor on three dummy indicator variables. The factor loading for one of the indicator variable is not significant ( Est./S.E. GEOMIN ROTATED LOADINGS is 1.629). Should I still use this indicator variable or should I replace it with some other indicator variables? Thanks in advance for your help! 


I would replace it if I could. I would recommend no fewer than three signficant factor indicators per factor. More than three is better. 

Back to top 