Which Method of Factor Extraction is Preferable With Communality Greater than 1.0?

[rossj81]

Hi all

I am trying to perform Exploratory Factor Analysis using proc factor with priors=smc, but am not sure which factor extraction method to use. O’Rourke and Hatcher (2013) suggest that Maximum Likelihood (ML) is the preferred method, but when I use that, I receive the "Communality greater than 1.0" error. Khattree and Naik (2000:150) say we can use the HEYWOOD or ULTRAHEYWOOD options to get past this error, but caution of improper solutions as a result.

1. Should I use ML with HEYWOOD, or should I rather use Principal Factor (PRIN) instead?

2. If I use ML with HEYWOOD, how will I know whether the resultant solution is valid or improper?

I'm using SAS release 3.4 University Edition.

Ross

[Reeza]

It also indicates that some of the reasons for the error may be with the data. 

Is that possible in your scenario? If so, is there anything you can do with the data to resolve this?

[rossj81]

I'm not sure what would cause errors in the data. I ran proc factor with priors=smc and method=ml on two similar but different datasets. I received "Communality greater than 1.0" in both cases. When using HEYWOOD to force them, I received very different results on the two datasets.