need some clarification PLS v/s LISREL

[radhika]

Dear Sir/Maam,

I was using LISREL for quite a longtime but after attending and meeting PRof. Joe's lecture on smart PLS, started using this software, its easy and good. But I need one clarification I have my 565 datapoints from new research, when i first used PLS it asked me to validate with some data and i did it. It perfectly gave path model etc.,

when I tried several times using the same data in LISREL just to compare and calculate GFI , I did not get pathdiagram and model did not converge at the end of the result i got the message saying : co-variance matrix is not positive definite and give some starting points...I tried many ways ...fixing some variables etc., then also model did not converge.

My question is if the SPLS validates the data and says valid....even then is their a possibility of occurance of the covariance problem. Then what is the meaning of data valid in smart PLS.

Hope you try help me ...

Regards,

Radhika

[erigdon]

It would be surprising to learn that a dataset that was well-behaved for PLS produced an empirical covariance matrix that was not positive definite. Do make sure that the LISREL message was about the input covariance matrix and not about Sigma-hat, the model-implied covariance matrix, or some parameter covariance matrix. It would not surprise me to learn that a model specified for PLS could misbehave very badly in a factor-based SEM analysis. As you know, software error messages tend to be misleading and not especially helpful.