topic Re: Principal Component Analysis Results difference in New SAS User

Principal Component Analysis Results difference

Jonison — Tue, 02 Jul 2019 10:48:23 GMT

Hello, all, I am currently transferring a analysis tools from SIMCA to SAS, and when I did PCA analysis comparison, I found the scores calculated by two software are same value but with different sign (same data and pretreatment), for example, the score for pc1 from SIMCA is 1.55133 but from SAS enterprise miner v14 is -1.55133 (same thing for loading results). I dont think they are wrong results, but need to identical as the results are used by other modelling tools that may impact the prediction results.

any suggestions that could make the PCA calculation identical?

Many thanks.

Re: Principal Component Analysis Results difference

PaigeMiller — Tue, 02 Jul 2019 10:52:28 GMT

There is no unique answer to the sign of principal components scores. Both of the results are equivalent. As far as I know, you can't force SAS to match the sign of some other principal components algorithm.

If you have a model that used the SIMCA scores but now will be using the SAS scores, you could multiply the scores by -1, or change the way the model handles them.

Re: Principal Component Analysis Results difference

Jonison — Tue, 02 Jul 2019 11:40:03 GMT

Thank you for your help!