CAMO Software

Functional


I've run the same set of data in The Unscrambler« (PCA) and in another software. The results on variable loadings and sample scores were very different. Why?

The differences encountered are very often due to different variable scaling. In The Unscrambler«, the PCA and Regression dialogs show a "Weights" button in the X-variables and Y-variables sheets. By default, the weights are set to 1.0 (that is to say: no weighting), and if left such the model will be built on the covariance matrix. Note that this should only be done if the variables were all measured in the same unit and scale.

If you adjust the weights to "1/SDev" the variables will be standardized, and the model will be built on the correlation matrix.
Also note that in The Unscrambler«, the data are centered by default, this unless the "Center Data" option was un-checked. Some software packages may require a separate centering.


others...