PCA in JMP 17

[#1]

In JMP 16 while doing Principal Component Analysis use some estimation methods only but in JMP 17 pro some new options are introduced like Standardize & method Family. If we have normal data [ No missing values & number of columns < 500] should we go for default selections?

[#2 · Accepted solution]

From what I can see, these are not so much new options, rather a reorganisation in the options that were previously available.

In version 16 of JMP after performing the PCA under the red triangle you could select:

Principle Components> 

     on Correlations

     on Covariances

     on Unscaled

The default was 'on correlations' and that makes sense in most circumstances.

In version 17 these 3 options are presented as Standardize options in the PCA dialog:

Standardize

    Standardized

    Unscaled

    Unscaled and uncentered

In fact version 17 still keeps the red triangle option.  So for example if you change the Standardized option to Unscaled then in the PCA you will see that the 'on Covariances" option has been selected.

As for the Method Family: in version within the PCA dialog there was an Estimation Option:

- Default

- REML

- ML

- Robust

- Row-wise

- Pairwise

- Wide 

- Sparse

In version 17 these accessed via the Method Family option.  This shows:

- Default

- Narrow data

- Wide data

But when you select Narrow Data you then see Variance Estimation methods:

- REML

- ML

- Robust

- Row-wise

- Pairwise

and if select Wide Data you see different Variance Estimation methods:

- Fast approximate

- Robust PCA

This has probably been done to allow more sub-options in relation to handling wide-data.  These options are new to version 17.