Interacting/saving Multivariate Nonpar Density data

Question

Am new to JMP, and its great that so many of the graphing options include 'Nonpar Density'

When using "Analyze -> Multivariate methods -> Multivariate"

The 'Nonpar Density' can be added to the Scatterplot Matrix

I would like to interact with these density contours like the bivariate analysis, especially in order to save the 'density grid'. (I would use the bivariate analysis if the nonpar density could be calculated for multiple variables.  I gather the scatterplot matrix nonpar density is actually doing this!?)

Happy to read papers, watch tutorials etc. but have been at this for days  - could anyone point me in the right direction?

Many thanks,

Peter

pliddi · Answer

Use R, or another package - I think you are totally right here.

The scatterplot 3d looks like the only JMP package to be performing multivariate density analysis.  After my initial impressions and training with JMP - where I was feeling spectacularly impressed - I'm now totally bemused and beside myself that after so much amazing work the authors would include features merely for 'glitter' without permitting associated data and measurements to back it up.

Cheers,

Peter

ms · Answer

I thought you referred to Scatterplot matrix, not 3D scatterplot. I can't spot any difference between the densities displayed in bivariate and scatterplot matrix platforms. I am also not able to spot any changes in the contours if excluding variables or adding random variables in the scatterplot matrix. I am not convinced these nonparametric densities are multivariate or different than the ones produced by the bivariate platform (apart from the latter being more detailed by default). But I may be using the wrong data to be able to detect any differences.

Try R package ks. It lets you do kernel estimation for up to six dimensions. And it's free.

pliddi · Answer

Thanks MS, you had me excited there for a moment!

Upon comparing the density or quartile values, they were not computed considering multiple variables, only ones chosen in X and Y.

The 3D scatterplot pretty clearly calculates a multi variable nonpar density.  It is so frustrating that we can't actually use the data that is obviously produced.

The visualization, while great, is not sufficient for professional purposes - it needs to be quantified!  What we see needs measurements.  Its like a creating a graph without any numbers on either axis.

(For the serious $$ involved, does it seem half-baked to anyone else?)

Regards,

Peter