Contribution Plots in PCA - Discovering Why Points Are Outliers

d_barnett — Tue, 10 Jan 2017 10:51:06 GMT

When using PC Analysis in JMP I would find it very useful to know what contribution a single point or points have compared to others. In other programs ( SIMCA) this is available as a 'Contribution plot' and is quick and easy to use and interpret.

This analysis method doesn't appear in JMP and I wondered if anyone else has used it and can share it as an add-in. Finding out why a sample/ measurement is different to another can be very useful for discovering the reasons for the observation being an outlier.

example which shows the type of graphic I'm after in JMP from 'Principal component analysis' by Rasmus Bro a and Age K. Smilde ab

http://pubs.rsc.org/en/content/articlehtml/2014/ay/c3ay41907j

This shows how easy it should be to find out how/ why one point is different from another.

Any help with this would be most appreciated.

Reagrds

David

Re: Contribution Plots in PCA - Discovering Why Points Are Outliers

stephen_pearson — Thu, 26 Jan 2017 13:32:57 GMT

You can get some of the same functionality using two-way clustering in JMP (Put the same columns in the clustering platform, and then pick Two-way clustering from the hotspot. You may also want to turn on colour clusters and alter the number of groups). Rather than bars with length you are relying on a coloured 'bar code'.

topic Re: Contribution Plots in PCA - Discovering Why Points Are Outliers in Discussions

Contribution Plots in PCA - Discovering Why Points Are Outliers

Re: Contribution Plots in PCA - Discovering Why Points Are Outliers