Hi SPHARM team,

Thank you very much for your excellent tool.

I'm aiming to run some correlations with covariates, however I'm running into some problems.

From my reading the Pearson & Spearman correlations only use uncorrected means, and this matches my trial runs where including covariates doesn't change the Pearson/Spearman parameter or significance maps. For example, both the command lines below give similar Pearson maps:
shapeAnalysisMANCOVA /path/to/inputfile.csv --interactionTest --simpleCorrs --numIndependent 3 --columnIndependent 1,2,3 --testColumn 1 
shapeAnalysisMANCOVA /path/to/inputfile.csv --interactionTest --simpleCorrs --numIndependent 1 --columnIndependent 1 --testColumn 1

What does change with the inclusion of covariates is the interaction test Raw P and FDR maps, however I'm not clear what exactly this interaction test is assessing as the significance maps don't overlap very well with areas of strong correlation on the Pearson R value maps.

Am I using the wrong command line options to use covariates for correlations, or is there something more fundamental to do with the interaction test/correlation difference I'm missing?

Thank you,
David.

Hi DJ,



I seem to have missed your message, I am sorry.

In my understanding, the changes in your correlation maps and associated p-value maps with inclusion of more or less covariates might depend in whether or not your covariates are related. If they are related between themselves the changes in the maps might not be very significant. Does that might sense?

On if I would change the wrong commands, I dont think so. They look good to me.


Hope that helps,
Beatriz

Agree with Bea. Also, the spearman/pearson correlation maps are independent of the covariates (that's why these are called "simple Correlations), where as the MANCOVA interaction tests are using those covariates.

Martin

Thanks Beatriz & Martin for these and your other recent responses.

Kind regards,
David.