t.jombart at imperial.ac.uk
Fri Jul 8 10:27:01 CEST 2011
that's basically the idea, yes. The global/local tests are not great. They have good type 1 error but lack power. If one is significant, then definitely you can look at global/local components of sPCA, but that won't tell you how many to retain. If it is not significant, then... it is not really helping, as you don't know if this is lack of power or just absence of structure.
From: Alastair Potts [potts.a at gmail.com]
Sent: 07 July 2011 18:50
To: Jombart, Thibaut; adegenet-forum at r-forge.wu-wien.ac.at
Subject: Re: sPCA
Ah, yes. I see that now. I was using my original object (prior to any
transformation). I was confused because it also included the $lw of the
spca object, but I see now that is just the sample connection network.
So, in that case I take it that the global.rtest only really helps with
a binary decision - is there global structure or not? However, this does
not help one selecting the number of sPCA axes that may be informative
or not. Correct?
Thanks for your help,
On 2011/07/07 05:42 PM, Jombart, Thibaut wrote:
> Thanks for the feedback.
> You can't use the global test, or any test of spatial structure, on sPCA components, since they are already optimized for spatial structure.
> The global test is meant to be performed on the raw dataset as a whole, or at least any preceding transformation should not optimize spatial structure to avoid circularity.
> Dr Thibaut JOMBART
> MRC Centre for Outbreak Analysis and Modelling
> Department of Infectious Disease Epidemiology
> Imperial College - Faculty of Medicine
> St Mary’s Campus
> Norfolk Place
> London W2 1PG
> United Kingdom
> Tel. : 0044 (0)20 7594 3658
> t.jombart at imperial.ac.uk
> From: Alastair Potts [potts.a at gmail.com]
> Sent: 07 July 2011 16:35
> To: Jombart, Thibaut; adegenet-forum at r-forge.wu-wien.ac.at
> Subject: sPCA
> Hi Jombart,
> You've put together a really nice vignette for sPCA (just downloaded the
> most recent version). It really explains the whole thought process very
> A question: can the global test be used on an axis by axis basis? More
> specifically, can one use it to test the first set of axes independently
> to decide which axes to keep and which to drop? At the moment I have a
> significant global test, but I'm not too sure where to stop when it
> comes to dropping and retaining axes.
> Alastair Potts
More information about the adegenet-forum