[Eeglablist] LIMO toolbox - difference between clusters surviving correction with R^2 vs. F statistics

[PERNET Cyril]

Hi Dan

Yes the new version is almost ready addressing most of your issues - and will patch current version too before releasing.

1st level - do you let LIMO zscore that binary variable? If so then the beta reflects the presence of 2 conditions (-0.5 and 0.5) via a difference (equivalent to a contrast) on which you can indeed 2nd level regress whatever. If it is not zscored, as a binary 0/1 variable only one condition is modelled relative to the biased constant (because that constant now include all variance of the missing condition + the mean across conditions).

Note the single continuous file is all that is needed to encode conditions as well... At import each value is used as a condition.

Best
Cyril


--
Dr Cyril Pernet,
Senior Academic Fellow
Neuroimaging Sciences

Centre for Clinical Brain Sciences
Chancellor's Building, Room GU426D
The University of Edinburgh
49 Little France Crescent
Edinburgh BioQuarter EH16 4SB

cyril.pernet at ed.ac.uk
tel:  +44 (0)131 465 9530
https://urldefense.com/v3/__http://www.sbirc.ed.ac.uk/cyril__;!!Mih3wA!QLGAEGnZABwWynXbkOWoDFOCd-qnl2U3OnOboztnzmybFa3OzTh_TTC-W-CR-HJpytqpLw$ 
https://urldefense.com/v3/__http://www.ed.ac.uk/edinburgh-imaging__;!!Mih3wA!QLGAEGnZABwWynXbkOWoDFOCd-qnl2U3OnOboztnzmybFa3OzTh_TTC-W-CR-HJVuCevuQ$ 
<https://urldefense.com/v3/__http://www.ed.ac.uk/edinburgh-imaging__;!!Mih3wA!QLGAEGnZABwWynXbkOWoDFOCd-qnl2U3OnOboztnzmybFa3OzTh_TTC-W-CR-HJVuCevuQ$ >


________________________________
From: Dan Kleinman <kleinman at gmail.com>
Sent: 01 August 2020 16:31
To: PERNET Cyril <cyril.pernet at ed.ac.uk>
Cc: EEGLAB List <eeglablist at sccn.ucsd.edu>
Subject: Re: [Eeglablist] LIMO toolbox - difference between clusters surviving correction with R^2 vs. F statistics

Hi Cyril,

Thanks so much for your quick response.

1) I have uploaded the folder containing the 2nd-level analysis to OSF and will send you a private link to it via email.

2) I *think* I downloaded this version of LIMO (v2.0) from GitHub, but it’s been a while.

3) Since my condition only had 2 factor levels, I coded them as 0 and 1 and then told LIMO to treat it as a continuous variable at the 1st level of analysis. I realize this gives up some functionality as far as contrasts, but I often add other continuous variables at the first level – trial number, for instance – and it’s (very slightly) easier to generate a single continuous variable file for each subject rather than creating both categorical and continuous variable files for each subject.

Thanks,
Dan

—
Daniel Kleinman, Ph.D.
Postdoctoral Fellow
Haskins Laboratories


On Aug 1, 2020, at 3:51 AM, PERNET Cyril <cyril.pernet at ed.ac.uk<mailto:cyril.pernet at ed.ac.uk>> wrote:

Hi Dan,


  1.  With a single varialbe, the covariate effect is the R^2 --> should give the same results. Can you give me access at that 2nd level analysis please? So I can check what is going on.
  2.  is LIMO toolbox (v2.0) from GitHub or the eeglab plug-in (might not be the same)
  3.  I don’t understand your 1st level, if that is conditions, how did you coded this as a continuous variable? (might actually be ok, all is a single linear model anyway – but that’s a separate issue from clustering above)

Thx
Cyril


From: Makoto Miyakoshi <mmiyakoshi at ucsd.edu<mailto:mmiyakoshi at ucsd.edu>>
Sent: 31 July 2020 22:52
To: EEGLAB List <eeglablist at sccn.ucsd.edu<mailto:eeglablist at sccn.ucsd.edu>>; PERNET Cyril <cyril.pernet at ed.ac.uk<mailto:cyril.pernet at ed.ac.uk>>
Subject: Re: [Eeglablist] LIMO toolbox - difference between clusters surviving correction with R^2 vs. F statistics

Dear Cyril,

When you have time, could you please take a look for your comment?

Makoto

On Thu, Jul 30, 2020 at 12:42 PM Dan Kleinman <kleinman at gmail.com<mailto:kleinman at gmail.com>> wrote:
Hello All,

I have a question about results obtained using the LIMO toolbox (v2.0). Specifically, I performed an analysis in two different – but, I think, equivalent – ways, and obtained substantially different results depending on how it was conducted. I am wondering if others who have experience using (or programming) the toolbox could please shed light on why the results are different.

At the first level, I coded for (binary) trial condition using a single continuous variable (no categorical variables) for all participants. At the second level, I conducted a Regression analysis to identify spatiotemporal clusters at which a continuous between-subjects variable correlated significantly with the effect of condition. Importantly, *I only entered one between-subjects variable at this stage*.

There are (at least) two ways to view the results:
(1) Show clusters at which r^2 is significant (by selecting R2.mat; “Model fit")
(2) Show clusters at which F is significant (by selecting Covariate_effect_1.mat; “F test for a continuous regressor")

If I do not apply a correction (MC Correction=None), the cluster maps with uncorrected thresholds look identical (as I would expect with only one regressor). However, if I apply a correction (MC Correction=Clustering), method (1) yields a significant cluster but method (2) does not. This pattern holds true across a number of different datasets, in that method (1) often yields a significant result even when (2) does not and only a single regressor is used.

Is this expected behavior? If so, how should I interpret the r^2 cluster results vs. the F cluster results with one regressor?

Many thanks,
Dan Kleinman

—
Daniel Kleinman, Ph.D.
Postdoctoral Fellow
Haskins Laboratories
_______________________________________________
Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu<mailto:eeglablist-unsubscribe at sccn.ucsd.edu>
For digest mode, send an email with the subject "set digest mime" to eeglablist-request at sccn.ucsd.edu<mailto:eeglablist-request at sccn.ucsd.edu>
The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.