Hi Wiebke, Boris, List,

well, the parametric modulator allows you to scale the amplitude per hemodynamic response with some factor you acquired for a response (eg RT, subject judgements, stimulus properties, whatever). This would ideally follow some quantatative model or expectation you have of such effects in the brain. Perhaps it makes more sense to dub it a polynomial expansion than a taylor series in the 0th order, sorry for the confusion.

As these scaling is modeled in an extra regressor with responses scaled with your factor per trial (with HRF amplitudes cenetered around 0), the beta represents the amount of scaling, which could be in units of milliseconds when using RT, or some other scale you want to apply. It does not make sense to compare the effect of this scaling factor, expressed in BOLD/ms in case of RT, to compare that to absolute signal amplitude.

So your GLM model would be like this:

X_n * B_n + X_p * B_p=Y+e where X_n is your normal regressor and B_n normal parameter estimate, and X_p for the extra parametrically modulated regressor and parameter estimage B_p. Y is your time series data and e your error term.

This is basically a common first order polynomial expansion for some factor, nothing more.

List: please correct me if im wrong, this is what I assume SPM does, and it corresponds to what I see in my results.

Therefore directly contrasting B_n with B_p would be the same as subtracting your cars speed from your milage traveled, eg apples and oranges (different units).

You can inspect your normal and parametric regressors in SPM5 when you want to visualize this.

I used parametric modulation in several papers. In this one I explained the method in some detail and a graph (fig 3) because not all reviewers appeared to be familiar with series expansions in modeling temporal signal changes at that time:

http://www.ncbi.nlm.nih.gov/pubmed/16473025

I did not invent parametric modulation (unfortunately ;-)  ), I do cite a few papers in the above manuscript that come closest to describing the thing first. You might also want to do the example data sets fropm Rik Henson provided at the SPM website, they also include parametric analyses.

Good luck,

Bas

--------------------------------------------------
Dr. S.F.W. Neggers
Division of Brain Research
Rudolf Magnus Institute for Neuroscience Utrecht University Medical Center Visiting : Heidelberglaan 100, 3584 CX Utrecht
          Room B.01.1.03
Mail    : Huispost B.01.206, P.O. Box 85500
          3508 GA Utrecht, the Netherlands
Tel      : +31 (0)88 7559609
Fax      : +31 (0)88 7555443
E-mail  : [log in to unmask]
--------------------------------------------------


-----Oorspronkelijk bericht-----
Van: Wiebke Johanna Trost [mailto:[log in to unmask]]
Verzonden: woensdag 15 april 2009 15:52
Aan: Neggers, S.F.W.
CC: [log in to unmask]
Onderwerp: Re: [SPM] Regressor Covariate or parametric modulate on first level analysis

Hi Bas,

I'm contacting you as I have a rather similar problem to Boris'.
Could you recommend perhaps some material explaining more in detail how spm handels parametic modulation?
I did not find any tutorials giving more information and it is still not clear to me why the beta of a regressor is not comparable to the beta of its parametric modulator.
As I understood it a Taylor approximation could be used for modeling changes of a variable in time, but I can't see the application for a normal parametric modulation.
Secondly, how is it possible to put a parametric modulator before the regressor that should be modeled?

Any help would be very much appreciated.

All the best,
Wiebke


Neggers, S.F.W. wrote:
> Hi Boris,
>
> no, that's not correct, contrasting different parts of a taylor polynomial expansion (what parametric modulation really is) does not make anyt sense, the 0th and 1st order components model different things (the constant term and the expansion by RT). Contrasting the parameter estimates for a normal and parametrically modulated regressor would therefore make as much sense as subtracting the speed of your car from the milage traveled.
>
> You already 'clean up' your data for RT-related amplitude modulation of your HRF by simply including the parametrically modulated regressor in your GLM in the first place. This is true for all GLM statistics in general, not only SPM.
>
> So when ' cleaning up' is what you want, use [1 0].
>
> Cheers,
>
> Bas
>
>
> --------------------------------------------------
> Dr. S.F.W. Neggers
> Division of Brain Research
> Rudolf Magnus Institute for Neuroscience Utrecht University Medical 
> Center Visiting : Heidelberglaan 100, 3584 CX Utrecht
>           Room B.01.1.03
> Mail    : Huispost B.01.206, P.O. Box 85500
>           3508 GA Utrecht, the Netherlands
> Tel      : +31 (0)88 7559609
> Fax      : +31 (0)88 7555443
> E-mail  : [log in to unmask]
> --------------------------------------------------
>
>
> -----Oorspronkelijk bericht-----
> Van: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] 
> Namens boris suchan
> Verzonden: woensdag 15 april 2009 10:16
> Aan: [log in to unmask]
> Onderwerp: [SPM] Regressor Covariate or parametric modulate on first 
> level analysis
>
> Hi all,
> we would like to model out some variables (like RT)  at the first level.
> If I would use RT as a regressor I must have the same number of variabels like I have scans.
> But I have as many variables as conditions.
>
> If I would use additionally parametric modulation and than set up a contrast between the "normal" regressor and the parametric modulated regressor (1 -1) and analyse the con image at the 2nd level. Does that mean that I have the activation "cleaned " by my additional variable?
> Or are ther anay other ways to do this?
> Many thanks in advance
> bs
>
> --
> Priv.- Doz. Dr. Boris Suchan
> Institute of Cognitive Neuroscience
> Department of Neuropsychology
> GAFO 05/613
> Universitätsstrasse 150
> 44780 Bochum
> phone: + 49 234 3227575
> mailto:[log in to unmask]
>
>