it's perfectly fine to test interaction using a T, it's
just that if you have more than 2 levels to a factor, the
overall effect is an F. I think the usual way of doing
analysis of variance is to first look for an overall effects,
and if this one is significant, use post-hoc T tests.
(see an analysis of variance table ...this is especially
true if there are a lot of levels in a factor eg > 3)
This raises the problem of doing a lot of t tests without
any correction for that, so it's worth doing the F first ...
the very best
> Dear Jean Baptiste and SPMers
> Do you mean that you can not test for an interaction using T-contrast ?
> The only trouble could be in the interpretation of the result or is there some
deeper reasons ?
> | Jack Foucher Universite Louis Pasteur
> | Institut de Physique Biologique UPRES-A 7004 du CNRS |
> | 4 rue Kirschleger Tel: 33 (0)3 88 77 89 90
> | 67085 STRASBOURG Fax: 33 (0)3 88 37 14 97
> | France
> | Faster E-mail: [log in to unmask]
> | Other [log in to unmask]
> |_______________________________________________________________ |
> -----Message d'origine-----
> De: Jean-Baptiste Poline [SMTP:[log in to unmask]]
> Date: mardi 7 decembre 1999 18:06
> A: [log in to unmask]
> Objet: Re: advice on contrasts
> Just a small add on to Jesper email : the "global" mood
> by verbal fluency interaction contrast would be
> an F-contrast of the form
> -1 1 1 -1 0 0
> 0 0 -1 1 1 -1
> and the "global" effect of mood
> 1 1 -1 -1 0 0
> 0 0 1 1 -1 -1
> In general, an interaction of effect1 X effect2 where either
> effect1 or effect2 has more than 2 level will express itself
> as an F-test, as in classical analysis of variance.
> ------------- Begin Forwarded Message -------------
> Date: Tue, 07 Dec 1999 16:33:43 +0000
> Subject: Re: advice on contrasts
> From: Jesper Andersson <[log in to unmask]>
> To: "McBride, Alan" <[log in to unmask]>, "[log in to unmask]"
> <[log in to unmask]>
> X-Unsub: To leave, send text 'leave spm' to [log in to unmask]
> X-List-Unsubscribe: <mailto:[log in to unmask]>
> Dear Alan and David,
> >Dear Spm experts,
> >We have performed an fMRI expt to examine the effects of experimental mood
> >induction on verbal fluency activations on bipolar patients and normal
> >controls. Three blocks of scans are obtained, one in each mood state (
> >happy, sad, neutral)while subjects perform a verbal fluency task constisting
> >of epochs of generation repetition and rest.
> >We have specified two conditions: repetition and generation, and examined
> >contrasts -1 1 in all three mood conditions, to examine main effect of
> >verbal fluency.( ie. -1 1 -1 1 -1 1)
> Looks fine.
> > To examine the modulation of fluency activations by mood, we have compared
> >contrast -1 1 for scans obtained in the individual mood states (eg -1 1 0 0
> >0 0, if the first series of scans is obtained in Happy mood, and compared
> >the activations with those associated with contrast (0 0 -1 1 0 0) where the
> >second set of scans were obtained following sad mood induction.
> What you describe here seem to be some form of anecdotal description of the
> task_by_mood interaction. While this mode of analysis may be helpful in
> understanding the different activation patterns, you may want to use the
> proper interaction contrast for the inferential assesment. This, as you
> probably know, is affected by e.g. the contrast (-1 1 1 -1 0 0) which will
> show areas exhibiting a larger activation due to generation in happy mood
> than in sad mood.
> > To attempt to identify the main effect of mood, we have compared all scans
> >obtained in one mood state vs other mood states, eg ( 1 1 -1 -1 0 0). This
> >contrast does not identify any regional differences- we have used scaling to
> >remove global effects, and wonder whether this may have anything to do with
> >the problem
> I dont think global scaling effects this. I think your problem may be that
> the main effect of mood has a very low frequency content (i.e. it changes
> very slowly with time). Either this has prevented you from using a high
> pass filter (HPF) in which case your sensitivity may be a bit poor.
> Alternatively you may have (erroneously) used an HPF wich may then have
> removed the main effects of mood.
> I the above I have assumed that you have scanned all moods in the same
> "session" such that mood effects and session effects are orthogonal.
> >Could you advise us on how best to examine the main effect of mood in this
> I think I would need some additional information about your design (e.g.
> length of epochs of both verbal task and mood) and analysis (e.g. HPF cut
> off frequency) to be able to give any advice.
> >We would also be interested in performing a further analysis, constraining
> >the comparison of fluency activations across all mood states to areas
> >activated by verbal fluency following neutral mood induction, could you also
> >help with advice about how best to do this.
> I suspect that what you want to do here is to mask the ineraction contrast
> (e.g. [-1 1 1 -1 0 0]) with the main effect contrast (i.e. [-1 1 -1 1 0
> 0]). The resulting (uncorrected) p-level is simply the product of the
> p-level threshold of the mask, and the display p-level of the resulting
> masked (interaction) contrast. You are likely to be most sensitive if you
> use a high threshold (small p-value) for the mask and a more liberal
> criterion for the interaction contrast itself.
> >With thanks, David Barbenel, Alan McBride
> Good luck Jesper
> Jesper Andersson
> Wellcome Dept. of Cognitive Neurology
> 12 Queen Square
> London WC1N 3BG
> phone: 44 171 833 7484
> fax: 44 171 813 1420
> ------------- End Forwarded Message -------------