I think that the 'extra' degrees of freedom come from the fact that the
scans at the 'intermediate' levels also contribute to the estimation of the
error term. It is as if you were doing a planned comparison between two
levels of a four-level factor and used the pooled error term from all four
conditions, not just from the two conditions in the comparison/contrast.
Ian
At 00:39 15/05/2002 +0100, [log in to unmask] wrote:
>I am hoping that possibly one of the SPM authors -- or anyone else who
>understands the issues -- could comment on this earlier unanswered post
>(quoted below).
>
>On Mon, 6 May 2002 15:21:16 -0500, Kevin Black <[log in to unmask]> wrote:
>
> > ...
> >Dear SPM folks and other statisticians,
> > I have a quantitative PET blood flow study in which subjects received
> >one of 4 doses of a drug, 0, 1, 10, or 100mg/kg.
> > I entered and coded all data into SPM using 4 conditions to represent
> >the 4 drug doses. I also included an additive subject effect to account
> >for differences in response between subjects. I did not have a strong
> >a priori hypothesis about the shape of the response curve, especially
> >[at what dose] the maximal change would occur; I did expect any relevant
> >effects to be monotonic. Because of this, my primary statistical
> >comparison was of the scans done after the highest dose to the scans done
> >after the no-drug control. There were ~25 scans total at each dose.
> > When I asked for a contrast of high-dose versus control scans
> >(i.e. ~25 vs ~25 scans), SPM calculated statistics based on just under
> >100 degrees of freedom. I would have thought a d.f. of just under 50
> > would be more appropriate. Which is correct?
> >
> >Please cc: [log in to unmask]
> > Thanks,
> >Kevin J. Black, M.D.
Ian Nimmo-Smith
MRC Cognition and Brain Sciences Unit
15 Chaucer Road
Cambridge UK
CB2 2EF
Tel +44 (0) 1223 355294 x 710
Fax +44 (0) 1223 359062
|