RE: Stephen, I agree with your reasoning. I was initially thinking along
the lines of the analysis done in Stonnington et al, NeuroImage 2008,
which obviously isn't a single group.
RE: Patrick, you can see this article for the reasoning behind a
factorial design in that situation (with 2 groups and 2 magnets).
Patrick Floyd wrote:
>
> Thank you for your feedback. As it turns out, we are also collecting a
> patient group on both magnets, and although patient collection on the
> second magnet is still ongoing, we would like to analyze those data
> together with the healthies eventually. Also, we do (or will) have
> balanced samples on both magnets.
>
> It sounds to me that you are suggesting an ANOVA/factorial design
> rather than a regression model (as did the other response). I'm
> unclear why this would be more appropriate than regressing out any
> between-magnet effects explicitly as a covariate? I can appreciate the
> factorial design would allow to test for any groupXmagnet effects, but
> assuming the groups themselves are the same across magnets, wouldn't
> splitting them up into different cells in a factorial design be
> treating them as independent subgroups rather than as part of the same
> patient or healthy group?
>
> Thanks again,
> Pat F
>
>
>
> ------------------------------------------------------------------------
> *From:* Stephen J. Fromm <[log in to unmask]>
> *To:* [log in to unmask]
> *Sent:* Wed, December 1, 2010 1:19:56 PM
> *Subject:* Re: [SPM] one group, two magnets
>
> With all due respect, I have to differ.
>
> There's a problem using two magnets when there's e.g. two groups,
> because any purported effect of group _could_ be confounded with the
> effect of magnet. In that case, my impression is that the appropriate
> remedy is to test the groupXmagnet interaction and show that it's not
> significant at a very liberal p level (at least in the areas of interest).
>
> Here, the OP states that he's looking at one group only. While the
> effect of magnet in this case isn't explicitly accounted for, in terms
> of means alone (not looking at df), it just means "here's the mean
> effect, where this fraction of the sample was collected on magnet 1,
> and this other fraction was collected on magnet 2." The lack of
> explicit accounting concerns what the proportion is.
>
> In your solution (which is correct apart from my concern here), the
> two subgroups are weighted equally. While one could argue that this
> is in some sense better, it's not clear to me why it really matters as
> far as inferences of scientific (biological) interest are concerned.
> Obviously which magnet used has implications for results (e.g.,
> raising field strength is usually thought to improve the ability of
> the instrument to detect results of interest), but usually not in ways
> which affect _validity_ of results (unlike the case of two groups).
> Of course, there are exceptions to this---in one dataset I worked on,
> the experiment was unfortunately not counterbalanced, and there was a
> peculiar artifact generated by that particular magnet and pulse
> sequence that led to an interaction with the presentation
> order...which generated faux results in the area of interest! But
> such concerns about validity even apply when using only one magnet.
>
> Finally, to elicit the effect of magnet itself, the OP could use the
> [1 -1 ...] contrast representing the differences between the subgroups.
>
> Cheers,
>
> S
>
|