I know this thread is a bit stale, but I wanted to mention (and get it
included into the thread) that betas are not simply the reciprocal of
each other if you swap an "IV" with the "DV". That is not the way that
the regression formulas work out.
On Mon, 2011-01-24 at 15:20 -0600, Darren Gitelman wrote:
> Dear DDW
> I am not surprised that while A to B is "significant", B to A might
> not be, because the variance may differ between the areas. However, I
> would be surprised if there was absolutely no result at A if you start
> with B as the source. You should see some activity at A starting with
> B as the source and the beta should be the reciprocal of starting with
> A as the source. (When selecting the voxels defining B as the source
> they should be the same voxels as the ones that came up as the target
> using A as the source.)
> Unfortunately you can neither use the difference in significance nor
> the values of the [reciprocal] betas to help you decide on the actual
> direction of the influence. You have to decide on the direction based
> on your concept or model of the brain's organization, and can then
> take the PPI result as a confirmation of an influence existing (and
> supporting your model) or not.
> On Mon, Jan 24, 2011 at 3:00 PM, D D W <[log in to unmask]> wrote:
> Dear Darren & list,
> > Exactly. You cannot use PPI to make an independent
> > statement about the absolute direction of the influence, but
> > shouldn't let that trouble you. This is why I brought up the
> > Grabenhorst and Rolls paper because they did a nice of
> showing how you
> > can pick a seed (source) region but have an inference about
> > directionality that is actually target to source (prefrontal
> > orbitofrontal) rather than the canonical source to target.
> As I said,
> > PPI allows an inference about directionality because you
> > specified the alternative model. Of course someone could
> disagree with
> > your model but that's ok.'
> I don't have much to contribute except to say that I am
> flummoxed by how to interpret such a finding. For instance,
> I'm sure I'm not alone in having observed a significant PPI
> from source to target that is not symmetric (i.e. the target
> to source PPI is non-significant). According to the line of
> reasoning you outlined above, I would still be allowed to
> argue, based on other evidence, that the direction of
> influence is from target to source.
> To be more specific, I've run PPI analyses on 4 apriori ROIs
> and found that ROI A has a significant PPI to ROI B, but not
> vice versa (all at 0.05, so it's not a threshold issue). It
> seems odd that I would still be allowed to conclude that the
> direction of influence is from B to A.
> I understand that the directionality of the regression
> doesn't buy you directionality of causal influence, but
> nevertheless it seems rather odd to me that you can infer a
> specific direction (based on a biological model) in the
> absence of PPI in that same direction.
> If I'm missing something glaringly obvious, apologies.
> Darren Gitelman, MD
> Northwestern University
> 710 N. Lake Shore Dr., 1122
> Chicago, IL 60611
> Ph: (312) 908-8614
> Fax: (312) 908-5073