On Mon, Nov 4, 2013 at 3:13 PM, bettyann BA <[log in to unmask]> wrote: > Don, thanks for your reply. > > Yes, my paradigm does have a third task C. Why would this third task > knock out Option 1 as a possible method? > It doesn't necessarily preclude option 1, but makes option 1 more challenging. As stated on the website, you also need several other regressors if you have multiple tasks. As described in my paper, under modelling the data can lead to false positive and false negatives. I've tried to come up with the regressors to properly model 3 conditions, but haven't found the solution yet. I'm sure there are others who could provide the solution. > > I was concerned about Option 2 because the "variance shared between your > regressors A*phys and B*phys will NOT be captured by either regressor". > Yes. This is a potential concern; however, the correlation between A*phys and B*phys is usually low. The more common problem is the collinearity of A*phys and A. This can occur with limited number of events per run. > I misspoke when I used the term 'correlation'. What I meant was that I > would run a regression/correlation to get the slope of the best-fit line. > > For didactic reasons, let's assume I use Option 1 after all. Do I > understand you correctly when I conclude that, yes, I can plot activity of > Region 1 v activity of Region 2 and look at the slope of the best-fit line. > If this single slope is statistically significantly different from zero, > then I can conclude that there is a PPI effect? > The real problem with plotting the data is that you don't know which timepoints are associated with which condition, so its hard to get the true slope of each effect. The plot is also not a good representation of the model. The closest plot would be to residualize A*psych and B*psych using all the other regressors for region1 and do the same for region 2. Then draw two lines. Then compute the difference in slope between the two lines. To reiterate, I would not use plots to determine the PPI effect, I'd just build the models. > > Thanks for your insight. I greatly appreciate it. > > Best, > - BettyAnn >