Dear Alfonso



I apologize in advance, the background is long and I hope you would manage to understand what I tried to convey...

question: How to model the regression that was specified by the contrast [1 0 0 0 0 0 0], using an ancova model?
Does [1 0 0 0 0 0 0; -1 0 0 0 0 0 0] actually do a regression? (it gives F values and not T)



background: My model consists of 3 groups (group1, group2, group3), one covariate of interest (emotion) and 4 covariates of no-interest (age, gender, meds1, meds2 ). I wanted to check which was more predictive of brain connectivity: group or emotion (after controlling for the 4 covariates of no-interest).

For that purpose, I originally performed two analyses to identify the roi-pairs that were significant for each model: one was an ancova comparing the 3 groups while controlling for emotion+ the 4 covariates of no-interest [1 0 -1 0 0 0 0 0; -1 1 0 0 0 0 0 0]. The second was a regression, controlling for group + the 4 covariates of no-interest [1 0 0 0 0 0]. I replicated my results in spss and thus far everything was great.

I then learned that I should use only one of the methods (i.e. regression or anova including all of my variables), because they are statistically the same (and indeed I got comparable results using the two aforementioned contrasts). So far so good. However,I still need to be able to identify which pairs are significant for emotion and which for group. I tried to go back and do an ancova instead of the regression (which should give the same results). I was able to get the same results as the regression gave using [1 0 0 0 0 0 0; -1 0 0 0 0 0 0], but I'm not sure that actually constitutes an ancova (?).

The only other way I can think of, is by splitting the emotion covariate into 3 covariates (i.e. emotion group1, emotion group2, emotion group3; using zeros to replace values of subjects not included in the covariate), and using [1 0 -1; -1 1 0] to contrast them. The whole contrast would than use 0 0 0 to control for the 3 groups (group1 group2, group3; coded as before) and 0 0 0 0 to control for the 4 covariates of no-interest. The entire contrast would read [0 0 0 0 0 0 0 1 0 -1; 0 0 0 0 0 0 0 -1 1 0]. However, that didn't give me the expected results. O also tried to control for group using a single covariate (coded with 1,2,3), but that didn't solve it either. Any suggestion you may have as to how to solve this would be greatly appreciated!!


Sorry again for the long question, I hope it made sense...
Thank you very much!!

Liron.