Hey AFNI crew!
I have a question best explained by a scenario. Imagine we have preformed a simple MID-task experiment. For simplicity we only focus on the feedback part: win_$5, win_$10, lose_$5 and lose_$10. In the future it would be interesting to look at contrasts between winnings/loosing different amounts (e.g. win_$10 vs win_$5) but for the moment let's imagine that we only have a couple subjects and want to make an early analysis to see if our design is working. I.e. we will only look at win vs loose.
These is (at least, I'm sure) two ways of doing this. Either you merge the 4 stimfiles into two simfiles by merging win_$5 with win_$10 and merge lose in the same way. With this approach we now have two regressors (win and lose). The downside with this is that you have to rerun the analysis later when you finally have a lot of subjects and for example want to investigate winning $10 vs winning $5. So, instead of merging, you can use all 4 stimfiles as 4 regressors in the analysis and then, in the gltsym, create the win vs lose contrast by adding the pairs:
-gltsym 'SYM: win_$10 + win_$5 - lose_$10 - lose_$5
-glt_label 1 win_vs_lose
So!
Question 1: Is it the same thing to combine two stimfiles into one stimfile as to use two seperate stimfiles and add their effects later in the analysis? In other words: Will the regressor coefficient for a big/merged regressor/stimfile be the same as the resulting coefficient you get from adding the coefficients of two smaller stimfiles/regressors? (I guess you loose two degrees of freedom when using 4 regressors instead of 2, but would not really impact a task based analysis that much?)
Question 2: The question above triggers another question. What is the difference between the effect coefficient in AFNI for a regressor and the actual Beta value? I have learned that they are not the same?
Thanks a bunch!
Edited 6 time(s). Last edit at 02/23/2016 10:35AM by Robin.