 |
 |
See below. On Wed, Oct 21, 2015 at 2:33 PM, Mike <[log in to unmask]> wrote: > Hi SPM experts, > > This question has puzzled me for a while and I hope some experts could > help. > > In my experiment, I have one decision task (i.e. D). I want to identify > neural correlates related to D. A hypothesis suggest that if a brain region > is related to decision, its activity should covary with task difficulty. So > I designed 2 types of difficulty levels (each repeated for 20 times) and > collected data from 20 subjects. Behavioral data showed that subjects did > report significantly different difficulty ratings for the 2 types of task > levels. Below is the proposed three methods to isolate decision-related > regions: > > Method (1): Subject's first-level model contained 1 EV representing D > epoch. Each subject has an average difficulty score obtained from the mean > of the 2 levels. In group-level model, I perform a regression analysis > using each subject's average difficulty score as a covariate. > > Method (2): Subject's first-level model contained 2 EVs corresponding to > the 2 difficulty conditions of D epoch. In contrast manager, I used "-1 1" > to identify activity related to task difficulty in each subject. Finally, > in group-level model, one-sample t-test was performed. > > Method (3): Subject's first-level model contained 1 EV representing D > epoch. In the first-level model, I entered each trial's difficulty rating > in "parametric modulations" under "condition" (i.e., 40 numbers were > entered in this EV; 1st order modulation selected). In contrast manager, I > used "1", which I think would identify activity correlated with > trial-by-trial task difficulty. Finally, in group-level model, one-sample > t-test was performed. > > Question 1: Which method is correct or better? > It depends on your hypothesis. If you want to know how difficulty correlates with activity, then #1 or #3 would be better. If you want to know what areas are significantly different for easy and hard trials, then #2 would be better. I would probably avoid #3 as I'm not sure how accurate the subject will be for identifying the difficulty of individual trials. As you designed your experiment with 2 levels, I would probably select model #2. > > Question 2: If I did not design 2 or more levels of task difficulty (i.e., > with only one type of task), can I still identify neural correlates related > to decision, based on the hypothesis mentioned above? I mean, even though > the design has just one difficulty level, different difficulty ratings > could be reported across subjects. Then I use Methods (1) and (3) to get > the neural correlates. > Yes. You could do this. From the way you phrased this, it sounds like model #1 would be better suited for the question. > > Thanks. > Mike >