Dear everyone, I was wondering if anyone would have any comments on this -- I'm trying to compare two models in terms of the fit they provide. In this case the two models have the same number of regressors, but they differ in the way the regressors have been constructed - in one case the canonical HRF was used to model each event (it's an event-related study) while in the second case, the canonical HRF was convolved with a square-wave modelling each trial's response duration (that is, each event was treated as a very short block, its duration was determined according the the subject's RT, and it was then convolved with the canonical HRF). Since trials vary considerably in response times (range apprx. 2 - 14 s), I strongly suspect that the second model should fit the data much better. But how would one test this formally? Maybe one possibility would be to sum the residual mean-square estimates for all subjects as given in the ResMS.img files, thus producing two images files: for the first model: Sum_ResMS_1 = (ResMS for Subj No1) + (ResMS for Subj No2) + ... and for the second model: Sum_ResMS_2 = (ResMS for Subj No1) + (ResMS for Subj No2) + ... and then perform an F-test, F = Sum_ResMS_1 / Sum_ResMS_2, with df1 = df2 = the sum of the error df from the subjects' analyses and then, for voxels where F > F_crit, conclude that the second model provides a better fit. Does that seem a valid possibility? One somewhat worrisome step is that summing the degrees of freedom for all subjects can inflate the df used for the fit test - for instance in my case I have to perform the F-test with df1 = df2 = 10410. Does this change the nature of inference? Any comments or thoughts on this would be strongly appreciated - thanks a lot. -k. _____________________________________________________________________________ Kalina Christoff Email: [log in to unmask] Office: Rm.478; (650) 725-0797 Department of Psychology Home: (650) 497-7170 Jordan Hall, Main Quad Fax: (650) 725-5699 Stanford, CA 94305-2130 http://www-psych.stanford.edu/~kalina/ _____________________________________________________________________________ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%