Hi, I wonder if you could elaborate on the views you expressed here with respect of the type of design in an fMRI experiment. That is to say, are you referring to a block design, or would you also apply this to an event-related design? Would you also plead for designing a baseline event? I am asking this because I am puzzled by the way in which contrasts in an ANOVA and in a regression task are handled as if they were equivalent. The underlying models surely are not the same. In a regression, a contrast [+1 -1] means a linear combination in which for the first regressor the predicted BOLD signal goes up, for the second regressor the predicted BOLD signal goes down relative to the intercept. You do not have a BOLD signal that goes up less than another. It's rather the direction of the linear dependency of the variance on the regressor that is contrasted, in contrast with ANOVA designs where you compare means. You aren't modelling a lower BOLD response, you are modelling a negative BOLD response. All the best Roberto Viviani Department of Psychiatry University of Ulm, Germany ----- Original Message ----- From: "Eric Zarahn" <[log in to unmask]> To: <[log in to unmask]> Sent: Monday, May 30, 2005 1:01 AM Subject: Re: [SPM] Inference fo contrast Hi Amit, ----- Original Message ----- From: "Anand, Amit" <[log in to unmask]> To: "Eric Zarahn" <[log in to unmask]>; <[log in to unmask]> Sent: Sunday, May 29, 2005 6:20 PM Subject: RE: [SPM] Inference fo contrast > Hi Eric, > > Thanks for you reply but I am still unclear about this. There are only two conditions A and B in our current experiment. One could take the resting state as condition C but what if you do not have that. Yes, I understand. But when you laid out your possibilities underlying a positive A minus B contrast value, you were assigning absolute levels of activation to A and B. But in BOLD fMRI there are no meaningful absolute levels activation (e.g., even the brain of a cadaver will have positive T2* weighted signal in its voxels); all meaningful activation is implicitly relative to some other condition (even when it is not stated as such). So, without some third condition C (even if just conceptual), then statements like "positive activity in A greater than negative activity in B" make no sense (or are at least ill-defined and unknowable) in BOLD fMRI. To hammer the point home, the only way to understand BOLD fMRI activation is in relative/comparative terms (unless there has been some important theoretical paper out there that I have missed!). > > By inhibited I do mean decrease from resting state. Right. So if B represents your resting state, B minus A will have positive values in voxels that have higher values in the resting state compared to A. Eric