Hi Gang,
Thank you for your prompt response. We should have been more clear in describing Method 1: we would use a two-pronged approach in which one 3dDeconvolve script would concern only correct and incorrect trials (6 conditions in the model) and another that concerns 'all' trials (3 conditions in the model). Will this two-pronged approach yield better data than collapsing across correct and incorrect trials as outlined in Method 2?
Our specific concern with Method 2 has to do with how conditions are weighted when collapsed. In our experiment, the number of correct and incorrect trials are quite disparate in some conditions; for example, one condition has 76% correct and 24% incorrect. When collapsing these two together using method 2, we are concerned that when assigning 1s to these two conditions, the 76% and 24% receive equal weighting when clearly more trials are contributing to the correct estimate. It was argued that creating a separate 'all' script avoids this problem by using all trials in the condition to create an estimate.
To be clear, Method 3 should be used in situations in which one seeks to determine whether the BOLD signal in one condition proportionally differs from the BOLD signal in another, right? In our study, we are simply interested in differences between experimental conditions and we have no ad hoc predictions regarding the proportionality of BOLD responses across conditions. Does this render Method 3 inappropriate for our experimental design?
Thank you for your assistance.
George