I encounter an question in doing MVPA that been bothering me for a long time, please let me know your comments, many thanks!
In short, the results generated by RSA and classification are different using the same raw data. How should I make sense of it?
In more details, there are 4 categories of stimulus in my task. I estimated the brain activation of each categories by a GLM with 4 task-relevant regressors, and got 4 beta maps in each subject.
In the follow-up RSA, I compared the similarity of brain activations and a theoretical RDM in which the same category was encode as high similarity. The whole-brain searchlight showed a significant cluster locating in dmPFC.
Then, I conducted a decode analyses with linear discriminant analysis. I also inputted beta-maps in the GLM and labeled each of them by the corresponding category. However, the whole-brain searchlight found a significant cluster laying in precentral cortex, and no overlap was found between the results of RSA and classification.
I know little about the mathematics underlies MVPA, and barely articles I could find which discuss the difference of these two MVPA methods, thus, I have no idea about how to interpret these results.
What’s your opinion. Do you think the results are incompatible or not? Any idea will be appreciated.