By Pim Mostert
Neural decoding, or multivariate pattern analysis (MVPA), is an advanced analysis technique that has steadily gained in popularity over the past decade. However, I have the impression that these analyses are often treated as black boxes: it is unclear what exactly they do, but as long as it works, it works.
I believe it’s worth spending some time on trying to understand what happens inside of the black box, what parameters are involved and how choices by the researcher may influence the results. This will help with choosing one’s analysis strategy, as well as with interpretation of the results.
To this end, I wrote a document in which I tried to write down what I learned about decoding analyses. Which method should I use? What exactly am I decoding? Why is noise covariance important to consider? Is ROI selection really a good idea?
Though I may not always provide concrete answers, I do hope that it’s useful and at least stimulate further thought and discussion!
Link to pdf: Opening the black decoding box