Power Analysis for Representational Similarity

Hi all,

Shot in the dark here - I haven’t been able to find any answers on this topic. I’m planning to conduct a study with the intention to run a representational similarity analysis with two conditions and several a-priori ROIs. Unfortunately because of the complicated nature of RSA it’s not clear to me run a regular power analysis. Does anyone have any experience with this or any tips?



Ahoi hoi @kylieisenburg,

sorry for the late reply. Also: not an expert, thus not sure how helpful my 2 cents will be, sorry.

First question: do you have a specific reason/motivation to utilize RSA? I’m asking because with 2 conditions, an RSA might seem like overkill and might not add a lot of information as compared to e.g. “classical stats” (t-test) or ROI decoding, no? Sorry, I don’t want to sound mean or so, I was just wondering.

Re power analysis: yeah, you’re definitely right. Power analyses are already something else but their application is especially elevated concerning analyses like RSA. Generally, you could look into resources like neuropowertools (preprint, website) which allows you to compute and evaluate not only power but also design efficacy. Re the latter: IIRC, one of the main advantages of RSA was its application to condition-rich designs within which other, more standard, approaches might be drastically limited. This aspect is actually nicely discussed in the OG RSA paper from 2008, specifically in the section “Core Concepts for Experimental Design” but also throughout the paper.
It might not be directly “power” but definitely very related.

HTH, cheers, Peer

Hi Peer,

Thank you so much for your response! I will check out neuropowertools and see what it looks like. By the way, I was wrong in saying two conditions. I have two over-arching conditions of my experiment but there are sub-conditions that fall under them so ultimately it will be 10 conditions that I will use RSA for.

Again, thanks for the recomendation!