I can’t get in to see the methodology, but I just fundamentally have a hard time buying that there’s going to be an approach that actually addresses the confounding variables in any way. The sample size is way too small for how massive the variance is between whatever you consider “extremely similar” games.