@haunted said: log transforming your variable is a pretty standard way to make the data look more normal—and it is important that it roughly follows a normal distribution, since the assumptions a linear regression makes work on that type of distribution

thanks for the suggestion! I tried a few transforms of the dependent variable which is messing things up (points of 0-100, ~60% are at 100 and some smaller peaks around, with still 10% at 0). I did a log(x+1) and looked at residuals resulting from that and square rooting and squaring. I may try to recast it as a binary (pass vs fail) and do a logistic regression for a coarser analysis instead. I am not convinced that the scores are that meaningful themselves.

Need to read up on this a bit more, I feel like I’m missing something.