I have the same problem with these analyses from interview.io every time I see them: they are making many of their inferences by fitting linear models to data that is non-linear. Is getting a 4 twice as good as getting a 2? If not, don't use a linear model!