The methodology feels odd. The results are essentially just errors in their prediction model. They're assuming that the model errors represent the school's co

The methodology feels odd. The results are essentially just errors in their prediction model. They're assuming that the model errors represent the school's contribution to median income. They're also assuming that those contributions are normally distributed around zero with constant variance. I understand why they made those assumptions but they're almost certainly not true, which isn't exactly ideal. Then again I can't quickly think of a better way to do it without more granular data. Maybe use a mixed effects model with multiple years of data from each school....