In some variables the severity impact was dependent on the trait variables in a nonlinear way
I should test for this using non-linear models because this could be a bombshell