Question 1

saving time, being faster, or more generally the efficiency, of
stepwise is not something that would enter into discussion with the
dean.


Question 2

df, or number of parameters, does not tell us how well one model or
the other does, in explaining the data




Question 3


it is true that RMSE is also SE of prediction, *as long as* the
predictions are unbiased.   



Question 4

which interactions would be helpful here?

At least, replace beautyf2upper with btystdave, since the beauty
scores given by 1 student shouldn't affect decisions for the whole
college...

are there any specific predictors or interactions that you would
consider adding in or taking out?