(1) make sure they discuss xforms somewhere in main text

(2) for question 2, ask if it matters whether you look at crimes or
crimerate = crimes/pop

(3) for question 3, make sure they combine judgement/intuition with at
least one automatic variable selection method. there's just too much
"shoot from the hip" judgement, that leads to small models that ignore
important demographic factors otherwise.

(4) Also for question 3, need a more explicit discussion of
transformations.  Make sure this gets discussed in the results section
and not in the methods section.