<p>
</p>
<p>This is a great point, and it could tilt the interpretation, IF we had data on it. The only data I can find is for all courses, not broken down by STEM vs non STEM, or in any other way. Similarly, the grade distribution is for all students, not just those in a particular SAT range. </p>
<p>If you have data suggesting that essenatially all of the, rare, low grades are in the premed science courses, then that might change the interpretation. Even better if we had the grade distributions for those same courses at Princeton.</p>
<p>The best we can do with available data is compare grades as a whole to SAT as a whole. Doing that we do not find evidence of tougher grading. Might there be two Berkeleys? One attended by “Princeton-like” students that grades harshly, and another attended by people who would not get in Princeton with generous grading? Then one could hide the tough grading for the minority in rich grading for the majority. One would have to postulate that the premeds are forced to take the tough courses and their grades are lower. </p>
<p>Of course lots of people take intro biology, chemistry, and calculus. Not just a select few. </p>
<p>This two Berkeley hypothesis would also make it hard to reconcile with the suggestion that the less capabable students promptly flunk out. With so few failing grades, it must be pretty unusual for people to flunk out, since I assume that requires failing more than one course…</p>
<p>The self selection is likely to be even more problematic for med school than law school. There are lots of law schools of varying selectivity, and a student who wants to go to law school can do so, even with pretty low grades and LSAT’s. They will not get in Yale, but they will get in somewhere.</p>
<p>For med school, although there is a range of selectivity, even the least selective school is still pretty tough. So there are plenty of people whose grades tell them not to waste their time with the MCAT, or whose grades plus MCATs tell them not to bother with actually applying. If you limit the data, assuming we had it, to those who apply, you have a classic restriction of range problem. This hardly invalidates the results, but needs to be viewed carefully.</p>