<p>Vossron: Large sample size does not solve bias. You could do a complete sample of all the Republican likely voters, millions of people, and you would still underestimate national support for Clinton. You would have a very precise estimate of support for her among Republicans, but not of her support overall.</p>
<p>If the sample of MCAT takers is not representative of students as a whole at each college, then it does not matter how many colleges you test. The MCAT-taker sample is limited in telling you what happens to students overall at that college. If colleges are homogeneous in the probability that a student at a given level of ability will pursue premed far enough to take the MCAT, then the sample still will not tell you about all the people who did not take the test, but it will capture the variation in preparation among the colleges. However, there is no reason to assume that colleges are homogeneous on this measure, and it would be very difficult to check.</p>
<p>Token: So are you converting the argument that “people get PhD’s because they cannot find work” into they “get PhD’s because they can afford to stay out of the job market”? </p>
<p>Certainly there are financial sacrifices in attending graduate school. However, absent evidence that the small LAC’s that have high PhD rates also have wealthier student bodies than many of the elite universities with lower PhD rates, not clear how the family income consideration explains the differences in PhD rates.</p>