I understand that everyone will have their opinions, and many will disagree with mine. None of you have presented any verifiable, rigorous evidence contrary to my points, other than an exploration of the weakness inherent to my evidence - sure there's plenty. I started off saying this is anecdotal; I have completed no large-scale studies. That said, there is some experience (again, anecdotes, many of them), talking to a lot of adcom members where I got in and where I didn't. My thoughts are the impressions I have formed.
If I were to design a study to explore whether higher != better scores (the null), I would want to collect info on accepted scores from, say, Harvard, WashU, etc. I would then mark the dependent variable binary (1 = accepted, 0 o.w.), and run a logit or probit regression (use STATA software), the latter if it could be shown that the Law of Large Numbers applied (which I think it likely could, though for a better, more cynical, discussion of the Gaussian Distribution, read up on Taleb's The Black Swan) . The independent variable would be the applicant's MCAT score. The regression coefficient's value and p-value would give insight to the problem at hand. I would then run additional regressions, adding in independent variables for hours of community service, GPA, shadowing hours, presence of research, presence of strong LORs, ECs, URM status, and leadership. Likely, the effects of MCAT scores would decrease. In the discussion section, I would note studies that have failed to demonstrate correlation between MCAT and USMLE performance, studies that have demonstrated lack of the MCAT's predictive power on curricular performance, and interviews with adcom members. All of these being reasons for adcoms to focus on factors besides the MCAT.
Luckily, much of this data has been collected and presented on the MSAR, where
Harvard's distribution of MCAT Scores is10th percentile: 32, Median: 37, 90th: 41, and
WashU's is 34, 38, 41 (respectively). If I were to construct an appropriate acceptance distribution for each of these (fatter right tails, and approximately symmetric between 10th and 90th percentiles), and generate applicants from a normal distribution, simulating their decisions based on their probability of success from the acceptance distribution, then run the above described regressions, I expect to see:
A sharp cutoff, followed by a diminishing rate of increase in the resulting binary sigmoid. I would also expect to see a good measure of fit.
Now, all I'm saying is: high MCATs will ensure a thorough inspection of your application, but make sure that when that happens, you've got a lot of other good stuff to sell yourself on. So, yes, study for the damned test, but don't think for a second if you don't get an interview at Harvard it was only because your score wasn't high enough. Also, don't expect to get in, just because it was. Build yourselves as strong, developed, human applicants, and I wish you all the greatest success!
