<p>“based on the means of accepted applicants from a given year”</p>
<p>If you have a Harvard undergrad grid, you don’t have a mean/median problem because they show the raw data for each individual (as well as that individual’s gender, race, where they matriculated, etc.). Outliers jump out at you. They’re relatively few in number, and the median’s very close to the mean.</p>
<p>Besides, n is so large – 100+ kids accepted at HLS each year, 30+ each at YLS and SLS – that outliers shouldn’t be such a big deal. It would be really unusual to see a significant shift in mean from year to year in a group that size, given a data set where the variance is low (i.e. ~98% of the accepted students each year have LSATs in the 96th-99th percentile). Note the stability of the mean LSAT at Harvard College: it’s been either 165 or 166 every year for the last 10 years, which isn’t surprising given the large n and the cap on scores at 180.</p>