Hold on now. Michigan is a top 30, same category as JHU. And I haven't claimed any school-specific trends, only looked at top 30 vs non top 30, and each of those two categories has n = plenty and from a good variety of schools. I also have made no claims about whether we are looking at top/middle/bottom students at each category, or whether they'll get accepted at significantly different rates, etc. I've never heard of midlevel schools which forgive low GPA in exchange for better MCAT more than other schools; I think it's very safe to assume applicants applied to this school because their two stats fit the interval for both. Neither sets of stats (3.5 and 33 or 3.6 and 30) is lopsided compared to the averages at those schools to imply a selected-for sample with unexpectedly strong MCAT or GPA from within the student population like you suggest.
The only claim I'm making, and one which the data backs up, is that a group of students who are of mean ~33 MCAT ability earned a lower GPA at top 30 schools than a group of students who are of a mean ~30 MCAT earned at non-top-30 schools. If the MCAT is truly an equalizing assessment of ability, and there was not a significant difference in difficulty between the groups of schools, you'd have predicted the opposite (higher MCATers earning higher GPAs).
You cannot hand waive away that this is a self-selecting group, and thus selection bias could just as easily explain the trend you are trying to explain as inherent trends in the student population. The only thing that [limited] data tells you is that at this particular school, they have more student applicants with certain MCAT/GPA combinations from School A vs School B.
If the MCAT is truly an equalizing assessment of ability, and there was not a significant difference in difficulty between the groups of schools, you'd have predicted the opposite (higher MCATers earning higher GPAs).
Again, this is one interpretation, but there are many others that could explain the data, and you don't have sufficient information to distinguish among them. If the MCAT is truly an equalizing assessment of ability, and there is not a significant difference in difficulty between the groups of schools, one must also ask whether these applicants are representative samples of A) the students from each school B) the applicants from each school. I would argue that no applicants to any
one med school is representative of all the students from all the schools represented in that medical school's application pool (simple thought experiment: are the applicants from Harvard to University of Nevada Medical School representative of all Harvard applicants at other schools?).
This is not "n = plenty and from a good variety of schools." I apologize for being blunt, but I don't have the time to go into detail explaining statistics, regression [which is basically what you're attempting to do], and how to interpret data to you. I'm not saying that there is no relationship between the school (your top 30 groupings appears arbitrary, but we can accept it for now) and an applicant's MCAT and the GPA. There may be, and if there is, the magnitude of the difference matters (is there a nominal difference that is basically being scapegoated here to explain away failures, or is there a large trend here that must be recognized in order for schools to avoid making perilous decisions?). What I
am saying, is that you do not at all have sufficient information to make any claim beyond that which is relevant to this
one school for which the data was obtained, and even for this one school, your sample size is incredibly limited. You cannot begin to control for factors like race, age, school, major, etc when some of your schools have less than 10 data points.
Having data is great, but knowing your limits is greater.