sillymike wrote:this is a great thread. props to nonexpostfacto for starting it, and for lsatscores2012 for posting that link for the analysis done by kappycaft1!
Not trying to be "that guy", but taking this post and kappycaft1' analysis together leads me to a question.
What kind of margin of error are we talking about here, if we're basing our data on lsn and most schools have 10% total applicants/20% admittances? Especially in the super splitter range, which is what I think the OP was driving at, it only takes one or two data points to completely change things.
Then again, maybe one or two data points won't change anything, because I'm SO not a stats guy.
Mostly, as a super splitter myself, I'm just wondering when people say things like "locked out of Penn", they're really saying "10% to 20% of the data shows you're locked out, but there's a 1/5 chance, or maybe less adjusting for applicant caliber, that they've actually dipped below your GPA this cycle and we just didn't know about it".
I love this thread. I hope it's one of the ones that never dies!
The fact that it "only takes one or two data points to completely change things" is why we don't use a strict definition of an absolute floor. Being below the commonly-discussed "floor" at a school does not mean you are absolutely 100% out, it just means your chances of being accepted are very, very low, probably under 5%, as best we can infer. It should be reiterated that there is no such thing as 100% certain to be accepted or 100% certain to be rejected--it's best to deal in probabilities rather than absolutes.
While LSN does have a demonstrable bias towards higher-number applicants (particularly high LSAT scores) when compared to their overall applicant pool, and LSN contains a bias within itself of people being more likely to report acceptances than rejections, or more likely to update their profiles with acceptances as opposed to rejections, it does at least roughly
approximate a sample of the total applicant pool. The sample is sufficiently large such that strong inferences can be drawn from it even if the sample is not a gigantic proportion of the population. That's the power of sampling, if it's truly random--600 people is a tiny subset of the ~130 million who will vote in a presidential election, but a poll of 600 randomly selected voters can nonetheless predict an election to within 3 percentage points with 95% confidence. You can predict a relatively slim margin of error for the applicant pool based on TLS data even if you can't say with 100% certainty that something is the case. I can't prove that something is impossible just because it hasn't been on LSN, but I can say with a high degree certainty that such a thing is very, very unlikely.
yeah. I always go back and forth on that point. Everybody says that this is good for splitters, and for sure some people are having mad cycles, but it's true that it might not be as great as it seems.
On the one hand, it does really just allow more people to be at median.
One the other hand, I would hope that if a schools lsat median starts dropping, they might be willing to take some lower GPAs to compensate or even try to raise the median. Numbers-wise, if a school's medians are 3.7 and 169, and they want a 3.7/170, isn't a 2.0/170 worth more for the median than a 3.69/169, since they will both drop the GPA median by the same amount (or not at all) but the 169 will (possibly) contribute to the median staying at 169 whereas the 170 could be the data point that secures the 170 median?
I mean, it looks pretty bad to take a 2.0/171 so i doubt many of the schools with a 170 median would actually see it that way, but from a pure math standpoint, isn't this the case? Sorry if I'm totally off, I'm still figuring out this whole median business
Is there a point where schools will start to think in terms of pure numbers, or do they care too much about their 25/75ths? I had assumed they don't care a huge amount about those numbers, but what do I know?
nothing. that's what.
Strategic median-gaming is certainly a thing. A 3.8/169 is essentially out at UVA and Penn despite being barely below both medians. A 3.9/165 and a 3.4/171 both have a better shot at being accepted despite being, in my opinion, much less qualified than the 3.8/169. Again, look at the 2.8/170 students being accepted to NU over the 3.5/169. That's the starkest example of a school that seems to have absolutely no concern for the quality of the applicant and is just attempting to game medians. So yes, the strategic admissions policy is to balance out splitters with reverse splitters to have artificially inflated medians. Medians are the big fish and 25th/75th scores, which have to be reported, matter at least a little. But outside of those, it doesn't matter. You could admit 24.9% of your class with godawful GPAs and another 24.9% with godawful LSATs and no one would know unless the students self-report on LSN.
Sure, some schools care about the overall numerical strength and do not partake in the strategic manipulation. HYS don't manipulate medians, nor do Columbia, NYU, Berkeley or Cornell. At each of those schools, the strength of your numbers is roughly determined by the "sliding scale" each school applies, which means there's no situation in which one group with stronger numbers is systemically disadvantaged compared to other groups with weaker numbers (as is the case when schools manipulate medians). You can tell when a school isn't manipulating medians when they fall roughly halfway between the 25th and 75th percentiles--contrast that with the manipulators, where the medians are much closer to the 75th percentile than the 25th.