ITT: I try to estimate U of C’s index formula
Posted: Thu Jan 20, 2011 7:21 pm
This is a non-technical summary. If you want the technical details, feel free to ask.
I tried to create a model that could predict admissions outcomes using the TLS stats data for the purposes of better understanding why I was rejected. The three outcomes I designated were rejections, waitlists, and acceptances. URMs were eliminated from the analysis after I found that including a URM variable messed up the predictors for reasons that are really complicated. PLUM-ordinal regression gave me the following equation:
(GPA * 3.726) + (LSAT *0.280) = index
and these thresholds
Index < 60.916 = Rejection
60.916 < Index < 62.757 = Waitlist
62.757 < Index = Accept
The model works (X2 = 64.921, df = 2; p < .000), but it is much better at predicting rejections and acceptances than waitlists. If a score falls within the reject range the probability of rejection is 71.4%, the probability of a waitlist is 23.2%, and the probability of an acceptance is 5.4%. On the other side, an index score in the accept range gives you a 64% chance of acceptance, a 31% chance of a waitlist, and a 5% chance of rejection. Waitlist range scores should be considered inconclusive. This didn’t really help me answer any of my questions (since my score falls within the waitlist range), but I thought it might be helpful for others.
I tried to create a model that could predict admissions outcomes using the TLS stats data for the purposes of better understanding why I was rejected. The three outcomes I designated were rejections, waitlists, and acceptances. URMs were eliminated from the analysis after I found that including a URM variable messed up the predictors for reasons that are really complicated. PLUM-ordinal regression gave me the following equation:
(GPA * 3.726) + (LSAT *0.280) = index
and these thresholds
Index < 60.916 = Rejection
60.916 < Index < 62.757 = Waitlist
62.757 < Index = Accept
The model works (X2 = 64.921, df = 2; p < .000), but it is much better at predicting rejections and acceptances than waitlists. If a score falls within the reject range the probability of rejection is 71.4%, the probability of a waitlist is 23.2%, and the probability of an acceptance is 5.4%. On the other side, an index score in the accept range gives you a 64% chance of acceptance, a 31% chance of a waitlist, and a 5% chance of rejection. Waitlist range scores should be considered inconclusive. This didn’t really help me answer any of my questions (since my score falls within the waitlist range), but I thought it might be helpful for others.