I'm planning out upgrades for Law School Predictor: Version 3.0. If you have any suggestions/comments, please let me know (via this thread is fine).
Hi, I was wondering why no one has collaborated with LSN yet to do regression analysis on the charts there... It seems to me that this would be the only way to get reliable feedback regarding splitters. On LSP it seems all the models are linear, while the real models are of course curved...
LSP works with published (and sometimes unpublished/estimated) admissions index formulas, which are linear.
Right, I knew that. But you must admit that the index formulas, as official as they may be, are not as reliable as a simple curve that could come from regression analysis on the real data, right? For example, for a GPA below 3.3, no LSAT score can really salvage a UCLA application... However, the linear formula puts you above the 95th percentile for matriculated students...
I'm sure there are other examples, but this is just one that I remember from the beginning of my own cycle; at first I thought I would have a shot at UCLA with 3.2 / 170 based on LSP, then I saw the graph.
Just wondering what you think about all this.
(By the way, the reason why I previously said that LSP seems
linear to me rather than stating it as a matter of fact is because I'm not sure how you have the splitter modifier built into your formulas... that could possibly make LSP not linear.)
The splitter modifier, in most cases, makes it non-linear, although this would only apply to roughly 10% of the applicant pool (based on LSN data). LSP started off based on index formulas, and I'm wary of drastically changing something that has worked well for the most part. That said, a number of users have commented on how some schools have "walls" or "cutoffs," and I hope to incorporate these suggestions into 3.0.
It's also worth noting that the % at/below field is a very rough estimate and should not be used to gauge one's chances; instead the textual prediction ("Admit, Strong Consider, etc.") is a better bet because I don't currently have the ability to be that precise with my predictions.