## Relative importance of numbers at T-14 (Layman data project)

(Applications Advice, Letters of Recommendation . . . )
elterrible78

Posts: 1120
Joined: Mon Mar 12, 2012 3:09 am

### Relative importance of numbers at T-14 (Layman data project)

Hey all,

I don't feel like I contribute too much around here, and I had some time on my hands today, so I started crunching some numbers to see about quantifying the importance of LSAT and GPA at T14 schools. A few caveats:

1. This is ONLY based on LSN data, and only for the last application cycle, so you should bear that in mind. I just don't have the time to cut and paste the data, and I don't know how to readily access the totality of LSN data in spreadsheet format.

2. The analysis is based on a very simple regression model, using outcome (accepted or rejected) as the dependent variable, with the only independent variables being LSAT and GPA. What I'm not trying to do is create a comprehensive model that explains admissions decisions. I'm simply trying to determine how much of the variation on the dependent variable (do I get in or not?) can be explained by LSAT and GPA alone.

3. I am the furthest thing from a stats major, so take this for what it's worth. It's imperfect, but I still think it may be instructive.

Notes:
- The parenthetical number in the first two tables is the pseudo R2ed. It is meant to represent the the variation on the dependent variable explained in terms of a percentage (e.g., 75 would mean that 75 percent of the admissions decision is based on LSAT and GPA alone). Because of a bunch of stats stuff that I don't want to get into, with a binary dependent variable (like the one we have here), this number becomes a little suspect, and as such is always to be taken with a grain of salt. But the best and brightest minds came up with it, so I think it's certainly meaningful.

- UVA, Michigan, and Penn are famous for suspected YP. If you look at the percentage change when waitlists are excluded, I think there's definitely some statistical evidence of this. The explanatory power of numbers alone increases by more than 4 times for UVA when you drop waitlists from the data. You can see that, while most schools have significant jumps when you exclude WLs, UVA, Michigan, and Penn are still in a class by themselves (maybe Chicago, too, but they seemed to have something weird against 177 LSAT scores last year...I'd have to dig into this more to say for sure).

- At the bottom of Table C, also in a class by themselves, we have Boalt and Yale. I think Boalt's lack of change can be chalked up to the fact that they just don't WL too many people, so including or excluding WLs makes little difference. Yale....is Yale.

I know I had stuff regarding importance of LSAT and GPA before, and I'll put that back up in a separate post. I'm messing with it right now.

Thanks for all the input, guys.
Last edited by elterrible78 on Fri Jan 25, 2013 12:48 pm, edited 2 times in total.

Cobretti

Posts: 2593
Joined: Tue Aug 21, 2012 12:45 am

### Re: Relative importance of numbers at T-14 (Layman data project)

I love this, very well done. Confirms some splitter friendly schools too... (NU, NYU). Surprised at UVA results though.

elterrible78

Posts: 1120
Joined: Mon Mar 12, 2012 3:09 am

### Re: Relative importance of numbers at T-14 (Layman data project)

mrizza wrote:I love this, very well done. Confirms some splitter friendly schools too... (NU, NYU). Surprised at UVA results though.

It's because of the YP. The whole idea is that higher numbers = better chances, but when you get a YP situation in which higher numbers = better chances UNTIL the numbers get too high, all bets are off with this type of thing. Glad you enjoyed it.

Crowing

Posts: 2631
Joined: Fri Feb 10, 2012 4:20 pm

### Re: Relative importance of numbers at T-14 (Layman data project)

UVA ED gets a lot of attention on TLS because of how easy it is to get in that way, but on the flipside UVA RD is actually really brutal.

megagnarley

Posts: 498
Joined: Sun Sep 30, 2012 1:58 am

### Re: Relative importance of numbers at T-14 (Layman data project)

Always feel like Boalt DGAFATLSAT

wert3813

Posts: 1409
Joined: Sat Oct 06, 2012 6:29 pm

### Re: Relative importance of numbers at T-14 (Layman data project)

Crowing wrote:UVA ED gets a lot of attention on TLS because of how easy it is to get in that way, but on the flipside UVA RD is actually really brutal.

This.

Also keep playing with it. Boalt seems high based off looking at their LSN charts. Also the fact that they could have a higher median if they want to suggests they aren't as worried with the numbers but could put your data in a weird spot (someone above both medians gets in because they are above both but this ignores that their 169 would actually be below if Boalt cared as much as Penn, UVA, etc.

warandpeace

Posts: 301
Joined: Sat Mar 12, 2011 1:43 pm

### Re: Relative importance of numbers at T-14 (Layman data project)

Looking good. Looking forward to watching this evolve.

elterrible78

Posts: 1120
Joined: Mon Mar 12, 2012 3:09 am

### Re: Relative importance of numbers at T-14 (Layman data project)

warandpeace wrote:Looking good. Looking forward to watching this evolve.

If I get really ambitious later, I'll build URM and non-trad status in. Not sure how much of a change it will make, but why not?

Micdiddy

Posts: 2230
Joined: Tue Mar 20, 2012 3:38 pm

### Re: Relative importance of numbers at T-14 (Layman data project)

Ts;Du
Last edited by Micdiddy on Thu Jan 24, 2013 6:51 pm, edited 1 time in total.

dnptan

Posts: 355
Joined: Fri Jan 11, 2013 5:33 pm

### Re: Relative importance of numbers at T-14 (Layman data project)

I can see this being very useful, keep up the good work!

Rahviveh

Posts: 2333
Joined: Mon Aug 06, 2012 12:02 pm

### Re: Relative importance of numbers at T-14 (Layman data project)

wert3813 wrote:
Crowing wrote:UVA ED gets a lot of attention on TLS because of how easy it is to get in that way, but on the flipside UVA RD is actually really brutal.

This.

Also keep playing with it. Boalt seems high based off looking at their LSN charts. Also the fact that they could have a higher median if they want to suggests they aren't as worried with the numbers but could put your data in a weird spot (someone above both medians gets in because they are above both but this ignores that their 169 would actually be below if Boalt cared as much as Penn, UVA, etc.

Boalt's charts really aren't all that different from its peers - they just like to reject people instead of waitlisting them. They admit most of their class early and keep a small waitlist for later.

Betharl

Posts: 233
Joined: Tue Jul 12, 2011 8:48 pm

### Re: Relative importance of numbers at T-14 (Layman data project)

Schools that rely on ED a lot like UVA/Penn might come up as being less numbers focused in those results, because at those school, ED can be a huge factor.

Ti Malice

Posts: 1948
Joined: Sat Mar 03, 2012 2:55 am

### Re: Relative importance of numbers at T-14 (Layman data project)

Good stuff. You should send a PM to LSATSCORES2012 to talk about this. He calculated Pearson correlation coefficients for the importance of numbers at the T14 schools here: viewtopic.php?f=9&t=188663&p=5753360&hilit=#p5753360. He chose to exclude WL data to avoid YP pollution. (There are obviously arguments to be made for both approaches to treating WL data.)

suralin

Posts: 18586
Joined: Wed Nov 14, 2012 1:52 am

### Re: Relative importance of numbers at T-14 (Layman data project)

Ti Malice wrote:Good stuff. You should send a PM to LSATSCORES2012 to talk about this. He calculated Pearson correlation coefficients for the importance of numbers at the T14 schools here: viewtopic.php?f=9&t=188663&p=5753360&hilit=#p5753360. He chose to exclude WL data to avoid YP pollution. (There are obviously arguments to be made for both approaches to treating WL data.)

Damn, that poster is fucking awesome. And so are you OP

elterrible78

Posts: 1120
Joined: Mon Mar 12, 2012 3:09 am

### Re: Relative importance of numbers at T-14 (Layman data project)

Ti Malice wrote:Good stuff. You should send a PM to LSATSCORES2012 to talk about this. He calculated Pearson correlation coefficients for the importance of numbers at the T14 schools here: viewtopic.php?f=9&t=188663&p=5753360&hilit=#p5753360. He chose to exclude WL data to avoid YP pollution. (There are obviously arguments to be made for both approaches to treating WL data.)

Hmmmm...I wonder what he used for variables, then, with the Pearson? Like I said, I'm not a stats whiz (I just picked up some stuff in grad school), but I think with Pearsons you need to use continuous variables, and admissions is an in/out proposition.

As far as excluding WL data, there are actually much better arguments for excluding it than including it. The YP really messes stuff up, so I'm going to go ahead and do this whole thing over again excluding WLs and see what happens. I think that will give a more accurate representation, especially as far as MVP are concerned. There is still the issue of ED, but I'm not willing to wade through the data right now and manually code each EDer, and again, getting it right is highly dependent on the LSN users reporting accurately, anyway. I think whether or not they were ED is one of those things that is maybe not most likely, but definitely more likely, to be left out by the user.

And thanks!

elterrible78

Posts: 1120
Joined: Mon Mar 12, 2012 3:09 am

### Re: Relative importance of numbers at T-14 (Layman data project)

Updated OP, and am continuing to work on some other stuff regarding LSAT and GPA (namely, doing some charts to get a visual representation of splitter-friendly schools, and creating a scale to compare apples-to-apples with the two numbers).

Thanks!

elterrible78

Posts: 1120
Joined: Mon Mar 12, 2012 3:09 am

### Re: Relative importance of numbers at T-14 (Layman data project)

As you can see, and as someone pointed out earlier, looking at these lists gives you a good idea of which schools tend to be more splitter friendly. Below is a scatterplot to give you an idea of where schools fall in terms of being relatively friendly to splitters and reverse-splitters:

I doubt this is helpful or interesting many people at all, but I feel better for doing it. For me, it beats the hell out of refreshing TLS and the status checker all day.
Last edited by elterrible78 on Mon Jan 28, 2013 10:59 am, edited 1 time in total.

Micdiddy

Posts: 2230
Joined: Tue Mar 20, 2012 3:38 pm

### Re: Relative importance of numbers at T-14 (Layman data project)

elterrible78 wrote:I doubt this is helpful or interesting

jetsfan1

Posts: 571
Joined: Thu Nov 01, 2012 8:14 pm

### Re: Relative importance of numbers at T-14 (Layman data project)

Very interesting- well done OP

LSATSCORES2012

Posts: 770
Joined: Mon Jul 02, 2012 2:12 pm

### Re: Relative importance of numbers at T-14 (Layman data project)

jetsfan1 wrote:Very interesting- well done OP

+1