Regressing LSAT Medians etc on ABA Data Forum

[Kimikho]

(I didn't really know where to put this, so I hope here was okay)

So at the beginning of the quarter my stats professor wanted us to have categorical and continuous data with more than 150 observations that we could run regressions on, so I downloaded some ABA data and used that. I ended up getting some pretty cool results and figured I'd put them up here.

A Regression
I can post the other regressions if people want, but most of them have pretty low R-squares
percent in big law = 1.2503 - 4.4002(scaled LSAT+GPA score) + 3.8276(scaled^2)
R-squared: 0.7044
196 degrees of freedom

Graphs

LSAT Medians


LSAT Median Percentile


GPA Medians


Scaled GPA and LSAT Medians


Scaled GPA and LSAT Medians
This is the one that had the highest R-squared.


The next post I'll post some T14s in unemployed and seeking, percent in bar-passage required school funded part time jobs, solo practioners, etc.

If anyone has any requests I'll pull them up, too, since I need practice with this stuff and have it all on my computer . If no one cares then that is okay too

Methodology
Employment is the total employed requiring bar passage divided by total graduates. This includes part time/short term/school funded positions. In retrospect I should have deleted at least the part time school funded positions, but I was worried this would skew the data and end up reflecting a little bit which schools can afford to hire graduates. There are a lot of state universities that didn't do any school-funded positions for c/o 2013. I could go back and add that in, but I've forgotten which schools I took out because of lack of data/r being weird so I'm not going to do that.

Percent in big law is percent in firms with more than 251 lawyers. I did this because my professor wanted us to make categories, and then pick one to look at more deeply and that's the one I picked.

The scaled LSAT+GPA is:
(LSAT percentile + (GPA/4.3)*.9)/2.
The .9 was because schools value LSAT scores more than GPA.
The 2 was so that it would be out of 1.

Unfortunately, the medians are from 2015 (I think) but the employment data is c/o 2013...I was too lazy to fix that once I noticed the mistake, but if there is high demand I will. I can't imagine it would change much.

Source: ABA and LSAC

[Kimikho]

Obviously this wasn't part of my hw but it takes like five seconds and I was curious. I liked to imagine how TTT schools would use to market the programs.

"We are in the top 14 for number of employed graduates!"
T14 in Total Number Employed:
GEORGETOWN UNIVERSITY
NEW YORK UNIVERSITY
HARVARD UNIVERSITY
GEORGE WASHINGTON UNIVERSITY
COLUMBIA UNIVERSITY
VIRGINIA, UNIVERSITY OF
MICHIGAN, UNIVERSITY OF
Thomas M. Cooley Law School
FORDHAM UNIVERSITY
TEXAS AT AUSTIN, UNIVERSITY OF
BROOKLYN LAW SCHOOL
SOUTH TEXAS COLLEGE OF LAW
MIAMI, UNIVERSITY OF
CALIFORNIA-BERKELEY, UNIVERSITY OF

"XXX School of Law is known for its professionalism. We are in the top fourteen for number of graduates employed in a professional position that does not require the bar."
T14 in Number employed in a non-bar professional position:
Thomas M. Cooley Law School
INTER AMERICAN UNIVERSITY OF PUERTO RICO
NEW YORK LAW SCHOOL
FLORIDA COASTAL SCHOOL OF LAW
MARYLAND, UNIVERSITY OF
PACE UNIVERSITY
NEW ENGLAND LAW | BOSTON
BROOKLYN LAW SCHOOL
SUFFOLK UNIVERSITY
RUTGERS UNIVERSITY-NEWARK
WIDENER UNIVERSITY-DELAWARE
TOURO COLLEGE
ARIZONA SUMMIT LAW SCHOOL
INDIANA UNIVERSITY - INDIANAPOLIS

"We are committed to getting our students jobs [we hire them]."
T14 in percent of grads in school-funded, part time (long term and short term), bar-required positions
NEW YORK LAW SCHOOL
SUFFOLK UNIVERSITY
DUKE UNIVERSITY
GEORGETOWN UNIVERSITY
JOHN MARSHALL LAW SCHOOL
VALPARAISO UNIVERSITY
NORTHEASTERN UNIVERSITY
CHICAGO-KENT COLLEGE OF LAW-IIT
MICHIGAN STATE UNIVERSITY
LEWIS AND CLARK COLLEGE
LIBERTY UNIVERSITY
CHICAGO, UNIVERSITY OF
NORTHWESTERN UNIVERSITY
INDIANA UNIVERSITY - INDIANAPOLIS

I think this one shows how difficult it is to parse out school funded jobs. I chose part-time because that seemed the most sketchy to me.

The Midlaw T14 (highest percentage of grads in firms 26-50 lawyers):
SOUTH DAKOTA, UNIVERSITY OF
FAULKNER UNIVERSITY
TULSA, UNIVERSITY OF
WYOMING, UNIVERSITY OF
SOUTH CAROLINA, UNIVERSITY OF
ARKANSAS, FAYETTEVILLE, UNIVERSITY OF
HAWAII, UNIVERSITY OF
OREGON, UNIVERSITY OF
FLORIDA A&M UNIVERSITY
NORTHERN ILLINOIS UNIVERSITY
MAINE, UNIVERSITY OF
VANDERBILT UNIVERSITY
APPALACHIAN SCHOOL OF LAW
ALABAMA, UNIVERSITY OF

[bugsy33]

Nice work. Very interesting to see the trends. Love that Cooley is T-14 for number of grads employed. LOL

[cotiger]

Holy endogeneity, batman. Need a preftige var up in therr

[Kimikho]

Holy endogeneity, batman. Need a preftige var up in therr

I was thinking about doing something with rank, but it means I need to manually add them in. I prob will though, I'm still curious.

[cotiger]

Holy endogeneity, batman. Need a preftige var up in therr

I was thinking about doing something with rank, but it means I need to manually add them in. I prob will though, I'm still curious.

Rank's not going to fix it. Preftige isn't linear, and neither is that error term correlation. Something like that biglaw partnership data would be much better.

Ultimately, though, it's all bullshit. Plug in a time lag and everything's going to get completely wiped out. Biglaw today is pretty much exclusively a function of biglaw yesterday. You'd need panel data going back quite some time to capture any effects from GPA/LSAT.

eta: i know this is just for kicks, just thought i'd add in a little RIGOR

[Kimikho]

Holy endogeneity, batman. Need a preftige var up in therr

I was thinking about doing something with rank, but it means I need to manually add them in. I prob will though, I'm still curious.

Rank's not going to fix it. Preftige isn't linear, and neither is that error term correlation. Something like that biglaw partnership data would be much better.

Ultimately, though, it's all bullshit. Plug in a time lag and everything's going to get completely wiped out. Biglaw today is pretty much exclusively a function of biglaw yesterday. You'd need panel data going back quite some time to capture any effects from GPA/LSAT.

eta: i know this is just for kicks, just thought i'd add in a little RIGOR

Time series would be better, for sure. The regression up there is more useful in figuring out if there is a random school with 1XX LSAT and X.XX GPA medians then what does this model predict biglaw percentage will be?

I actually don't agree that rigor would play a huge part, because the model squares LSAT and GPA. I think that reflects prestige--there is a linear increase with higher LSAT and GPA medians, and then on top of that, at some point prestige kicks in, too. Schools with higher LSAT medians tend to have more prestige, so adding in a prestige factor would add in some pretty severe collinearity. So it's probably not super accurate to say:

"I'm going to make a for-profit law school right now but only accept people with above a 173 and 3.9 and it will have about 39% in BL," because the model assumes that any school with 173/3.9 medians is relatively well-regarded. But saying...

"If Princeton opened a law school right now with medians of 173/3.9 it would have about a 39% BL placement once it gains the same amount of alumni/prestige/etc as Harvard" might be. Or at least, might be more so.

[cotiger]

Ohhh I was just looking at your individual graphs. Didn't see that overall regression. Does that last graph have your regression line on it? If so, might wanna think about exponential, which makes more sense anyway.

Still, (assuming that you're trying to make an actual model ), I don't know that entry scores is the strongest instrument for prestige, which with the Princeton vs for profit you seem to indicate you're trying to do. Straight USNWR (perhaps some sort of historical average) would probably be a stronger instrument.

But if you're trying to indicate some sort of causal relationship between entry scores and employment.. http://www.top-law-schools.com/forums/v ... 2&t=225983

[Kimikho]

Ohhh I was just looking at your individual graphs. Didn't see that overall regression. Does that last graph have your regression line on it? If so, might wanna think about exponential, which makes more sense anyway.

Still, (assuming that you're trying to make an actual model ), I don't know that entry scores is the strongest instrument for prestige, which with the Princeton vs for profit you seem to indicate you're trying to do. Straight USNWR (perhaps some sort of historical average) would probably be a stronger instrument.

But if you're trying to indicate some sort of causal relationship between entry scores and employment.. http://www.top-law-schools.com/forums/v ... 2&t=225983

Yah, the last line is the regression and it's exponential.

I wasn't looking for a causal anything just doing my hw. There is for sure a correlation though.

[Pragmatic Gun]

Dear Lord, the idea of Princeton having a law school freaks me out! The Northeast Corridor is already a hypercompetitive legal market.