What are the statistical odds of class rank based on LSAT? Forum

[kryoung]

The LSAT has a certain correlation to actual law school grades. How does this relate to your score in reference to a school's median. If you have a median LSAT what are your chances of finishing top 10%? Top third? Median? Bottom third? Etc. What if you have a 75th LSAT or 25th? I guess grades could factor in here as well but to a much lower extent because the correlation isn't as strong.

I feel like this information should not be overlooked when deciding on schools. We are not special snowflakes. If you are accepted to a school which has 50% strong job placement but your LSAT indicates you have a thin chance of actually reaching minimum, wouldn't this be useful to take into account?

[t-14orbust]

Based on what I've read here your chances for a certain percentage rank are that same percentage. Chance for top 10%=10%, top third = ~30%, median = 50% etc

[kryoung]

Based on what I've read here your chances for a certain percentage rank are that same percentage. Chance for top 10%=10%, top third = ~30%, median = 50% etc

The LSAT has some type of correlation and because of it I think this would clearly be false for some people. A 25th% LSAT is going to have lower odds than a 75th%. Any statistical people want to run some analysis of this?

[rad lulz]

m

[kryoung]

There's a study you can google ab this

But lemme save you some hassle and give you the takeaway

Although an LSAT way higher than everyone else's increased your chance of doing better relatively

Job prospects drop off a lot faster than LSAT medians and 75ths

So it basically doesn't make sense to go to a worse school and try to kill it there

I understand that but I am primarily concerned about this question for the top schools. Columbia, Chicago, Harvard, and Stanford do not seem to have the immense employment drop off between schools. At least, not when you throw in money for Columbia and Chicago. Is a 170 going to have much worse odds of hitting median or some other platform at Harvard than at Stanford?

[t-14orbust]

There's a study you can google ab this

But lemme save you some hassle and give you the takeaway

Although an LSAT way higher than everyone else's increased your chance of doing better relatively

Job prospects drop off a lot faster than LSAT medians and 75ths

So it basically doesn't make sense to go to a worse school and try to kill it there

I understand that but I am primarily concerned about this question for the top schools. Columbia, Chicago, Harvard, and Stanford do not seem to have the immense employment drop off between schools. At least, not when you throw in money for Columbia and Chicago. Is a 170 going to have much worse odds of hitting median or some other platform at Harvard than at Stanford?

lol no

[Pneumonia]

Also keep in mind that, while schools place a lot of weight on 25/75's, objectively the difference is minimal. In admissions the difference between a 168 and a 172 is huge, but don't let that full you into thinking you are smarter or better than someone who has an LSAT 5 points below you or whatever. It's 98th vs 96th percentile, so really only a 2 percent difference in apptitude (read: no difference at all) vs what you might be thinking of as a 50% difference if your just looking at the 25/75's.

Also- what Rad said.

[patogordo]

LSAT scores are actually score bands, so "statistically" your 172 is actually no different than your classmate's 170.

[Clearly]

LSAT scores are actually score bands, so "statistically" your 172 is actually no different than your classmate's 170.

You can't just assume his aptitude is at the high end of his range, and your aptitude is at the low end of your range. "no different" isn't the same thing as "some overlap"..

[nebula666]

I think most would agree that someone above the top 75% percentiles at one school would have a marginally better shot to be above median than at a school where they are below both 25%. Short of that, I've really seen too many examples of people with high scholarships slacking and people who were waitlisted making law review. People come from entirely different educational and other backgrounds. There is no real metric to determine how you will fare head to head over three years of classes and exams.

[patogordo]

LSAT scores are actually score bands, so "statistically" your 172 is actually no different than your classmate's 170.

You can't just assume his aptitude is at the high end of his range, and your aptitude is at the low end of your range. "no different" isn't the same thing as "some overlap"..

i meant that the difference is insignificant considering the not particularly strong correlation between LSAT and law school gpa, hence the scare quotes around "statistically"

[kryoung]

Also keep in mind that, while schools place a lot of weight on 25/75's, objectively the difference is minimal. In admissions the difference between a 168 and a 172 is huge, but don't let that full you into thinking you are smarter or better than someone who has an LSAT 5 points below you or whatever. It's 98th vs 96th percentile, so really only a 2 percent difference in apptitude (read: no difference at all) vs what you might be thinking of as a 50% difference if your just looking at the 25/75's.

Also- what Rad said.

A 2 percent difference in aptitude can be enormous at the higher end. The person with the higher score isn't necessarily better than the other person but they seemingly would have better odds in law school. I think Stanford might be one of the better examples of a wide score range at the top. A great deal of low 170's and high 160's would be competing with a bunch of high 170's.

[Pneumonia]

Also keep in mind that, while schools place a lot of weight on 25/75's, objectively the difference is minimal. In admissions the difference between a 168 and a 172 is huge, but don't let that full you into thinking you are smarter or better than someone who has an LSAT 5 points below you or whatever. It's 98th vs 96th percentile, so really only a 2 percent difference in apptitude (read: no difference at all) vs what you might be thinking of as a 50% difference if your just looking at the 25/75's.

Also- what Rad said.

A 2 percent difference in aptitude can be enormous at the higher end. The person with the higher score isn't necessarily better than the other person but they seemingly would have better odds in law school. I think Stanford might be one of the better examples of a wide score range at the top. A great deal of low 170's and high 160's would be competing with a bunch of high 170's.

Nope, sorry. Some differences would start to emerge between 178/168, but those aren't the 25/75's. There really isn't much difference in aptitude between 25/75's. It seems like you're more interested in the predictive value of your score in which case I would again refer you to what Rad Lulz said.

[patogordo]

A 2 percent difference in aptitude can be enormous at the higher end. The person with the higher score isn't necessarily better than the other person but they seemingly would have better odds in law school.

no. this is dumb. this are the results your opinion is based on:

During 2010, validity studies were conducted for 189 lawschools. Correlations between LSAT scores and first-year lawschool grades ranged from .12 to .56 (median is .36). The correlations between UGPA and first-year law grades rangedfrom .09 to .45 (median is .28).

you simply cannot make any significant comparison between two law students with similar LSAT scores, especially if you completely ignore other factors like UGPA.

[ph14]

Based solely on anecdotal evidence, I would say very little.

[Tiago Splitter]

Both LSAT and GPA have predictive value. It's higher for the LSAT overall, but that may not be the case at any one school. In any event, most people with LSAT scores well above the median have below-median GPAs and vice versa, so whatever you can predict from LSAT scores alone is largely lost when you factor in uGPA.

In another thread someone calculated that someone at Harvard's medians (3.9/173), even assuming the most powerful correlations, could only expect to move up about 5% if they went to Georgetown. So if they finished their first year at median at Harvard finishing at about the 45th percentile at Georgetown would be expected. In other words, when deciding between two legitimate options, these factors should be ignored.

[cotiger]

Here is the LSAT validity study: http://www.lsac.org/docs/default-source ... -13-03.pdf

The average correlation between LSAT alone and first-year grades is .36
The average correlation between LSAT/UGPA combined and first-year grades is .47

The correlations vary between schools, however.
In 2012, the range of correlations for LSAT vs FYA was .19 - .55
The range of correlations for LSAT/UGPA vs FYA was .31 - .63

Other things of note in this report:
-People have pointed out that there is are range restrictions that decrease the predictive abilities of your numbers (e.g. a school only has students with LSATs from 160-180 instead of 140-180, so the real differences between students decrease). LSAC attempts to correct for that, and finds that without that range restriction, the correlation coefficient between LSAT and FYA would increase to .58 from the .36 observed.
-Schools that have a higher correlation between LSAT and UGPA have higher validity coefficients. That is, the more splitters a school has, the less predictive numbers become.

[cotiger]

Additionally, a .47 correlation between LSAT/UGPA and FYA means that 22% of the variation in first-year grades is explained by your numbers. The range of correlations of .31-.63 gives a range of 10%-40% for explanatory power.

Using that average number of 22%, if your entering numbers are at the 90th percentile of your class, you would expect to be at the 59th percentile for first-year grades (so 41% of the class would be above you).

Higher ranked law schools will tend to be at the low end of those correlations, however, as they have much more compressed classes, numbers wise. For instance, Harvard's 170-175 LSAT interquartile range is only 2 percentiles wide. The University of Houston, to pick a random school, also has a 5 point interquartile range (157-162), but that encompasses 15 percentiles worth of test-takers. Schools with wider ranges, such as the University of Minnesota (IQR of 28 percentiles), will tend to be at the higher end of those correlations.

[JCougar]

There's a study you can google ab this

But lemme save you some hassle and give you the takeaway

Although an LSAT way higher than everyone else's increased your chance of doing better relatively

Job prospects drop off a lot faster than LSAT medians and 75ths

So it basically doesn't make sense to go to a worse school and try to kill it there

I understand that but I am primarily concerned about this question for the top schools. Columbia, Chicago, Harvard, and Stanford do not seem to have the immense employment drop off between schools. At least, not when you throw in money for Columbia and Chicago. Is a 170 going to have much worse odds of hitting median or some other platform at Harvard than at Stanford?

At these top schools, you have a restriction of range problem coupled with what I would guess is a ceiling effect with the criterion measure (first year grades). As for the ceiling effect, LSAT measures intelligence with logical reasoning, but you only need a certain floor level of reasoning ability to do well on the logical challenges you'll encounter on law school exams. I don't have any data, but I'd bet with LSAT scores above the 95th percentile, it's predictive power drops pretty close to zero. And at all the schools you mention, virtually all of the students are above the 95th percentile.

Even without the ceiling effect, though, you have a restriction of range problem. LSAT has been shown to predict about 35% of the variance in your first year grades on average. But that predictive power is virtually nothing if you're talking about it across 2 or 3 LSAT points. LSAC even admits that LSAT scores have a 2.7 point confidence interval, so any two LSAT scores within 5 points of each other are statistically insignificant. If you're comparing to an entire class' median, than that confidence interval shrinks as the sample size increases, but your own LSAT score still retains its 2.7-point CI.

Where LSAT has predictive power is if you go to a lower-ranked school with a score substantially above the school's median...like somewhere around 5-10 points. And even then, it has limited predictability. Like if your LSAT score is something like 10 points above median, there's a good chance you'll end up above median, but it's still unclear whether you will more likely than not finish in the top 10%.

tl;dr version: for the schools you are looking at, I doubt there's going to be any measurable effect at all.

[cotiger]

At these top schools, you have a restriction of range problem coupled with what I would guess is a ceiling effect with the criterion measure (first year grades). As for the ceiling effect, LSAT measures intelligence with logical reasoning, but you only need a certain floor level of reasoning ability to do well on the logical challenges you'll encounter on law school exams. I don't have any data, but I'd bet with LSAT scores above the 95th percentile, it's predictive power drops pretty close to zero. And at all the schools you mention, virtually all of the students are above the 95th percentile.

Even without the ceiling effect, though, you have a restriction of range problem. LSAT has been shown to predict about 35% of the variance in your first year grades on average.

It's actually not even that much. It has a .36 correlation with first year grades. That means that it only explains 13% of the variance in first-year grades.

And you're absolutely right about the restriction of range reducing the predictive quality even further. The top schools are most likely going to be the ones with the correlations of around .2, which means that the LSAT only explains a measly 4% of the variance in first-year grades.

[09042014]

Range restriction would likely increase the correlation.

However, if you studied anything more than average ( real average not TLS average) for the LSAT your predictiveness is zero. You gamed the test.

[kingjones59]

I've said it before and I'll say it again. Law school tests (all that matter) are not hard. Anyone with a 160+ can make the "logical leaps" needed to get an A on a T-14 exam. Anecdotal Evidence: 165, Top 25%, my buddy with a 178 is bottom 10%. Work hard and get over the TLS-circle jerk that is LSAT scores.

[JCougar]

At these top schools, you have a restriction of range problem coupled with what I would guess is a ceiling effect with the criterion measure (first year grades). As for the ceiling effect, LSAT measures intelligence with logical reasoning, but you only need a certain floor level of reasoning ability to do well on the logical challenges you'll encounter on law school exams. I don't have any data, but I'd bet with LSAT scores above the 95th percentile, it's predictive power drops pretty close to zero. And at all the schools you mention, virtually all of the students are above the 95th percentile.

Even without the ceiling effect, though, you have a restriction of range problem. LSAT has been shown to predict about 35% of the variance in your first year grades on average.

It's actually not even that much. It has a .36 correlation with first year grades. That means that it only explains 13% of the variance in first-year grades.

And you're absolutely right about the restriction of range reducing the predictive quality even further. The top schools are most likely going to be the ones with the correlations of around .2, which means that the LSAT only explains a measly 4% of the variance in first-year grades.

Hmmm...I read somewhere else that it was r^2 = .35, not r = .36.

Edit: I guess you linked to the official source. So the correlations are from individual schools (so range restriction isn't an issue--although I guess it still could be at splitter schools where the top half of the class is jammed into a 2-3 point LSAT range, but the bottom half of the class spans about 10 points, in which case you'd have range restriction problems within the above median section of the class). But the number reported was the correlation, not the coefficient of determination. So these two misconceptions of mine kind of cancel each other out...sort of...the r^2 thing weights a bit more in this sense.

I wonder if LSAC would ever just release the numbers for each individual school. That would be really interesting.

[JCougar]

Range restriction would likely increase the correlation.

It depends on which correlation you're initially talking about. If you're talking about each school's individual correlation in the first place, then yes it would. If you're talking about the general population correlation, than each school's individual correlation would be lower.

It looks like the numbers LSAC released were based on each school's own correlation, before adjusting for range restriction...which surprised me, because I always thought the numbers they released were for the general population.

[spleenworship]

Based solely on anecdotal evidence, I would say very little.