## Better 25th% at CLS or 75th% at Umich

(Rankings, Profiles, Tuition, Student Life, . . . )
JG Hall

Posts: 362
Joined: Mon Jan 25, 2010 11:18 pm

### Re: Better 25th% at CLS or 75th% at Umich

dabbadon8 wrote:Think of it in terms of percentiles, when your talking about LSAT scores that high you are talking about the 99th percentile. I don't think the LSAT can be that predictive when your already so far right on the bell curve. Also, Imagine an applicant who took the LSAT 3 time after a full year of studying and got a 162, 167, 174, and then some one who took it after a month of prep got a 169 and called it a day. Both had a 3.8, one is going to UMich another to columbia. I couldn't tell you who I'd rather compete against. I can tell you that I am going to umich and am 2 points over their 75th from last year and I am worried about competing against such a qualified pool.

The answer to your question is the kid with the 169, because the kid with the 162, 167, and 174 is probably going to try a lot harder.

Patriot1208

Posts: 7023
Joined: Tue May 18, 2010 11:28 am

### Re: Better 25th% at CLS or 75th% at Umich

d34dluk3 wrote:
Patriot1208 wrote:
d34dluk3 wrote:
Patriot1208 wrote:I've taken three statistics classes, I understand it would be considered relevant. But, a regression analysis that gives us a .4 correlation is one that i'm not using as evidence for, pretty much anything other than I need more variables.

Actually, in a complex system like this, a correlation that strong is a huge predictor (since even with all possible input variables included, you typically have 0.3-0.6 of the regression going to random noise).

This is where your assumption dies. This isn't a complex system. It's a fairly simple analyiss based on 1 or 2 variables. If you were an economist that was studying all the variables in consumer preferences, i'd agree with you. Since you are taking one to two variables and simply comparing them to class rank, a .33 correlation (according to the LSAC article) is not very good.

No, the system of 1L grades is complex. Obviously there are more inputs than we can account for and the human factor involved in the performance.

But the analysis is not. They just took the grades and matched them up with their LSAT anc gpa scores. It's about as easy of a statistical analysis as you can do. Any undergraduate who takes a couple statistics course could do this Nor is a .33 correlation a good predictor in such a simple linear analysis.

The Real Jack McCoy

Posts: 279
Joined: Wed Aug 18, 2010 3:55 pm

### Re: Better 25th% at CLS or 75th% at Umich

Alex-Trof

Posts: 525
Joined: Fri Jan 07, 2011 1:42 am

### Re: Better 25th% at CLS or 75th% at Umich

The Real Jack McCoy wrote:http://www.lsac.org/LsacResources/Research/TR/tech-reports.asp#correlation

Thanks. I looked through those articles and it looks like the correlation for LSAT score is .33. The articles also seem to indicate that correlations vary by the schools. I was wondering if anybody had any idea on whether this correlation would be weaker or stronger for top schools.

P.S. I was not really worried reading this thread, but after googling this question came across a 2004 discussion from LSD where everyone many seemed to believe that people who performed below medium were less capable to do well. The idea sort of was that LSAT shows schools maximum possible performance while GPA shows effort.

d34d9823

Posts: 1879
Joined: Wed Apr 14, 2010 2:52 pm

### Re: Better 25th% at CLS or 75th% at Umich

Alex-Trof wrote:The idea sort of was that LSAT shows schools maximum possible performance while GPA shows effort.

This is one of those stupid pop culture generalizations. Intelligence is too complicated to be quantified in a single way, and both GPA and LSAT require both intelligence and effort.

JG Hall

Posts: 362
Joined: Mon Jan 25, 2010 11:18 pm

### Re: Better 25th% at CLS or 75th% at Umich

d34dluk3 wrote:
Alex-Trof wrote:The idea sort of was that LSAT shows schools maximum possible performance while GPA shows effort.

This is one of those stupid pop culture generalizations. Intelligence is too complicated to be quantified in a single way, and both GPA and LSAT require both intelligence and effort.

meh... usually, not necessarily

ihp12

Posts: 247
Joined: Sun Jun 20, 2010 5:33 pm

### Re: Better 25th% at CLS or 75th% at Umich

There might be something to the idea that the rigidity of 1L grading curves prevents better measurement of different variables' effect upon actual 1L performance. Generally speaking, the majority of students are lumped together in the C-B range, making it hard to measure the effect of variables upon performance since performance itself is being poorly measured.

That said, I think that the formula for predicting 1L grades is very complex (ie there are alot of variables interacting). The fact that the best variable we have predicts only 35 % of the variance in the data, and the next (GPA) is something like .15, says that there is a a lot going on that we can't account for.

Getting to the heart of the question of whether you can expect to do better at a school at which you have an LSAT score that is a few points higher than the median student at that school, since the standard error is 2.6 points, a difference in 2-3 points would not, statisiticly speaking, have any predictive power. Would 4 points be significant? On average yes, but only slightly so.

ihp12

Posts: 247
Joined: Sun Jun 20, 2010 5:33 pm

### Re: Better 25th% at CLS or 75th% at Umich

Patriot1208 wrote:
But the analysis is not. They just took the grades and matched them up with their LSAT anc gpa scores. It's about as easy of a statistical analysis as you can do. Any undergraduate who takes a couple statistics course could do this Nor is a .33 correlation a good predictor in such a simple linear analysis.

I think what he means is that .33 is a relatively strong predictor, i.e. it is strong relative to any other variable in the equation. Also I could be wrong, but I'm pretty sure that this is a high correlation for a social science type inquiry. In the hard sciences they may get above .75, but in social sciences they tend to be weak.

Agreed that deriving the correlation is easy

d34d9823

Posts: 1879
Joined: Wed Apr 14, 2010 2:52 pm

### Re: Better 25th% at CLS or 75th% at Umich

ihp12 wrote:
Patriot1208 wrote:
But the analysis is not. They just took the grades and matched them up with their LSAT anc gpa scores. It's about as easy of a statistical analysis as you can do. Any undergraduate who takes a couple statistics course could do this Nor is a .33 correlation a good predictor in such a simple linear analysis.

I think what he means is that .33 is a relatively strong predictor, i.e. it is strong relative to any other variable in the equation. Also I could be wrong, but I'm pretty sure that this is a high correlation for a social science type inquiry. In the hard sciences they may get above .75, but in social sciences they tend to be weak.

Agreed that deriving the correlation is easy

Generally, yeah. The important thing to realize is that the sum of the correlation factors for all completely independent variables is 1. If you're having trouble seeing this, note that a system with a correlation of 1 is completely determined and thus has no remaining independent input variables that are significant at any level. Given a real-world, complex problem like this, it is a good assumption that 0.3-0.6 of the correlation will be indeterminate due to unmeasurable variables and noise. Thus, even a 0.33 represents 50-80% of the entire available correlation strength of the system. It may not be considered strong in the absolute sense, but it's a huge proportion of the strength that is available.