Law School Discussion

LSAT/IQ Conversion Table

Hocine

  • ****
  • 336
    • View Profile
Re: LSAT/IQ Conversion Table
« Reply #50 on: October 29, 2004, 12:58:22 PM »
Ah okay, "If IQ doesn't strongly enough predict LSAT performance such that a correlation between the two can be drawn, how can it be said to measure anything meaningful? " tripped me up, but I see what you're saying now.

Sure IQ tests test some of the basic skills that everyone has, to some degree, and which are required (to some degree) for the LSAT

For example, spatial reasoning skills can be helpful on games sections, and can save you a lot of time (you can do more work in your head)

Memory is required for all the sections, and having an extremely good memory could save you a LOT of time

All of the parts of the IQ test are timed, and your ability to think quickly is factored in to your score.. which is obviously helpful on the LSAT


So you see how IQ tests test the basic things that make it easier for smart people to get good scores, but don't necessarily result in good scores?

Re: LSAT/IQ Conversion Table
« Reply #51 on: October 29, 2004, 12:59:01 PM »
I don't give a sh*t about all this nonsense.

All I know is, according to MENSA (and to every test I've taken that MENSA lists), I'm a genius.  This is all I care about. Who's with me?  GRE, LSAT, SAT, MAT.  I'm a f-in' genius, bitches.  And this confirms my actual IQ test (143).  Genius!

If you don't like my post, you must not be a genius.

:)

Re: LSAT/IQ Conversion Table
« Reply #52 on: October 29, 2004, 12:59:41 PM »
Look. The correlation between the LSAT and IQ tests is very high, on the order of .8 or .9. It's not perfect, but if you know statistics, you'll know that a .8 or .9 correlation is huge. The old SAT, which was a de facto IQ test and was designed as such, had a lower IQ correlation than this. Many IQ tests themselves have a lower correlation with each other than this. The LSAT, for all intents and purposes, quite simply is an IQ test.

As long as two measures are correlated in any way at all, it's possible to derive a formula that "converts" between them. This does not mean that you can plug in your LSAT score and out pops your IQ, accurate to within four significant figures. It just means that the formula is able to predict scores better than any other formula. If there were no correlation, this wouldn't be possible. The higher the correlation, the more accurate the predictions. With a .8 correlation, the predictions should be pretty damn accurate for a lot of people.

Thus, if a formula doesn't predict your score, or even if it's *way* off, that does not mean there is no correlation to be found between LSAT and IQ or LSAT and SAT. In any statistical sampling, there will be outliers. You may be one of them. I just don't understand how supposedly intelligent people, bound for law school and some even in MENSA (gasp!), can fail to grasp this. Anecdotal evidence in the form of single data points that don't fit the curve prove nothing. It's that sort of thinking that causes someone to bomb the LSAT in the first place.

Re: LSAT/IQ Conversion Table
« Reply #53 on: October 29, 2004, 01:09:47 PM »
Another thing: some people aren't fully grasping the idea of correlation. If we want to convert between scores, all we care about is correlation. Simple vocabulary tests have one of the highest correlations with raw IQ at over .9. If I gave a good enough vocabulary test to 1,000 people, the distribution of scores would match the distribution of their IQ's very closely. Obviously though, vocabulary is not IQ. Word knowledge is a very tiny part of what we want to call "intelligence," and it's pretty easy to improve one's vocabulary too. There's certainly no a priori reason why vocabulary tests should correlate so highly with IQ tests, but the fact is they do, and so they have very good predictive value for IQ. Same idea with the LSAT. It doesn't matter if the LSAT tests a very small subset of what appears on IQ tests -- hell, it could test movie trivia for all we care -- because as long as the correlation is high enough, it suits our purposes, regardless of the content.

Hocine

  • ****
  • 336
    • View Profile
Re: LSAT/IQ Conversion Table
« Reply #54 on: October 29, 2004, 01:28:26 PM »
I agree that smarter people tend to do better on this test, since it's a test of mental ability, and not of knowledge (except for knowledge of the english language)..

That's all fine and dandy, but you can't apply it to individuals.
"So, Mr. Johnson, you got a 135 on the LSAT?  There is a 73.2% chance that you are mentally retarded."

All you can really do with the correlation is make broad overgeneralizations about groups of people...

Re: LSAT/IQ Conversion Table
« Reply #55 on: October 29, 2004, 01:37:28 PM »
I agree that smarter people tend to do better on this test, since it's a test of mental ability, and not of knowledge (except for knowledge of the english language)..

That's all fine and dandy, but you can't apply it to individuals.
"So, Mr. Johnson, you got a 135 on the LSAT?  There is a 73.2% chance that you are mentally retarded."

All you can really do with the correlation is make broad overgeneralizations about groups of people...

Bayesian reasoning. If the formula works for most people, and you're a person, it probably works for you as well. Everything is within a margin of error too, of course. LSAT score reports come with a scoring band -- the range you're likely to score in if you took the test again -- and if you convert that band into an IQ range, it's a reasonable bet that your IQ falls somewhere in there.

Hocine

  • ****
  • 336
    • View Profile
Re: LSAT/IQ Conversion Table
« Reply #56 on: October 29, 2004, 01:51:41 PM »
Short story: The confidence interval is too small, it needs to be at least 95% for this to be statistically significant.

I find it rather improbable that research on this subject would yield such a conversion system without at least one of the two following problems:

An extremely broad IQ range (i.e. "your IQ is x, plus or minus one standard deviation")

or

a very very high "p" value (low confidence interval), which would render it not statistically significant .. The highest you can get away with is .05 (95% confidence), but to be taken seriously .01 or .001 are preferred.


And since you've said that the test which most highly correlates with IQ (spelling?) only has a confidence interval of 90%, IT isn't even statistically significant..

So, in order to make this statistically significant, you'd have to have each LSAT score correspond to a very very large range of IQ scores, just to reduce the number of errors you get..

And that would render the conversion useless..


Hocine

  • ****
  • 336
    • View Profile
Re: LSAT/IQ Conversion Table
« Reply #57 on: October 29, 2004, 01:52:02 PM »
Oops, double posted.

I'll edit this and continue here.

So, since this can not be statistically significant without a huge range of IQ scores for your LSAT to correspond to, you're relegated to telling people that they have a 95% or 99% chance of falling into a HUGE HUGE range of IQ scores, or telling them that they have a much lower % chance of having a certain IQ.

Either way, this is not helpful when dealing with individuals.


Re: LSAT/IQ Conversion Table
« Reply #58 on: October 29, 2004, 02:56:45 PM »
Short story: The confidence interval is too small, it needs to be at least 95% for this to be statistically significant.

...

And that would render the conversion useless..

You're confusing confidence interval with correlation. The .9 for vocabulary tests and the .8 and .9 for the LSAT are the correlation coefficents, NOT the confidence intervals. I didn't do the original studies. I don't know what the confidence intervals were, but it's safe to assume they were above 95%. Psychologists don't publish findings with lower confidence intervals and try to pass them off as significant. That doesn't do good things for one's career.

With a .9 correlation between LSAT and IQ, that means that .9 x .9 = 81% (or 64% for a correlation coefficient of .8 ) of the variance in one test can be predicted by the other. Any correlation above .5 is generally considered strong. .8 or .9 are very strong.

That said, this is all rather tangential to the point. I didn't come up with these equations through regression analysis, and they're not intended as serious statistical predictors. I did it as a diversion, and it's a simple matching up of percentiles. The assumption is that with a high enough correlation (which I have reason to believe exists), the percentile pairings will yield fairly accurate results for most people.

More importantly, the fact that IQ is nothing but one's score on any one of several qualifying tests, none of which correlates with the others much better than the LSAT does, gives no good reason not to take the LSAT as a valid indicator of IQ as well. If I get an IQ score of 140 on the Raven's Progressive Matrices, and you get a 140 on the Stanford-Binet, why are you willing to consider these scores commensurable but not ones obtained through LSAT conversions? If the correlations are the same, the scores should be treated the same.

Bisquick

  • ****
  • 1660
    • View Profile
Re: LSAT/IQ Conversion Table
« Reply #59 on: October 29, 2004, 03:04:38 PM »
you people are all confusing IQ with mattering!