What Does Your Major Matter? Forum

[FiveSermon]

But you post your bullshit EVERYWHERE. Not reading what you write would require me to stop reading TLS completely.



Which, actually, would probably be a good thing. Maybe I should thank you.

This made me snicker.

[spacepenguin]

But you post your bullshit EVERYWHERE. Not reading what you write would require me to stop reading TLS completely.



Which, actually, would probably be a good thing. Maybe I should thank you.

r6 actually has a very good grasp of statistics--though I think he's living in the statistical dark ages of bias sampling, or the people he worked with all got their Phds in statistics in the 1970s..


I think people are being kind of harsh either way

[r6_philly]

Have you ever worked with determining insurance premiums as a function of risk/genes? Actuarial science is constantly in the world of 'could.' Either way, unbiased sample>>>>>>>>biased sample. However, there are correcting mechanisms one can implement to 'correct' for bias...just saying.

Yes there are, but not on LSN so it is not within the scope of what I am arguing. It would be hard to correct LSN because it's the only sample. If you had multiple samples we can test/cross section but not in this case.

Insurance companies can take multiple random samples from their biased sample and develop a distribution. But their bias isn't a self-selection bias. LSN self-selection bias is bad because people who self-select into LSN sample are likely to do better in school and have better GPAs (now this is a pure hunch)

[r6_philly]

But you post your bullshit EVERYWHERE. Not reading what you write would require me to stop reading TLS completely.



Which, actually, would probably be a good thing. Maybe I should thank you.

This made me snicker.

Yes it would help you both. Glad to be of service.

[spacepenguin]

Have you ever worked with determining insurance premiums as a function of risk/genes? Actuarial science is constantly in the world of 'could.' Either way, unbiased sample>>>>>>>>biased sample. However, there are correcting mechanisms one can implement to 'correct' for bias...just saying.

First time I've ever seen "I'm an actuary" used to boost E-peen.

Thinking I'm an actuary is the biggest compliment I've ever been given.. The jocks of the math world. The rich mathematicians. The smart kids who know applicable math. No, but seriously, I have a raging math boner.

[r6_philly]

LSN may not be perfect, but I agree with the sentiment that some data, even if it is flawed to a certain extent, is better than no data at all.

That is a sentiment that is not shared by people who actually use data because you can't determine how flaw it is. How do you determine the extent it is flawed? If you know it, then you know how to correct it (or know the actual data), then you wouldn't need to use the flawed data.

[spacepenguin]

Have you ever worked with determining insurance premiums as a function of risk/genes? Actuarial science is constantly in the world of 'could.' Either way, unbiased sample>>>>>>>>biased sample. However, there are correcting mechanisms one can implement to 'correct' for bias...just saying.

Yes there are, but not on LSN so it is not within the scope of what I am arguing. It would be hard to correct LSN because it's the only sample. If you had multiple samples we can test/cross section but not in this case.

Insurance companies can take multiple random samples from their biased sample and develop a distribution. But their bias isn't a self-selection bias. LSN self-selection bias is bad because people who self-select into LSN sample are likely to do better in school and have better GPAs (now this is a pure hunch)

All I really wanted was your reasoning, and I'm glad you've given it. Awesome answer. Though, just for curiosity sake, I'd like you to answer whether the differences between published numbers and LSN are statistically significant.

[FiveSermon]

LSN may not be perfect, but I agree with the sentiment that some data, even if it is flawed to a certain extent, is better than no data at all.

That is a sentiment that is not shared by people who actually use data because you can't determine how flaw it is. How do you determine the extent it is flawed? If you know it, then you know how to correct it (or know the actual data), then you wouldn't need to use the flawed data.

I'm quoting this so I can just point out how ridiculous a claim you make. So people who "actually use data", will NEVER use flawed data? So every data they use must be perfect if every data they use is not flawed. I think reality is proof enough that you are wrong.

[sundance95]

LSN may not be perfect, but I agree with the sentiment that some data, even if it is flawed to a certain extent, is better than no data at all.

That is a sentiment that is not shared by people who actually use data because you can't determine how flaw it is. How do you determine the extent it is flawed? If you know it, then you know how to correct it (or know the actual data), then you wouldn't need to use the flawed data.

Disagree. Tell me Duke doesn't have a hard 3.5 GPA floor when you look at this.
http://duke.lawschoolnumbers.com/stats/?whichCycle=1011

We aren't conducting regression analysis, we're just trying to get a picture into what admissions committee's think. LSN is imperfect but it's definitely the best we have, and it's definitely better than nothing.

[FiveSermon]

LSN may not be perfect, but I agree with the sentiment that some data, even if it is flawed to a certain extent, is better than no data at all.

That is a sentiment that is not shared by people who actually use data because you can't determine how flaw it is. How do you determine the extent it is flawed? If you know it, then you know how to correct it (or know the actual data), then you wouldn't need to use the flawed data.

Disagree. Tell me Duke doesn't have a hard 3.5 GPA floor when you look at this.
http://duke.lawschoolnumbers.com/stats/?whichCycle=1011

We aren't conducting regression analysis, we're just trying to get a picture into what admissions committee's think. LSN is imperfect but it's definitely the best we have, and it's definitely better than nothing and hunches/speculations of R6 based on very little info. Although R6 ardently argues that his hunches are superior..

[r6_philly]

All I really wanted was your reasoning, and I'm glad you've given it. Awesome answer. Though, just for curiosity sake, I'd like you to answer whether the differences between published numbers and LSN are statistically significant.

I didn't model it because well the cardinal sin in my work is to start with only a biased sample. I just read 25/median/75 publsihed and had a hunch. I didn't test the hunch so it stays a hunch/guess/untested null hypothesis lol

Ok I just went and modeled 2009-10 cycle for CLS

GPA 25/50/75 (217 samples)
3.72/3.82/3.92

Published
3.58/3.72/3.82

DISCUSS!

[sundance95]

All I really wanted was your reasoning, and I'm glad you've given it. Awesome answer. Though, just for curiosity sake, I'd like you to answer whether the differences between published numbers and LSN are statistically significant.

I didn't model it because well the cardinal sin in my work is to start with only a biased sample. I just read 25/median/75 publsihed and had a hunch. I didn't test the hunch so it stays a hunch/guess/untested null hypothesis lol

Ok I just went and modeled 2009-10 cycle for CLS

GPA 25/50/75
3.72/3.82/3.92

Published
3.58/3.72/3.82

DISCUSS!

I don't think we're using this data to try and determine the answers to questions that already have answers, i.e., have been published. We're just trying to get a window into adcomms' thinking.

[r6_philly]

LSN may not be perfect, but I agree with the sentiment that some data, even if it is flawed to a certain extent, is better than no data at all.

That is a sentiment that is not shared by people who actually use data because you can't determine how flaw it is. How do you determine the extent it is flawed? If you know it, then you know how to correct it (or know the actual data), then you wouldn't need to use the flawed data.

Disagree. Tell me Duke doesn't have a hard 3.5 GPA floor when you look at this.
http://duke.lawschoolnumbers.com/stats/?whichCycle=1011

We aren't conducting regression analysis, we're just trying to get a picture into what admissions committee's think. LSN is imperfect but it's definitely the best we have, and it's definitely better than nothing.

Actually, scroll up, I said it could be a flawed individual predictor. The line between in and out could be more relevant in a biased sample, but it is more usable because you have 1 biased IN sample and 1 biased OUT sample. So you can cross the 2 samples. So if you want to see someone with your numbers would be in or out (as in testing the floor) you are not trying to predict the whole population - it isn't the same analysis.

[r6_philly]

I don't think we're using this data to try and determine the answers to questions that already have answers, i.e., have been published. We're just trying to get a window into adcomms' thinking.

Agree. But this stupid argument was about whether LSN is representative of the whole or not. (it is not)

I use LSN all the time. You don't need statistically valid anything when you are guessing what other people are thinking.

[r6_philly]

LSN may not be perfect, but I agree with the sentiment that some data, even if it is flawed to a certain extent, is better than no data at all.

That is a sentiment that is not shared by people who actually use data because you can't determine how flaw it is. How do you determine the extent it is flawed? If you know it, then you know how to correct it (or know the actual data), then you wouldn't need to use the flawed data.

I'm quoting this so I can just point out how ridiculous a claim you make. So people who "actually use data", will NEVER use flawed data? So every data they use must be perfect if every data they use is not flawed. I think reality is proof enough that you are wrong.

People in the business are very careful about how they collect data. It sound's ridiculous, but 1 basis point on the model can be millions of dollars. You can't afford to be off. By using valid data, you are more likely to end up near the mode of the probability distribution, which gives you the best odds of being right.

[spacepenguin]

All I really wanted was your reasoning, and I'm glad you've given it. Awesome answer. Though, just for curiosity sake, I'd like you to answer whether the differences between published numbers and LSN are statistically significant.

I didn't model it because well the cardinal sin in my work is to start with only a biased sample. I just read 25/median/75 publsihed and had a hunch. I didn't test the hunch so it stays a hunch/guess/untested null hypothesis lol

Ok I just went and modeled 2009-10 cycle for CLS

GPA 25/50/75 (217 samples)
3.72/3.82/3.92

Published
3.58/3.72/3.82

DISCUSS!

Without doing any statistical test on this..it seems like there's a positive bias--which fits the idea that LSN is populated with overachieving individuals. However, you can still extrapolate information from that sample considering that positive bias would only benefit those who look on LSN as a source of information. As long as the bias is uniform, you could still calculate population parameters and create a model as long as you keep that in mind.

[retake]

Glad to see a completely original topic on TLS that has never, ever been discussed before or anywhere else. Keep racking up the pages, people. You're doing it right.

[nshapkar]

once you are in a quartile it should not matter. A 3.9 mathematician at Princeton > 4.0 woman's studies major at shit state school

No. Remember these are law schools. A 3.9 mathematician could be very, very unsuitable to study law. It depends on the person. 4.0 is also not a good indicator to compare because the 4.0 student may not be anywhere near the upper bound of his/her potential.

3.8 Princeton math > 3.9 state BFA I will give you.

My roommate is a math major not sure exactly what his GPA is but I know its above a 3.5. One day he came home and asked to take one of my practice tests for kicks. He had never seen an LSAT before that or known the breakdown of the test. He took it in the living room where our other roommate was cooking, playing music, and talking to him while he was taking the test. I timed my buddy, he finished every section with 7-10 minutes to spare, scored a 170. I can't even imagine what he would've gotten in complete silence. Either way, math majors are really suited for the LSAT. In fact, I can't find the study at the moment, but math and philosophy majors score highest on LSAT. History and PoliSci are around 8-9th.

[FiveSermon]

once you are in a quartile it should not matter. A 3.9 mathematician at Princeton > 4.0 woman's studies major at shit state school

No. Remember these are law schools. A 3.9 mathematician could be very, very unsuitable to study law. It depends on the person. 4.0 is also not a good indicator to compare because the 4.0 student may not be anywhere near the upper bound of his/her potential.

3.8 Princeton math > 3.9 state BFA I will give you.

My roommate is a math major not sure exactly what his GPA is but I know its above a 3.5. One day he came home and asked to take one of my practice tests for kicks. He had never seen an LSAT before that or known the breakdown of the test. He took it in the living room where our other roommate was cooking, playing music, and talking to him while he was taking the test. I timed my buddy, he finished every section with 7-10 minutes to spare, scored a 170. I can't even imagine what he would've gotten in complete silence. Either way, math majors are really suited for the LSAT. In fact, I can't find the study at the moment, but math and philosophy majors score highest on LSAT. History and PoliSci are around 8-9th.

There is some person on TLS who has a 3.9 as a math major and can't get above a 158 after multiple LSATs.

[nshapkar]

once you are in a quartile it should not matter. A 3.9 mathematician at Princeton > 4.0 woman's studies major at shit state school

No. Remember these are law schools. A 3.9 mathematician could be very, very unsuitable to study law. It depends on the person. 4.0 is also not a good indicator to compare because the 4.0 student may not be anywhere near the upper bound of his/her potential.

3.8 Princeton math > 3.9 state BFA I will give you.

My roommate is a math major not sure exactly what his GPA is but I know its above a 3.5. One day he came home and asked to take one of my practice tests for kicks. He had never seen an LSAT before that or known the breakdown of the test. He took it in the living room where our other roommate was cooking, playing music, and talking to him while he was taking the test. I timed my buddy, he finished every section with 7-10 minutes to spare, scored a 170. I can't even imagine what he would've gotten in complete silence. Either way, math majors are really suited for the LSAT. In fact, I can't find the study at the moment, but math and philosophy majors score highest on LSAT. History and PoliSci are around 8-9th.

There is some person on TLS who has a 3.9 as a math major and can't get above a 158 after multiple LSATs.

He may not be representative of all math majors and my roommate may just be a genius.