Yes, and I’m familiar with your IQ-related links in the OP*. But what’s the opposite here? Let me make sure my position is clear: I agree that the people who post on LW are noticeably cleverer than the people that post elsewhere on the internet.
The narrow claim that I’m making is that the average self-reported IQ is almost definitely an overestimate of the real average IQ of people who post on LW, and a large change towards the likely true value in an unreliable number should not be cause for alarm. The primary three pieces of evidence I submit are:
On this survey, around a third of people self-reported their IQ, and it’s reasonable to expect that there is a systematic bias, such that people with higher perceived IQs are more likely to share them. I haven’t checked how many people self-reported on previous surveys, but it’s probably similarly low.
When you use modern conversion numbers for average SAT scores, you get a reasonable 97th percentile for the average LWer. Yvain’s estimate used a conversion chart from two decades ago; in case you aren’t familiar with the history of psychometric testing, that’s when the SAT had its right tail chopped off to make the racial gap in scores less obvious.
The correlation between the Raven’s test and the self-reported IQ scores is dismal, especially the negative correlation for people without positive LW karma. The Raven’s test is not designed to differentiate well between people who are more than 99th percentile (IQ 135), but the mean score of 127 (for users with positive karma) was 96th percentile, so I don’t think that’s as serious a concern.
* I rechecked the comment you linked to in the OP, and I think it was expanded since I read it first. I agree that more than half of people provided at least one IQ estimate, but I think that they should not be weighted uniformly; for example, using the self-reported IQ to validate the self-reported IQ seems like a bad idea! It might be interesting to see how SAT scores and age compare- we do have a lot of LWers who presumably took the SAT before it was dramatically altered, and with younger LWers we can compare scores out of 1600 to scores out of 2400. It’s not clear to me how much more clarity this will give, though, and how much the average IQ of LW survey responders actually matters.
What IQ would you correlate to the SAT numbers, considering?
As for the Raven’s numbers, I am not sure where you’re getting them from. I don’t see a column when searching for “raven” in the 2012 spreadsheet, nor do I see “raven” on the survey result threads.
What IQ would you correlate to the SAT numbers, considering?
SAT scores, combined with the year that it was taken in, give you a percentile measure of that person (compared to test-takers, which is different from the general population, but in a fairly predictable way), which you can then turn into an IQ-equivalent.
I say equivalent because there are a number of issues. First, intelligence testing has a perennial problem that absolute intelligence and relative intelligence are different things. Someone who is 95th percentile compared to high school students in 1962 is not the same as someone who is 95th percentile compared to high school students in 2012. It might also be more meaningful to say something like “the median LWer can store 10 numbers in working memory, compared to the general population’s median of 7” instead of “the median LWer has a working memory that’s 95th percentile.” (I also haven’t looked up recently how g-loaded the SAT is, and that could vary significantly over time.)
Second, one of the main benefits may not be that the median LWer is able to get into MENSA, but that the smartest LWers are cleverer than most people have had the chance to meet during the lives. This is something that IQ tests are not very good at measuring, especially if you try to maintain the normal distribution. Reliably telling the difference between someone who is 1 out of 1,000 (146) and someone who is 1 out of 10,000 (155) is too difficult for most current tests; how many people would you have to base your test off of to reliably tell that someone is one out of a million (171) from their raw score?
As for the Raven’s numbers, I am not sure where you’re getting them from. I don’t see a column when searching for “raven” in the 2012 spreadsheet, nor do I see “raven” on the survey result threads.
iqtest.dk is based on Raven’s Progressive Matrices; the corresponding column, CV in the public .xls, is called IQTest. I referred to the scores that way because Raven’s measures a particular variety of intelligence. It’s seen widespread adoption because it’s highly g-loaded and culture fair, but a score on Raven’s is subtly different from a total score on WAIS, for example.
This might be virtuous doubt. Have you considered the opposite?
See Also:
Luke’s article on overconfident pessimism.
My IQ related links in the OP.
Yes, and I’m familiar with your IQ-related links in the OP*. But what’s the opposite here? Let me make sure my position is clear: I agree that the people who post on LW are noticeably cleverer than the people that post elsewhere on the internet.
The narrow claim that I’m making is that the average self-reported IQ is almost definitely an overestimate of the real average IQ of people who post on LW, and a large change towards the likely true value in an unreliable number should not be cause for alarm. The primary three pieces of evidence I submit are:
On this survey, around a third of people self-reported their IQ, and it’s reasonable to expect that there is a systematic bias, such that people with higher perceived IQs are more likely to share them. I haven’t checked how many people self-reported on previous surveys, but it’s probably similarly low.
When you use modern conversion numbers for average SAT scores, you get a reasonable 97th percentile for the average LWer. Yvain’s estimate used a conversion chart from two decades ago; in case you aren’t familiar with the history of psychometric testing, that’s when the SAT had its right tail chopped off to make the racial gap in scores less obvious.
The correlation between the Raven’s test and the self-reported IQ scores is dismal, especially the negative correlation for people without positive LW karma. The Raven’s test is not designed to differentiate well between people who are more than 99th percentile (IQ 135), but the mean score of 127 (for users with positive karma) was 96th percentile, so I don’t think that’s as serious a concern.
* I rechecked the comment you linked to in the OP, and I think it was expanded since I read it first. I agree that more than half of people provided at least one IQ estimate, but I think that they should not be weighted uniformly; for example, using the self-reported IQ to validate the self-reported IQ seems like a bad idea! It might be interesting to see how SAT scores and age compare- we do have a lot of LWers who presumably took the SAT before it was dramatically altered, and with younger LWers we can compare scores out of 1600 to scores out of 2400. It’s not clear to me how much more clarity this will give, though, and how much the average IQ of LW survey responders actually matters.
What IQ would you correlate to the SAT numbers, considering?
As for the Raven’s numbers, I am not sure where you’re getting them from. I don’t see a column when searching for “raven” in the 2012 spreadsheet, nor do I see “raven” on the survey result threads.
SAT scores, combined with the year that it was taken in, give you a percentile measure of that person (compared to test-takers, which is different from the general population, but in a fairly predictable way), which you can then turn into an IQ-equivalent.
I say equivalent because there are a number of issues. First, intelligence testing has a perennial problem that absolute intelligence and relative intelligence are different things. Someone who is 95th percentile compared to high school students in 1962 is not the same as someone who is 95th percentile compared to high school students in 2012. It might also be more meaningful to say something like “the median LWer can store 10 numbers in working memory, compared to the general population’s median of 7” instead of “the median LWer has a working memory that’s 95th percentile.” (I also haven’t looked up recently how g-loaded the SAT is, and that could vary significantly over time.)
Second, one of the main benefits may not be that the median LWer is able to get into MENSA, but that the smartest LWers are cleverer than most people have had the chance to meet during the lives. This is something that IQ tests are not very good at measuring, especially if you try to maintain the normal distribution. Reliably telling the difference between someone who is 1 out of 1,000 (146) and someone who is 1 out of 10,000 (155) is too difficult for most current tests; how many people would you have to base your test off of to reliably tell that someone is one out of a million (171) from their raw score?
iqtest.dk is based on Raven’s Progressive Matrices; the corresponding column, CV in the public .xls, is called IQTest. I referred to the scores that way because Raven’s measures a particular variety of intelligence. It’s seen widespread adoption because it’s highly g-loaded and culture fair, but a score on Raven’s is subtly different from a total score on WAIS, for example.
It’s the “IQTest” column, corresponding to scores from iqtest.dk.