It seems likely you’re going to get skewed answers for the IQ question. Mostly it’s the really intelligent and the below average who get (professional) IQ tests—average people seem less likely to get them.
I predict high average IQ, but low response rate on the IQ question, which will give bad results. Can you tell us how many people respond to that question this time? (no. of responses isn’t registered on the previous survey)
I think it would be more informative to ask people to take one specific online test, now, and report their score. With everyone taking the same test, even if it’s miscalibrated, people could at least see how they compare to other LWers. Asking people to remember a score they were given years ago is just going to produce a ridiculous amount of bias.
Are there any free, non-spam-causlng, online IQ tests that produce reasonable results (i.e. correlate strongly to standard IQ tests)?
No chance.
To calibrate a serious IQ test, you need to test (1) many (2) randomly selected people in (3) controlled environment; and when the test is ready, you must test your subjects in the same environment.
Online calibration or even online testing fail the condition 3. Conditions 1 and 2 make creating of a test very expensive. This is why only a few serious IQ tests exist. And even those would not be considered valid when administered online.
And there is also huge prior probability that an online IQ test is a scam. So even if they would provide some explanation of how they fulfilled the conditions 1, 2, 3, I still would not trust them.
To calibrate a serious IQ test, you need to test (1) many (2) randomly selected people in (3) controlled environment; and when the test is ready, you must test your subjects in the same environment.
If you have a test thus calibrated, you can use it to evaluate tests that can’t be calibrated in the same way.
Here’s one that closely imitates Raven’s Progressive Matrices and claims to have been calibrated with a sample of 250,000 people: http://www.iqtest.dk/
Here’s another one: http://sifter.org/iqtest/ . I can’t find any mention of where the questions came from or how it’s calibrated, but it’s shorter and doesn’t require Flash.
Neither one asks for an e-mail address or any identifying information. They might be too easy for some on LW, but harder ones tend to cost money. As Viliam_Bur pointed out, any free online test’s validity is questionable, but the first one is basically a direct copy of a “real” test, and neither one has any apparent ulterior motive. Anecdotally, they were both within 10 points of each other and my “real” score.
The first test gave me a score a few points below that on the Mensa site I did a few years ago, but I gave up early on a few questions (I had about 10 minutes left when I finished).
One weird thing about it is that there were so many questions based essentially on the same idea, which makes me think it would be possible to have a test with not-too-much-worse accuracy but half as many questions (unless they intended to test ‘stamina’ as well—but I’d guess that that varies more for a same person depending on how much they’ve slept recently than across people).
Some data points:
IQ (age 7, 14, 20) = ~145-150 S-B
SAT (age 16) − 1590 = ~150 S-B
iqtest.dk (age 29) = 133 S-B
sifter.org/iqtest (age 29) = 139 S-B (159 euro scale)
I don’t use my spacial skills in my daily work they way I used to use them in my daily school work, and both online tests seem to measure only that.
I found the second test much more difficult—there wasn’t enough information to derive the exact missing item, so you had to choose things that could be explained with the simplest/least rules. There were some where I disagreed that the correct answer had a simpler rule-set. The problem style is also highly learnable, and I question the diagnostic value of “figuring out” that you’re looking at a 3x3 matrix where operations occur as you move around it, but various cells have been obscured to make the problem harder. Not including instructions makes it feel like there’s a secret handshake to get in.
I got 130 on the first one and 156⁄137 on the second.
Going with the lower result for the purpose of Yvain’s survey. I found the second result a little suspect because a lot of questions on the second test made little sense to me. I would often see 2-3 possible answers that made more or less equal (small) sense to me, and had to take a gut feeling guess on which the author might have possibly meant.
Maybe I just got lucky. Or my gut is a better thinker than I suspected.
Got 135 on the first test. Got 139 on the Stanford-Binet/USA scale (stdev 16) in the second. This seems about right.
But since the second one was polite enough to tell me which answers I got wrong, I have to call bullshit on it: some of the “correct” answers it claimed made no sense, and seemed more wrong and illogical than the ones I had placed.
I tried the second one after reading this and had similar results: 118 on the first one (implausibly low); 137 (stdev16) on the second one (sounds about right).
Though if I was taking this more seriously I’d probably have to weigh the facts that my kids were being more distracting when I took the first one, and I ate flaxseed shortly before taking the second one.
I took the first one under reasonably good conditions, and the second under about the same conditions a little while afterwards.
The first one seemed like a test of endurance as much as anything—it was as though my ability to focus was running out on the last ten questions or so, and possibly as though it would have been somewhat easier if I’d been in better physical condition.
General question about that sort of puzzle—how much can effort help with them? Can they be solved reliably given more time (and probably a chance to write down theories and guesses), or does inspiration have to strike fairly quickly?
Interesting question. On the first test, I went through many of them quickly—some of them obviously pattern-matched to the same kind of a puzzle—but also solved a number by staring at them for a few minutes, refusing to give in to my brain’s “I don’t see any patterns, this doesn’t make any frakking sense, can we do something else now?”. I’m certain given 10 or 20 more minutes I’d have done better. And come out with a headache, probably.
My eyes were hurting after the first test, and this continued (less intensely, I think) into the second, even though reading on the monitor isn’t generally a problem for me. There may also be sensory issues involved in scores—I was running into trouble anyway, but having to distinguish between very dark gray squares and black squares in one of the later puzzles didn’t help. If I had more of a different sort of intelligence, I would have thought of fiddling with my monitor settings.
I’m inclined to think that practice/information could help a lot with the puzzles—having a repertoire of possible patterns is going to make solutions easier than trying to find patterns cold.
Possibly as a result of not being entirely pleased at that 107 score, I’m doubting the whole premise of IQ testing—that it’s important to find out what can’t be improved about people’s minds.
Part of this is the arrogance problem—how complete is your knowledge of the possibility of improvement, anyway?-- and the other part is wondering whether all those resources could be better put into learning how to improve what can be improved.
The other thing is that I’ve had some recent evidence that the ways the parts of the mind are interconnected aren’t completely obvious. I’ve been doing some psychological work on fading out self-hatred, and the results have been being less frightened about what I post (I decided before taking the IQ tests to post my scores, but there was still a bit of a pang), easier and faster typing—not tested, but I do seem somewhat apt to write at greater length (this seems to be the result of feeling less need to over-monitor so that typing can be a low-level habit), less akrasia (still pretty bad, but the desire to do things is happening more often), and the ability to walk downstairs more easily (I have some old knee injuries which can be ameliorated by better coordination—but I haven’t been working on coordination).
In this type of test, I can solve generally about all except about 4 of them almost immediately with some seconds of thought. I skip those few, then return to them at the end, and in the minutes that remain manage to make an educated guess for say two of them, while having to leave two more to complete chance.
Interesting. Did you find the questions in the first test more difficult than the second? I did notice that the first test relies a lot on mental rotation.
With everyone taking the same test, even if it’s miscalibrated, people could at least see how they compare to other LWers.
There are two ways an IQ test can fail:
a) it can be miscalibrated;
b) it can measure something else than IQ.
If you only want to know your percentile in LW population, (a) is not a problem, but (b) remains. What if the test does not measure the “general intelligence factor”, but something else? It can partly correlate to IQ, and partly to something else, e.g. mathematical or verbal skills.
Also you have a preselection bias—some LWers will fill the survey, others won’t.
Don’t forget those of us who aren’t native English speakers. Didn’t try it again recently, but I used to have a 5-10 points difference between an IQ test in French (my native language) and English. Word-related questions are of course harder, but even for the rest, I’m not sure if it’s because it took me longer to process the English (while the IQ is time-limited), or just that decoding a non-native language use more brain power (leaving less for solving the problem). But anyway, I score better in my native language than in English, and I answered with my score in native.
Yes—I’m quoting an IQ test I did as a kid which had a suspiciously high score, I’m pretty confident I’d get a much less spectacular score if I did one today.
Yes—I’m quoting an IQ test I did as a kid which had a suspiciously high score, I’m pretty confident I’d get a much less spectacular score if I did one today.
Awesome. Definitely don’t do another one then. (Unless you need to diagnose something of course!)
I underwent a real IQ test when I was young, and so I can say that this estimation significantly overshoots my actual score. But that’s because it factors in test-taking as a skill (one that I’m good at).
Then again, I’m also a little shocked that the table on that site puts an SAT score of 1420 at the 99.9th percentile. At my high school there were, to my knowledge, at least 10 people with that high of a score (and that’s only those I knew of), not to mention one perfect score. This is out of ~700 people. Does that mean my school was, on average, at the 90th percentile of intelligence? Or just at the 90th percentile of studying hard (much more likely I think).
If you’re in the median age band for Less Wrong, you misread the estimator. The “SAT to IQ” table is for the pre-1995 SAT, which had much more rarefied heights. The “SAT I to IQ” table is for the 1995-2005 SAT.
And of course, there are also SAT prep services which offer guarantees of raising your score by such and such an amount (my mother thought I ought to try working for one, given my own SAT scores and the high pay, but I don’t want to join the Dark Side and work in favor of more inequality of education by income,) and these services are almost certainly not raising their recipients’ IQs.
I’ve never taken an IQ test, so when I was responded to the survey I considered estimating my IQ based on my SAT and GRE scores. The result, according to the site torekp linked to, is surprisingly high (150+). I think I’m smart, but not that smart. Anyone have any idea if these estimators should be trusted at all?
My IQ according to the estimator would put me in the 99.995th percentile, but it seems to me that at least 5% of my friends and acquaintances are at least as smart as me. Part of this is probably selection bias, but I doubt that could account for it completely. I don’t move in particularly exalted circles.
EDIT: If you had asked me to estimate my IQ before I consulted the website, I would have said 135. I’d probably still say that, actually. I’m guessing the GRE-to-IQ conversion is useless above some ceiling.
FYI, if you’re in the median age band for Less Wrong, you misread the estimator—I know, because I made the same mistake. Clicking “SAT to IQ” on the left shows a table for the test prior to a re-centering in 1995, whereas “SAT I to IQ” shows the table for tests given between 1995 and 2005. The latter’s top end is much less exceptional.
GRE quantitative scores are not useful for high-IQ estimates because 6% of people get perfect scores.
A perfect GRE verbal score is roughly the 99.8th percentile, as can be inferred from the charts in this pdf: http://www.ets.org/Media/Tests/GRE/pdf/994994.pdf It shows that the percent of people with a perfect scores varies between less than 0.1% and 1.5%, depending on field, but it is usually 0.1% or 0.2%. (The 1.5% field was philosophy.) Because many non-native English speakers take the test, it’s likely that one ought to adjust that percentile a bit lower.
That’s among people applying to grad school, which is a higher-IQ group than the general population, but not by so much that 99.8th percentile among grad school applicants correlates to the 99.996th percentile among the general population, as that site (http://www.iqcomparisonsite.com/GREIQ.aspx) claims. That would be impossible assuming more than one in fifty people in the applies to grad school.
If we attribute a perfect GRE score to the 99.8th percentile, then looking up that percentile on the chart on the same page, we get an IQ score >142 for 1600 on the GRE.
I’ve only got the one data point, but my tested IQ is within a couple points of what that site predicts from my SAT score. I took the tests almost a decade apart, though, so this could be coincidental; scores for both tests aren’t that stable over that kind of timeframe, I don’t think.
My (limited) background knowledge is that SATs, GREs, etc. are designed for people near the average, and give imprecise results for the highest IQs. You’re probably in that range the tests aren’t very good for.
I was wondering if the IQ-calibration question was referring to reported or actual IQ. It seems to be the latter, but the former would be much more fun to think about.
Also, are so many LWers comfortable estimating with high confidence that they are in the 99.9th percentile? Or even higher? Is this community really that smart? I mean, I know I’m smarter than the majority of people I meet, but 999 out of every 1000? Or am I just being overly enthusiastic in correcting for cognitive bias?
I’d estimate with high confidence that I’m higher than that. Subjectively, I’ve only met a couple of people in my life who seem definitely smarter than me. And I’ve barely met anyone who was malnourished or lacking in education. That said, there is the “everyone else is stupid” bias.
ETA: In case it wasn’t clear from the outset, on the outside view, most people with this notion are wrong, and there’s a recursive problem in justifying that I’m special. But intelligence tests, though imperfect, are a good hint.
I’m not contradicting you at all, but I’m just curious: how do you know that you are smarter than virtually everyone you meet? If there is anything more to it than an intuition, I’d love to know about it. I’ve always wondered if there was some secret smart-person handshake that I wasn’t privy to.
Personally, I’d say the lower 80 or 90% immediately identify themselves as such, but beyond that I try to give others the benefit of the doubt. Maybe they aren’t interested in the conversation, don’t want to seem intelligent, or or just plain out of my leauge. I don’t value humility very highly at all; but there aren’t many things that would convince me I or someone else was demonstrably in the top fraction of the top percentile.
Also, I’ve been intuitively aware of the optimism bias for as long as I can remember, and estimates like ”.1% and 99.9%” trigger my skepticism module hard.
Personally, I’d say the lower 80 or 90% immediately identify themselves as such, but beyond that I try to give others the benefit of the doubt.
I’d agree with that statement, revising it up to at least 95%. Once you’ve got it down to more than 19 in 20 people you meet being obviously-dumb, it’s worth the effort to inspect the others more carefully, since it’s always good having really smart people around.
Also, I’ve been intuitively aware of the optimism bias for as long as I can remember, and estimates like ”.01% and 99.99%” trigger my skepticism module hard.
I’m much more familiar with people thinking 95% is an orders-of-magnitude higher estimate than 80%, and so I tend to adjust others’ carefully-thought-out estimates outward rather than inward, unless they are 0 or 1.
ETA: It’s worth noting that one of the huge signals smart people give off is the “OMG you’re talking about something that requires intelligence I’m so happy to have met a smart person because that happens to me less than 5% of the time” reaction, which if rarer than I think would significantly throw off my estimates.
Seeming “obviously” dumb and actually not being in the top 5% are very, very different. A person might just be tired, or stressed, or distracted and so not exude intelligence. Or, they might be acting a little less intelligent than they actually are, maybe for social reasons.
I predict high average IQ, but low response rate on the IQ question, which will give bad results.
I predict with 70% certainty that we will get an IQ in the range of 140-145 again, though I think it will be a bit lower than last time. I’m very surprised if it’s outside 130-150.
(Also took the survey. Would like more “other” options so I can ramble about my totally different opinions on many issues, but whatever.)
IQs (warning: self-reported numbers for notoriously hard-to-measure statistic) ranged from 120 to 180. The mean was 145.88, median was 141.50, and SD was 14.02. Quartiles were 155.
For myself I took my result to the Mensa online pre-test, that I did for the purpose of calibrating myself a few years ago. It’s not a fully professional test (and not done in test situation), but I consider it valid enough to be more than pure noise.
This is great! I hope there’s a big response.
It seems likely you’re going to get skewed answers for the IQ question. Mostly it’s the really intelligent and the below average who get (professional) IQ tests—average people seem less likely to get them.
I predict high average IQ, but low response rate on the IQ question, which will give bad results. Can you tell us how many people respond to that question this time? (no. of responses isn’t registered on the previous survey)
I think it would be more informative to ask people to take one specific online test, now, and report their score. With everyone taking the same test, even if it’s miscalibrated, people could at least see how they compare to other LWers. Asking people to remember a score they were given years ago is just going to produce a ridiculous amount of bias.
Are there any free, non-spam-causlng, online IQ tests that produce reasonable results (i.e. correlate strongly to standard IQ tests)?
Mensa organizes cheap standardized IQ testing worldwide with many available dates.
I don’t care for everything else they’re doing, but at least that is a very valuable service to the world.
No chance.
To calibrate a serious IQ test, you need to test (1) many (2) randomly selected people in (3) controlled environment; and when the test is ready, you must test your subjects in the same environment.
Online calibration or even online testing fail the condition 3. Conditions 1 and 2 make creating of a test very expensive. This is why only a few serious IQ tests exist. And even those would not be considered valid when administered online.
And there is also huge prior probability that an online IQ test is a scam. So even if they would provide some explanation of how they fulfilled the conditions 1, 2, 3, I still would not trust them.
If you have a test thus calibrated, you can use it to evaluate tests that can’t be calibrated in the same way.
Will this evaluation include giving both tests to many randomly selected people and comparing the results?
It’s a bit late now, but if you recommend a particular test that’s valid, short, and online, I can try that on the next survey.
Here’s one that closely imitates Raven’s Progressive Matrices and claims to have been calibrated with a sample of 250,000 people: http://www.iqtest.dk/
Here’s another one: http://sifter.org/iqtest/ . I can’t find any mention of where the questions came from or how it’s calibrated, but it’s shorter and doesn’t require Flash.
Neither one asks for an e-mail address or any identifying information. They might be too easy for some on LW, but harder ones tend to cost money. As Viliam_Bur pointed out, any free online test’s validity is questionable, but the first one is basically a direct copy of a “real” test, and neither one has any apparent ulterior motive. Anecdotally, they were both within 10 points of each other and my “real” score.
Incidentally, I keep a list for DNB purposes in http://www.gwern.net/DNB%20FAQ#available-tests focused on matrix-style tests. Doesn’t include that
sifter.orgone, though.Wow. Wish I would’ve thought to google ‘iq site:gwern.net’.
Wouldn’t necessarily have helped—Google’s excerpt for the DNB FAQ doesn’t mention the list of tests. Kind of have to know it’s already there.
The first test gave me a score a few points below that on the Mensa site I did a few years ago, but I gave up early on a few questions (I had about 10 minutes left when I finished).
One weird thing about it is that there were so many questions based essentially on the same idea, which makes me think it would be possible to have a test with not-too-much-worse accuracy but half as many questions (unless they intended to test ‘stamina’ as well—but I’d guess that that varies more for a same person depending on how much they’ve slept recently than across people).
Some data points: IQ (age 7, 14, 20) = ~145-150 S-B SAT (age 16) − 1590 = ~150 S-B iqtest.dk (age 29) = 133 S-B sifter.org/iqtest (age 29) = 139 S-B (159 euro scale)
I don’t use my spacial skills in my daily work they way I used to use them in my daily school work, and both online tests seem to measure only that.
I found the second test much more difficult—there wasn’t enough information to derive the exact missing item, so you had to choose things that could be explained with the simplest/least rules. There were some where I disagreed that the correct answer had a simpler rule-set. The problem style is also highly learnable, and I question the diagnostic value of “figuring out” that you’re looking at a 3x3 matrix where operations occur as you move around it, but various cells have been obscured to make the problem harder. Not including instructions makes it feel like there’s a secret handshake to get in.
I got 130 on the first one and 156⁄137 on the second.
Going with the lower result for the purpose of Yvain’s survey. I found the second result a little suspect because a lot of questions on the second test made little sense to me. I would often see 2-3 possible answers that made more or less equal (small) sense to me, and had to take a gut feeling guess on which the author might have possibly meant.
Maybe I just got lucky. Or my gut is a better thinker than I suspected.
Got 135 on the first test. Got 139 on the Stanford-Binet/USA scale (stdev 16) in the second. This seems about right.
But since the second one was polite enough to tell me which answers I got wrong, I have to call bullshit on it: some of the “correct” answers it claimed made no sense, and seemed more wrong and illogical than the ones I had placed.
I got 107 on the first test (which seems implausibly low), and 138 on the second (which seems reasonable).
I tried the second one after reading this and had similar results: 118 on the first one (implausibly low); 137 (stdev16) on the second one (sounds about right).
Though if I was taking this more seriously I’d probably have to weigh the facts that my kids were being more distracting when I took the first one, and I ate flaxseed shortly before taking the second one.
I took the first one under reasonably good conditions, and the second under about the same conditions a little while afterwards.
The first one seemed like a test of endurance as much as anything—it was as though my ability to focus was running out on the last ten questions or so, and possibly as though it would have been somewhat easier if I’d been in better physical condition.
General question about that sort of puzzle—how much can effort help with them? Can they be solved reliably given more time (and probably a chance to write down theories and guesses), or does inspiration have to strike fairly quickly?
Interesting question. On the first test, I went through many of them quickly—some of them obviously pattern-matched to the same kind of a puzzle—but also solved a number by staring at them for a few minutes, refusing to give in to my brain’s “I don’t see any patterns, this doesn’t make any frakking sense, can we do something else now?”. I’m certain given 10 or 20 more minutes I’d have done better. And come out with a headache, probably.
My eyes were hurting after the first test, and this continued (less intensely, I think) into the second, even though reading on the monitor isn’t generally a problem for me. There may also be sensory issues involved in scores—I was running into trouble anyway, but having to distinguish between very dark gray squares and black squares in one of the later puzzles didn’t help. If I had more of a different sort of intelligence, I would have thought of fiddling with my monitor settings.
I’m inclined to think that practice/information could help a lot with the puzzles—having a repertoire of possible patterns is going to make solutions easier than trying to find patterns cold.
Possibly as a result of not being entirely pleased at that 107 score, I’m doubting the whole premise of IQ testing—that it’s important to find out what can’t be improved about people’s minds.
Part of this is the arrogance problem—how complete is your knowledge of the possibility of improvement, anyway?-- and the other part is wondering whether all those resources could be better put into learning how to improve what can be improved.
The other thing is that I’ve had some recent evidence that the ways the parts of the mind are interconnected aren’t completely obvious. I’ve been doing some psychological work on fading out self-hatred, and the results have been being less frightened about what I post (I decided before taking the IQ tests to post my scores, but there was still a bit of a pang), easier and faster typing—not tested, but I do seem somewhat apt to write at greater length (this seems to be the result of feeling less need to over-monitor so that typing can be a low-level habit), less akrasia (still pretty bad, but the desire to do things is happening more often), and the ability to walk downstairs more easily (I have some old knee injuries which can be ameliorated by better coordination—but I haven’t been working on coordination).
In this type of test, I can solve generally about all except about 4 of them almost immediately with some seconds of thought. I skip those few, then return to them at the end, and in the minutes that remain manage to make an educated guess for say two of them, while having to leave two more to complete chance.
Interesting. Did you find the questions in the first test more difficult than the second? I did notice that the first test relies a lot on mental rotation.
I found the last third or so of the questions in the first test much more difficult than almost anything in the second.
There are two ways an IQ test can fail: a) it can be miscalibrated; b) it can measure something else than IQ.
If you only want to know your percentile in LW population, (a) is not a problem, but (b) remains. What if the test does not measure the “general intelligence factor”, but something else? It can partly correlate to IQ, and partly to something else, e.g. mathematical or verbal skills.
Also you have a preselection bias—some LWers will fill the survey, others won’t.
Don’t forget those of us who aren’t native English speakers. Didn’t try it again recently, but I used to have a 5-10 points difference between an IQ test in French (my native language) and English. Word-related questions are of course harder, but even for the rest, I’m not sure if it’s because it took me longer to process the English (while the IQ is time-limited), or just that decoding a non-native language use more brain power (leaving less for solving the problem). But anyway, I score better in my native language than in English, and I answered with my score in native.
Yes—I’m quoting an IQ test I did as a kid which had a suspiciously high score, I’m pretty confident I’d get a much less spectacular score if I did one today.
Awesome. Definitely don’t do another one then. (Unless you need to diagnose something of course!)
Are we encouraged to estimate IQ from SAT tests and the like? That’s what I did. That could reduce the excluded-middle bias that Gedusa mentions.
I underwent a real IQ test when I was young, and so I can say that this estimation significantly overshoots my actual score. But that’s because it factors in test-taking as a skill (one that I’m good at). Then again, I’m also a little shocked that the table on that site puts an SAT score of 1420 at the 99.9th percentile. At my high school there were, to my knowledge, at least 10 people with that high of a score (and that’s only those I knew of), not to mention one perfect score. This is out of ~700 people. Does that mean my school was, on average, at the 90th percentile of intelligence? Or just at the 90th percentile of studying hard (much more likely I think).
If you’re in the median age band for Less Wrong, you misread the estimator. The “SAT to IQ” table is for the pre-1995 SAT, which had much more rarefied heights. The “SAT I to IQ” table is for the 1995-2005 SAT.
(I did the same thing.)
You are quite right. My scores correlate much better now; I retract my confusion.
And of course, there are also SAT prep services which offer guarantees of raising your score by such and such an amount (my mother thought I ought to try working for one, given my own SAT scores and the high pay, but I don’t want to join the Dark Side and work in favor of more inequality of education by income,) and these services are almost certainly not raising their recipients’ IQs.
I didn’t think of that—given that a huge chuck here have probably taken such tests, if Yvain allowed such an estimation, it would be very helpful.
Yes! That’s what I was thinking of :)
I’ve never taken an IQ test, so when I was responded to the survey I considered estimating my IQ based on my SAT and GRE scores. The result, according to the site torekp linked to, is surprisingly high (150+). I think I’m smart, but not that smart. Anyone have any idea if these estimators should be trusted at all?
What is your evidence?
I am not trying to convince you either way, but in my experience people aren’t very good at estimating their own IQ.
My IQ according to the estimator would put me in the 99.995th percentile, but it seems to me that at least 5% of my friends and acquaintances are at least as smart as me. Part of this is probably selection bias, but I doubt that could account for it completely. I don’t move in particularly exalted circles.
EDIT: If you had asked me to estimate my IQ before I consulted the website, I would have said 135. I’d probably still say that, actually. I’m guessing the GRE-to-IQ conversion is useless above some ceiling.
FYI, if you’re in the median age band for Less Wrong, you misread the estimator—I know, because I made the same mistake. Clicking “SAT to IQ” on the left shows a table for the test prior to a re-centering in 1995, whereas “SAT I to IQ” shows the table for tests given between 1995 and 2005. The latter’s top end is much less exceptional.
GRE quantitative scores are not useful for high-IQ estimates because 6% of people get perfect scores.
A perfect GRE verbal score is roughly the 99.8th percentile, as can be inferred from the charts in this pdf: http://www.ets.org/Media/Tests/GRE/pdf/994994.pdf It shows that the percent of people with a perfect scores varies between less than 0.1% and 1.5%, depending on field, but it is usually 0.1% or 0.2%. (The 1.5% field was philosophy.) Because many non-native English speakers take the test, it’s likely that one ought to adjust that percentile a bit lower.
That’s among people applying to grad school, which is a higher-IQ group than the general population, but not by so much that 99.8th percentile among grad school applicants correlates to the 99.996th percentile among the general population, as that site (http://www.iqcomparisonsite.com/GREIQ.aspx) claims. That would be impossible assuming more than one in fifty people in the applies to grad school.
If we attribute a perfect GRE score to the 99.8th percentile, then looking up that percentile on the chart on the same page, we get an IQ score >142 for 1600 on the GRE.
That link should probably point to this (without the dot at the end).
thanks
I’ve only got the one data point, but my tested IQ is within a couple points of what that site predicts from my SAT score. I took the tests almost a decade apart, though, so this could be coincidental; scores for both tests aren’t that stable over that kind of timeframe, I don’t think.
My (limited) background knowledge is that SATs, GREs, etc. are designed for people near the average, and give imprecise results for the highest IQs. You’re probably in that range the tests aren’t very good for.
I wouldn’t trust it. My GRE estimated IQ by that is wildly higher than my professionally measured IQ.
Also check out:
The scores are highly correlated. One must assume those charts are from a reliable source. So… yes?
Does the correlation remain if you conditionalize on, say, having an IQ higher than 130?
Well, not with that attitude.
I was wondering if the IQ-calibration question was referring to reported or actual IQ. It seems to be the latter, but the former would be much more fun to think about.
Also, are so many LWers comfortable estimating with high confidence that they are in the 99.9th percentile? Or even higher? Is this community really that smart? I mean, I know I’m smarter than the majority of people I meet, but 999 out of every 1000? Or am I just being overly enthusiastic in correcting for cognitive bias?
I’d estimate with high confidence that I’m higher than that. Subjectively, I’ve only met a couple of people in my life who seem definitely smarter than me. And I’ve barely met anyone who was malnourished or lacking in education. That said, there is the “everyone else is stupid” bias.
ETA: In case it wasn’t clear from the outset, on the outside view, most people with this notion are wrong, and there’s a recursive problem in justifying that I’m special. But intelligence tests, though imperfect, are a good hint.
I’m not contradicting you at all, but I’m just curious: how do you know that you are smarter than virtually everyone you meet? If there is anything more to it than an intuition, I’d love to know about it. I’ve always wondered if there was some secret smart-person handshake that I wasn’t privy to.
Personally, I’d say the lower 80 or 90% immediately identify themselves as such, but beyond that I try to give others the benefit of the doubt. Maybe they aren’t interested in the conversation, don’t want to seem intelligent, or or just plain out of my leauge. I don’t value humility very highly at all; but there aren’t many things that would convince me I or someone else was demonstrably in the top fraction of the top percentile.
Also, I’ve been intuitively aware of the optimism bias for as long as I can remember, and estimates like ”.1% and 99.9%” trigger my skepticism module hard.
I was mostly going by the handshake.
I’d agree with that statement, revising it up to at least 95%. Once you’ve got it down to more than 19 in 20 people you meet being obviously-dumb, it’s worth the effort to inspect the others more carefully, since it’s always good having really smart people around.
I’m much more familiar with people thinking 95% is an orders-of-magnitude higher estimate than 80%, and so I tend to adjust others’ carefully-thought-out estimates outward rather than inward, unless they are 0 or 1.
ETA: It’s worth noting that one of the huge signals smart people give off is the “OMG you’re talking about something that requires intelligence I’m so happy to have met a smart person because that happens to me less than 5% of the time” reaction, which if rarer than I think would significantly throw off my estimates.
Seeming “obviously” dumb and actually not being in the top 5% are very, very different. A person might just be tired, or stressed, or distracted and so not exude intelligence. Or, they might be acting a little less intelligent than they actually are, maybe for social reasons.
I predict with 70% certainty that we will get an IQ in the range of 140-145 again, though I think it will be a bit lower than last time. I’m very surprised if it’s outside 130-150.
(Also took the survey. Would like more “other” options so I can ramble about my totally different opinions on many issues, but whatever.)
Awesome. I said my IQ was 140 and 50% probability that I was higher than average, because I figured I’d be almost exactly average.
(I hope you didn’t actually put “140ish”, right?)
I’m actually surprised the lower bound on the previous survey was 120. I would have figured more of a U-shaped curve.
I put 140. Fixed.
Oh wow, is that what the IQ average was last time? Can I update my probability that mine will be higher?
Last survey in 2009:
For myself I took my result to the Mensa online pre-test, that I did for the purpose of calibrating myself a few years ago. It’s not a fully professional test (and not done in test situation), but I consider it valid enough to be more than pure noise.
Same here. (I rounded the result to the nearest ten, also because I don’t remember the last digit for sure.)