IQs (warning: self-reported numbers for notoriously hard-to-measure statistic)
Yeah, I’m extremely skeptical of the IQ data. Assuming a standard mean=100 SD=15 test (although at least one respondent says he took a test with SD=24), our reputed median is above the 0.003th percentile. I don’t think any public blog is that elite.
I’m skeptical of the IQ data because of the number of IQs above 140. Most IQ tests don’t measure well above IQ 140, and so even if we have that many truly exceptional people, I would not expect it to show up in their measured IQs.
Most human beings in any forum, anywhere, will be more obsessed with signaling and other concerns than The Truth—even in a pseudo-anonymous survey—and will be subject to most of the standard cognitive biases that bodybuilders will be, even if to a lesser degree. Being obsessed with The Truth does not mean never lying or exaggerating (or reporting just that one internet IQ test you took that was 1 std dev higher than your real-world test).
If a lot of people actually got scores outside the calibration range of whatever IQ test they took, they could have answered honestly and the resulting numbers still be as bogus as Eliezer suggests.
We had similar data on the survey I ran (which I still need to write up the results of). I don’t know that the numbers past 140 are intelligence-indicative, but I suspect people really did get their reported scores on IQ tests.
Also, in the responses to my survey, people who said they were from the USA were no more or less likely than people who said they weren’t to report scores over 140. Which argues against regional variation in what IQ tests mean. Although I don’t know how consistent the meaning is of IQ tests within the USA; anyone have knowledge, here?
The person who administered my test told me it was inaccurate above 150, and then told me my result was high enough to be somewhere in the inaccurate range, so I explicitly mentioned that it was an “at least” figure.
If we were to assume a test with a standard deviation of 24, a median of 141.5 would be just below the 96th percentile. That still seems too high for the median user, but it’s almost plausible—much more so than 99.7th percentile.
It’s also quite likely that LW readers with abnormally high IQs (relative to LW) are (A) much more likely to have been tested and to know (and remember) the result, and (B) include the score on the survey.
It doesn’t strike me as all that implausible, given how many other indicators of quirkiness we have as a group (e.g., the 95-97% male, the 12% with PhDs (and 23% of members over 35 with PhDs), the portion with advanced math/compsci skill, etc.).
Math is not my strong suit, but my arithmetic comes out differently on the PhD bit. Are you counting as PhDs the people who have “student” in the “degree status” field?
Yeah, I’m extremely skeptical of the IQ data. Assuming a standard mean=100 SD=15 test (although at least one respondent says he took a test with SD=24), our reputed median is above the 0.003th percentile. I don’t think any public blog is that elite.
ERRATUM: Oh, dear. I meant 99.7th percentile.
I’m skeptical of the IQ data because of the number of IQs above 140. Most IQ tests don’t measure well above IQ 140, and so even if we have that many truly exceptional people, I would not expect it to show up in their measured IQs.
But if so many lied, it would also be a surprising fact, that doesn’t seem to be a better explanation.
It’s only a little more surprising than somebody at an online forum for bodybuilders lying about how much they can bench press.
I take it the reason it’s not equally surprising is that few bodybuilders are as monomaniacally obsessed with The Truth as we are?
Most human beings in any forum, anywhere, will be more obsessed with signaling and other concerns than The Truth—even in a pseudo-anonymous survey—and will be subject to most of the standard cognitive biases that bodybuilders will be, even if to a lesser degree. Being obsessed with The Truth does not mean never lying or exaggerating (or reporting just that one internet IQ test you took that was 1 std dev higher than your real-world test).
If a lot of people actually got scores outside the calibration range of whatever IQ test they took, they could have answered honestly and the resulting numbers still be as bogus as Eliezer suggests.
We had similar data on the survey I ran (which I still need to write up the results of). I don’t know that the numbers past 140 are intelligence-indicative, but I suspect people really did get their reported scores on IQ tests.
Also, in the responses to my survey, people who said they were from the USA were no more or less likely than people who said they weren’t to report scores over 140. Which argues against regional variation in what IQ tests mean. Although I don’t know how consistent the meaning is of IQ tests within the USA; anyone have knowledge, here?
Did you ever write up your results? They would make a valuable addition to the historical data.
The person who administered my test told me it was inaccurate above 150, and then told me my result was high enough to be somewhere in the inaccurate range, so I explicitly mentioned that it was an “at least” figure.
If we were to assume a test with a standard deviation of 24, a median of 141.5 would be just below the 96th percentile. That still seems too high for the median user, but it’s almost plausible—much more so than 99.7th percentile.
It’s also quite likely that LW readers with abnormally high IQs (relative to LW) are (A) much more likely to have been tested and to know (and remember) the result, and (B) include the score on the survey.
It doesn’t strike me as all that implausible, given how many other indicators of quirkiness we have as a group (e.g., the 95-97% male, the 12% with PhDs (and 23% of members over 35 with PhDs), the portion with advanced math/compsci skill, etc.).
Math is not my strong suit, but my arithmetic comes out differently on the PhD bit. Are you counting as PhDs the people who have “student” in the “degree status” field?
I was working off the 233 people who filled out my earlier survey. I haven’t analyzed Yvain’s data; what percentage do you get there?
I didn’t write it down and I don’t want to count them up again, sorry.