My example may have been idealized, but it doesn’t apparently matter. Looks like the raw scores for modern IQ tests are transformed to fit a normal distribution with mean 100 and standard deviation 15. (Meaning they basically subtract the average score from raw scores, divide by the deviation of the raw score, multiply by 15, and add 100. Basically they just scale the data from every test so number of questions doesn’t really matter. This also might reduce bias from people in a given location or who took a specific test if those are the groups for normalization).
The only real question for confidence in the scale is then the number of people who have taken the test. So let’s say we want to be 8sigma sure (likelyhood were wrong is about 1x10^-15) that a person’s IQ is correct to ±1 point.
For this confidence interval we have 8 as the critical z value and 15 as the standard deviation. A 1 point error in score means we’d only need a sample size of 14,400 people.
In other words, you only need to have 14,400 people take the test (or an equivalent one with the same normalization) in order to dettermine with ~99.999999999999999% confidence that someone’s score is between 200 and 202.
I’d imagine that’s not an unreasonable number of samples for MENSA or WAIS. Ergo IQ scores of extraordinarily high values are not necessarily signs of defects in the math of the test.
Given the world population there is zero reason to go to sigma 8. That’s more than 12000x the population of earth.
But, anyway. If the theoretical limit to IQ for the current population is 201 and a test even has the potential to give a result of 202 the test is defective. And people that make IQ tests have to know that because any result outside of theoretical limits is going to get them laughed out of town.
You mentioned Wikipedia earlier. The test that gave the 210 result is known to be defective.
My example may have been idealized, but it doesn’t apparently matter. Looks like the raw scores for modern IQ tests are transformed to fit a normal distribution with mean 100 and standard deviation 15. (Meaning they basically subtract the average score from raw scores, divide by the deviation of the raw score, multiply by 15, and add 100. Basically they just scale the data from every test so number of questions doesn’t really matter. This also might reduce bias from people in a given location or who took a specific test if those are the groups for normalization).
The only real question for confidence in the scale is then the number of people who have taken the test. So let’s say we want to be 8sigma sure (likelyhood were wrong is about 1x10^-15) that a person’s IQ is correct to ±1 point.
For this confidence interval we have 8 as the critical z value and 15 as the standard deviation. A 1 point error in score means we’d only need a sample size of 14,400 people.
In other words, you only need to have 14,400 people take the test (or an equivalent one with the same normalization) in order to dettermine with ~99.999999999999999% confidence that someone’s score is between 200 and 202.
I’d imagine that’s not an unreasonable number of samples for MENSA or WAIS. Ergo IQ scores of extraordinarily high values are not necessarily signs of defects in the math of the test.
Given the world population there is zero reason to go to sigma 8. That’s more than 12000x the population of earth.
But, anyway. If the theoretical limit to IQ for the current population is 201 and a test even has the potential to give a result of 202 the test is defective. And people that make IQ tests have to know that because any result outside of theoretical limits is going to get them laughed out of town.
You mentioned Wikipedia earlier. The test that gave the 210 result is known to be defective.