8 Comments

Reliability of FSIQ typically around .90-.95 depending on the measure. https://arthurjensen.net/wp-content/uploads/2020/04/Bias-in-Mental-Testing-Arthur-R.-Jensen.pdf chapter 7. However, this reliability is mostly concentrated among the average person. You see, when you use a fixed length test, the scores are more reliable around the mean as most items have difficulties near 0. Scores further from the mean are less reliable, and thus show much more regression. I don't know what the information figure looks like for WAIS, but a typical result is like this: https://www.researchgate.net/publication/320132198_Item_Response_Theory_for_Medical_Educationists/figures?lo=1 figure 3.

The second issue is that stability of intelligence is lower from childhood to adulthood, and of course, the achievement tests have a true g loading of .80. So SMPY and various gifted problems are not that predictive of later life intelligence as one would think. Early screening for giftedness is not that important.

I did some SAT math here but these are for regular test takers, not early test takers. Current max SAT corresponds to about 135 IQ. https://emilkirkegaard.dk/en/2022/04/iqs-by-university-degrees/

Expand full comment

Suppose instead of multiplying the test score by reliability/g-loading, you simply divided the standard deviation by the same?

The z-score would become the same either way, but you wouldn't have to multiply the test score of every individual.

For example, instead of saying that a 143 on the WAIS is actually a 0.92*43+100 = 139.56

a 143 is always a 143, but with an SD of 16.3 instead.

Expand full comment
Mar 29, 2023Liked by Joseph Bronski

This a great post. Perfect length. No excess tangents. Doesnt just say “existing position is gay”, rather also includes “and here is an updated more accurate position”. 10/10 would read again.

Expand full comment

I agree that good estimates of intelligence should be holistic. That's why I prefer to judge people by talking to them and observing their general behaviour. Eliezer Yudkowsky strikes me as the kind of rigid-brained sperg the term "skullwork" was made for. A natural slave who can do math. At best we might say he's intelligent like a pocket calculator. Only we know better than to ask those for their opinions. His particular claims of meaningful intelligence are all rooted in abstract measurements that can be quibbled over endlessly. What else could they be in? Nobody's going to show off things he's said to prove this. I do not take the man seriously.

Yarvin, by contrast, seems like a broadly curious and driven character who has achieved notable success in different fields over the course of his life. Yes we have test results, STEM history, etc. But we don't need it. I can point anybody I consider truly intelligent towards his writing and they'll just get it. The clarity, wit, and humour are stronger proof of exceptional intelligence than any test score. Yes we can quibble the true value of any particular piece of writing, but that's not much of an objection when this entire piece, the twitter engagements that prompted it, and its comment sections are an extended quibbling on the value of what are supposed to be far more reliable measurements. I take Yarvin seriously.

And of course you, Joseph, came to my forum once. I do not take you seriously.

Expand full comment

As someone who has been consistently annoyed by people estimating their IQ by just taking their standard deviation-score on their SAT, or equivalent (often their best attempt) I appreciate this post, thank you. Was thinking of writing something similar but I think I'll skip it now.

Expand full comment

My comment is what prompted this article.

Yarvin's number is not from SMPY, as far as I know. He mentioned in an post recently he had the same IQ as the Unabomber, which Google says is 167.

As for Yudkowski, I think you are underestimating his intelligence. The 3.5 SD 1600 isn't adding anything. But yeah, he is almost certainly not 4.66 SD above the mean in G.

Expand full comment

We know that the g loading of a test is 0.5. We want to find the SD (Standard Deviations) above the mean at that metric to claim that their expected g is 145.

First, we need to calculate the g score difference between their expected g (145) and the mean g score (100):

Δg = 145 - 100 = 45

Now, we can use the g loading to find the equivalent number of standard deviations (SD) for the specific test:

SD = Δg / g_loading

SD = 45 / 0.5 = 90

So, someone must score 90 points above the mean on the specific test with a g loading of 0.5 for their expected g to be 145.

Now, we can convert this test score difference (90 points) to the number of standard deviations above the mean, assuming a mean score of 100 and a standard deviation of 15 for the general population:

SD_above_mean = (Test_score_difference) / Standard_deviation

SD_above_mean = 90 / 15 = 6

Therefore, you need to find someone who is 6 SD above the mean at that metric to claim that their expected g is 145.

Now, to find the rarity, we can use a standard normal distribution table or a calculator. A Z-score of 6 has a rarity of approximately 1 in 506,797,346.

Expand full comment