Okay, so the IQ tweet storm has been turned into a Medium post.https://medium.com/incerto/iq-is-largely-a-pseudoscientific-swindle-f131c101ba39 …
Show this thread
-
The range restriction is bizarre, yes.
Presumably he used data based on a test with a low ceiling, e.g. WORDSUM.
-
-
I'd guess it's based on one of the NLSY samples and the AFQT. The regression of AFQT on SAT looks as below in the NLSY79. The slope levels off at the high end, but I'd think it's b/c the AFQT lacks discrimination there; in military use, everyone scoring >122 gets the same score.pic.twitter.com/N6NyXSfIat
-
Looking at this plot again, it looks like there is a 145 ceiling affect which is bending the top tail down. But also looks heteroskedastic (noisier at the top). Suppose it is; how should I roll that into my understanding? Should I do regressions on the logged variables instead?
-
It means that your prediction intervals will be larger at the high end. It can also result from the test properties, as pnin says. Most tests have bad reliability in high range because of little practical relevance and SLODR effect.
-
I'm surprised you would suggest that in the high range tests have "little practical relevance."
- 1 more reply
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.