> Studies pretty consistently show that IQ is a valid predictor of performance all the way up the scale, so that, for example, someone with an IQ of 145 is likely to perform (at school, on the job, etc) better than someone with an IQ of 135.
To be fair, he doesn’t say that high IQ are not correlated with high performance, he conjectures that they ‘decorelate’, by which I presume he means the correlation weakens. (It’s pretty obvious at the end of the post that he’s arguing that).
If true, that is concerning for a measure. Heteroscedasticity is a serious problem for accurate prediction - in particular it tends to conceal the inaccuracies of predictions.
He also points out other issues, non-monotinicity and intransitivity which would be problematic. If your measure predicts candidate A is better than B and candidate B is better than C, you generally want it to predict candidate A is better than C. If it predicts candidate C is better than A, it creates confusion.
To be fair, he doesn’t say that high IQ are not correlated with high performance, he conjectures that they ‘decorelate’, by which I presume he means the correlation weakens. (It’s pretty obvious at the end of the post that he’s arguing that).
If true, that is concerning for a measure. Heteroscedasticity is a serious problem for accurate prediction - in particular it tends to conceal the inaccuracies of predictions.
He also points out other issues, non-monotinicity and intransitivity which would be problematic. If your measure predicts candidate A is better than B and candidate B is better than C, you generally want it to predict candidate A is better than C. If it predicts candidate C is better than A, it creates confusion.