A test, ANY TEST, claiming to make prediction of X must be accompanied by two metrics: sensitivity (detection of X) & specificity (exclusion of non X). Without those two metrics we would have no medical tests. Only BS. Exactly like the IQ-test literature: BS pseudoscience.
Statistical correlations (if not faked) often linked to high sensitivity / low specificity. It looks impressive… with no value. Example: we find students that excelled in math in Africa had a pencil. No pencil = awful at math. So, pencil is the secret of a math genius!
The second problem is tautological correlation. If you measure how much you can run in 60 seconds, when you’re 10 years old, it may be correlated with survival in the special forces at 20 years old. Why? Cause the context is almost tautological: physical exercise.
The same with IQ and a system were education & finance are becoming more abstract and complex. The same root causes that increase excellence in processing visual/abstract/verbal information (school, academy, video games) are pushing up the IQ, a measure also devised by academics.
Tautological correlation gives the impression of “discovery” of causality. It’s an illusion. IQ-test doesn’t even define the “X” that it predicts, let alone sensitivity or specificity of X. I’m not persuaded it’s is a test, in stat terms.
In the “correlation is not correlation” issue I noticed something but I don’t know of any particular technical term, so I call it: “activated correlation”.
Skin cells & muscle cells have exactly the same genes. All cells have the same genes, 100% “DNA correlation”. But different genes are activated in different times in different cells, a timely effect so important that is the “secret” of complex life sharing same information.
And we can see the same phenomenon, much more complex, in social life. Major aspects of behaviour (e.g. violence, morality, religiosity) are “activated” in certain times in certain ways.
And that’s why common sense beats the “evidence” of correlation in social science again and again. Example: It’s naive to make generalisations on any social behaviour, by studying population in times of peace & prosperity.
In biology this is called differentiation, and maybe we need this mental tool in statistics as well. Social differentiation: social correlations come and go, appear and disappear based on context.
A typical example would be the correlation between “faith in God” and “morality”. Social scientists were sure there is no correlation there. They even did a documentary to inform the people that there is no correlation between ethics and faith.
And… suddenly… the millennial generation appears. And in this generation (raised in a context of atheism) we see academic honesty, idealistic consuming behaviour, prosociality & other aspects of morality being linked to…. faith in God in multiple research articles.
Activated social correlation : “the context of the correlation (time & space) carries more information than the correlation itself.”
In more general terms in logic: “the context of any statement carries more information than the statement itself”. Or “you can create a logical paradox by changing the context of the statement” Tunnel Vision: focusing solely on a single statement (here: a correlation)