[dllthomas]

> That word "Predicts" is dangerous. It strongly suggests causation, though it only reflects a statistical correlation.

REALLY? What do you WANT them to say? Accuracy of a prediction implies some sort of causal relationship, but not a direct one by any means. All of the thing you suggest are causal links. A genetic disorder causing a decrease in both lifespan and creativity is causation. "A predicts B" does NOT AT ALL imply "A causes B" - tinsel, wreaths, and increased toy buying predict Christmas.

[tlb]

"is correlated with".

If you're not convinced that the reporters are trying to suggest causation with the word "predicts", try swapping the two terms. Would they write a headline, "Longer life predicts creativity"?

[meric]

If the statistics show that creative people are correlated with longer life, but that longer life is only correlated with creativity if factors X, Y, Z are also present (but not necessarily just X, Y, Z without creativity), then it would be correct to say "creativity predicts longer life" but not "longer life predicts creativity"; while not implying any causation.

Also it could be that some long living people are creative and all creative people are long living; while creative still not being a cause of the longer life. (let's say all creative people eat peanut butter sandwiches, and it was infact the peanut butter that caused a longer life.)

"The weatherman predicts the weather". Are you saying in this sentence, causation is implicitly referenced?

...

[dllthomas]

It doesn't work the other way because of temporal constraints, not causality. A person's creativity happens before their longevity is determined. Similarly, you wouldn't write "election predicts poll results."

That said, if I was looking at records of the deceased and first discovered a person's age at time of death, then I could use the correlation to predict that I am likely to further find that they were creative.

[lutusp]

>> That word "Predicts" is dangerous. It strongly suggests causation, though it only reflects a statistical correlation.

> REALLY?

Yes -- really. It is very misleading.

> What do you WANT them to say?

Replace "predicts" with "is correlated with." That's the meaning of the work being described -- a correlation between A and B has been observed. But this doesn't mean that A predicts B, or A causes B, or even that B causes A. Both may result from some unevaluated cause C.

Science is not description, it is explanation. The linked article describes, and then draws a conclusion that goes far beyond the description. But it never presumes to explain why the author's conclusion might be so.

> "A predicts B" does NOT AT ALL imply "A causes B"

It certainly does. Rain predicts puddles. Puddles do not predict rain, they follow from it.

[dllthomas]

"A predicts B" means "A precedes B, and knowing A will substantially improve your guess about B." Obviously, puddles do not predict rain - they happen after. As someone else said, though, the weather man predicts rain, but rain doesn't follow from the weatherman's statements - the causal relationship there is substantially indirect; measurements of various phenomena are fed into models, themselves developed from past observation, which causes various information to be printed on a screen which causes the weatherman to say certain things that, yes, correlate strongly but not perfectly with the weather tomorrow. This is not hugely different, in terms of information flow, than us predicting a lower longevity for Steve based on observation of his lesser creativity, both caused perhaps by the same gene (leading to the correlation we are exploiting to make the prediction).

[lutusp]

> This is not hugely different ...

You're missing the crucial distinction between describing and explaining. A correlation is a description, and descriptions aren't science. Only by proposing an explanation, then testing it, do we enter the domain of science. We also have the chance to turn a correlation into something more than a coincidence.

[dllthomas]

"The article is simply reporting a correlation, which is not good science" is a separate objection from "predicts implies causation, and none exists here." The latter was what I was objecting to; specifically the first part.

If you want to separately discuss the former, I don't have a strong opinion about it any which way. We do need to document correlations somewhere, because 1) they are a useful starting point when looking for causal relationships, and 2) we might be able to make use of them before we understand why they work. We do need some quality controls to ensure that we are finding correlations that really exist, and I don't see a problem with using the infrastructure around "properly" scientific experiment to this end, but if you propose we move it somewhere else or simply call it something else I don't see any big problems with that.