Re:Lung Cancer and the Power of Suggestion
by ceejayoz on Sunday February 10, 2013 @01:06PM
(#5197177)
> After all, if A & B are correlated, then one of these statements must be true: "A causes B", "B causes A" or "there is something else, C, which causes A and B".
You missed "it's purely coincidence", which is an entirely valid option.
You should do enough experiments to be able to state with statistical confidence that it's not a coincidence.
Re:Lung Cancer and the Power of Suggestion
by greenyoda on Sunday February 10, 2013 @05:35PM
(#5198280)
The correlation could also be due to a flaw in your experimental design (e.g., something in your lab is contaminating your samples, or the researcher has an unconscious bias that skews measurements in a particular direction), in which case repeating the same experiments will not solve the problem.
Sure, there's loads of ways an experiment can be flawed. But also, it might not be flawed. If you're got a good experiment that shows correlation, then it has to be one of the 3 results above.
Re:Lung Cancer and the Power of Suggestion
by ceejayoz on Monday February 11, 2013 @05:16PM
(#5203692)
No, it is statistically most likely to be one of the 3 results above. "Has to be" doesn't have much of a place in a statistics discussion. There is a non-zero probability of the correlation being due to chance.
Yes this is correct. And to go off on a general tangent, correlation tells you where to start looking. After all, if A & B are correlated, then one of these statements must be true: "A causes B", "B causes A" or "there is something else, C, which causes A and B".
Just because you can find a correlation, doesn't mean that finding is irrelevant. It is very interesting and tells you where to start looking.