Everybody knows correlation isn’t causality, but often the statement of causality creeps back in to our thinking and writing nevertheless
Kaiser Fung’s book, Numbersense, contains a nice example of causal creep from a paper on obesity. First we have the cautious statement:
We…analyze the association [not the cause-effect] between BMI and stroke …. The mechanism by which BMI affects stroke risk independent of established risk factors, such as hypertension and diabetes is not fully understood.
But eventually a strong implication of causality creeps into the paper, in the last sentence:
Prevention of obesity should help prevent risk of stroke in men.
Notice the creep from the careful “association, not cause-effect” to the straightforward claim that “prevention of A should change B”.
It’s particularly hard to catch causal creep when it occurs in a context in which we think we know the answer. How else could the world be?
Well, the world might be this way (in which prevention of obesity would help prevent risk of stroke in men), but it might be another way. Let’s look at two plausible alternatives – maybe NOT as plausible as the idea that preventing obesity would prevent risk of stroke, but plausible nevertheless.
1. Perhaps measures to prevent obesity could actually make things worse – starvation or excessive exercise may prevent obesity, but may actually raise the risk of stroke, not lower it.
2. We should remember that smoking for years was touted as a weight control method, with unfortunate results. OK, this is lung disease rather than stroke, but let’s not get overly picky about our method of getting gravely ill.
So the world could be different. The medical literature abounds in things that seem like they should work, but turn out not to. We have to be careful about letting correlational association creep into causal prescription.
(As for me, I know I should lose weight for various good reasons. Working on it.)
Post a Comment