10 July 2014

Lies, damned lies, and...

Statistics. I am so tired of hearing shitty science extolled as good science by people who are trying to point out shitty science. Convoluted, no?

Okay, so it's like this. Some study comes out showing a correlation between X and Y. Not a causal relationship, but a strong correlation and a note in the study saying more research is necessary.

Crazy Nutbars Who Can't Distinguish Between Correlation and Causation (CNWCDBCC) jump on this study and scream WE TOLD YOU X CAUSES Y!
Snarky Motherfuckers Who Know Less About Stats Than They Claim (SMWKLASTTC) start snarking that correlation != causation, and conclude that X cannot cause Y, because crappy understanding of Science.

Listen SMWKLASTTC, in an observational study, a strong correlation is required for proof of causation. It is not the only requirement by a longshot. You also need to have a well-designed study, consistency (hence the "more research is necessary" in the study), dose-response relationship (more people exposed, more people affected), reversibility (remove the potential cause and the incidence rate should decline), biological plausibility and coherence with known facts.[1]

It's those last two that have the SMWKLASTTC crowd screaming. It's not plausible, blah blah blah. Remember when rheumatic fever was most emphatically NOT caused by strep? Yeah. That was what science said then. Now it doesn't. Because they proved it, and changed the known facts.

And what's really important is controlling your study. Removing or randomizing associated factors and doing so with a large enough sample isn't easy. Especially when studying the cause of Y (Cancer, autism, OCD, roseacea, obesity, warts, whatever). Because human genetics aren't easy to control for if you don't know what you're looking for.

Suppose for the sake of simplicity that we're looking at whether the sun causes skin to burn. We get a nice random sample of people from all around the world, and they're all sorts of natural shades, from ivory to black coffee. We've got hundreds of thousands of people, because we're awesome. But we don't know anything about melanin, because we're not scientifically advanced enough yet. All we know is that in 15 minutes in the sun, only a tiny fraction of our millions in the sample are burnt. So we conclude that the sun doesn't cause burns. Nope. Because science! Because stats!

Oh, but we left them out for another 15 minutes, and all of a sudden, a lot more of them are burned. Oh, huh. And it's only the pale ones. But there's no reason for this (remember, we still don't know about melanin), so the pale ones must be having a psychological reaction. Because science! Because stats! But the CNWCDBCC are screaming WE TOLD YOU X CAUSES Y! Crazy nutbars.

And then someone comes up with the idea of melanin. Crazy bastard! EVERYONE knows that pale redheads are just psychologically weak. But eventually, the idea catches on, and he proves that it exists. Huh. Cool. So the sun does cause burns in short periods of time. But only in people who are genetically sensitive to it. Well, I'll be damned! Science was wrong! New info! The correlation between being in the sun and getting a burn was actually causal all along. But until they knew about melanin, they couldn't prove it. And the crazy nutbars? Well, lucky guess, right?*

And what's worse, is that the same motherfuckers who swear that X doesn't cause Y are the same ones that love to jump into an argument about obesity and scream that overeating is the cause of obesity because of that whole calories in/calories out bullshit lie. They holler that eating too much and obesity cause diabetes, even though there is no causal link proven. There is a strong positive correlation. But they pick and choose. Much like we all do, I suppose. Some correlations we assume to be causal and some we don't, based on our beliefs. But if you're going to get sanctimonious with me about suggesting that I'm going to stay away from X because of its correlation with Y, and claim I fail at science, I am so going to do it right back at you.

So stop that shit. If someone tells you vaccines or glyphosate cause autism, strep causes OCD, aspartame causes cancer, eating sugar causes diabetes, sniffing farts prevents cancer, or that wearing shoes will give you cancer*, feel free to tell them that current scientific consensus says that's not true. And when they yank out some study that shows a correlation, feel free to remind them in as condescending a tone as you can manage that correlation doesn't imply causation. But do try to remember that correlation doesn't exclude causation either. There may be some factor missing in the studies and those crazy nutbars may one day be vindicated. You know, like the ones who insisted that cigarettes cause cancer. The ones who were certain that epilepsy was physical, not psychological. The ones who said BPA was dangerous, even in tiny amounts - especially in tiny amounts. The ones who screamed that the poor air quality was giving them asthma. All of those crazy nutbars who couldn't prove causation until they did.

*And of course, sometimes the crazy nutbars are just crazy nutbars.
[1] http://learnandteachstatistics.wordpress.com/2013/10/21/proving-causation/
I picked my links more or less randomly. Fair warning! :)