Post hoc ergo propter hoc (Or, does shit happen because Zeus is angry?)

On Nov 01, 2012 In Tags: , , ,

I’m sitting at Equinox on 44th Street (Thanks, Equinox for letting me work out at another gym other than my own in this mess!) writing this because I have to wait another hour for everything I own to be recharged (thanks, Sandy). However, it’s amazing what pops into one’s head in silence, darkness and no Internetz. Unfortunately, it also means no actual study review, since I can’t seem to reliably get online (that, and I’m feeling a bit lazy, so someone send me a link, eh?)

In case, you haven’t figured it out, my life in on hold while they try to restore power to my area of Manhattan. One of the hospitals I work at has actually evacuated all the patients, and the other main one I work at is running on emergency power only (no labs, no computers, minimal lights).

Of the news that trickles in on an unreliable 3G connection (I’m probably going to move to a friend’s place with power soon-ish), is the highly un-useful news that a few religious preachers of varying denominations have concluded that hurricane Sandy was caused by gay people, and more specifically, the passing of same-sex marriage laws. Now I’m not saying that it’s right or wrong. I mean, technically, I _was_ in Montreal during the great Ice Storm of 1997, a tornado touched down on a small village kinda close to Calgary when I was there, and every year that I lived in Halifax there was at least one hurricane that swept through the region; and well, here I am in New York. You put the pieces together…

But this actually got me thinking in the darkness: Post hoc ergo propter hoc, “After this, therefore because of this,” which basically posits that if A happened before B, therefore A caused B. Most regard this is as a common fallacy, in which the popular phrase, “correlation not causation” comes forth like a rising flood in a parkade.

So let’s examine this bit a little more. I’ve written before about how using post hoc propter hoc in correlational research is a bit of a low blow. But why is it always so disappointing?

The more I think about it, the more I wonder whether massive-population correlational study follows along the same vein as the ancient Greeks. That is to say, it offers AN explanation which can neither be truly verified, nor seems to get at the root of the problem.

The ancient Greeks believed that natural phenomena occurred because of the gods of Olympus. The Greek gods were, however, well-known to be petty and fickle. Athena was just as likely to help you slay a Medusa as she was apt to turn you into a spider. Posideon was well-known to sink your ship if you forgot to make a sacrifice; and if you had already made one, well, maybe it just wasn’t enough; or maybe you did something else to summon his ire. Lightning was the tool of Zeus. Volcanos were the smithy’s of Hephastus.

However dramatic these stories were, for lightning, volcanos, sunk ships or spiders, what they lacked was actual explanation. It’s all fine to say that lightning bolts were the weapons of Zeus and that he must not have liked that tree growing your backyard on account of it being nothing but a charred remains, but the story fails to explain why Zeus decided THAT tree, or how you might keep him happy and thus keep yourself lightning bolt-free.

The parallels of greek mythology and correlational research can be depressing. In particular, is the retrospective perspective. The Greeks observed a chain of phenomenon (the lack of sacrifice and subsequent loss of a ship; or the presence of sacrifice and the safe arrival of goods) and then after observing it, assigned a story to explain it. The main difference in their case being that they really had no way of observing the gods to see whether they got it right.

But that’s not a whole lot different than correlational research. We collect data on exposures and then outcomes, punch it all into a computer, spit out a regression analysis and then using the fine lens of reverse-time, look to make a story about, say, why red meat will kill you. In reality, the most we can use the data for is to make the linkage between exposure and outcome as an observation. It’s useful because it gives a hook to other researchers to latch onto when teasing the finer details of WHY apart, but in and of itself, is just as useful as saying eating red meat makes Zeus angry and thus, he smites carnivores earlier than vegetarians. Why then, do some vegetarians die early? Or better yet, why do some meat-eaters live longer? Are they somehow pleasing to Zeus in ways that we do not understand?

Thus, behaviour modification based on correlational data may be mal-founded (yes, I made that word up.) Some questions, however, can’t be answered through any other means for practical reasons. There will always be some level of Zeus-y-ness, particularly at the human-research (which some people call “clinical research”) level. I’m the first to promote the strength of human research design in its ability to definitively answers certain types of questions despite NOT knowing the mechanism by which something works (after all, I AM a methodologist). However, I will also be the first to acknowledge its limitations in being able to elucidate these precise mechanisms. The WHY becomes important when the Zeus-level is high. We still lack enough bridging between “basic” and “clinical” sciences, where we, as a research community can take true advantage of the strengths of both to make some real strides in human health. However, it’s an easy pick for a media outlet to pull the Zeus trick (knowingly or not) and spin everyone into a frenzy, because it makes for good headlines, but unfortunately not for any actionable information.

So should you always throw out correlation data? My approach has always been to regard it with extreme caution. I try not to modify my behaviour based only on correlation data, but in a dire situation (which doesn’t really happen when it comes to fitness and nutrition (let’s keep some perspective here people); I’m talking about my other hat as a surgeon), in the absence of any other data, it can be the only evidence to support a course of action.

PS. Please take the time to fill out my poll on my FB page

Click Here to view the Full Version of our Website