curiosity across disciplines

Clocks cause correlations?

I'd like to share a helpful metaphor I came up with[1] for differentiating between correlation and causation.

Correlation is ceremoniously trotted out for examples such as:

  • increased ice cream consumption is associated with greater numbers of shark attacks, and
  • from 1996-2000 increasing imports of fresh lemons from Mexico were tightly linked with a proportionate decrease in US highway fatalities.

There's all manner of reasons for correlations, ranging from the "shared third factor" seen in seasonal effects of more ice cream consumption during warmer swimming months relating to more shark deaths, to "if you have a big enough pile of data, a random statistical lemon is bound to show up."

However, it's easy to mentally intermingle correlation and causation. Too easy.

So, as a helpful shorthand for myself, I've been thinking of this dividing line in the sand using a clock.

Clocks, as we traditionally imagine them, are made of a bunch of internal gears of various sizes.

The gears turn, which moves the hour, minute, and second hands along their winding path to infinity. Aided by a power source, the gears cause the hands to rotate. They're mechanistically linked.

In contrast, that's not necessarily the case with correlation.

It's theoretically possible that sharks, having existed for millions of years longer than humans, already cracked unified field theory and are altering ice cream at the subatomic level to lure more of us chum into their intelligent-but-content-with-the-simple-life jaws.

However, we've only got evidence for the "more frozen snacks, more sweet-ish fish" bit.

So, to avoid such causal confusions, I propose the following: correlation is observing the clock from the outside. Causation is understanding something about the gears.

Watching the clock from the outside, we notice when the patterns sync up, but that doesn’t give us any insight into the working of the interior mechanisms. Correlations are like that. Causation, in contrast, is ripping the clocks guts out and seeing what it’s made of.

This heuristic is surprisingly helpful, despite how basic it is, specifically because, rather topically, humans are lazy, pattern-seeking machines. It's unavoidable that we conflate cause and correlation.

Pausing to mull over the question, “is this a clock-from-the-outside, or is it a clock-from-the-inside situation?” seems to be just enough space to not leap immediately from "monkey see pattern" to "monkey know pattern."

You can use this in a variety of situations, ranging from being a more clever news consumer, not getting swept up in overreaching narratives, to better contextualizing research.

Nutrition research, for example, is notorious for muddying the waters between correlational and causal studies because of how difficult it is to isolate specific variables.

To not-really summarize: if time is money, then correlations are a dime a dozen. And causation is...this sentence grinding your gears to induce your current eye rotation.

  1. I'm one step removed from certain that someone already came up with this, but I'm far too momentarily enamored with the possibility that I had an original™ idea that I refuse to check. ↩︎