## When a bunch of things are correlated, one of which is time

We know this is a fallacy:

1. X has increased over the last 50 years.
2. Y has increased over the last 50 years.
3. Therefore either X causes Y, or Y causes X.

If we add an exit clause like “or there’s confounding” to 3., we weaken the argument to uselessness.

Now, although we can’t eliminate the possibility of confounding, we can get interesting evidence if there’s more to the data than “both increase”. If the peaks and troughs are simultaneous, then there’s some kind of strong relationship between the variables, whether causal or not. If one variable consistently leads the other, this can suggest direction of causation, though it’s easy to kid ourselves.

Barring such clear pattern, we need a more complicated causal model. It doesn’t have to be too complicated: something as simple as “X causes Z, which in turn causes Y” is interesting enough to have implications, if true, and is somewhat more susceptible to falsification. We can add more detail to the model as necessary. But you need to be explicit about pathways. Draw a picture if necessary.