A recent New York Times article on beer consumption and academic performance has a good attempt at analyzing a correlation.

It starts with the negative correlation between beer drinking and publishing papers of merit, and goes downhill from there. They mention that beer drinking may cause non-publishing, and the converse that non-publishing may cause beer drinking. However...

Any correlation may have a "hidden cause" that was not measured but in fact drives the observed correlation. With beer-drinking and paper-writing, we note that both take time. We also note that we all have a finite amount of time (which differs per person but appears normally distributed).

So, this correlation may show that we all have a little slider for allocating time: the more beer we drink, the less time we have for publishing; likewise, the more publishing we do, the less beer-drinking we can do.

While that may sound obvious, we could test it by 1) measuring paper-publishing versus horse-riding or some arbitrary activity not related to academics, and 2) measuring the amount of time an academic spent on their work (difficult methodology as many academics would inflate the time spent working as their core identity probably revolves around working on papers).

The import of all this lies in establishing the next place to look. Just as we only have a finite amount of time for beer-drinking, we also have a limited number of resources for our next experiment, so we want to develop a theory of how the world works and then test it where our theory seems to have problems.