I was thinking about how I might conceptualize the increase in death rates that Anne Case and Angus Deaton observed last year, so I drew this monstrosity:
The funny thing here is not only that you could add more boxes/potential causal factors ad infinitum, but that it’s awfully hard to rule out one chain of arrows versus another. For example, this picture assumes that the way that social and technological changes increase obesity is by increasing inactivity and changing diet. But there’s a fair amount of evidence that the micro-biome (bacteria in the gut) might affect obesity directly as well as via changed eating and exercise habits, and I even have some crank hypotheses about atmospheric changes affecting obesity and metabolism that I haven’t entirely sworn off.
There’s a lot to be said for the Rubin Causal Model, where you imagine causes as the treatment in an ideal randomized experiment. Thinking in terms of counterfactuals and real-life treatment effects is very useful for sharpening what you mean when you say “X causes Y,” and is at the very least a good guideline for public policy, which is almost always in the business of considering marginal rather than average treatment effects. But there’s a long way from any of those kinds of “causes” to what we mean in our everyday way of asking what causes a social problem. Our ability to notice that the engine of our society is leaking oil or puffing out smoke is considerably ahead of our ability to open up the hood and figure out what is going on.