I often hear that people do not want to use causal inference because they don’t believe they can come to a causal conclusion. This fundamentally misunderstands what causal inference is for and what a causal conclusion is.
When should you use causal inference? When your question is causal. It’s as simple as that. If what you want to know involves a causal quantity, then you need to use causal inference. In case you’re not sure if your question is causal take a look at this paper by Ito et al or this paper by me and Katrina Kezios. One rule of thumb (I repeat RULE OF THUMB) is that if the word confounding appears in your paper, you are almost certainly trying to answer a causal question. Same with if you’re adjusting for more than three varaibles. Remember…RULE OF THUMB. Best thing is to sit down with your question and figure out what you want to know.
Still worried you’ll be over-interpreting your result by using causal inference? Good news! Or, it’s more like bad news. The same way buying a lottery ticket almost never results in you winning the lottery, using causal inference never results in being able to say with certainty that something is cause. Causal inference tells you how to conduct your study and tells you what is needed to interpret your results causally. And usually what is needed is to believe some pretty hard to believe assumptions.
If the assumptions are so hard to believe, why should you use causal inference? Because there’s no other way to study causes? Because causal inference is really hard and requires very careful thinking to decide how much causal evidence (if any) a study actually contributes. Sure, you can just interpret everything as an association and you don’t have to ever worry about being wrong because you can’t be wrong. Should you act on that associational evidence though? There’s only one way to know and that is…[drum roll]…to use causal inference to think through and interpret the results of your “associational” study.
