I’m reading an interesting article which Marc Lipsitch tweeted about.

The C-Word: Scientific Euphemisms Do Not Improve Causal Inference From Observational Data by Miguel Hernan.

The main take-away messages for me are that

1. Almost all scientific studies are aiming for causal inference, but people working on non-intervention, non-randomised studies (aka association/descriptive/exploratory studies) are generally discouraged from talking about causation.
2. Even randomised trials, which are generally accepted to infer causation, are still ‘just’ association studies which aim to randomise all relevant confounders between treatment and control groups, thereby minimising confounders.
3. Observational studies and intervention studies are on the same spectrum (albeit at opposite ends) when it comes to causal inference, both use the same idea of association.
4. Accepting that we are doing causal studies (even when just observational) will lead to better formulated questions/hypotheses, better adjustment for confounders.

It’s great to have this article by one of the grand fromages of causal inference to provide cover in case we want to talk about causation in an observational study.