One of the first things taught in statistics, is that correlation does not imply causation.
Indeed, to say something about causation, one basically needs experimental or quasi-experimental design. Unfortunately, this can be difficult, or impossible in many contexts. Randomly assigning study participants to have the trait “extraversion”, is challenging, while randomly assigning employees to harassment, is unethical.
As a result, a lot of the data we use is observed, and the analysis correlational. Following from this, we are taught not to claim causality, and avoid it in our writing.
However, as Miguel Hernan, a biostatistician at Harvard, writes in a commentary, using alternative terms, like “associated” or “linked”, does not improve our writing, accuracy or communication. Some points to consider:
- When arguing for a theoretical causal model; you should use causal language, it adds clarity. And while we might be able to prove causality, the aim should be clear.
- There are very few designs which can prove causality in the social sciences; so it is a question of where we draw the line of acceptable uncertainty.
- By using causal language when formulating questions, sharpens the focus on the paper. By also discussing the quality of data needed to test such a question, and the difference between presented data, and that which was needed, clarifies the strength of the results.
- Control variables only make sense when testing for a causal claim, they are there to rule out alternative explanations and factors. If only examining associations, control variables should not be used.
The commentary is three pages long, and well worth reading! It is much better written than my notes above 😀
The American Journal of Public Health (AJPH) from the American Public Health Association (APHA) publications
The American Journal of Public Health (AJPH) from the American Public Health Association (APHA)
If you want to go deeper, Miguel Hernan is writing a book on Causal Inference; and a complete draft of the book is available here: