First, the decision to use causal language despite the weak id strategy is conscious. I think it is the most honest approach to make explicit causal claims in the theory because the theory is causal.
Comments
Log in with your Bluesky account to leave a comment
Many authors simply avoid using causal language and depict their research as „descriptive“ to avoid this line of critique. But, if you carefully read their theories, they remain causal (as most theories on the relationship between concepts are).
Substituting “drives” for “associates with”, etc., strikes me as obfuscation. This is why I decidedly use causal language – it is not because I fully believe in the validity of causal id assumptions underpinning the design.
However, I fully agree that causal language should be avoided in the presentation of the empirical results, and I edited the manuscript accordingly (precisely because of the weak identification).
Next, assuming that the research topic is worthwhile and that it addresses an important topic (which I would argue the present paper does), cross-sectional analyses may be the only approach to study the topic in a broad range of settings.
In an optimal world, there would be a natural experiment available for analysis. But I found no case where there is also access to suitable survey data. In my opinion, the best approach is to take the next best solution, that is, use a cross-section.
The alternative would be to drop the research topic, but this creates a world where researchers only look for evidence where there is an identification strategy, which leaves many blind spots akin to the drunk who only looks for his/her keys under the street light.
Comments