I am comparing (between countries) how housing conditions mediate the relationship between working conditions and health.

Variables are: Work (Exposure = E), living conditions (Mediator = M) and health (Outcome = O).

The problem comes with the working condition indicator. In this case, I use the temporality of work (having a permanent/temporary contract) as an indicator of better/worse working conditions. However, the people within the temporariness is very different from country to country. In Northern European countries, temporariness is limited to very vulnerable profiles, while in the South it is more widespread .

This means that the regression estimates capture the compositional effect, and not the effect of the context (which is important to me). How can I solve this problem?


1 Answer 1


When you're measuring something, you need to ensure that the measure is equivalent in every group you are measuring.

It sounds like you're not measuring the same thing in every country. This is a problem.

If the measure doesn't measure the same thing, at best your results are doing to be difficult to interpret. At worst, they will be meaningless. You can hope that they lean more towards 'difficult'. Imagine something like asking an English and a German speaker if they run 'fast'. In English, that means run with speed. In German, it means 'almost' - so they can almost run. It looks like the same question, but it's not, and you can't use it to compare across two groups that speak different languages.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.