0

We know correlation is not causation. Consider following example.

Let's say we observe that on average people who smoke, die earlier that those who don't. This is correlation right? But how do you design a study that would exclude that above correlation exists because of following factor: people have a need to smoke because they have condition A, condition A is what makes them die early.

4
  • Welcome to Medical Sciences! Please take the tour and read the help center. For reasons mentioned in this post and in How to Ask, we require prior research information when asking questions. See this list of helpful resources. Please help us to help you and edit your question to provide more information on what you have read on this subject, what made you ask this question, and any problems you are having understanding your research. If you found nothing, what did you Google? – Carey Gregory Nov 19 '20 at 22:05
  • @CareyGregory I did research to find what is correlation, that correlation isn't causation. But can't think of an example how to design study which would determine that the example in my question is false correlation! – user20793 Nov 20 '20 at 9:14
  • It seems hypothetical and bootless if condition A is not specified. You would need to be very much more specific - after all data analysis to eliminate confounding factors relies very much on details. – A Rogue Ant. Nov 21 '20 at 3:37
  • Only just seen this. I have to agree with @Tantalustouch. What would an example be of condition A requiring someone to smoke? Lack of research information aside, this is a hypothesis which has no solid basis. More clarity is required on condition A example(s) and information on prior research will also be required. – Chris Rogers Nov 28 '20 at 11:32
1

I think this question is too vague for a definite answer, but areas to consider would include:

  • confounding vs effect modifier: where do smoking and condition A lie in relation to the outcome of mortality?
  • matched pairs study: if it is ethical, you could get a simple random sample consisting of people with condition A that do not smoke and pair each individual in that sample with another individual in a simple random sample consisting of people with condition A that do smoke.
  • case-control study: this study design is useful for if condition A is a rare disease (note: not rare exposure; in that case you would use a cohort study). You would then adjust for confounding variables based on your study population.

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