Hi All,
I received the following question via email and thought that everyone might benefit from my response. We encourage all of you to use this forum to post any questions that are relevant to the whole class. You then have the opportunity to consult one another + the entire instructional team monitors this forum closely.
Best,
Rebecca
Question
I have a question about exchangeability.We define exchangeability as 2 groups being interchangeable with respect to their response to treatment. how does this work in the situation where the exposure does not have an effect on the outcome (as in MMR vaccine and ASD)?Applied to the MMR article - those who receive the vaccine are different from those who do not receive the vaccine (likelihood of ASD as indicated through presence of sibling with ASD). But, the vaccine does not cause ASD in anybody so the effect of the vaccine in all people is going to be the same (i.e. no effect).Should we be sticking to the impact of the exposure on the treatment (i.e. all groups are exchangeable when an exposure doesn't have an effect) or recognizing those underlying group differences when thinking about exchangeability?
Response
Whether or not two groups are exchangeable does not depend on whether or not there is a causal effect of exposure on outcome. In the absence of a causal effect, lack of exchangeability can result in an association between exposure and outcome. Even if no association is present, however, that doesn't mean that exchangeability holds. Confounding factors with opposing effects could render a null association.
For the purposes of the homework, consider whether or not the authors have accounted for the variables that you think would likely confound the relationship of interest.