Dear RG members. When studying concepts about confounding one can find that a possibility about "cofounding due to random associations" may exist.
My question is:
How a process of random sampling could create an imbalance between comparison groups with regard to a given variable? and how this situation could lead to confounding? I mean, how a random distribution of the experience of exposure and/or disease could create confusion in an observational design?
Also... How to identify confounding in a clinical trial?
I would appreciate your comments and response (also if you can provide me with some bibliography). Thanks!
Kind regards.
Ramiro
confounding, epidemiology, bias, analysis
I'm curious as to where you read about this. It doesn't show up in a couple of quick literature searches.
In small studies, the natural random variation in covariates can potentially create confounding, but I'd like to see a real example.
Dear Ronan.
I read it in "Epidemiology Beyond the Basics; Szklo and Nieto" page 157. You can see it from the screenshot I am attaching with this message. Also please look at this reference: http://www.bmj.com/content/345/bmj.e7951 ... Precisely I'm looking for a real example too.
Looking forward to your response.
Kind Regards
Ramiro
If you think it's easy to remove confounders, remember that you have to measure them first, and, second, model them out.
Dear Ronán Michael Conroy I greed with you that we have to measure them first, and, second, model them out
The real problem is that there is no 'test' for confounding. I teach my students that they should examine correlations and consider as a potential confounder any variable that is correlated with both the exposure and the outcome. But only subject matter expertise (clinical understanding) can determine whether confounding is likely. This is particularly important, since a mediator (on the causal pathway) will show the same correlation patterns. And while it's easy to 'adjust' for confounders using a variety for statistical models, this would not necessarily be appropriate for causal mediators.
Dear responders.
Thank you for your answers. I appreciate your effort in making knowledge sharing possible.
Kind regards.
Ramiro
the problem of confounding does exist as we can adjust only known factors but there are various unknown factors that can't be adjusted in observational studies for that randomization is necessary