I am new to logistic regression and I am trying to determine whether it makes sense to weight a particular case in my data which is oversampled in order to model my data better. I'm not even sure if this is something that can be done, but I was wondering if anyone knew for sure. I know you can apply a greater weight to one of your binary labels (dependent variables) in order to make that class more important. But I'm actually trying to weight particular cases in the data - more from the standpoint of features in the model. So for example if I was modeling whether or not someone is going to pass a kidney stone and my inputs were age and amount of caffeine consumed, and it so happens I have sampled a ridiculously large amount of data from people who are over the age of 60 and drink more than 4 cups of coffee a day, and I think that is going to really inflate that "scenario", could I put a slightly lesser weight on all of these cases and mitigate the problem?

Sorry if this question is confusing or less technical than it should be for this site. I tried googling but had a tough time finding information about the use cases for applying weighting in logistic regression.

Thanks, Christie


The answer is yes.

Two important aspects of modeling: internal and external validity.

To help to ensure internal validity, include the age variable. To help ensure external validity (so that you results can be extrapolated to the population) you must weight your sample by these known age population proportions using inverse probability of selection weights.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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