I have a question. While performing logistic regression, I noticed that the odds ratio of an event happening can sometimes be confounded by other variables, which means certain variables may not actually be predictors of an event happening. I would like to know if there is a statistical procedure I can use that solves this problem.
Here is an example of what I mean: Say we know that women over 50 are more likely to choose breast conserving therapy. But in the logistic regression model, it also says that if you are Dr. Smith's patient, you are more likely to choose breast conserving therapy vs. being Dr. Jones's patient. However, a frequency table reveals that all of Dr. Smith's patients are over the age of 50, so it's not necessarily true that being Dr. Smith's patient means you are more likely to choose breast conserving therapy if Dr. Smith has little to no patients under the age of 50.
So is there a statistical method to figure out what the true predictors of an event happening are?
