I am estimating a logistic regression model to investigate how various factors influence second-year college retention. I include a variable indicating whether a student is in a minority racial-ethnic group and, as expected, find that this negatively impacts the probability of second-year retention. However, when I introduce first-year GPA into the model, the sign of the minority variable flips —- minority status positively influences probability of retention. I’m stumped. Any help appreciated.
Asked
Active
Viewed 31 times
2
-
2Welcome to Cross Validated! This sounds like something called Simpson’s Paradox. Perhaps you will have some luck searching that term on here, as this is a topic that tricks pretty much everyone at some point. – Dave Oct 21 '22 at 02:02
-
@Sycorax This question has been asked and answered? Regression coefficients that flip sign after including other predictors – dipetkov Oct 21 '22 at 07:04
-
@dipetkov That's a good example of a duplicate. You can vote to close questions as duplicates using the "Close" dialogue box. – Sycorax Oct 21 '22 at 08:13