We use the WOE (Weight on evidence) method on one variable on the logistic regression. And we find the sign of the coefficient of that variable does not make sense. But when I use the origin value of that variable, instead of the WOE one, the sign of the coefficient changes and start to make sense. In my understanding, WOE is just the simplified snapshot of the original variable. Keep all other constant, how can it happens that the coefficient of the WOE has totally opposite signs from the original one when we replace the WOE with the original one in the model?