Misclassification Rate

[BTAinRVA]

Does anyone know if it is possible to control the direction of misclassification in a logistic regression model? I would rather misclassify a good customer as a bad customer rather than misclassify a bad customer as a good customer

Thanks for any insight!

[FeloniousPunk]

The assigned classification is dependent on the probabilities generated by the model. By default, if the probability is greater than .50, the classification is issued in the affirmative.

proc logistic data=CustomerData outmodel=logitModel;
 model Customer(Event='Good') = ShopFrequency MoneySpent;
 score data=NewCustomerData out=NewCustomerData;
run;

In this example, predicted values of I_Customer = 'Good' when P_Good > .50.

You can reassign the predicted value I_Customer by processing the data through another data step and requiring a lower threshold.

For example:

data NewCustomerData;
	set NewCustomerData;
	if P_Customer > .4 then I_Customer = 'Good';
	else I_Customer = 'Bad';
run;

I don't think artificially lowering the classification threshold has a methodological justification, but it sounds like this is a low-stakes application.