How to Correct Sample Bias: Ignoring it can undermine the 'prediction' in predictive models

File uploaded by on Feb 19, 2015
Version 1Show Document
  • View in full screen mode

Author:   Nina Shikaloff





Assessment of credit risk or worthiness is one area where credit decisions frequently reflect selection bias, resulting in a smaller population identified as ‘credit-worthy’—that is, consumers who are approved for credit. Accordingly, models developed only from data on consumers already approved for credit are ideal to demonstrate the dangers of sample bias, and the tools that can be used to correct it. In this paper we discuss both those topics, and use a case study with a population of credit applicants with a simulated credit decisioning process to demonstrate how different corrected and uncorrected outcomes can be.