Selection Bias

Selection bias is a broad category of errors that arise from how data are selected. Sampling bias and survivorship bias are specific types of selection bias.

For example, suppose a bank wants to know whether a new savings product leads customers to save more. It compares customers who chose the product with those who did not and finds the first group saves more. But customers who chose the product may already have been keener savers. The difference may reflect who chose the product, not the effect of the product itself.

In finance, selection bias can appear in many ways. Credit scoring models built only on borrowers who were approved for loans may not work well for applicants who would previously have been rejected. Studies of companies that voluntarily disclose certain information may not represent companies that choose not to.

Researchers reduce selection bias through random selection, randomised experiments where possible and statistical methods that account for how the data were selected. Recognising potential selection bias is essential before drawing conclusions from any data.