In the CIA.Bias practice questions sections, there’s a question that asks about instances in which a model can be biased but fair. The answer provided mentions a model which charges different rates based on age. However, it says that the model might do so because its actuarially justified. Would that not make it unbiased? Wouldn’t bias be if the model was charging different rates for different age groups without actuarial justification?
Bias just means systemic discrimination which exists in pretty much any model as that is the purpose of segmentation. For example, you could have more young drivers in your dataset who happen to have more claims which means you are biased against them, but in a fair manner since this is actuarially justified. Your example would be one where the model is biased and unfair. A lot of times you can say young drivers are riskier, which we know from anecdotal evidence but it is often very difficult to point to a causal relationship but we accept correlation as sufficient. This is still biased in a sense.