Group Fairness refers to?

Get ready for the GARP Risk and AI Exam with flashcards and multiple choice questions. Each question comes with hints and explanations. Prepare for success!

Multiple Choice

Group Fairness refers to?

Group fairness is about ensuring outcomes do not differ systematically across predefined groups defined by protected attributes. The goal is to prevent disparities that arise purely from group membership, by using criteria that compare how different groups are treated. A common concrete form is demographic parity, where the rate of positive decisions is the same across groups, so no group is favored or penalized at the overall outcome level. Another related form is predictive rate parity (predictive parity), which aims for the same positive predictive value across groups, so a positive prediction is equally likely to be correct regardless of group. Individual fairness, by contrast, focuses on treating similar individuals similarly, which is a different approach that doesn’t hinge on group-level comparisons. The idea that specific groups are systematically disadvantaged by an algorithmic outcome signals bias or unfairness, not a fairness criterion itself.

Group fairness is about ensuring outcomes do not differ systematically across predefined groups defined by protected attributes. The goal is to prevent disparities that arise purely from group membership, by using criteria that compare how different groups are treated.

A common concrete form is demographic parity, where the rate of positive decisions is the same across groups, so no group is favored or penalized at the overall outcome level. Another related form is predictive rate parity (predictive parity), which aims for the same positive predictive value across groups, so a positive prediction is equally likely to be correct regardless of group.

Individual fairness, by contrast, focuses on treating similar individuals similarly, which is a different approach that doesn’t hinge on group-level comparisons. The idea that specific groups are systematically disadvantaged by an algorithmic outcome signals bias or unfairness, not a fairness criterion itself.

More practice questions