Report subgroup-specific false-negative rates alongside: Which analysis is decisive?
Aggregate accuracy can mask unequal harms; compare the relevant error rate within each subgroup as well as overall performance.
The question
A public-benefit eligibility model achieves 96% overall accuracy, but its false-negative rate is substantially higher for one subgroup. Officials need evidence that average performance has not hidden unequal access. Which analysis is decisive?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Report subgroup-specific false-negative rates alongside aggregate performance. ✓Subgroup false-negative rates directly test whether missed eligible applicants are unevenly distributed despite strong average accuracy.
- Compare total errors across the entire applicant population.Total errors can remain dominated by larger groups and obscure materially different error rates for smaller subgroups.
- Publish the overall accuracy with its confidence interval.Overall uncertainty does not reveal whether errors are concentrated in a subgroup facing greater risk of wrongly denied benefits.
- Replace the model with a simpler ruleset before measuring subgroup outcomes.Changing the system before measuring subgroup outcomes removes the evidence needed to understand the current model’s disparate error pattern.
The trap
Match the metric to the harm described, then calculate it separately for relevant subgroups. How to remember it
Aggregate accuracy can mask unequal harms; compare the relevant error rate within each subgroup as well as overall performance.
How many of these would you get right?
One of 1581 AIGP questions on Certsqill. Take a free five-minute check and see your score per domain — not one number, but which section to open tonight.
Test your AIGP readiness — freeMore Understanding How to Govern AI Development questions
- Split by policyholder before testing: What correction is decisive? →
- Reduce false negatives first: Which evaluation priority follows? →
- Conduct separate security testing for access: Which evidence is still required? →
- All 426 Understanding How to Govern AI Development questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Development ·
Every answer, right and wrong, comes with its own explanation.