A model's precision is computed separately for each demographic subgroup. A stakeholder wants the aggregate precision across all groups. Why can't they simply average the per-group precisions?