Викторина, связанная с руководством · Средний Уровень
Викторина о дискриминации по доверенности в ценообразовании на страхование
Изучите прокси-переменные, аудит цен на страхование, актуарную поддержку и почему законы штата важны для разных продуктов и юрисдикций.
Вопрос 1 из 8
Почему удаление защищенного поля не доказывает справедливость модели ценообразования?
Предварительный просмотр каждого вопроса8 вопросы
Ответы и пояснения появляются по мере прохождения теста.
- Почему удаление защищенного поля не доказывает справедливость модели ценообразования?
- Other variables may still correlate with and act as proxies for that trait
- Removing one field guarantees identical outcomes across all groups
- Any model without race as an input is exempt from insurance rules
- A protected trait can only be predicted from the exact same field
- Что должен исследовать страховщик, если функция коррелирует с защищенной группой?
- Whether the overall model has a high accuracy score
- Whether the vendor says the input is neutral
- Its provenance, relationship to insured outcomes, and effects on rates and eligibility
- Only whether the feature name contains a protected term
- Является ли наблюдаемая разница в тарифах сама по себе нарушением закона?
- Yes, if a model rather than a person set the rate
- No; it signals a need for analysis under applicable law and evidence
- Yes; every difference proves unlawful discrimination
- No; disparities can never matter in insurance
- Как следует описать юридическую силу типового бюллетеня для штата, оценивающего рекомендации NAIC?
- It automatically becomes binding law in every state
- It overrides every state insurance statute
- It applies only when a consumer signs consent
- Its legal effect depends on adoption and applicable state requirements
- Какой подход к аудиту наиболее полезен для обнаружения прокси-эффектов?
- Trace features and compare outcomes across relevant groups and geography
- Check only whether sensitive fields were deleted
- Review only average model accuracy
- Compare the model with a second black-box model
- Why include missing-data handling in an insurance pricing review?
- Missing data always means the applicant has low risk
- A model cannot use any record with a missing field
- Missing values have no effect after normalization
- Missingness may vary by group or location and affect resulting risk tiers
- What does a high predictive accuracy score establish about fairness?
- It proves no proxy is present
- It eliminates the need for state-specific review
- It does not by itself establish that outcomes are fair or legally compliant
- It proves every feature has an actuarial basis
- What should insurers do after identifying a plausible proxy problem?
- Treat one audit as permanent approval
- Document the finding, assess legal and actuarial support, correct as needed, and monitor
- Ignore it if the sensitive field was excluded
- Delete every feature correlated with any group without review