The underwriter and the models - solo dances or pas-de-deux? What policy data can tell us about how underwriters use models