Does the selective attrition of a panel survey of older people affect the multivariate estimations of subjective well-being?