MCAR auxiliary-variable efficiency
ID: mcar-auxiliary-variable-efficiency
With a fully observed covariate taking finitely many values of positive probability, response of finite variance, and an independent response indicator with probability , let . A covariate-standardized mean has asymptotic variance , whereas the complete-case mean has variance . The difference is . Thus missing completely at random does not justify discarding predictive observed covariates on efficiency grounds.
New to topics? Read the docs here!