A Nonparametric Approach to Matched Pairs with Missing Data

Abstract

The matched-pairs t-statistic on the overall ranks is extended to data with observations missing at random. Either one of the two variables is allowed to be missing. The procedure is completely nonparametric. Comparisons with the likelihood ratio test for normal data indicate that the proposed method fares well when the data are normal and outperforms it in other cases. Simulations also confirm that the proposed method has higher power than common nonparametric complete-pairs tests for observations missing completely at random. Finally, a data set on the delinquent values of boys released from correctional institutions is analyzed and discussed.