A method for comparing two normal means using combined samples of correlated and uncorrelated data.

A method is proposed for comparing two normal means when the data consist of some observations that are correlated and others that are uncorrelated. Typically, such data will consist of one subsample in which the observations for treatment 1 and treatment 2 are independent of each other, and another subsample which consists of paired observations taken under both treatments. The proposed method is developed using asymptotic results and is evaluated using simulation. The simulation results indicate that the proposed method can provide substantial improvement in type I error rate and power when compared with standard methods of analysis. Copyright 2003 John Wiley & Sons, Ltd.