Analysis of means for experimental designs with missing observations