Errors of Treatment Comparisons when Observations are Missing

Abstract

IN a randomized block experiment With J blocks and K treatments, for which σ ² is the variance of the random component of the observations and for which no observations are missing, the estimated difference of the means of two treatments has variance 2σ²/J. For the case where some observations are missing, Yates¹ has shown how the missing values may be estimated, a valid analysis of variance carried out, and an unbiased estimate of the difference of means of two treatments obtained. He has given the variance of this estimate for the case of only one missing observation, and has suggested an approximation for use in other cases which may be expressed as follows.