The analysis of change-over trials

Abstract

The analysis of simple change-over trials based on Latin squares is considered. It is shown that several difficulties are encountered when estimates of residual effects are required. In particular, there may be biased error variances. In certain circumstances the amount of bias in the error variance can be estimated in a proportion of the experiments based on a given design. If this bias is sufficiently large to be of importance an average correction can be calculated for use with further experiments of the same type.