Bias in estimation of error in change-over trials with dairy cattle

A study has been made of the bias in error which occurs when Latin-square change-over trials conducted on dairy cattle are analysed by the usual method for Latin-square experiments, with modification for carry-over effects. Bias is present for adjusted direct effects and for permanent effects (direct plus carry-over), but does not exist for unadjusted direct effects. Analyses of fifteen sets of experimental data showed that the bias is of no importance in 3 × 3 designs but might be serious for some practical situations in the 4 × 4 designs. A tentative factor to correct for bias was given for the latter case.