The errors of long-term experiments

SUMMARY Yield (or other variate) values from crops recurring on the same plots in long-term experiments are subject to correlations known as serial plot correlations. Neglect of these plot correlations can result in loss of efficiency in the estimation of treatment effects and bias in the estimation of error. Serial plot correlations are calculated from 12 experiments on arable crops in England and Wales, for yields 4, 6 or 8 years apart. They are usually positive, with average value 0·2. In one set of experiments the losses in efficiency caused by plot correlations of this magnitude are shown to be small, but some of the biases in the estimates of error are large. Biases in the variances of treatment means can be wholly or partly eliminated by separating two components of error, plot error and plot x year error.