Resumo:
This paper was written primarily to illustrate the process of fitting constants to non-orthogonal data. The design of the experiment, on the pruning of lower branches of coffee-trees, was systematic (see plan at the beginning). Three constants were fitted, representing respectively the treatment contrast, the difference between upper and lower half of the field and the linear gradient of fertility across the field. The use of the residual sum of squares for the estimation of error is not, of course, logically correct but it is unlikely to lead to seriously wrong conclusions if caution is exercised. In the event, the procedure failed to eliminate any soil heterogeneity (see Analysis of Variance, p. 288). Inspection of the data showed that much of the heterogeneity would have appeared in a quadratic term in the gradient. When, as here, the number of data is very small, it is particularly dangerous to allow the data to suggest the scheme of adjustment, because this will lead to an over-estimation of the precision of the experiment. The conclusion was that not pruning the lower branches of tree gave an increase in yield of 104 ± 50 kg (dry fruits) per plot, with a general mean of 743 kg per plot.
