ABSTRACT
The marginal pattern in the two houses is very similar, high weight being achieved for soyabean, low level of protein and high level of fish solubles. Analysis of variance in such situations serves two purposes. One is to determine an estimate of variance for assessing the precision of the contrasts of means. The second is to ensure that no contrast estimable from the design is overlooked. Usually in factorial experiments it is hoped that main effects and perhaps some two-factor interactions will turn out to be the only appreciable contrasts. Three of the factors represent treatments and one, 'houses', replication of the experiment and choice of an estimate of error is most reasonably based on the interactions with houses. This essentially hinges on the supposition that the treatment effects are the same in the two houses; even so it will be wise to check that the various component interactions with houses are roughly comparable.
