ABSTRACT
A first crucial step is the choice of response variable or variables. In effect, taking different response variables amounts to posing different questions and it is not to be supposed that there is just one allowable choice. Inspection of the proportions is an essential step in descriptive analysis of the data. For more formal analysis it would be possible to apply analysis of variance directly to these proportions, ignoring changes in precision across the data and measuring effects directly in terms of differences of proportions. Partly because the proportions vary over a substantial range, the main analysis has used a logistic model. The data are fitted adequately by a linear logistic model containing the three main effects, with no interactions; the difference in 2 log between this and the full model giving perfect fit is 25.24 with 24 degrees of freedom. Each of the main effects is significant.
