Introduction

The ideal form of statistical consultation starts with a discussion between the research worker and the statistician on the design of the whole research project, as well as the design of any component experiments or surveys. In this way, the statistician can ensure that all the data that are collected can be fitted together into a comprehensive model of the processes that underlie the phenomena being investigated. Only then does he help the research worker to identify the factors to be controlled or varied throughout the research, and the variables to be measured. The methods of mathematical analysis that will be used to test hypotheses, or to estimate parameters, then follow logically from the design of the whole project and its component experiments.

Unfortunately, very little consultation with statisticians follows this ideal pattern. Indeed, few research organisations have enough statisticians for more than a very small proportion of research projects to be planned with the active collaboration of the statisticians. What frequently happens, therefore, is that the statistician is consulted long after the data have been collected, and, by definition, long after the design of the investigation has been fixed and partly executed. It is also only very seldom that much thought will have been given to the integration of several (or many) experiments or surveys, so that, if the statistician is consulted at all, it will usually be about the analysis and interpretation of a single experiment or survey, without much regard for the whole investigation into which the individual research activities fit.