Matching and Prediction on the Principle of Biological Classification

Matching criteria have often been selected on the basis of either custom or hunch, and their application usually involves discarding the test results of those who do not fit into the matching pattern. There are, however, better ways of selecting the matching criteria and more efficient ways of applying them. The all-important point in matching is the use or development of relevant matching criteria. A simple test of the relevance of a matching criterion is whether or not it is associated with or predictive of whatever is being studied. Here are some examples: (i) Suppose that the whole purpose of a particular matching operation is to study the relative effects of two different stimuli upon 'activity X'. If the matching criteria used are correlated with 'activity X', they qualify as relevant. If not, they serve no useful purpose. (ii) Similarly, if consumer or listening panels are to be made representative of the public (i.e. matched to them), the matching criteria used will be relevant only if they are associated with the measure or measures which the panel is meant to provide. (iii) The same applies to the controls used in quota surveys. Relevance is, however, a matter of degree and the best matching criteria are those which, taken together, have the highest available matching power (granted, of course, adequate reliability). This can only be achieved, with any approach to certainty, by empirical methods. These involve establishing the predictive power of each of a wide range of possible matching variables and then developing from them the best possible predictive composite. In earlier work' r had established the relative predictive power of the proposed matching variables by ordinary correlation methods. After this, the selection of the best composite of these had been effected by means of the Wherry-Doolittle technique.2 This procedure was a considerable departure from 'matching by hunch', but it had certain methodological weaknesses and it was tedious in the extreme. The method which is described here was subsequently developed to simplify the prodecure and to increase the efficiency of the selective process.