Relevance Trees by Empirical Method.

With relevance trees, doubts about the adequacy of the tree structure or the validity of relevance numbers can seriously weaken inference from an analysis. The sophistication of methods for calculating relevance is evidence of the care needed if overall results are to remain pertinent. In this paper we present a method of carrying out a comprehensive relevance analysis by empirical means. With this approach, both the tree network, relevance numbers and supplementary statistics on cross-impact are obtained directly from initial data. Data may be quantitative and qualitative and we demonstrate the generality of our method by considering the problem of prioritising training needs in the retail trades. Results from this application suggest the method is robust across a range of technical options. Also where there are inadequacies or duplications in the original data it is likely they will be brought out in the analysis, improving the prospects of subsequent revision. The method is particularly well-suited to problems in the early stages of formulation.