Cluster Analysis in Educational Research

Abstract

AbstractThis paper examines the use of cluster analysis in educational research. Its first conclusion is that the important issues are not technical but interpretative. While we must, of course, reject clusters which fail to meet minimum technical requirements, it does not follow that those which are technically admissable are therefore meaningful explanatory constructs. In general, I will argue, they are not.The principal difficulty with cluster analysis is that it introduces a concept of ‘type’ which is unscientific. As a result, we become embroiled with character‐sketches and pen‐pictures more appropriate to literature than to research. This point has been touched on by critics who find cluster analysis to be ‘gross' in certain respects: in its “failure to pinpoint significant teaching variables”, in its tendency to “intensify” what are described as “unproductive polarizations” (Elliott, 1978, p. 77), in its propagation of “emotionally laden catch‐all terms” (Wragg, 1976, p. 285). The second conclusion of the paper is that when the ‘grossness' of cluster analysis is understood and accepted, the technique can nonetheless serve an important heuristic function, provided certain conditions are met. The paper refers to the Lancaster and ORACLE studies to illustrate the problems and possibilities for cluster analysis in educational research. It ends with some recommendations.