Generalizing from "Imperfect" Replication

At the core of research on most marketing problems lie attempts to assess relationships between variables which are important to understanding how choice processes work and how they are affected by various factors which are either controllable or assessable by the marketing manager. These relationships may be aggregate and may involve hard, economic measurement (sales and advertising by brand, for example), or they may be softer and more disaggregate (response of different segments of consumers to certain advertising copy, for example). As understanding of buyers' choice processes builds up, so do collections of studies which use similar research techniques to approach similar sorts of problems. While these bodies of literature, published and unpublished, generally differ in enough important respects that they cannot reasonably be viewed as replications in a technical sense, they often use approaches similar enough to one another that they might be viewed as imperfect replications. In fact, methodological differences in a set of studies may allow insights not available from either a set of perfect replications or a single study. This paper proposes a procedure for grouping comparable studies into a quasi-experimental design and using analysis of variance procedures to An analysis of variance approach is suggested for evaluating the impact of various salient interstudy differences on the results of comparable but not perfectly replicative studies. As an application, multivariate analysis of variance is used to assess differences in parameter estimates across 37 quasireplicative studies based on the Fishbein Behavioral Intention Model.