Issues in Classification in Meta-Analysis in Substance Abuse Prevention Research

Abstract

Meta-analysis as a method holds great promise for allowing fields of research to accomplish synthesis and integration of findings. This goal must be compared to experimental research, which is inherently reductionistic in its approach. Because of this divergence in methods and implicit goals, meta-analysts are often faced with a need to reconceptualize original research in order to fit it into a method that allows comparison. The authors have identified two such issues (classification of variables and classification of correlational results) that will pose continued dilemmas for meta-analysts. This chapter has two goals. The first is to discuss strategies to create schema for classifying independent variables. The second is to discuss issues of classifying types of correlational relationships between independent and dependent variables. Both have practical relevance for incorporating theory into meta-analytic practice. CLASSIFYING INDEPENDENT VARIABLES Creating a classification schema for independent variables is a major dilemma for meta-analysts. Yet, inevitably, completing such work is among the first steps that one must take in beginning a meta-analysis. The resulting schema will ultimately determine much of the meaning that emerges from subsequent analyses. An extensive literature about analytic methods has emerged. Topics include attention to effect size estimates (Glass et al. 1981; Hall et al. 1994) and controls for methodological problems (Cook et al. 1992; Wortman 1994). What is being analyzed, which lies at the root of wanting to complete analyses in the first place, has unfortunately received less attention (Cooper 1990; Orwin 1994; Stock 1994). It is the authors' observation that creating links with theory in completing meta-analyses gives meaning and value to the methods. Categorization of independent variables is challenging because no single theory captures all available variables. Theories that guide reductionist research focus only on relevant variables, ignoring variables that are