Performing within and between analysis (WABA) in Stata

The “fallacies of the wrong level” (e.g., ecological fallacy) are a concern in much of social-science research. When appropriate data are available, researchers frequently use intra-class correlations or ANOVAs to decide whether to use individual-level data or whether to aggregate and analyze data at some higher level, such as workgroups, schools, or locations to examine the associations among theoretical constructs. However, this approach disregards either lower-level information through aggregation or information on group membership by solely using individual-level data. Within and between analysis (WABA), on the other hand, provides a detailed picture of the correlation structure among constructs at different levels by partitioning correlations among constructs into between-groups and within-group components. WABA thus represents a more informative and efficient technique when interested in associations among constructs beyond mere within-group homogeneity. When dealing with more than two hierarchical levels, multiple within and between analysis indicates at which level associations are strongest. Therefore, WABA can also be fruitfully employed to inform model specification in HLM/MLM. In this presentation, I aim to outline the basic ideas and concepts of WABA and to introduce the user-written wabanova program to perform basic single and multiple WABA in Stata.