Chapter Summary

Set-theoretic methods analyse social reality using properties of cases as set memberships and their relationships, which may be seen in terms of intersection, union, negation and subset. Configurational data analysis is a particular realization of such methods which focuses on combinations of set memberships that may be associated with outcomes that are the focus of the research. Analyses that use only crisp (binary) sets can simply look at case membership or non-membership of categories of potential causal conditions and outcomes, and then investigate which combinations of category memberships (configurations) are prevalent in a set of cases, which potential configurations have no empirical existence, which configurations appear to be sufficient to give rise to an outcome and which individual conditions appear to be necessary for an outcome to happen. Fuzzy set analysis extends binary set configurational analysis by allowing for degrees of category membership. However, not only is it still possible to evaluate which potentially causal configurations may be sufficient but not necessary for an outcome to occur, or which individual condition may be necessary but perhaps not sufficient, but also, as pointed out by Ragin (2000), the analysis in fact becomes sharper and more demanding.


The processes of data analysis outlined in Chapter 1 apply as much to configurational analysis as they do to variable-based analyses. Data still need to be prepared in various ways, for example, records need to be checked for usability and edited for consistency and accuracy. Each property then needs to be given a set membership value. Such values may be generated directly, indirectly or derived from traditional variables. Variables that are already in binary form should be given a value of [1] for the presence of a characteristic and [0] for its absence. Nominal variables will need to be converted into binary form, each category being given a value of [1] for membership or [0] for non-membership. Ordered categories or rankings will usually be converted into a single fuzzy set. Metric variables will need an algorithm to define an upper value beyond which full category membership is being proposed, a lower value below which non-membership is proposed and a crossover value of [0.5] for maximum ambiguity.

Data description will take the form of examining the nature of diversity among the configurations, while interpretation will be needed to link empirical diversity with theoretical expectations. Relating variables together takes the form of selecting an outcome that is the focus of the investigation, then looking at subset relationships to establish which single conditions may be causally necessary and which conditions or configurations of conditions may be causally sufficient. Evaluation is then required of causal statements that meet an acceptable level of raw consistency. For sufficiency statements where there are several pathways to an outcome, they are first minimized and then consistency and coverage are recalculated on the minimized statements. Levels of consistency and coverage will need to be tested for sensitivity to different frequency cut-offs, for different raw consistency cut-offs, for different assumptions about remainders and for the negation of the outcome. Researchers also need to check for triviality and irrelevance arising from skewed set memberships and for contradictory statements.

The explanation and presentation of results will require the researcher to bear in mind the likely familiarity or unfamiliarity of the audience with configurational analysis, linking the results back to theory and, ideally, comparing these results with other forms of data analysis. In many ways the application of configurational analyses to business strategy or public policy is much clearer than for variable-based analyses, since causally sufficient configurations can be seen as alternative ‘recipes’ for bringing about a desired outcome.

The results of a fuzzy set analysis of the alcohol marketing dataset show results that are very different from those that would have been obtained from a variable-based analysis. Configurational analysis has its strengths and its weaknesses. These are explained in detail and compared with the strengths and weaknesses of variable-based approaches in Chapter 9. The researcher, however, does not have to choose between these approaches; they can be mixed in various ways, which are explored also in Chapter 9.