Chapter Summary

Cluster analysis uses traditional variables to analyse potential grouping structures among cases, based on measures of ‘distance’ between them. While it is perhaps not as essentially case-based as configurational analysis, it is more case-based than variable-based in the sense that the focus is on predicting groupings for individual cases, as evidenced by the fact that SPSS will assign – if asked – each case to a grouping, creating a new variable in the process. The same is true for discriminant analysis; the difference is that cluster analysis generates the groupings from the data, while in discriminant analysis the researcher selects a nominal or ordered category variable as the basis for groupings, and a sample or sub-sample is used to estimate how good a selection of metric variables are in predicting those groupings. In terms of the different types of multivariate analysis suggested in Chapter 6 in Kent (2015), cluster analysis is an interdependence technique in that no distinction is drawn between dependent and independent variables, while in discriminant analysis, the grouping variable is always dependent on the metric predictor variables and so is a dependence technique.

Both methods are complex in the variety of algorithms that may be used and the SPSS outputs of both can be quite daunting to interpret. Both use the equivalent of crisp set (binary) category membership of the groupings, except for fuzzy cluster analysis, which allows for degrees of membership. Its advantage over fsQCA, however, is unclear, particularly since interpretations in terms of asymmetrical relationships, necessary and sufficient conditions cannot be made. The procedures for doing fuzzy cluster analysis are not, in any case, available in SPSS.