14. Using Cluster Analysis, Qualitative Comparative Analysis and NVivo in Relation to the Establishment of Causal Configurations with Pre-existing Large-N Datasets: Machining Hermeneutics

14. Using Cluster Analysis, Qualitative Comparative Analysis and NVivo in Relation to the Establishment of Causal Configurations with Pre-existing Large-N Datasets: Machining Hermeneutics

In the social sciences, with the availability of online datasets we often have large-N – up to several thousand – datasets that contain large amounts of quantitative information about sets of cases. With all the appropriate and necessary provisos about casing (see Ragin 1992) taken into account, the cases in these datasets are typically rather tightly bounded institutions or administrative units, which can be taken as cases for the purpose of exploration and causal analysis. Typically, these datasets are available as spreadsheets, usually in some format that can be read into Excel and hence into statistical packages such as SPSS (Statistical Package for the Social Sciences) and into qualitative comparative analysis (QCA) software. A range of such sets is typically the product of the administrative collections of statistical information. Here, we can illustrate how we can type data using cluster analysis in SPSS and then use QCA as a preliminary quantitative exploratory tool in relation to outcomes for institutions and then move on to further qualitative [Page 261]investigation – if you like a tool-based version of the hermeneutic circle. The essential basis of such exploratory use revolves around identifying contradictory configurations in a first pass with QCA and the moving from these contradictory configurations into qualitative further investigation. This is often surprisingly easy, as, for a range of public institutions, not only do we have online datasets but also online qualitative reports based on inspections. These reports represent ‘found’ field studies and can be interpreted in order to establish further distinctive aspects among the cases within a contradictory configuration. In effect, we can use the packages and the methods contained within them in a machine shop fashion. SPSS allows us to create types, which is particularly useful in relation to outcome variables. As Rihoux and Ragin (2004, p. 18) have remarked:

Studies of this kind are particularly valuable when we are trying to assess achievement of clearly defined objectives for case entities in the form of public institutions that have such objectives specified for them by governments. However, we can extend the idea beyond this to consider outcomes as state conditions for any set of entities for which we have this kind of mix of quantitative and qualitative information. An example of the first kind that will be used to illustrate the procedure here is the set of English state secondary schools. For this set, we have a range of statistical descriptors and inspection reports available online. We can define outcomes in terms of the clear and specific goals associated with maximizing performance by cohorts of pupils in public examinations. An example of the second kind would be ‘post-industrial’ cities in the European Union that we can classify using a clustering technique and contemporary data, and then use historical data and qualitative materials to try to establish multiple causal paths towards different post-industrial citysystem states in the present.

...