Rethinking data analysis (2): Alternatives to frequentist approaches

Abstract

In ‘Rethinking data analysis (1) The limitations of frequentist approaches’ (Kent 2008) it was argued that standard, frequentist statistics were developed for purposes entirely other than for the analysis of survey data; when applied in this context, the assumptions being made and the limitations of the statistical procedures are commonly ignored. This article examines ways of approaching the analysis of datasets that can be seen as viable alternatives. It reviews Bayesian statistics, configurational and fuzzy set analysis, association rules in data mining, neural network analysis, chaos theory and the theory of the tipping point. Each of these approaches has its own limitations and not one of them can or should be seen as a total replacement for frequentist approaches. Rather, they are alternatives that should be considered when frequentist approaches are not appropriate or when they do not seem to be adequate to the task of finding patterns in a dataset