Rethinking data analysis (1): The limitations of frequentist approaches

Abstract

The analysis of data from market research has, until fairly recently, been reliant upon statistical techniques that were developed during the nineteenth and early twentieth century for uses entirely other than the analysis of survey and other types of observational, non-experimental data. Such techniques rely on reviewing and relating the frequency distributions of variables that have been concocted and measured by researchers. This article argues that key features of such ‘frequentist’ statistics are also limitations that need to be recognized by academics, market research practitioners and the managers to whom they report findings. By focusing on variable distributions across cases, they overlook patterns of within-case configuration; they seek out only symmetrical, linear patterns by reviewing the ‘net effects’ of individual variables; they rely on a very circumscribed view of statistical inference from samples to populations; they are not good at demonstrating causal connections between variable or at handling system complexity. A follow-up article, ‘Rethinking data analysis (2) Some alternatives to frequentist approaches’ examines ways of approaching datasets that can be seen as viable alternatives.