SAGE Journal Articles
This article reminds us that the trend is towards huge, exhaustive datasets, resulting in a paradigm shift so that instead of seeking out data to test theories, analyses are data driven so that datasets are scoured for patterns that might be useful; in short, a modern version of data dredging. So, this phenomenon is not just limited to over-active bivariate analyses, but is a growing feature of multivariate analyses as well.
Although written for psychologists 16 year ago, this article discusses the possibilities for univariate, bivariate and multivariate analysis of nominal variables. It argues that quite a lot can be achieved by examining relative frequencies of the various categories. It then discusses the phi, Cramer’s V and lambda statistics for measuring the degree of bivariate association and then goes on to consider multivariate procedures including the use of dummy variables in regression analysis and discriminant analysis, which is considered in Chapter 8 of Kent (2015).
In this article, Schrodt is strongly criticising accepted ways of analysing data in political analysis, but his criticisms are appropriate to all the social sciences. He refers to ‘dysfunctional habits that have ‘rendered much of contemporary research more or less meaningless’. Thus he refers to ‘garbage can’ models that ignore the effects of collinearity, pre-scientific explanation in the absence of prediction, misinterpreting frequentist (variable-based in Kent (2015)) statistics, a monoculture that relies on regression-based linear analyses, and confusing statistical controls with experimental controls.