Statistical Models, Scientific Method and Psychosocial Research

This piece is a compilation of a number of short class-notes I wrote in 1987 and 1988 as a result of discussions with Ben and fellow students whilst I was a student at the University of Chicago. At that time Ben was pushing us to consider why progress in the psychosocial sciences seemed to be so frustratingly meagre when compared to progress in the ‘hard’ sciences. In discussions with Ben it seemed, to me at least, that central to his argument was a view that much of ‘so-called’ statistical modelling was unscientific—that it focussed on the description of ad-hoc collections of existing data, rather than proposing and rigorously testing of models and theories through the analysis of measures with well understood properties. Ben was very critical of exploratory statistical analysis, made a clear distinction between measurement models and analytic models and was always reluctant to fit statistical models to data—he wanted to use statistics as a tool to test whether data were consistent with theoretically posited models, he wanted to fit data to models. One wonders how he would have felt about the current big data and data mining movements.