The interplay between subjectivity, statistical practice, and psychological science

Bayesian inference has been advocated as an alternative to conventional analysis in psychological science. Bayesians stress that subjectivity is needed for principled inference, and subjectivity by-and-large has not been seen as desirable. This paper provides the broader rationale and context for subjectivity, and in it we show that subjectivity is the key to principled measures of evidence for theory from data. By making our subjective elements focal, we provide an avenue for common sense and expertise to enter the analysis. We cover the role of models in linking theory to data, the notion that models are abstractions which are neither true nor false, the need for relative model comparison, the role of predictions in stating relative evidence for models, and the role of subjectivity in specifying models that yield predictions. In the end, we conclude that transparent subjectivity leads to a more honest and fruitful analyses in psychological science.