Statistical Inference in Behavioral Research: Traditional and Bayesian Approaches
Null hypothesis significance testing (NHST) has long been a mainstay of scientific research, more in some scientific fields than others. It persists despite numerous calls across multiple scientific disciplines to abandon or at least modify the practice. In 2016, the American Statistical Association issued a statement decrying the use of the “bright line” p < 0.05 criterion as leading to a “considerable distortion of the scientific process.” There are a number of alternatives to NHST that don’t share its logical and practical deficiencies. First among them is Bayesian inference, which can be viewed as both a calculus of evidence and of belief. The Bayesian definition of “evidence” differs profoundly from what the p-value represents. In this chapter, we review deficiencies in NHST and provide an introduction to Bayesian reasoning, with particular attention to its relationship to the truth of scientific claims.