This paper is from Session 8A: Inference in Times of Crisis, Part 1
which comes under Topic 8: New approaches to research in statistics education
Paper 8A3 (Friday 13th, 11:00-12:30)
Confidence intervals and replication - the reality
Presenter
- Ian Gordon (University of Melbourne, Australia)
Abstract
In the climate of “replication crises” across many disciplines, but particularly psychology, there has been a focus on the reproducibility of statistical inferences in various forms. Confidence intervals and the perspective of estimation have long been regarded as more insightful than P-values and hypothesis testing, for representing simple statistical inferences. It has been argued that P-values should be abandoned altogether. Two recent arguments in favour of this proposition, are that confidence intervals have (1) superior “replication” to, and (2) less variation than, P-values. I discuss these arguments and show they are unsound, and present a different statistical perspective.