A Look at p-values from Another Angle (Subtitle: Do Not Take Theory Too Seriously)

Abstract

The journal, Basic and Applied Social Psychology, announced in an editorial in 2015 that it would ban the reporting of p-values in its articles. It was partly prompted by an increasing concern in the scientific community that most current published research findings are false. The editorial started a lively discussion in the statistical community, with some blaming the problems on improper use of p-values (such as failing to control for multiple testing) and others arguing that p-values should be replaced by confidence intervals. Some claimed that the rate of false positive results will lessen if sample sizes are increased. This talk will address the problem by revisiting the origin of hypothesis testing and confidence interval estimation and show that if the theory is taken seriously, p-values and confidence intervals are often meaningless, regardless of sample size.