The first part of the talk discusses the problems with Fisher’s approach to significance testing and the practice in psychology to publish only significant results. I then discuss Neyman-Pearson’s alternative approach, statistical power, and Cohen’s seminal meta-analysis of power in social/abnormal psychology. I then point out that questionable research practices must have been used to publish 95% significant results with only 50% power.
The second part of the talk discusses Soric’s insight that we can estimate the false discovery risk based on the discovery rate. I discuss the Open Science Collaboration project as one way to estimate the discovery rate (prettty high for within-subject cognitive psychology, terribly low for between-subject social psychology), but point out that it doesn’t tell us about clinical psychology. I then introduce z-curve to estimate the discovery rate based on the distribution of significant p-values (converted into z-scores).
In the empirical part, I show the z-curve for Positive Psychology Interventions that shows massive use of QRPs and a high false discovery risk.
I end with a comparison of the z-curve for the Journal of Abnormal Psychology in 2010 and 2020 that shows no change in research practices over time.
The discussion focussed on changing the way we do research and what research we reward. I argue strongly against the implementation of alpha = .005 and for the adoption of Neyman Pearson’s approach with pre-registration which would allow researchers to study small populations (e.g., mental health issues in the African American community) with a higher false-positive risk to balance type-I and type-II errors.