Neuroscience Needs to Test Both Statistical and Scientific Hypotheses

Experimental neuroscience typically uses "p-valued" statistical testing procedures (null hypothesis significance testing; NHST) in evaluating its results. The rote, often misguided, application of NHST (Gigerenzer, 2008) has led to errors and "questionable research practices." Although the problems could be avoided with better statistics training (Lakens, 2021), there have been calls to abandon NHST altogether. One suggestion is to replace NHST with "estimation statistics" (Cumming and Calin-Jageman, 2017; Calin-Jageman and Cumming, 2019). Estimation statistics emphasizes the uncertainty inherent in scientific investigations and uses metrics, e.g., confidence intervals (CIs), that draw attention to uncertainty. Besides procedural steps and methods, the Estimation Approach prefers expressing "quantitative," rather than "qualitative" conclusions and making generalizations, rather than testing scientific hypotheses. The Estimation Approach embodies a philosophy of science-its ultimate goals, experimental mindset, and specific aims-that diverges unhelpfully from what laboratory-based neuroscience needs. The Estimation Approach meshes naturally with, e.g., clinical neuroscience, drug development, human psychology, and social sciences. It fits less well with much of the neuroscience published in the Journal of Neuroscience, for example. In contrast, the philosophy behind NHST fits naturally with traditional, evaluative testing of scientific hypotheses. Finally, some Estimation Approach remedies, e.g., replication, ideally with "preregistration," are incompatible with much experimental neuroscience. This Dual Perspective essay argues that, while neuroscience can benefit from practical aspects of estimation statistics, entirely replacing conventional methods with the Estimation Approach would be a mistake. NHST testing should be retained and improved.SIGNIFICANCE STATEMENT Experimental neuroscience relies on statistical procedures to assess the meaning and importance of its research findings. Optimal scientific communication demands a common set of assumptions for expressing and evaluating results. Problems arising from misuse of conventional significance testing methods have led to a proposal to replace significance testing with an Estimation Statistics Approach. Practical elements of the Estimation Approach can usefully be incorporated into conventional methods. However, the prevailing philosophy of the Estimation Approach does not address certain important needs of much experimental neuroscience. Neuroscience should adopt beneficial elements of the Estimation Approach without giving up the advantages of significance testing.