Moving beyond p<0.05 in ecotoxicology: A guide for practitioners
Statistical inferences play a critical role in ecotoxicology. Historically, Null Hypothesis Significance Testing (NHST) has been the dominant method for inference in ecotoxicology. As a brief and informal definition of the NHST approach, researchers compare (or test) an experimental treatment or observation against a hypothesis of no relationship or effect (the null hypothesis) using the collected data to see if the observed values are statistically significant given predefined error rates. The resulting probability of observing a value equal to or greater than the observed value assuming the null hypothesis is true is the p-value. Historically, criticisms of NHST have existed for almost a century and more recently these have grown to the point where statisticians, including the American Statistical Association, have felt the need to clarify the role of NHST and p-values in science beyond their current, common use. These limitations also exist in ecotoxicology. For example, a review of the 2010 Environmental Toxicology & Chemistry (ET&C) volume found many authors did not correctly report p-values. We repeated this review looking at the 2019 volume of ET&C and the incorrect reporting of p-values still occurred almost a decade later. Problems with NHST and p-values highlight the need for statistical inferences besides NHST, something that has long been known in ecotoxicology and the broader scientific and statistical communities. Furthermore, concerns such as these led the Executive Director of the American Statistical Association to recommend against use of statistical significance in 2019. In light of these criticisms, however, ecotoxicologists require alternative methods. In this paper, we describe some alternative methods including confidence intervals, regression analysis, dose-response curves, Bayes factors, survival analysis, and model selection. Lastly, we provide insights for what ecotoxicology might look like in a post-p-value world.