Planning Falsifiable Confirmatory Research

Abstract

Falsifiable research is a basic goal of science and is needed for science to be self-correcting. However, the methods for conducting falsifiable research are not widely known among psychological researchers. Describing the effect sizes that can be confidently investigated in confirmatory research is as important as describing the subject population. Power curves or operating characteristics provide this information and are needed for both frequentist and Bayesian analyses. These evaluations of inferential error rates indicate the performance (validity and reliability) of the planned statistical analysis. For meaningful, falsifiable research, the study plan should specify a minimum effect size that is the goal of the study. If any tiny effect, no matter how small, is considered meaningful evidence, the research is not falsifiable and often has negligible predictive value. Power ≥.95 for the minimum effect is optimal for confirmatory research and.90 is good. From a frequentist perspective, the statistical model for the alternative hypothesis in the power analysis can be used to obtain a p value that can reject the alternative hypothesis, analogous to rejecting the null hypothesis. However, confidence intervals generally provide more intuitive and more informative inferences than p values. The preregistration for falsifiable confirmatory research should include (a) criteria for evidence the alternative hypothesis is true, (b) criteria for evidence the alternative hypothesis is false, and (c) criteria for outcomes that will be inconclusive. Not all confirmatory studies are or need to be falsifiable.