Critical Replications for Statistical Design

Abstract

While writing a new book [Wiley ’87] I found several distinct statistical problems, each with a separate solution, but all in need of a higher principle of justification. I looked in vain into statistics for a higher principle or criterion which we need to serve as a common foundation for those similar solutions. Then I thought that “falsifiability” could be borrowed from the logic of scientific discovery, from the philosophy of science, if adopted in a suitably modified form to our needs in statistical design. However, discussions and correspondence with several philosophers of science and several statisticians suggested that I was wrong in trying to stretch falsifiability to cover our needs in statistics. I may also have been wrong in believing that falsifiability was both well known, well established, and well accepted as a logical principle of scientific discovery. Thus I am forced to propose a new name: Critical Replication. I can suggest several alternative names, some suggested by others, and ask for your preferences: sturdiness, robustness, resilience; sturdy conditioning, (re)conditioning, probing, replicability, generalizability, multiplicity.