Preregistrations without Code do not Prevent P-Hacking

Preregistrations have been suggested as a tool to prevent p-hacking. Most researchers have used semi-structured preregistration templates to deprive themselves of researchers’ degrees of freedom. I show how such “narrative preregistrations” do not fully prevent p-hacking for even very simple statistical techniques (e.g., correlations) unless the analysis code is part of the preregistration.