A Method to Reduce Type 1 Error while Maintaining Power Using Split Samples

Abstract:

We discuss a method aimed at reducing the risk that spurious results are published. Researchers send their datasets to an independent third party who randomly generates training and testing samples. Researchers perform their analysis on the former and once the paper is accepted for publication the method is applied to the latter and it is those results that are published. Simulations indicate that, under empirically relevant settings, the proposed method significantly reduces type I error and delivers adequate power. Unlike alternative approaches such as the registration of a pre-analysis plan, this method allows researchers to learn from the data.