Pre-registration of data collection strategies, research designs and model specifications is said to enhance the credibility of published empirical results. This claim can only be true if analyses in published articles stick closely to the pre-registration plan and explain and justify any deviations from this plan.

This article examines a sample of pre-registered studies published in political science journals to assess how closely researchers adhere to their pre-specified hypotheses. The analysis shows that a majority of pre-registered studies introduce unregistered hypotheses, omit pre-registered ones, or substantially revise their original formulations. Substantive changes to the hypotheses allow fishing strategies and p-hacking in pre-registered studies.

As a consequence, the credibility and validity of results from pre-registered studies cannot be assumed simply because a research plan has been pre-registered.