Mitigating LLM-based p-Hacking by Preregistering for the Next LLM
For researchers using LLMs in hypothesis testing, this protocol provides a practical method to reduce p-hacking, a growing concern in LLM-based research.
The paper proposes a protocol to mitigate p-hacking in LLM-based research by preregistering experiments and running them on the first eligible LLM released after preregistration. The protocol blocked successful transfer of p-hacks in 73.9% and 72.7% of cases across two tasks, and in a preregistered experiment, hacking failed to carry over in 6 out of 7 configurations.
Large language models (LLMs) are increasingly used to generate, classify, and annotate data whose outputs feed downstream hypothesis tests. However, LLM-based research is easy to p-hack: a researcher can tune the prompts, decoding parameters, or output format until a desired result is reached. We propose a protocol to mitigate p-hacking in LLM-based research: preregistering the experiment and eligible models, and then running it on the first eligible LLM that is released after the preregistration. The researcher finalizes the procedure on current models, preregisters the analysis plan together with a set of eligible future models, and runs the confirmatory analysis on the first eligible model released afterward. Because this model does not exist at commitment time, it cannot be hacked against; furthermore, configurations that hack one model frequently do not transfer to the next. We evaluate the protocol on two tasks whose true values are known. Across 20 models from four providers and 11 LLM-analysis configurations, the protocol would have blocked successful transfer of the p-hack in 73.9% and 72.7% of cases in the two tasks. Additional analyses reveal that mitigation remains substantial under several stress tests. Finally, putting money where our mouth is, we followed our own protocol and preregistered our experiment. The preregistered experiment confirmed the protocol's effectiveness: out of the 7 configurations that hacked the prior model, the hacking failed to carry over in 6 configurations on the first eligible model released afterward.