Poldrack, Russell A

Machine learning (ML) methods are proliferating in scientific research. However, the adoption of these methods has been accompanied by failures of validity, reproducibility, and generalizability. These failures can hinder scientific progress, lead to false consensus around invalid claims, and undermine the credibility of ML-based science. ML methods are often applied and fail in similar ways across disciplines. Motivated by this observation, our goal is to provide clear recommendations for conducting and reporting ML-based science. Drawing from an extensive review of past literature, we present the REFORMS checklist (recommendations for machine-learning-based science). It consists of 32 questions and a paired set of guidelines. REFORMS was developed on the basis of a consensus of 19 researchers across computer science, data science, mathematics, social sciences, and biomedical sciences. REFORMS can serve as a resource for researchers when designing and implementing a study, for referees when reviewing papers, and for journals when enforcing standards for transparency and reproducibility.

The ability to regulate behavior in service of long-term goals is a widely studied psychological construct known as self-regulation. This wide interest is in part due to the putative relations between self-regulation and a range of real-world behaviors. Self-regulation is generally viewed as a trait, and individual differences are quantified using a diverse set of measures, including self-report surveys and behavioral tasks. Accurate characterization of individual differences requires measurement reliability, a property frequently characterized in self-report surveys, but rarely assessed in behavioral tasks. We remedy this gap by ( i ) providing a comprehensive literature review on an extensive set of self-regulation measures and ( ii ) empirically evaluating test–retest reliability of this battery in a new sample. We find that dependent variables (DVs) from self-report surveys of self-regulation have high test–retest reliability, while DVs derived from behavioral tasks do not. This holds both in the literature and in our sample, although the test–retest reliability estimates in the literature are highly variable. We confirm that this is due to differences in between-subject variability. We also compare different types of task DVs (e.g., model parameters vs. raw response times) in their suitability as individual difference DVs, finding that certain model parameters are as stable as raw DVs. Our results provide greater psychometric footing for the study of self-regulation and provide guidance for future studies of individual differences in this domain.