Excellence in science depends on getting the right balance between curious flexibility (e.g. we must follow the intriguing hints in our data) and rigorous hypothesis-testing (e.g. we must ensure our claims are replicable). Many scientific projects have phases of open ended exploration, and phases of careful confirmatory testing. However, in the past decade it has become clear that, without heightened vigilance, these two phases often get mixed up. Exploratory analyses, fishing through protocols and datasets until an interesting result appears, are often reported as confirmatory analyses; this process is analogous to training and testing our ideas on the same dataset, and leads to well-understood inflation and biases in reported effect sizes.

A tool that can help to reduce these biases, and thus increase the credibility, replicability, and impact of scientific claims, is pre-registration. Pre-registration is a formal, time-stamped, externally-verifiable record of planned confirmatory analyses. Pre-registrations can be registered privately, and then made public when a manuscript is later submitted. Alternatively, pre-registrations can be submitted for external review, and given in-principle acceptance for publication as a journal article before the results are known, a format called “Registered Reports”, now accepted at over 300 journals.

In recent studies, neuroscientists have expressed three concerns about pre-registration: (1) Will it limit the flexibility of my research in the exploratory phase? (2) Is it too time-consuming, or even impossible, given the complexity of my analyses? and (3) it sounds like a good idea, but I don’t know how to get started. In this presentation, we will discuss and address these concerns, based on our personal experiences. This presentation will be led jointly by Rebecca Saxe and a team of instructors and students from this Fall's graduate seminar on Tools for Robust Science, in which we discussed pre-registration and registered reports, and their application to the disciplines represented in BCS. 

Zoom link: mit.zoom.us/s/94578220794.

MIT login is required to attend this event.