Cross-validation: Addressing the replicability crisis
Time & Location
About the Event
Concerns over replicability are, in part, concerns about the generalizability of our (statistical) models to new samples and populations. In this talk, Dr Serena Wee from the University of Western Australia will discuss why model fitting implies model overfitting, when problems of overfitting are most likely to occur, and how cross-validation can help us (a) evaluate model generalizability, and (b) develop models that are more likely to generalise. I will also demonstrate how to conduct cross-validation using the R caret package. This presentation is from the perspective of a non-expert, and based on what I have learned in an attempt to address comments, rightly raised by reviewers, that my work might have capitalised on chance (Wee, Newman, & Joseph, 2014; Song, Wee, & Newman, 2017).