SODAS Lecture: Estimating the replicability of research findings in psychology - From manuscript text
Estimating the replicability of research findings in psychology - From manuscript text.
The replication crisis is one of the biggest headaches in many empirical disciplines and particularly in Psychology. Scholars are eager to distinguish robust findings from irreplicable ones. Our research team conducted the first discipline-wide replication census. We trained a machine learning model to estimate a paper’s likelihood of replication from the text in its manuscript. The model was applied to more than 14,000 papers, covering nearly all papers published in six top-tier Psychology journals over the past 20 years. The algorithm-estimated replicability could be used to prioritise expensive manual replications of certain studies, accelerating the goal of (re)building a solid literature base.
Dr Youyou Wu is an Assistant Professor at University College London (UCL), Faculty of Education and Society (IOE), Department of Psychology and Human Development. She has a PhD degree in Psychology from the University of Cambridge. Dr Wu mainly studies personality expression in the digital space, using computational methods like machine learning and natural language processing. Her most prominent work shows that personalities inferred from people’s Facebook profiles are more accurate than judgements made by friends. As a psychologist of this generation, she is also drawn to metascience and methodology, and particularly reproducibility. She is interested in predicting replication outcomes before an actual replication takes place. Her work has appeared in journals like PNAS, Nature Communications, and Psychological Science. Her research has been covered by global outlets like BBC, New York Times, and the Economist.