Introduction to Multiverse Meta Analysis

Filippo Gambarota, psychologist, postdoctoral researcher in Psychometrics at the University of Padova, Department of Developmental Psychology and Socialization (DPSS) and member of the Psicostat research group.
In this session we will introduce the concept of multiverse analysis applied to meta-analysis from an exploratory and inferential point of view. Multiverse analysis is a recently developed approach where given a certain research question and a dataset, authors conduct and report all plausible statistical analyses. Usually only one analysis is reported and the impact of other plausible alternatives (researcher’s degrees of freedom) on the final results is often neglected. We will see some exploratory statistics and plots and the implementation in R. The increase in complexity due to reporting multiple analysis results on the same dataset need not only descriptive methods but valid inferential approaches. We will present some methods for statistical inference in a multiverse analysis such as the specification curve, and the PIMA (post-selection inference in multiverse analysis) with the related R code. Knowledge of the R software, multiple testing and p-value adjustment methods is helpful but not required to attend these sessions. Attendees with biostatistics and computational biology background would find this applicable to their work.
Our #DataScience Training Sessions offer hands-on training in #statistics, #computationalbiology and #machinelearning by experts at the Department of Data Science at Dana-Farber Cancer Institute.