Instructor: Shravan Vasishth
A recent analysis of publicly released data accompanying published papers in Cognition showed that not all published numbers could be reproduced, even though the data and code were available (https://royalsocietypublishing.org/doi/full/10.1098/rsos.180448). The authors state that: "...suboptimal data curation, unclear analysis specification and reporting errors can impede analytic reproducibility, undermining the utility of data sharing and the credibility of scientific findings."
In this workshop, I will suggest one way to minimize the chances of producing irreproducible results, focusing on repeated measures 2x2 factorial designs as a case study.
The steps I will discuss are:
- Experiment design, and planning sample size using simulated data
- Defining the analysis plan using simulated data
- Checking that your experiment software actually collects the data you need
- Once data are collected, visualizing and summarizing the data
- Creating an R package to document and release your data and analyses
- Code refactoring
- Integrating the data analysis into the manuscript
- Releasing data and code: a suggested checklist
Materials for the workshop are available here: https://vasishth.github.io/MPILeipzig2019/