Abstract

Twitter data are used as alternative metrics (altmetrics) to measure the impact or attention of research. Tweets are used to communicate about papers. However, Twitter data can only be used for research evaluation purposes, if biases do not influence tweet decisions on papers. The existence of biases can only be reasonably investigated using an experimental design with controlled (marginal) manipulations. In this comment, we propose to undertake an experimental approach to study the decision of scientists to ‘tweet’ on a paper. We describe the design of a study that might allow the experimental investigation of tweet decisions including randomized variations and theoretically derived mechanisms for explaining the empirical results. The described study design should be adaptable to other social media platforms (e.g., Facebook or ResearchGate). This comment is intended to be a plea for using an experimental design to investigate biases in tweet decisions. It is an advantage of tweets –in contrast to citations– that an experimental approach can be applied to investigate the decision of scientists to communicate on papers.