While most scientists work ethically, science is dealing with serious issues related to experimental replication, data quality, and literature completeness, which are often addressed under the heading of a reproducibility crisis. For example, as much as 90% of all biomedical research may be irreproducible and up to 30% of it is fraudulent. In surface analysis, 40% of the x-ray photoelectron spectroscopy data analysis (peak fitting) in the scientific literature may be incorrect and another 40% of it is suspect. We present and discuss five aspects of the reproducibility crisis: how our biases and preconceived notions affect what we publish, cherry-picking of data, the value of collaboration, linking related articles including with commentary by artificial intelligence (AI), and data availability and analysis transparency. Perspectives in this article come from surface science, biological sciences, and educational publishing. What and how we select materials for publishing has a significant impact on reproducibility challenges, particularly with the expanding uses of AI.
Pinder et al. (Mon,) studied this question.