Transparency, bias, and reproducibility across science: a meta-research view
Transparency, bias, and reproducibility across science: a meta-research view John P.A. Ioannidis
It is a great honor to deliver the AAP Presidential Address. Let me start with disclosures. My main conflict of interest is that I try to be a scientist. This means I am probably biased and often wrong, but hopefully not totally resistant to the possibility of getting corrected. Let me also make some preemptive comments. First and foremost, science is the best thing that can happen to humans, and research should be supported with heightened commitments. You have probably heard this too many times, but it is worth repeating. However, most research done to date has used nonreproducible, nontransparent, and suboptimal research practices. Science is becoming more massive and more complex. Scientific publications (about 200 million already, with 7 million more added each year) are mostly advertisements (“trust me, this research was done”); raw data and experimental materials and algorithms are not usually shared. Moreover, our reward systems in academia and science are aligned with nonreproducible, nontransparent, and suboptimal research practices. Can we do better? Even though we all use the scientific method, maps of science may visualize many thousands of clusters representing different scientific disciplines (1). The research practices in these many disciplines vary substantially in both expectations and implementation. However, some features are all too common. Notably, the quest for significance is almost ubiquitous. Significance takes many […] It is accessible in this page.