Published: August 26, 2016
https://doi.org/10.1371/journal.pbio.1002547
"The other pillar of scientific truth, reproducibility, means that another scientist can repeat an experiment and arrive at the same results or, conversely, show that the results are not reproducible. Just as for peer review, multiple problems increasingly imperil reproducibility. For example, it’s not uncommon for an initial high-profile study to report amazing results with a stunning conclusion. Then, when the experiments are repeated, only trends toward the same conclusion are observed with increasingly smaller effect sizes. This outcome neither contradicts nor confirms the original study but is a dead end, and the original paper is slowly forgotten. As discussed above, the problem is not that the initial paper is fraudulent, but that the results were “tweaked” or selected, or represented a statistical outlier, leading to a misleading conclusion.
A second emerging reproducibility problem is that many experiments are by design impossible to repeat. As formalized by Karl Popper [Keuth Herbert (Ed.): Popper Karl. Logik der Forschung. Akademie-Verlag, Berlin 2004, ISBN 3-05-004085-8.], scientific truth requires interpersonal reproducibility. Based on this postulate, any conclusion that cannot be falsified because the underlying experiment cannot be repeated in exactly the same way is not a scientific conclusion. Many current experiments are so complex that differences in outcome can always be attributed to differences in experimental conditions ... If an experiment depends on multiple variables that cannot be reliably held constant, the scientific community should not accept the conclusions from such an experiment as true or false. Such conclusions are simply non-scientific, even if based on an experiment."
Let us remember that the "falsifiability", according to the philosopher Karl Popper, defines the inherent testability of any scientific hypothesis.
The Philosophy of Karl Popper, by Herbert Keuth
Cambridge University Press, 2005
