- Failure to test and refute prominent hypotheses reduces confidence in the reliability of scientific results and hinders scientific progress.
Across many scientific fields there is a well-documented reproducibility crisis that is damaging trust in the reliability of research data. In a recent article published in eLife, Dr Sarah Rajtmajer and co-authors discuss how failure to falsify (refute) strong hypotheses through direct testing has contributed to the problem.
As a case study, the authors highlight two prominent and seemingly contradictory hypotheses in the field of connectomics:
- Hyperconnectivity hypothesis: brain injury results in an enhanced functional network response.
- Disconnection hypothesis: brain injury results in reduced functional connectivity.
Instead of deliberate attempts to challenge either of these positions, the research area has seen the publication of a large number of small studies examining under-specified hypotheses, which has done little to bring clarity to the existing body of literature. The authors argue that the ‘science-by-volume’ culture, coupled with the overuse of inappropriate statistical tests and lack of falsification attempts, fosters a research environment in which the quantity of scientific findings continues to grow, but the depth of understanding remains stagnant.
The article calls out the big data revolution as a factor adding to these concerns. The ability to analyse large datasets in different ways can produce false or coincidental correlations, particularly if the statistical methodologies used are not robust.
The strongest hypotheses are specific, easily testable, and clearly indicate the evidence needed to disprove their predictions.
According to Rajtmajer et al., the strongest hypotheses are specific, easily testable, and clearly indicate the evidence needed to disprove their predictions. The authors suggest embracing a ‘team science’ approach, where groups of scientists work together to form opposing hypotheses, design experiments to test them, and agree on the outcomes that would support or refute them.
Implementing a falsification approach, whereby every observation confirms or refutes a hypothesis, would be challenging in everyday research practice. However, the authors believe that regular attempts to falsify a hypothesis could guide the direction of scientific research and enhance the reliability of published science, particularly if combined with other processes aimed at improving data transparency.
Regular attempts to falsify a hypothesis could guide the direction of scientific research and enhance the reliability of published science.