experimental economist | associate professor at
MUNI | Management Science Data Associate Editor | coffee lover
Miloš Fišar
Loading...
Link to the paper 👉 www.nature.com/articles/s41...
One striking result: only 34% of reanalyses landed within a narrow tolerance of the original effect size (±0.05 Cohen’s d). With a wider tolerance (±0.20), that rose to 57%. Reanalysis effect sizes also tended to be smaller than the originals.
504 reanalyses by 457 independent reanalysts, with at least 5 reanalyses per claim. The core question was simple: how much do our findings depend on analysts’ choices rather than the data alone?
