Reproductibility, notion very important in science, isn’t always real.
There are obvious errors and data manipulation, but another explanation of this non-reproductibility could be that the value of 0.05 for the famous threshold p-value, validating the significance of a result, might be not so significant after all …
According to Valen E. Johnson, between 17 and 25% of results of scientific articles may be false, which is consistent with his estimate of non-reproducibility.
His solution would be to reduce this threshold (go to 0.005 or even to 0.001), so it would be necessary to increase the sample size, which would be quite expensive for the research (and for data management, and a lot of studies to revalidate/check) …
Real problem ?
For more information, here is the original article.
Source : Passeur de Science