The Production of "Truth"

Here's an article by Marcus Munafò and Jonathan Flint on how scientists end up publishing papers that does not represent truth, pressed by financial interests and the pressure to publish in prestigious journals.

An excerpt:

"Outright scientific fraud is rare, but less deviant behavior may be much more common. For example, researchers may run multiple statistical tests on their data: they keep analysing the results in slightly different ways (known as "data mining") until they get a P-value less than 0.05. This is tempting because it is much easier to get one's research published if the findings are "statistically significant" (i.e. the P-value is less than 0.05) – a phenomenon known as "publication bias".

With enough data, and by running enough statistical tests, it is easy enough to find a significant effect, given the probabilistic nature of the statistical methods used. And with enough people trying, this effect might even be found more than once, giving the appearance of replication. The problem is that the results almost certainly won't be true."