A hallmark of science is replicability. Another team of researchers following the same methods should be able to reproduce the original results. As Jonah Lehrer writes in The New Yorker, there may be a crippling flaw in this principle.
Lehrer gives examples from studies of the benefits of antipsychotics to the powers of ESP where subsequent experiments yield decreasing effects. This could be a case of muting the influence of outliers on data. But it’s so widespread that the decline effect points to something intrinsic to the practice of science.
John Ioannidis has written about the inherent biases in science. One article in PLoS Medicine entitled, Why Most Published Research Findings Are False, points to the way scientists influence data through their expectations. The preference of publishers for significant results also leads to inaccuracies. I would add that the mantle of objectivity also impedes scientists’ ability to accept the more qualitative elements of their craft.