Experimental science demands repeatability of results, but many experiments are not repeatable due to fraud or error.
[1][2] The list of papers whose results were later retracted or discredited, thus leading to invalid science, is growing.
[3] Some errors are introduced when the experimenter's desire for a certain result unconsciously influences selection of data (a problem which is possible to avoid in some cases with double-blind protocols).
[4] There have also been cases of deliberate scientific misconduct.
[5]