Prediction and theory evaluation: the case of light bending

Is a theory that makes successful predictions of new facts better than one that does not? Does a fact provide better evidence for a theory if it was not known before being deduced from the theory? These questions can be answered by analyzing historical cases. Einstein's successful prediction of gravitational light bending from his general theory of relativity has been presented as an important example of how "real" science works (in contrast to alleged pseudosciences like psychoanalysis). But, while this success gained favorable publicity for the theory, most scientists did not give it any more weight than the deduction of the advance of Mercury's perihelion (a phenomenon known for several decades). The fact that scientists often use the word "prediction" to describe the deduction of such previously known facts suggests that novelty may be of little importance in evaluating theories. It may even detract from the evidential value of a fact, until it is clear that competing theories cannot account for the new fact.