STATISTICAL PYROTECHNICS

[摘要] The term medical research dignifies a heterogeneous group of activities. These range from well-designed studies intended to test clearly articulated hypotheses to anecdotal accounts of arbitrarily selected cases. The former speak eloquently for themselves, and of the latter the least said the better. Somewhere between lies a category of studies known disparagingly as "fishing expeditions." These arise when investigators have no clear idea of what is being sought but collect data in the hope of finding some hitherto unsuspected associations between measurable variables . . . the cynic may perhaps be forgiven for objecting that fishing expeditions often beg far more questions than may feasibly be answered . . . .Data generated by fishing expeditions do not have any logical structure imposed upon them by prior hypotheses and therefore they tend to be numerous, messy, and difficult to interpret. Generally help is sought from statistics, in the form of significance tests which are performed on all conceivable associations in the data. Associations found to be statistically significant at, say, the 5% level are reported and all too often these are widely and uncritically accepted as genuine. Regrettably, the dangers of data-dredging by significance tests are insufficiently recognised.

[学科分类] 儿科学