ABSTRACT

Null-hypothesis significance testing is surely the most bone-headedly misguided procedure ever institutionalized in the rote training of science students. But you don’t need me to tell you that. Ever since its inception, thoughtful statisticians and data analysts have decried this doctrine’s more unsightly flaws (see Cohen, 1994, for a recent review); and although it is a sociology-of-science wonderment that this statistical practice has remained so unresponsive to criticism, recognition that confidence intervals, when feasible, are a much superior way to exploit our standard models of sampling noise finally appears to be growing with sufficient vigor that with luck, research reports in our professional journals will soon no longer be required to feign preoccupation with H

However, significance testing is just the statistical excrescence of a deeper malaise that has pervaded our discipline’s education in research design and data analysis, namely, the hypothetico-deductive (HD) model of scientific method. When this modern orthodoxy, that natural science is best advanced by empirical tests of

theoretical speculations, is construed loosely as just advice to continue improving our provisional generalizations and explanations of observed phenomena in light of performance evaluations, it is a tolerably benign way to apprise outsiders that the epistemic management of statements we honor as “scientific” differs considerably from that of religious scriptures. But it is worse than useless as an operational guide for the conduct of research. The HD program’s failure to advise on which hypotheses are worth testing becomes an evasion of educational responsibilities when science students are admonished that design of their research projects must begin with choice of a hypothesis to test. And far more damaging than this mild sin of omission are the HD outlook’s commissional sins in misdirecting the interpretation of experimental1 results.