Hypotheses, errors, and statistical assumptions

摘要: SEAMAN AND JAEGER (1990) contend that presumptuous use of parametric statistical methods to test hypotheses can lead ecologists astray; I wholeheartedly concur. As they point out, the usual argument against nonparametric alternatives is lack power-there a high probability failing reject false null hypothesis (type II error). To increase power particular (lower type error), however, one must generally error, rejecting true (Conover, 1980). Seaman and Jaeger suggest view tests are less powerful than often untrue rests in many instances on assumptions about shape an underlying distribution either incorrect or untestable. In addition, even situations which somewhat ones, there may still be good reason for favor former-relative costs two kinds errors (Connor Simberloff, 1986; Toft Shea, 1983). It important distinguish between scientific (Boen, 1989; Connor 1986). Scientific phenomena nature. Statistical properties populations based samples. Thus, quantified application specific set data. Rejection more would constitute piece evidence weighed deciding whether hypothesis. A related distinction global local (Dolby, 1982). applies all nature, while systems. unless have been sampled. Error classification was developed specifically testing hypotheses, but terms used metaphorically, appropriately, global, hypotheses. case, if some process phenomenon has no effect, then error consists concluding does effect when, fact, it not. same case when actually does. Finally, made classical hypothesis-testing decision theory (Kyburg, 1974). Testing aid inferring false, as noted above. not explicitly take account errors. By contrast, settings tested over again, results acted each time. For example, samples from batches computer chips, save discard entire results. This problem motivated development theory, acting rationally (with respect expected losses gains) face uncertainty, rather inference If statistics this purpose, attending various assessed. Below argue also considered part Perusing medical literature nowadays, easily conclude greater scourge entails larger (e.g., Freiman et al., 1978; Marks 1988). Bourne (1987) went so far title paper, "No Statistically Significant Difference?: So What?" His hypothetical example treatment disease efficacious, he betrayed his reasoning at outset (p. 40):