Academia.eduAcademia.edu

Verification versus falsification of existing theory

1982, Journal of Chemical Ecology

Abstract

A disturbing feature in science is the frequent emphasis on verification of popular theories rather than on falsification of hypotheses. As Dayton and Oliver (1980) stressed recently "The verification of ideas may be the most treacherous trap in science, as counterexamples are overlooked , alternate hypotheses brushed aside, and existing paradigms manicured. The successful advance of science and the proper use of experimentation depend upon rigorous attempts to falsify hypotheses." While all disciplines of science suffer from this problem, the reliance of behavioral research on observational techniques requires that one exercise extreme caution in data interpretation. To avoid compromising the conclusions of field and laboratory studies, it is necessary to test rigorously alternative hypotheses and to rely on valid statistical techniques. In his recent review of a 1981 paper by Itagaki and Thorp, Rose (1982) concluded that the earlier paper contained "... misconceptions concerning the nature of pheromones and intraspecific communication and misinterpretations of results within the paper." From our perspective the only potentially significant criticism concerned our general approach in evaluating experimental results. The opposite approach advocated at least de facto by Rose is illustrative of the problem mentioned previously. The specific criticisms by Rose and our opposite approaches to data interpretation are discussed below. Although theoretically it takes only one case to reject a "properly framed" hypothesis, one must be sure that the results of a test are real (with regard to type I errors), exclusive of alternative hypotheses, and directly applicable to the overall question. The overall null hypothesis (H0) in our study was that long-distance chemical communication of sexual identity, agonistic state, and stress condition does not occur among adult crayfish. To falsify the overall null hypothesis, it was necessary to show that (1) statistically significant results led to rejection of H0, (2) these results were consistent with other data, and (3) the data were not equally well explained by alternative hypotheses. Two alternative hypotheses were that (1) the number of I073