In our 1998 paper (Eberhard et al. 1998), we tested several hypotheses regarding the possible selective factors involved in the evolution of animal genitalia. We compared the slopes of log-log ordinary least squares (OLS) regressions of genitalia on indicators of body size with the slopes of other body parts on the same indicators in 20 species of insects and spiders. Our major conclusions regarding the rejection of male-female conflict, good viability genes, and lock and key arguments (see Abstract) are unaffected by the reanalysis proposed by Green (1999), even if it were more appropriate than ours, which we doubt (see below).In neither our OLS regressions nor Green's reduced major axis (RMA) regressions are the slopes of genitalia greater than those of other body parts, as would be expected if genitalia were used as weapons in forceful intraspecific battles (the male-female conflict hypothesis) or as signals of male size (good viability genes hypothesis). Instead, the slopes show a statistically significant trend to be lower in both analyses. In addition, neither we nor Green found differences on comparing the slopes of the genitalia of the species in which lock and key considerations might be important (species in which male genitalia fit against rigid female genitalic structures) with the slopes in species in which lock and key can be ruled out because of the mechanical mesh of the male's genitalia with those of the female. So, conservatively, we conclude that the major conclusions of our paper are not affected under Green's reanalysis. It is important not to lose sight of the biological questions being tested in debates over statistical methods.There are a number of reasons, however, to doubt several of Green's points. Green's claim that reanalysis is needed, the analysis he performed, and the additional explanations that he proposed in preference to those we mentioned in our original paper all have serious problems. We will discuss first the statistical questions, and then the more directly biological questions.
Is RMA Regression More Appropriate?Many of Green's objections (see the first half of his final, summary paragraph) hinge on the different values he obtained when he used RMA regression analyses. We struggled with the question of whether we should use RMA rather than OLS regressions while we were preparing our original paper, and finally decided against RMA for two reasons. The first is that it is not obvious which of the two types of regression is more appropriate, despite the impression given by Green. For instance, Sokal and Rohlf (1995, p. 544) mention that the RMA regression technique has been the subject of "serious criticisms" (see also reservations expressed by MeArdle 1988; Martin and Barbour 1989). LaBarbera (1989) states that "OLS regression is appropriate only when the goal is to ... allow prediction of expected values given one of the two variables" (which is the "good genes" hypothesis we wanted to test). Sokal and Rohlf also note that if a causality relation exists between variable...