COMMENTS ON THE CAUSES OF NATURAL SELECTION.

Abstract

Wade and Kalisz (1990) discuss approaches to identifying of natural selection. They state that previous statistical analyses (e.g., Mitchell-Olds and Shaw, 1987) give the impression that quality of an analysis is equated with sophistication in multivariate statistics, and have overshadowed fact that analysis of selection is essentially an ecological and not a statistical question. Wade and Kalisz recommend several procedures (experimental manipulation, consideration of environmental effects, and understanding of ecological mechanisms), which are billed as an empirical complement to MitchellOlds and Shaw (1987). have previously discussed interpretation and limitations of selection analyses (Mitchell-Olds and Shaw, 1987; all page references are from this paper). In addition to statistical issues, we advocated experimentation as an essential ingredient for biological interpretation. For example, we and many others have noted that correlation, even in a multivariate regression analysis, is not sufficient to imply causation (Perhaps most desirable to model validation is experimental manipulation to avoid weaknesses inherent in a correlational approach, p. 1159). also suggested explicit consideration of environmental effects, . . . [which] permits detection of environment specific differences in relationship of characters to (p. I 1 5 5). Beyond experimental manipulation, we also have suggested that functional analyses of ecological importance of characters may be used to support inferences from regression approach (p. 1 159). When results from correlational or regression analyses are in accord with known physiological or ecological mechanisms, then biological inferences are greatly strengthened. Biologists have long recognized limitations on biological interpretations based on correlational analyses, and importance of experimentation for resolution of these questions. These points were emphasized in our 1987 paper, and have been discussed by experimentalists for many years (e.g., causes and mechanisms must also be demonstrated with critical experiments, Endler, 1986, p. 201; additional references in Mitchell-Olds and Shaw, 1987). Wade and Kalisz present these ideas as their own original invention, with no reference to previous workers. Furthermore, their discussion mentions our 1987 paper, then appears to contrast their own views: We [Wade and Kalisz] are concemed with research program after this estimate of fitness function is obtained. suggest experimental manipulations.... too are interested in best unbiased estimate .... However, we treat even best estimate of selection gradient through multivariate analysis as an 'hypothesis'.... It can only suggest causation. applaud Wade and Kalisz's suggestions for consideration of environmental effects, experimental manipulation, and understanding of ecological (as well as physiological) mechanisms, having previously advocated this approach. Wade and Kalisz build on previous work of many biologists, and suggest useful extensions to provide improved ecological understanding. Regression analysis is an important method for obtaining partial understanding of possible pattems of natural selection, and for suggesting hypotheses for future research. However, such correlational analyses are never sufficient for inference of causal pattems of natural selection. Attempts to understand causation are reinforced by experimental manipulations or examination of physiological or ecological mechanisms. Our understanding of evolutionary processes will be greatly strengthened by intensive, long-term studies of wild populations employing a variety of techniques (Mitchell-Olds and Shaw, 1987, p. 1160).