Reply to Houston, McNamara and Thompson

Abstract

In commenting on the need for sensitivity analyses of dynamic optimization models (Gladstein et al. 1991), we did not argue against the validity of such models, nor did we claim that authors do not perform sensitivity analyses of their models. Indeed, we are certain that they must do so, in order to be sufficiently confident of their work to submit it for publication. Our concern was for the readers of these publications, who have no way of evaluating a model unless sensitivity analyses are provided with it. Such analyses are essential not merely as a form of confirmation of the model's correctness, but because they are really descriptions of the model's properties, as indispensable for the reader's information as are the equations themselves. We attempted to demonstrate the incompleteness of a model presented without sensitivity analyses, in our re-examination of Clark's (1987) model of lion foraging group sizes. As practicing ecologists we wish to know not only that a model can be constructed to calculate an optimal solution for a given biological problem, but whether the answer provided is a plausible explanation for that phenomenon in reality. Consider the inevitably subjective assessment of the requisite sharpness of peaks on an adaptive landscape. Population genetics demonstrates that, all else being equal, even slight selection pressure will alter organisms' behavior toward an optimum. But in the noisy real world, all else is never equal. Each reader must judge the degree of superiority over suboptimal alternatives that he or she expects in order to accept that the modelled explanation is indeed the probable cause of lion group sizes or the dawn chorus. The proponents of competing hypotheses, especially, are entitled to see the sensitivity of the outcome to strategy choice, and will not wish to take the authors' word for it. Analyses published elsewhere subsequently, or remaining unpublished, do not suffice. Houston et al. raise the valid point that journal editors may refuse space for extensive sensitivity analyses. We would be delighted if our paper were cited to persuade them that their readers learn far more from sensitivity analyses than from the model alone. If this attempt fails, authors could prepare a supporting report such as Houston et al. describe, summarizing the report's results briefly in the main paper and noting which author will provide it on request. These reports should also be submitted to journals for distribution to reviewers, along with the manuscripts.