MEASURING TOLERANCE TO HERBIVORY WITH NATURAL OR IMPOSED DAMAGE: A REPLY TO LEHTILÅ

The ability of plants to tolerate herbivory without a decrease in plant fitness has important implications for the ecology and evolution of plant herbivore interactions (Rosenthal and Kotanen 1994; Strauss and Agrawal 1999; Stowe et al. 2000; Juenger and Lennartsson 2000). Because tolerance is a measure of the effect of herbivore damage on plant fitness, it cannot be measured on a single plant. Rather, tolerance is usually measured as the slope of a linear regression of fitness versus herbivory for a related group of individuals (i.e., families , clones, or inbred lines). Researchers have used both naturally occurring herbivory and imposed herbivory to evaluate the effects of different levels of damage on plant fitness. In our previous paper, we drew attention to some of the biological, logistical, and statistical advantages and disadvantages of using natural or imposed damage for estimating tolerance (Tiffin and Inouye 2000). We also pointed out that covariances between unmeasured environmental variables, fitness, and herbivory can affect estimates of tolerance. Specifically , because experimentally imposed damage removes the covariance between herbivore damage and unmeasured environmental variables, randomly assigned imposed damage can provide more accurate (unbiased) estimates of tolerance. However, under some conditions, this greater accuracy comes at the expense of a less precise estimate. Lehtilä (2003) pointed out that Tiffin and Inouye (2000) assumed the distribution of levels of herbivory (H) is the same for naturally occurring and imposed herbivory. Relaxing this assumption questions the conclusion that estimates of tolerance based on naturally occurring damage will often be more precise than estimates based on experimentally imposed damage. We agree with Lehtilä that the statistical power of experiments can often be improved through the thoughtful allocation of experimental effort to different treatments (Inouye 2001). An experimenter who uses imposed herbivory can choose an efficient allocation of levels of herbivory, whereas an experimenter who uses natural levels of herbivory must accept whatever distribution the herbivores provide. Thus, experiments that use imposed herbivory can potentially gain enough statistical power to provide more precise and more accurate estimates of tolerance. Lehtilä (2003) makes a valuable contribution to the discussion of methods for estimating tolerance. We think, however , that the importance of relaxing the assumption about the distribution of levels of herbivory was overstated in concluding ''the precision of experiments with natural damage are usually very low compared to experiments with imposed damage'' (without citations). In cases in which natural levels …