Optimum Sample Size to Detect Perturbation Effects: The ...

6 Ortiz
Fig. 2. Power curves for ANOVA design for rocky shores south habitat with a. 10 % (n = 112) and b. 25 %
(n = 18) of variability (dispersion coefficient). Dotted line represents the minimum acceptable power of 0.80.
Based on these considerations, it is demonstrated that the procedure for estimating
the sample size is more complex than that proposed by Mouillot et al. (1999). Definitively,
estimating the magnitude of perturbation, i. e. effect sizes, becomes essential not
only to increase the robustness of the statistical test, but also to optimise the costs of the
sampling program. For instance, in a situation with large magnitudes of effect size (e.g.,
0.40) it would only be necessary to have a maximum variability of 25 %. This would
significantly reduce the costs of sampling and study time. Finally, once effect size rates
are roughly estimated, based on previous sampling or a detailed review of the scientific
literature (for comparable situations), a priori power analysis can be implemented to
estimate the optimum sample size (Cohen, 1988; Peterman, 1990a; Underwood, 1997).