Detection of Environmental Impacts: Natural Variability, Effect Size, and Power Analysis

Abstract

The power of any test of an environmental impact is simultaneously constrained by (1) the variability of the data, (2) the magnitude of the putative impact, and (3) the number of independent sampling events. In the context of the Before–After—Control–Impact design with Paired sampling (BACIP), the variability of interest is the temporal variation in the estimated differences in a parameter (e.g., population density) between two unperturbed sites. The challenges in designing a BACIP study are to choose appropriate parameters to measure and to determine the adequate number and timing of sampling events. Two types of studies that are commonly conducted can provide useful information in designing a BACIP study. These are (1) long—term studies that provide estimates of the natural temporal and spatial variability of environmental parameters and (2) spatial surveys around already—perturbed areas ("After—only" studies) that can suggest the magnitude of impacts. Here we use data from a long—term study and an After—only study to illustrate their potential contributions to the design of BACIP studies. The long—term study of parameters sampled at two undisturbed sites yielded estimates of natural temporal variability. Between site differences in chemical—physical parameters (e.g., elemental concentration) and in individual—based biological parameters (e.g., body size) were quite consistent through time, while differences in population—based parameters (e.g., density) were more variable. Serial correlation in the time series of differences was relatively small and did not appear to vary among the parameter groups. The After—only study yielded estimates of the magnitude of impacts through comparison of sites near and distant from a point—source discharge. The estimated magnitude of effects was greatest for population—based parameters and least for chemical—physical parameters, which tended to balance the statistical power associated with these two parameter groups. Individual—based parameters were intermediate in estimates of effect size. Thus, the ration of effect size to variability was greatest for individual—based parameters and least for population and chemical—physical parameters. The results suggest that relatively few of the population and chemical—physical parameters could provide adequate power given the time constraints of most studies. This indicates that greater emphasis on individual—based parameters is needed in field assessments of environmental impacts. It will be critical to develop and test predictive models that link these impacts with effects on populations.