Application of statistical power analysis for scaling and scoring metrics for multimetric rapid bioassessment protocols (RBPs)

B.D. Marshall and B.L. Kerans. Department of Ecology, Montana State University, Bozeman, Montana 59717

For bioassessment, Type-2 error is arguably more important than Type-1 because the costs of failing to detect impairments exceed the costs of “detecting” false impairments. We applied statistical power analysis to two aspects of RBP development: (1) metric selection and (2) metric scaling and scoring. Although both aspects incorporate measures of α and β, this presentation focuses on scaling and scoring techniques. We selected metric batteries and allowed α = β for two scoring criteria, based on the probability of observations deviating from a regional reference population. Thus, we defined a “moderate deviation” threshold at α = β = 0.15 and an “extreme deviation” threshold at α = β = 0.01. A sample with a metric value exceeding the “extreme” threshold is >99%-likely to be significantly different from the reference population and was scored “2”; values between the two thresholds were moderately different from the reference population (0.01<P<0.15) and scored “1.” Values below the moderate threshold were not different from the reference condition and scored “zero.” Summing the scores for all metrics resulted in “condition scores” that increase as the probability of deviation from reference conditions increases. This technique improves multimetric scoring methods because β is incorporated into criteria definition.