Alternative approach for establishing acceptable thresholds on macroinvertebrate community metrics

We demonstrated the use of statistical tolerance intervals as a method for deriving acceptable thresholds for benthic macroinvertebrate community metrics. Tolerance intervals are simply confidence intervals based on percentiles, and they allow selection of acceptable limits (referred to as tolerance limits) and a desired level of statistical confidence for a metric distribution (e.g., of a reference population). We used benthic macroinvertebrate community data from several long-term monitoring projects for streams on the US Department of Energy's Oak Ridge Reservation in eastern Tennessee, USA, for the demonstration. We focused on 3 benthic macroinvertebrate community metrics: density, total taxonomic richness, and taxonomic richness of Ephemeroptera, Plecoptera, and Trichoptera (EPT) taxa. Tolerance intervals yielded less restrictive thresholds than those produced by simple percentiles because the former approach includes variation of the reference data, whereas percentiles are distribution-free. The less restrictive thresholds produced by tolerance intervals will decrease the frequency with which metric values from test sites will be classified as unacceptable because data variation is included, but thresholds calculated using the tolerance interval approach may be better suited for studies that require a greater level of statistical rigor than routine monitoring or general screening surveys (e.g., biocriteria, environmental impact assessments). Conversely, approaches that use simple percentiles may be more appropriate for screening studies. Greater accuracy of tolerance limits can be achieved by increasing sample size, reducing variation (e.g., removing outliers, data transformation), and including data that incorporate both spatial and temporal variation. However, alternative approaches should be used if the data are not normally distributed. Tolerance limits can be adjusted easily to achieve the level of environmental protection desired or required, while providing a level of statistical confidence in derived thresholds. For this reason, a tolerance interval approach should be considered seriously, particularly if the study objectives require a known level of statistical certainty.