Comment:

The range is the distribution of metric values across all of the available data. The goal is to eliminate metrics that have very small ranges (e.g., richness metrics based on only a few taxa) or that have similar values at most sites (e.g., most sites have values ¼ 0). A small range could indicate that a metric might not vary sufficiently across sites to discriminate among sites in different conditions.

Comment:

S/N= Signal to Noise The use of metrics that have relatively stable values at individual sites helps ensure that between-site differences in individual samples are caused by differences in stream condition rather than by sampling variation within a site. Sampling variation is estimated from repeat visits to individual sites. Available measures of sampling variation reflect several sources of variability (i.e., short-term indexperiod temporal variability, spatial variability within the reach, and laboratory variability). Low sampling variation is necessary if a metric is to have a highprobability of discriminating between sites in good and poor condition, and sampling variation should be small relative to the size of the among-site differences to be discriminated. We quantify metric reproducibility with a variant of the signal:noise ratio (S/N). S/N is the ratio of the variance among all sites (signal) to the variance of repeated visits to the same site (noise) (Kaufmann et al. 1999). Metrics with high S/N values are more likely to show consistent responses to stressors than are metrics with low S/N values. A metric that is perfectly correlated with a hypothetical stressor and that has no sampling variability will have an R2 ¼ 1.0 for that stressor. As S/N decreases, the maximum possible R2 value of the regression decreases because the sampling variability of the metric increases. When S/N ¼ 1, a perfect correlation between the metric and the stressor would produce R2 ¼ 0.5 (Fig. 2). We have no fixed threshold below which we eliminate metrics based on S/N. However, S/N values 1 indicate that visiting a single site twice yields as much metric variability as visiting 2 different sites.

Comment:

The ultimate test of the effectiveness of a metric is its ability to distinguish degraded from relatively undisturbed streams. A more general approach is to base the evaluation of responsiveness on the ability of a metric to distinguish least-disturbed (reference) from most-disturbed sites.