Using mean and quantile regression with broken-stick models to develop stressor – response relationships to assess the potential causes of biological impairment in streams

USEPA’s CADDIS approach uses many types of evidence; one type discussed here compares stressors and responses at a site to stressor–response relationships developed from other sites.  We developed models for % sand & fines and biotic metrics using Minnesota monitoring data and broken-stick models in mean and quantile (i.e., 90^th percentile) regressions with prediction limits.  Broken-stick models estimated where the slope changed. If the slope change was not significant, a linear model was tacit.  Then, we used the models to assess 12 impaired sites not included in model development.  For metrics exhibiting various models along the 90^th percentile (e.g., % abundance, intolerant invertebrates; % abundance, clingers), sediment fines were a likely cause at two sites, because the metric was within the quantile prediction limits and % sand & fines was greater than any threshold.  At nine sites, the metric was less than the quantile lower prediction limit; other stressors were also likely causes. At one site, where benthic insectivore richness exhibited a threshold, sediment fines were discounted as a likely cause, because % sand & fines were less than the threshold.  Thus, we illustrate that field data and a clear inferential process can help identify causes of biotic impairments.