From causal inference to regulatory thresholds

Causal inference is increasingly used to identify stressors and their sources that cause biological impairment in waterbodies.  Large regional biomonitoring databases are important in developing plausible thresholds and associations between biological indicators and stressors from which causality is inferred.  These relationships can illustrate whether current chemical and physical water quality criteria are truly protective of aquatic systems.  We examined the associations between candidate stressors and benthic macroinvertebrates in a large monitoring dataset from West Virginia using several analytical methods, including conditional probability analysis, logistic regression, propensity functions, and multivariate approaches. We examined stressor-response relationships by partitioning out other stressors and compared them with the propensity function approach. Causal inferences were strong for stressors that are highly toxic (e.g., dissolved aluminum) but were weak for some other stressors for which criteria have been recommended (e.g., iron, nutrients).  A third group was stressors that consistently and strongly associated with biological degradation (especially ionic strength, measured as conductivity, TDS, or individual ions), but are only rarely regulated by agencies. These results suggest that using causal inferences based on associations between stressors and biological responses are a necessary step in defining chemical water quality criteria which would be protective of aquatic life uses.