A proposed method for improving whole effluent toxicity data interpretation in regulatory compliance

Abstract

We describe three statistical models for assessing whole effluent toxicity data for National Pollutant Discharge Elimination System compliance purposes. The first model, on which the current regulatory approach is based, only considers intratest variation. The second and third models consider both intratest and intertest sources of variability. We perform a variance components analysis with reference toxicant data to show the relative magnitudes of the intratest and intertest variation. The results indicate that the third model, which is a mixed two‐way ANOVA design with test crossed with concentration, is the most appropriate model. With simulations, we show that when using the first statistical model instead of the more suitable third model, the actual Type‐1 error can exceed the nominal Type‐1 error by a large amount. This means that tests with the first model will tend to produce an excessive number of false indications of toxicity by chance alone. We present and demonstrate a statistical test, called the reliable toxicity detection level (RTDL), that is based on the third statistical model. The RTDL is compared with other statistical measures used to determine compliance to toxicity and chemical effluent limits.