Statistical analysis of sediment toxicity by additive monotone regression splines

Modeling nonlinearity and thresholds in dose-effect relations is a major challenge, particularly in noisy data sets. Here we show the utility of nonlinear regression with additive monotone regression splines. These splines lead almost automatically to the estimation of thresholds. We applied this novel method to explore the relation between the toxicity of aquatic sediments, as observed in bioassays with Daphnia magna, Chironomus riparius and Vibrio fischeri, and the degree of contamination of the sediments. Despite the low signal-to-noise ratio in the data, some interesting thresholds and (non)linear effects were found. The method has added value compared to the linear multivariate methods applied earlier to these data. Percentages of explained variance remained low, but could be doubled by diminishing the effect of local variability.