Sources of variability in microcontaminant data for Lake Michigan salmonids: statistical models and implications for trend detection

Abstract

Variability in Lake Michigan salmonid microcontaminant data may arise from differences in contaminant concentrations in the water column, fat content, diet, age, location, and species-specific factors. Polychlorinated biphenyl (PCB) concentrations for salmonids in Lake Michigan have been determined for fish collected from a wide variety of locations and sizes for each of five species. Lake managers and policy makers wishing to evaluate trends in these concentration data must either remove this variability before trend assessment or leave it and settle for less statistical power for identifying the underlying time trend. Classification and regression trees (CART) have proven to be a useful tool for identifying nonlinear patterns of variability in these data and portraying them graphically. We used CART to study patterns of variability in PCB concentrations in five species of Lake Michigan salmonids, using data collected from 1972 to 1994 by both the Michigan and Wisconsin Departments of Natural Resources. For most of the species studied here, length was the major source of variability, followed by location.