Statistical methods for estimating ecological breakpoints and prediction intervals

Abstract: The relationships among ecological variables are usually obtained by fitting statistical models that go through the conditional means of the dependent variables. For example, the nonparametric loess and the parametric piecewise linear regression models, which pass through the conditional mean of the response variable given the predictor, are used to analyze simple to complex relationships among variables. We used loess and bootstrapped confidence interval to subjectively identify the number and positions of potential ecological breakpoints in a bivariate relationship, and a piecewise linear regression model (PLRM) to quantitatively estimate the location of breakpoints and the associated precision. We also estimated breakpoint location and precision using a piecewise linear quantile regression model (PQRM), which is fitted to the quantiles of the conditional distribution of the response variable given the predictor and provides much richer information in terms of estimating relationships and breakpoints. We compared the precision of breakpoints estimated by PQRM relative to PLRM. We compared the precision of the methods using two examples from the ecological literature suspected to exhibit multiple breakpoints: relating a Fish Index of Biotic Integrity (an index of wetlands' fish community 'health') to the amount of human activity in wetlands' adjacent watersheds; and relating the biomass of cyanobacteria to the total phosphorus concentration in Canadian lakes. Statistically significant breakpoints were detected for both datasets, demarcating the boundaries of three line segments with markedly different slopes. We recommend the piecewise linear quantile regression as an effective means of characterizing bivariate environmental relationships where the scatter of points represents natural environmental variation rather than measurement error.