Application of robust regression to the analysis of BMP effects in paired watersheds

A routine is developed in which known information from the pre-BMP phase of a paired watershed monitoring project is used to make inferences about the minimum level of change that would be significant after a specified duration of post-BMP monitoring, or to determine the minimum duration of post-BMP monitoring to be able to detect a specified level of change. The inferences are based on the ability to establish a linear relationship between water quality variables from the paired watersheds. In addition to the usual least squares regression procedure, the routine also estimates the linear relationship using the reweighted least squares procedure, a robust (nonparametric) procedure that reduces the effects of outliers. The routine was applied to three pairs of subwatersheds. In some instances, the reweighted least squares procedure results in an increased sensitivity of the statistical procedures for detecting change.