Testing for trends when there are changes in methods

Data from monitoring projects often include sampling or analytical changes that preclude trend analysis on the entire period of record. A modification of the nonparametric Kendall's test for monotonic trends, which accounts for such changes in the period of record, is described here. This approach blocks the data so that only data collected or analyzed under similar circumstances are compared. Alternatively, when appropriate data exist, data collected using the old method may be calibrated to values expected from the new method. Traditional trend tests may then be applied to resulting data sets. Results from simulations assessing both the power of the blocked test and the standard test performed on calibrated data are presented. The power of the blocked test exceeded the power of the calibration approach only when the calibration error was extremely large. Both the blocking and calibration approaches were applied to and compared for chemical data from Vermont lakes.