Water quality trend detection in the presence of changes in analytical laboratory protocols

Changes in analytical procedures, and detection limits with sub-detection limit data censoring, can have profound effects on trend analysis and add tedium and uncertainty to the process. Less accurate methods early in the record may produce not only high-biased data but also greater variance. The consequence of this is an artificial induction of a down-trend. In datasets with multiple procedure changes embracing relatively short time periods, there can be considerable difficulty in judging whether the data from adjacent time blocks can be coalesced to increase the length of the time period being analyzed and hence increase trend detection power. In the absence of paired comparisons between 'old' and 'new' procedures, combining adjacent data blocks adds uncertainty to any conclusions drawn. The consequence of changing (usually lowering) detection limits with data censoring at the detection limits may induce artificial (usually down) trends. Data censoring may be of concern when using non-parametric statistics because multiple tied values may result, the consequence of which may be the paradoxical situation of significant trend but a zero slope. For trend detection, it is preferable that there should be no analytical procedure changes without compelling reason, and networks should be designed with this in mind. If a thorough assessment of the consequences of multiple analytical procedure changes and multiple detection limits with sub-detection limit data censoring is not made during data analysis, the consequence may be that the data analyst is merely measuring changes in the monitoringprogram not the environment.