U.S. Geological Survey

Data Series 152

National Water-Quality Assessment Program

Water-Quality, Streamflow, and Ancillary Data for Nutrients in Streams and Rivers Across the Nation, 1992–2001

Modification to Data Censoring

In order to avoid false-positive quantification of a constituent, very low concentrations are censored, reported as a “less than” value by the laboratory. Censoring levels, generally referred to as “reporting limits,” are specific to analytic methods for individual constituents and can change over time as methods change. The U.S. Geological Survey National Water-Quality Laboratory (NWQL) uses several types of reporting limits. The most basic is the method detection limit (MDL), defined as the minimum concentration that can be measured, with 99-percent confidence, to be significantly greater than zero. At the MDL concentration, the risk of a false positive is no more than 1 percent.

For ammonia, the MDL was determined to be 0.01 mg/L from 1992 to 1994, then increased to 0.015 from 1995 to 1997, and 0.02 from 1996 to 1998 (depending on analytical method). In 1998, the NWQL reevaluated the historical data and recommended that all reported ammonia concentrations less than 0.02 prior to September 30, 1997, be recensored to <0.02. This change was made for almost 3,400 samples in the environmental data file.

Beginning in 1996, the NWQL began to change from censoring at the MDL to censoring at a laboratory reporting level (LRL), as described by Childress and others (1999). The LRL for an analyte generally is equal to twice the MDL and is intended to protect against false negatives (reporting a nondetection when the analyte is actually present) over a small range of low concentrations. Values measured less than the MDL are reported as less than the LRL. Values measured between the MDL and LRL are reported but are given an “E” remark code to indicate they are semiquantitative. This use of LRL and “E” values is an example of “informative censoring,” which can bias results if used with standard statistical techniques for censored data (Helsel, 2005). The possibility of a few false negatives is less important than the problems caused by such a bias. To avoid these problems, data can be recensored at the MDL; therefore, in the environmental data file, all values reported as less than the LRL were changed to less than the MDL. Values with an “E” remark code were left unchanged. All revisions to censoring levels are recorded in the data file in the column labeled “Nutrient National Synthesis Team Comments.”