Water quality data are often collected in monitoring programs to serve as a basis for the estimation of trends. A problem arises in trend estimation when data series contain observations reported as below detection limit (BDL). The several published methods that deal with BDL observations are generally oriented towards obtaining the 'best value' to substitute for the censored values. When the definition of a detection limit associated with a datum is unknown, a conservative lower bound (with a theoretical justification) for the precision of the observation is given. The results show that weighted regression estimates of linear trends are much less sensitive to the method of substitution for BDL values than unweighted regression trend estimates. The results also indicate that use of weighted regression in multiply censored data series eliminates the need to apply the highest detection limit to all data in the series (when data below this highest limit exist in the data series) in order to avoid trends due to changing detection limits.

"June 1993." Includes bibliographical references (pages 20-21). U.S. EPA. "CBP/TRS 76/93"--Cover.