Interlaboratory Bias and Uncertainty in the Analysis of Metals and Polychlorinated Biphenyls in Used Oil Samples

Abstract

Abstract Blank and used oil samples spiked with PCBs at two different concentrations and heavy metals were sent to various analytical laboratories based upon national reputation, recognition, and applicable accreditation (EPA or NELAC). The study, implemented in a double-blind manner, was designed to simulate a client sending actual used oil samples to a certified laboratory for the quantitative analysis of PCBs and heavy metals for regulatory purposes. The study was also designed for comparing and contrasting the reported results in a statistically meaningful manner. The primary purpose of the study was to assess the accuracy and precision of each of the laboratories. Sending multiple, double-blind samples of the same mixture allowed an estimate of uncertainty to be determined. A large bulk sample (∼10 L) of used crankcase oil was prepared. From this base of used oil, spiked bulk samples were prepared. A bulk sample composed of 2 mg/kg 1260 PCBs and a bulk sample composed of 50 mg/kg 1260 PCBs were prepared. Additionally, a bulk sample spiked with 5 mg/kg arsenic, 2 mg/kg cadmium, 10 mg/kg chromium, and 100 mg/kg lead was prepared. From these unspiked and spiked bulk samples, aliquots were poured into specialty containers certified to be free of trace metals and organic compounds. The aliquoted samples were then sent to ten different laboratories across the United States for the quantitative determination of PCBs and trace toxic heavy metals. For the analysis of blank oil samples, one laboratory reported values an order of magnitude lower relative to the other laboratories for most of the metals. These results were disregarded. Two laboratories reported detecting trace amounts of arsenic in the blank oil samples. The remaining laboratories reported the values below their reporting limits. There was fairly good agreement across the laboratories for the analysis of chromium. With leads, however, the laboratories could be grouped into three statistically different groups. The spiked oil results indicated that laboratories could be grouped into statistically similar groups based upon individual metals. The lack of a grouping pattern across metals indicate the grouping to be metal-specific and hence, the differences in the laboratories may be the result of some unknown random process that cannot be directly measured. Regarding the analysis of PCBs spiked at 2 mg/kg, there was relatively good agreement across laboratories in that most could not quantitate PCBs at this level. At the regulatory level of 50 mg/kg, 33 % of the laboratories could not detect the PCBs. Across the measured values at the regulatory limit, a very significant amount of variation was observed, in agreement with earlier observations. In an attempt to account for these differences, as the factors responsible for the variances are not known, a random effect paradigm is introduced along with a gaming strategy based upon a time-constant model.