Dioxin Test Method Precision

Sampling, sample recovery, laboratory preparation and analysis and performing calculations all introduce errors into a final test result. Even without any mistakes, random errors will remain. The precision of a method -the size of likely random errors from all sources from sampling tech­ niques through data reduction -is determined by analyzing the differences between simultaneous measurements. For total dioxins, the error increases with concentration and at 26 ng/dsm3, the highest concentration with suitable data, 99 percent of individual measurements are likely to be within ±75 percent of the true value. For ITEQ dioxins using a combination of USEPA Method 23 and European data, the precision of individual measurements also changes with concentration and is ± 110 percent at 0.1 ng/dsm3; when just the Method 23 data are used, precision does not change with concentration and it is ±105 percent at 0.1 ng/dsm3• Diluent correction introduces an additional source of error. The error contribution associated with the oxygen measure­ ment is usually much less than 1 percent. So, it can be safely ignored and diluent corrected measurement imprecision can be determined by mUltiplying the uncorrected method preci­ sion by the same dilution correction factor used to correct the concentration. CONCEPTUAL FRAMEWORK Errors are an inherent part of measurement. We are not talk­ ing about blunders and known mistakes that invalidate the results. Mistaken and biased measurements are not the sub­ ject of precision; precision applies to irreducible random errors. For example, if we are measuring the length of a board and know that the tip of the ruler has been snapped off, this re­ sult is biased by however much of the ruler was lost. Preci­ sion, on the other hand, describes the spread in results we are likely to get measuring that board a number of times. No matter how carefully we align the ruler and read the scale, the answers will be slightly different. We cannot align the ruler exactly the same way each time. The board will not be perfectly square and true. We will interpolate slightly differ­ ently each time we read the scale. If we take many meas­ urements, a distribution will result. 73 The spread of the distribution is affected by both the in­ strument resolution and measurement technique being used. Clearly, interpolation errors alone will be smaller when we are using a scale graduated to 1/32 of an inch instead of one marked every Y4 inch. If we have a short ruler that has to be moved many times to measure the board, then we will have more error as well. The error term associated with air pollution measurements is usually adequately represented by a normal distribution like the one shown in Figure I (Kennedy, et. aI., 1995). Figure 1. Normal distribution curve for individual meas­ urements. 1 � 99% between f I , j J / �68% betwee� L J L I _V 95% between 4 -3 -2 -1 0 -2.576 standard deviations 1\ � � � -.. -