Statistical Analyses and Plotting of Control Sample Data

Once the preliminary error checking of the raw data has been done, the control samples should be separated from the normal samples for more detailed examination. This process of separation is greatly aided by the inclusion of the STD SAMP field in the field database (see earlier) and a comprehensive sample list that identifies control samples and their relationships (Fig. 5.3). Control sample results can then be subjected to a number of statistical and plotting procedures that determine the accuracy and precision of results. These processes give an indication of the levels of uncertainty that are associated with the results, information that is essential to interpret the data and present it in a meaningful manner. 

In instances where only a few duplicate—replicate pairs are available, individual results can still be plotted to see where they fall relative to the line of gradient 1. For larger surveys, collecting many duplicate—replicate pairs, a good spread of values can be produced, giving plots such as those shown in Fig. 5.8. 

Thompson (1983) and Thompson and Howarth (1978) describe a method of estimating analytical precision using duplicate pairs. This is not a procedure routinely used by the G-BASE project but is a particularly useful way of estimating the analytical precision when no truly representative reference materials are available. 

In the previous section, the duplicate-replicate control data set was used to give graphical representation of sampling and analytical variability. A statistical procedure referred to as analysis of variance (ANOVA) analysis can be done on the same data set to give a more quantitative statement on variability. Sinclair (1983) describes this method that compares variations that arise from different identifiable sources, 

The G-BASE project has used several statistical packages to perform this nested ANOVA analysis (e.g., Minitab and SAS). It currently uses an MS Excel procedure with a macro based on the equations described by Sinclair (1983) in which the ANOVA is performed on results converted to logio (Johnson, 2002). Ramsey et al. (1992) suggest that the combined analytical and sampling variance should not exceed 20% of the total variance with the analytical variance ideally being 4%. 

---

Statistical Analyses and Plotting of Control Sample Data 104... 

---