A value further info of k > 1 indicates a laboratory whose replicates vary more than the average. A value of k < 1 indicates that the laboratory has lower-than-average interreplicate variability. These values are plotted in a similar way to the h values. Critical values can be obtained to quantify significant departures from average behavior at a chosen level of significance. Because this was designed as a descriptive study with a small number of laboratories per method, we did not stratify by different laboratory methods in the analysis. We also performed Cochran��s test for variance heterogeneity, but we will not report these statistics because most of the variance can be attributed to different methods rather than to experimental variability.

Also, because of the nature of the experimental design, no outlier investigation was conducted and all observations were kept in the final analysis. Therefore, we have not reported any outcomes for Grubbs�� test, which looks for outlier laboratory averages. Outlier exclusions for any reason are not generally recommended (Mandel, 1998). Results All measured values (blank adjusted as necessary) reported by the laboratories participating in this study are given in Table 3. The values observed for each of the pools were in good agreement among laboratories, and, as expected, according to currently accepted cutoff values, there would be no difficulty in separating smoker from nonsmoker samples in the observed results. All seven laboratories were capable of measuring the smoker-level samples in Pools C, D, and H, and five laboratories also reported results for the two pools with relatively high cotinine concentrations in the passive exposure range (pools B and G).

Four of seven laboratories reported results for the moderate exposure Pool A, with a concentration of approximately 0.5 ng/ml, and three laboratories were able to measure all eight pools down to the lowest concentration level, which was approximately 0.05 ng/ml. Table 3. All serum cotinine data (ng/ml) The mean, standard deviation, and range of values for each pool as measured by the participating laboratories are provided in Table 4. In general, there was excellent agreement among the laboratories for the pools that each laboratory could analyze. In Table 5, the final 95% CIs for all pools with known spiked target values (A, B, C, and D) indicated that there was no significant bias at the 5% significance level among the laboratories reporting results for a pool.

As recommended by ISO 5725-2, we calculated Mandel��s h and k statistics by laboratory and pool and these results Batimastat are given in Figure 1 with the 5% significance levels indicated on the charts with dotted lines. Only three pools exceeded the 5% limits in the h statistic plot, including one pool in Laboratory 1-B and two pools in Laboratory 3.