Standard Reference Material 2881 - National Institute of Standards ...

The second level of experiment at which we can look at the ANOVA is the individual oligomericO ctylA βirepeatability amongst Octyl A samples run in different Polymer-Pair-runs . Foreach Polymer-Pair-run about 6 to 15 sample preparations are made from the same solution ofeach of the two polymers, silver and matrix. The average of each βifor each oligomer of theOctyl A is obtained. The sum of within variances of all the Polymer-Pair-run can then becompared to the variances between all Polymer-Pair-run used in the average βiused to certifythe sample. Assuming the oligomerβ are independent one would expect 5% of the computediF’s to be above F critical . Again often as many as 50% of the F’s were above F critical suggesting thatthere is a significant variation between each sample preparation.Two other methods were used to check the reproducibility of the data. First, linearity ofconcentration versus signal intensity was found for all data sets. Second, all data sets had a few(at least two) sets of data points in which the concentration ratios of the two polymers wererepeated. In all these the noise within one repeat of five data sets was larger than therepeatability between the solutions made up the same—indicating the repeatability of our makingof the solutions was excellent. This showed that the sample preparation method used within oneset of experiments was good.The data was analyzed in two additional ways to check for self-consistency. In the first, asingle peak was given a fixed value (the average value for that peak of all spectra) and the rest ofthe peaks the spectrum under consideration were scaled accordingly. This did not change theoverall distribution of values of the p-value computed indicating that spectrum-to-spectrumdifferences were a leading cause of type A uncertainty. In the second, four adjacent data pointsin a spectrum were summed to one reducing the number of points per peak from 64 points to 16points. Assuming the oligomer βiobtained with this course graining of the spectrum (orsmoothing of the spectrum) are independent one would expect 5% of the computed F’s to beabove F critical obtained for these data. Again often as many as 50% of the F’s were above F criticalsuggesting that there is a significant variation between each sample preparation.31