Lisa M. Gill
Statistical Engineering Division, ITL
Analytical Chemistry Division, CSTL
In 1997, NIST released Standard Reference Material 2383, Baby Food Composite, which can be used as a control material when assigning values to in-house control materials and for validation of analytical methods for the measurements of proximates, vitamins, and minerals in baby foods and similar matrices. The ``recipe'' for SRM 2383 was developed at NIST. One thousand pounds of material were prepared by combining ingredients that go into commercially available jars of Gerber baby food; prepared creamed spinach and infant formula were added. The mixture was pumped into jars that held 70 g.
The certification of this SRM, which has 56 analytes, was broken up into three main categories: 1) Fat Soluble Vitamins and Cartenoids, 2) Water Soluble Vitamins, and 3) Proximates and Trace Minerals. The first category had measurements made by NIST and two outside laboratories. The other categories had data collected from a round-robin study and some additional participating laboratories. There was no data collected at NIST for these two areas.
The certification method used for the fat soluable vitamins and cartenoids has been applied to many chemical multimethod reference materials. For all the reporting methods, an equally weighted mean of the measurements is calculated with an expanded uncertainty at the 95% level of confidence with an additional bias allowance added on linearly. This method provides conservative results that have proven to be useful over time.
Due to the large number of participating laboratories for the water soluble vitamins, proximates and minerals, the certified value and associated uncertainty were derived by taking the mean of each of the lab means and then calculating the variance of the lab means. This method works well except when there are too few labs reporting data for a particular analyte; in these cases, the degrees of freedom is small, which in turn results in a large t-multiplier.
Three criteria were used for excluding data from the calculation of assigned values: 1) if for a particular analyte a laboratory's results did not agree with other laboratorys' results and that particular laboratory had poor quality control (QC) data; 2) data was not used if a laboratory did not provide QC data; and 3) outliers were determined based on the QC data and comparison with other laboratories utilizing the studentized residuals.
The uncertainty in the reference values is expressed as an expanded uncertainty according to
the ISO Guide to the Expression of Uncertainty in Measurement.
Figure 6: Comparison of the final assigned value (denoted .Cert) and its associated expanded uncertainty and the mean with expanded uncertainties for individual laboratory data for iron.
Date created: 7/20/2001