Samples from a hazardous waste site contaminated with lead and cadmium were analyzed by four independent laboratories, each using a different technique:  atomic absorption spectroscopy (AAS), X-ray fluorescence (XRF) spectroscopy, inductively coupled plasma−atomic emission spectroscopy (ICP-AES), and potentiometric stripping analysis (PSA). The four data sets were retrospectively analyzed to (1) establish the magnitudes of uncertainty in the measurements, (2) evaluate the comparability of the four instrumental methods, and (3) determine if any significant correlations existed between individual sets of data. In general, the four techniques gave comparable results for the analysis of lead and cadmium, with the best agreement between PSA and AAS. Concentrations determined by PSA were higher than those measured by ICP-AES, AAS, and XRF, while concentrations determined by XRF were lower than or equal to recoveries determined by ICP-AES and AAS. Principal component analysis determined that the two major principal components in the sample space of the data set were analyte concentration and sample preparation. The ICP-AES data were used to look for correlations among other elements in the samples. It was shown that concentrations of four of these elements (aluminum, zinc, iron, and calcium) were significantly higher than 19 other elements determined by ICP-AES. Principal component analysis on those 19 elements showed a first-component variation attributable to an analyte concentration effect and a second-component variation attributable to an analyst-day effect.

No abstract

Chemometrics is defined as the application of mathematical and statistical methods to chemical systems. Systems theory is seen to be useful for organizing and categorizing the inputs to and outputs from chemical systems. Advances in measurement science in the 1950s and 1960s, particularly in analytical chemistry, created a need for a multivariate approach to data analysis. Early chemometrics emphasized the use of structure‐finding methods for existing data sets. In many instances, data sets can be obtained from designed experiments. Such data sets are more likely to contain the desired information and the data can usually be acquired at less cost. Renewed interest in statistical process control will provide many new, more robust data sets in the future.

No abstract

No abstract