Detection Limits in Chemical Analysis
Tuesday 3 June, 10:20--12:10

Organizer: Ha NGUYEN
Merck Research Laboratories
Moderator: Ha NGUYEN
Merck Research Laboratories

Nonparametric Maximum Likelihood Estimation From Samples with Unverified Measurements, With an Application to Love Canal Data

Diane LAMBERT
Bell Laboratories, Lucent Technologies, Murray Hill, NJ, USA
Luke TIERNEY
University of Minnesota, Minneapolis, MN, USA
The first step in many environmental measurement processes is to separate the pollutant of interest from every other compound in the sample. This step is not error-free, however. Thus, in gas chromatograph/mass spectroscopy, for example, the isolated compound is tested to verify that it is the pollutant of interest. If the isolated compound passes the tests, then its measurement is reported numerically. Otherwise, the measurement is reported as ``nondetect'' and its numerical value may (or may not) be suppressed. In at least some applications, the verification tests are designed so that there are no false positives, but there may be false negatives. Usually, the probability of verifying a measurement depends on its size. The question is how can the mean concentration of the pollutant be estimated when some measurements are known to be from the pollutant, others may or may not be from the pollutant, the probability of verifying a measurement on the pollutant depends on its size, and neither the shape of the distribution of the pollutant nor the shape of the probability of verification function is known? This talk proposes new nonparametric answers to this question and applies them to data from a Love Canal study.

[Diane Lambert, Bell Laboratories, Lucent Technologies; Room 2C-256, 600 Mountain Ave., Murray Hill, NJ USA 07974, dl@bell-labs.com]


Reinventing the Measurement Process; the Synergism between Statistics and Chemistry

L. A. CURRIE
National Institute of Standards and Technology, Gaithersburg, MD, USA
When one wishes to perform "Measurements below the detectable limit," two broad issues must be addressed. First, what is the "detectable limit;" second, how can you proceed if the measurement requirements exceed the detection (or quantification) capabilities? The first question has been the subject of debate for decades, both in terms of nomenclature and in terms of evaluation. Happily, the resolution to this matter appears close at hand, as a result of deliberations by the International Union of Pure and Applied Chemistry (IUPAC) and the International Organization for Standardization (ISO). IUPAC and ISO addressed this issue in a joint meeting in 1993, and since that time IUPAC has published recommendations (Pure and Applied Chemistry v. 67, pp 1699-1723, 1995), and ISO has prepared a draft document on the topic. The talk will begin with a brief review of these international deliberations and recommendations.

The second question is, in many respects, the more challenging and interesting; it will represent the focus of the talk. Approaching measurements beyond the state of the current art offers special opportunities for synergistic collaboration between scientists and statisticians, through careful analysis of the structure of the measurement process, to designed experiments to assess statistical and physicochemical limitations, and especially to open the way for "statistical thinking" and "scientific intuition" to define totally new ways of approaching the measurement problem, yielding significantly reduced detection (quantification) limits. Illustrations will be given of successful reinvention of the measurement process, drawn from fields ranging from radiocarbon dating to the measurement of low levels of atmospheric radioactivity. Time permitting, the talk will conclude with a brief introduction to the use of "Standard Test Data" to reveal the real measurement limitations of chemical and physical measurement systems incorporating "black box" software and subconscious operator decisions.

[L. A. Currie, National Institute of Standards and Technology, Gaithersburg, MD 20899]


Back to JRC program
Back to JRC Home Page