2pSPa1. The need for efficient monitoring of data.

Session: Tuesday Afternoon, May 14

Time: 1:05

Author: Edith L. R. Corliss
Location: Forest Hills Lab., 2955 Albemarle St. NW, Washington, DC 20008


``Shirtsleeve Statistics'' is a name that has been applied to simplified methods for inspecting statistical properties of experimental data. Many of these techniques are in the nature of ``nonparametric'' statistical operations. Because of their simplicity, they are particularly valuable for monitoring the results of an experiment in progress. Also, study of the inherent relations of the variables can make for a more efficient layout of an experimental procedure. One vital result of this study of observations and their uncertainties is the elucidation of the quantity N, the number of independent variables involved in an experiment. Merely counting the number of observations does not suffice. Internal correlations can affect the degree of independence of the observations and thus exaggerate their estimated precision. Techniques are available for simple and timely monitoring of measurement consistencies. It is exceedingly important to keep a running statistical surveillance of data. Inherent variations in the phenomena being measured may obscure temporarily a failure in the measuring or data reduction equipment. Much time and data can be lost. (Remember: Dust swept under the rug raises a lump.)

from ASA 131st Meeting, Indianapolis, May 1996