| Zugriffsnummer | 13626 |
| Dokumenttyp | Zeitschriftenartikel |
| Sprache | Englisch |
| Titel | Bayesian-statistical decision threshold, detection limit, and confidence interval in nuclear radiation measurement |
| Autor(in); Institution |
Weise, Klaus; PTB-Braunschweig, retired staff
|
| Quelle/Jahr | Kerntechnik: 63 (1998), 4, 214 - 224 |
| ISSN | 0932-3902 |
| Zusammenfassung | When a contribution of a particular nuclear radiation is to be detected, for instance, a spectral line of interest for some purpose of radiaton protection, and quantities and their uncertainties must be taken into account which, such as influence quantities, cannot be determined by repeated measurements or by counting nuclear radiation events, then conventional statistics of event frequencies is not sufficient for defining the decision threshold, the detection limit, and the limits of a confidence interval. These characteristic limits are therefore redefined on the basis of Bayesian statistics for a wider applicability and in such a way that the usual practice remains as far as possible unaffected. The principle of maximum entropy is applied to establish probability distributions from available information. Quantiles of these distributions are used for defining the characteristic limits. But such a distribution must not be interpreted as a distribution of event frequencies such as the Poisson distribution. It rather expresses the actual state of incomplete knowledge of a physical quantity. The different definitions and interpretations and their quantitative consequences are presented and discussed with two examples. The new approach provides a theoretical basis for the DIN 25482-10 standard presently in preparation for general applications of the characteristic limits. |
