| Zugriffsnummer | 47271 |
| Dokumenttyp | Zeitschriftenartikel |
| Peer Review | unbekannt |
| Sprache | Englisch |
| Titel | Distribution-free confidence intervals |
| Autor(in); Institution |
Weise, Klaus; 7.4, Theorie und Prozeßdatenverarbeitung, PTB-Braunschweig
|
| Quelle/Jahr | PTB-Mitteilungen: 93 (1983), 6, 390 - 394 |
| ISSN | 0030-834X (PRINT) |
| Verlag | Braunschweig: Vieweg |
| Zusammenfassung | If only the resulting value of a measured physical quantity and its empirical standard deviation are given and no further information then, without any a priori assumption, it is impossible to determine the uncertainty of measurement in form of a confidence interval which, with a fixed given probability , embraces the true (expectation) value of the quantity. The optimistic assumption that the estimator of the expectation value is normally distributed implies very many degrees of freedom, and on the other hand the Chebychev inequality leads to an unrealistic width of the confidence interval. Assuming all possible distributions of the estimator to prossess equal a priori probabilities, a confidence interval is derived which is in a certain sense distribution-free and which has a width in between that one of the two other cases mentioned. The width exceeds by inly 6 % to 23 % the width obtained with the normal distribution for confidence levels from 68 % to 95 %; the minimum value holds for nearly 85%. |
