Statistical Methods for Monitoring Nuclear Tests,
Abstract
In previous papers, these authors have developed general methodology for detecting outliers. In particular, given a training set of, say, earthquake data, a test of hypothesis for testing whether or not a new observation should be classified as an earthquake was developed. The method was based on a generalized likelihood ratio test which did not require normality assumptions or even continuous data. In the most recent papers the technique was modified to allow missing data. In this paper the methodology is greatly extended to address a variety of issues that arise in a multistation environment. In particular, the method is generalized to allow the inclusion of expert opinion as part of the data and it is extended to allow an outlier to be an outlier to more than one population. Thus, for example, one might have training data from earthquakes and mining explosions and desire to know if an observed event should be classified as belonging to either of these groups or not. With the results of this paper, one can test, at a specified level, whether or not the new observation belongs to the joint population. Alternative approaches to the generalized likelihood method are also considered for the problems addressed. However, in general, the generalized likelihood approach seems to be the best method.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 14, 1995
- Accession Number
- ADP204506
Entities
People
- H. L. Gray
- W. A. Woodward
Organizations
- Southern Methodist University