Power of Statistical Tests Used in Correlation Techniques for Battlefield Identification.
Abstract
This report is the second in a series presenting a study of self-correlation algorithms in intelligence systems. These algorithms use multivariate statistical tests to determine the equality of mean vectors from two different data sets. For example, tests are used to determine the equality of location vectors from two different data sets (Are the data from the same emitter?). This report considers estimation of the probability that these tests may lead to an incorrect decision. Possible test errors are studied under different assumptions concerning: 1) the distribution of the data, e.g. normal error, skewed error, etc.; 2) the estimated location of the emitter, e.g. mean of the data, most frequent value, etc.; 3) the variability of the error, i.e. the variance-covariance of the data; and 4) the amount of data, i.e. sample size. frequency of test errors were estimated by simulation for most of the cases studied. The results indicate that in some of the cases the error rate is high enough to be of possible concern. Keywords: Robustness; Statistical distributions.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1985
- Accession Number
- ADA166477
Entities
People
- Daniel Hockman
- Janet Myhre
- Michael Rennie
- Will Duquette
Organizations
- Jet Propulsion Laboratory