Development of Statistical Techniques to Better Utilize Data Characterized by Being Below Instrument Detection Thresholds and by Small Sample Size.
Abstract
Estimation of parametric families for small data sets where a significant portion of the data lay below fixed instrument detection thresholds was investigated. Thus the number of data points was random (an example of Type I censoring). Both analytic and simulation procedures were utilized. In particular, maximum likelihood techniques, order statistic techniques, truncation techniques, fill-in with constants, and fill-in with expected values of the missing points were investigated. For exponential data, truncation seemed most appropriate while for normal and log-normal data, fill-in with expected values (modified to correct for conditioning on the number of data points) was best. The criteria for selection was the total square error. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 24, 1983
- Accession Number
- ADA135408
Entities
People
- A. S. Gleit