SED navigation bar go to SED home page go to SED projects page go to NIST home page SED Home Page SED Contacts SED Projects SED Products and Publications Search SED Pages

Education and Training: Statistical Analysis of Incomplete Data for Scientists & Engineers
Time and Location Statistical Analysis of Incomplete Data for Scientists & Engineers
Grace Yang and Hung-Kung Liu
Statistical Engineering Division, NIST
Monday/Wednesday December 13/15 2004, 9:00 am-12:00am
Administration Building, Lecture Room B and E
Abstract In many scientific investigations, the collected data, for a variety of reasons, may be incomplete. The data set thus consists of some complete observations, some partially observed values and some missing ones. Examples abound, including under-counts in population census or in the number of particles recorded by a counter; partially observed lifetimes in reliability studies, in clinical trials, or in radioactive decay experiments. Statistics theory shows that one cannot throw away partially observed data without introducing bias in the data analysis.

In this short course, we consider both censored and missing data. We discuss the Kaplan-Meier (KM) estimator for the reliability function (or survival function) using right-censored data. Martingale calculus provides an analytical method for establishing asymptotic properties and for calculating the variance of the KM estimator. The EM algorithm computes numerical results of the KM estimator. Some more general types of censoring and regression analysis with censored data are also discussed. The analysis involving missing values requires different methods. A partially observed Poisson process is used for modeling missing data. The maximum likelihood estimators of parameters of interest and their uncertainties are derived. Imputation methods for missing data will also be discussed.

NIST case studies, including electromigration in microelectronics reliability, deadtime in the phase Doppler interferometry (PDI) recordings, partially observed neutron lifetime using magnetically trapped neutron, software reliability, and quality assurance for software embedded systems are studied.

Comments on Course REGISTRATION FEE IS $100.

A set of notes will be provided for the class.

The class size is limited to 20 students.

Further Information For further information, contact or register online (once the course is scheduled) at http://www-i.nist.gov/cgi-bin/training.cgi

Date created: 12/3/2004
Last updated: 12/3/2004
Please email comments on this WWW page to sedwww@cam.nist.gov.

[ SED Home |  Education Home |  Previous ] Next ]