A Minitab Macro for the Kaplan-Meier Product Limit Estimator and the Tarone-Ware Class of Tests.

Abstract

Minitab is a user-friendly statistical computer package widely used in industry and academia. The Kaplan-Meier ple and the Tarone-Ware class of tests are often used in biostatistics and reliability theory to compare two survival distributions. The Tarone-Ware class of tests is generated by allowing the power of the sample size to run from zero (inclusive) up to and including one. The value 0 corresponds to the Mantel-Haenszel test, .5 to the Tarone-Ware and 1 to the Tarone-Ware version of Gehan's test. This power parameter is specified by the user in the MINITAB constant k4. Miller asserts that the Tarone-Ware class of tests allows the censoring distributions to be different, while Gehan's test does not. However, he also cites evidence that Gehan's tests appears to be insensitive to differences in censoring distributions, so it is not clear which test should be used. A MINITAB macro to calculate Gehan's test has been given by Soms. The main idea of the program is to first break ties between censored and completed observations by adding to the censored observations one-half of the smallest non-zero difference for the ordered combined observations. This does not change the K-M ple or Tarone-Ware tests.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Feb 01, 1988
Accession Number
ADA194298

Entities

People

  • Andrew P. Soms

Organizations

  • University of Wisconsin–Madison

Tags

Communities of Interest

  • Biomedical
  • Energy and Power Technologies

DTIC Thesaurus Topics

  • Biostatistics
  • Computers
  • Computing-Related Activities
  • Data Science
  • Estimators
  • Information Science
  • Interdisciplinary Science
  • Mathematical Analysis
  • Mathematics
  • New York
  • Observation
  • Plastic Explosives
  • Statistics
  • Survival
  • Universities
  • User Friendly
  • Wisconsin

Fields of Study

  • Mathematics

Readers

  • Regression Analysis.
  • Statistical inference.