IJPAM: Volume 22, No. 1 (2005)


Adebowale Olusola AdejumoDepartment of Statistics, University of Ilorin, Ilorin, NIGERIA
Department of Statistics
LMU, University of Munich
Ludwigstrasse 33/I, Munich, D-80539, GERMANY
e-mail: ao123adejumo@yahoo.co.uk

Abstract.In any experiment that involves measurement, counting, or diagnosis, there is the likelihood for some elements of missing observation which can be as a result of any of foreseen or unforeseen circumstances. Virtually in almost all life or social science researches, subjects are classified into categories by raters, interviewers or observers. Over the last four decades, Cohen kappa statistic has been the major statistic for measuring the overall level of agreement that exits between two raters. In this research work, we examine the effects on this statistic in a situation where certain percentages of the counts are missing, by assuming the ignorability criteria mechanism, along the main diagonal as well as off the diagonal cells of the resulting cross-classified table of the ratings by the two raters. We observed that missingness in the resulting cross-classified table has great effect on the Cohen kappa statistic by reducing the level of agreement and in some situation improving the strength of agreement depending on the structure of the missingness in such table, and also change the class of strength of agreement in some situations.

Received: March 6, 2005

AMS Subject Classification: 62K99, 47H40, 60H25

Key Words and Phrases: agreement, kappa statistic, missing at random (MAR), raters

Source: International Journal of Pure and Applied Mathematics
ISSN: 1311-8080
Year: 2005
Volume: 22
Issue: 1