ABSTRACT
In business data analysis, it is well known that the comparison of several means is usually carried out by the F-test in analysis of variance under the assumption of independently collected data from all populations. This assumption, however, is likely to be violated in survey data collected from various questionnaires or time-series data. As a result, it is not justifiable or problematic to apply the traditional F-test to comparison of dependent means directly. In this article, we develop a generalized F-test for comparing population means with dependent data. Simulation studies show that the proposed test has a simple approximate null distribution and feasible finite-sample properties. Applications of the proposed test in analysis of survey data and time-series data are illustrated by two real datasets.
ACKNOWLEDGMENT
This research was supported by a University of New Haven 2006 Summer Faculty Fellowship.
Dr. Jiajuan Liang is an Associate Professor in statistics and has been doing research in statistics and its applications in business since he joined the University of New Haven in 2001.
Dr. Linda Martin is a full Professor in quantitative business at the University of New Haven and has been doing research in quantitative methods and their applications in business.