A Monte Carlo Comparison of Four Estimators of the Dispersion Matrix of a Bivariate Normal Population, Using Incomplete Data
J. N. Srivastava, M. K. Zaatar
Journal of the American Statistical Association · 1973 · Volume 68 · Issue 341
DOI: 10.1080/01621459.1973.10481359
Abstract
Consider a random vector ( X 1 , X 2 ) distributed as a bivariate normal with mean vector zero, and dispersion matrix Σ = [(σ ij )]. Suppose we are given samples of sizes n 1 and n 2 , respectively, from the marginals of X 1 , X 2 , and a sample of size n 3 from the bivariate population of ( X 1 , X 2 ). The problem is to obtain a good estimator of Σ based on this (incomplete) sample. In this article, four estimators of Σ are compared using Monte Carlo methods, and it is found that a certain relatively simple estimator of Σ is the “best” or close to the best in almost all situations.
Enable JavaScript for citation tools and additional paper interactions.