Multivariate Estimation from `Two Variables at a Time`Observations

Suppose that we wish to estimate the mean µ and the covariance Cof a random p-vector X with p > 2, but we can only sample from thevector X two of its p components at a time. We give both nonparametric estimates and the maximum likelihood estimates insert ignore into journalissuearticles values(MLEs); undernormality, and their covariances.