A Note on Some Methods for Regression Analysis with Incomplete Observations

摘要

Some related proposals for estimating regresssion coefficients with incomplete experimental observations are discussed. The proposals include 'approximate standard errors' for the estimators. It is shown that the estimators of the regression coefficients are consistent under fairly weak conditions, but that only under rather strong ones can be usual (asymptotic) tests of significance be validly based on the estimated coefficients and the computated standard deviations. The covariate-sequence of assuming a random or fixed specification for the covariates are also discussed.