Small sample estimation of regression parameters in the three-variable linear model, with incomplete observations

This paper describes a new method for estimating the parameters, in linear regression models containing one dependent and two independent variables, when some observations are incomplete. The proposed technique is a markedly improved version of an approach described in previous papers [2, 3]. Results of Monte Carlo experiments suggest that the adopted modifications enhance considerably the small sample performance of the estimators. Cet article decrit une nouvelle methode pour estimer les partmetres de la regression lineaire lorsque certaines observations sont incompletes. Le Modele considere concerne le case ou la regression contient, en plus de la variable dependante, deux varibalbes independentes. L'approche suggeree constitue une amelioration importante par rapport a d'autres techniques similaires proposees anterieurement {2,3}. Les resultats d'experiences de Monte Carlo suggerent en effet que les nouveaus estimateurs se comportent nettement mieux, dans les petits echantillons.