(Received May 22, 1996; revised March 10, 1997)
Abstract. The paper discusses a likelihood based method of estimation which allows for a small amount of misspecification in the assumption of normality. Asymptotic results suggest that the new method can give an estimated model which is closer to the true model. An application to hearing threshold data is discussed.
Key words and phrases: Local likelihood, semi-parametric inference, robust estimation, model misspecification.