Abstract
The effects of model misspecification and non-normality of errors on estimates of coefficients are investigated in the multiple regression model.The expressions for the asymptotic expectation and variance of the M estimators, based on the Huber function, and a modified version of the M estimators are presented for a misspecified multiple regression model. The small sample properties of the least squares, M and modified M estimators are examined by Monte Carlo techniques when the true model is quadratic but a simple linear model is assumed and the errors are non-normal. Results show that the M estimators are sensitive to the misspecification in terms of their bias and variance where the degree of sensitivity is dependent on the choice of design, values of the misspecified parameters, and the error distribution. For the situations examined, the modified M estimators achieved the best overall performance.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call