Impact of Departure from Normality on the Efficiency of Estimating Regression Coefficients.pdfVIP

Impact of Departure from Normality on the Eciency of Estimating Regression Co ecients when Some Observations are Missing H Toutenburg  V K Srivastava Septemb er Abstract This article considers a linear regression mo del in which some obser vations on an explanatory variable are missing and presents three least squares estimators for the regression co ecients vector One estimator uses complete observations alone while the other two estimators utilize repaired data with nonsto chastic and sto chastic imputed values for the missing observations Asymptotic prop erties of these estimators based on small disturbance asymptotic theory are derived and the impact of departure from normality of disturbances is examined Intro duction During the pro cess of data collection we often encounter situations where some observations cannot b e recorded for one reason or the other Such instances o ccur quite frequently in mail surveys opinion surveys crop surveys so cio economic enquiries and planned exp erimentation in biological industrial and medical sciences Consequently the traditional statistical analysis cannot b e conducted due to some missing observations Now there are two alternatives One is to conne attention to complete observations alone and to discard the remaining incomplete observations The other alternative is to repair the data following some imputation pro cedure for lling the missing values and then to conduct the analysis Both the strategies have their own limitations and qualications When few values of some explanatory variables in a linear regression mo del are missing there are various ways to nd