Regression analysis for complex survey data with missing values of a covariate

Incomplete observations with missing values of a covariate may be incorporated into the fitting of a linear regression model by maximum likelihood methods. This paper considers the extension of these methods to accommodate a complex sampling design. Point estimators are weighted within a pseudomaximum likelihood framework. Standard errors are estimated by a jackknife method. The approach is applied to the fitting of a linear regression model to data from the British Household Panel Survey, where the response variable is a measure of stress and the covariate with missing values is income.