Roderick J. A. Little reviews methods for regression analysis with missing independent variables. Six classes of procedures are distinguished: complete case analysis, available case methods, least squares on imputed data, maximum likelihood, Bayesian methods, and multiple imputation. The paper compares and illustrates these methods when missing data are confined to one independent variable and discusses extensions to more general patterns. It emphasizes the importance of methods that account for missing data not being missing completely at random (MCAR). Likelihood-based methods are preferred as they provide more accurate estimates than methods that only use data on the independent variables. The paper also discusses Bayesian simulation methods and multiple imputation as promising avenues for future research. It highlights the need for more widely distributed software that goes beyond complete-case and naive imputation methods. The paper provides examples of different missing data patterns and discusses the performance of various methods under different assumptions about the missing-data mechanism. It concludes that maximum likelihood methods are generally more efficient and accurate than other methods, especially when data are missing at random (MAR). The paper also discusses Bayesian methods and multiple imputation as alternative approaches to handling missing data.Roderick J. A. Little reviews methods for regression analysis with missing independent variables. Six classes of procedures are distinguished: complete case analysis, available case methods, least squares on imputed data, maximum likelihood, Bayesian methods, and multiple imputation. The paper compares and illustrates these methods when missing data are confined to one independent variable and discusses extensions to more general patterns. It emphasizes the importance of methods that account for missing data not being missing completely at random (MCAR). Likelihood-based methods are preferred as they provide more accurate estimates than methods that only use data on the independent variables. The paper also discusses Bayesian simulation methods and multiple imputation as promising avenues for future research. It highlights the need for more widely distributed software that goes beyond complete-case and naive imputation methods. The paper provides examples of different missing data patterns and discusses the performance of various methods under different assumptions about the missing-data mechanism. It concludes that maximum likelihood methods are generally more efficient and accurate than other methods, especially when data are missing at random (MAR). The paper also discusses Bayesian methods and multiple imputation as alternative approaches to handling missing data.