Missing covariable values in linear regression models (Q2760418)

Incomplete data are a problem met in almost all areas of statistics. What should be done about missing data in multiple regression studies? The answer depends very much on the mechanisms that cause values to be unobserved. This booklet is devoted to the problem of missing values in linear regression models. The linear regression model without missing values, various estimate methods like least squares and maximum likelihood, just as the theory of mixed and weighted mixed estimates are introduced. The various missing data mechanisms, procedures for the treatment of missing values and diagnostic strategies are examined. In particular, two methods of the treatment of missing values are considered, which are dispersed in the standard software: `casewise deletion' and `pairwise deletion', also named as complete case and available case methods, respectively. Bootstrap methods in linear models, maximum likelihood methods, pattern mixture models and multiple imputation are presented as basic procedures for the treatment of missing values. Diagnostics to detect non-MCAR mechanisms are considered, and graphics for the detection of missing data mechanisms are proposed. The considered methods are implemented in C\(++\). NEWLINENEWLINE{ Contents}: Introduction. Linear Regression Model. Missing Data Mechanisms. Missing Values in \(Y\). Models with Missing \(X\) Values. Complete Case and Available Case Methods. Base Model Methods. Imputation Methods. Overview of Methods. Bootstrap Methods. Diagnostic Procedures for Discovery of non-MCAR Mechanisms. Graphics for Diagnostics of Missingness Mechanisms. Simulation Studies. Literature.