A primer on the use of modern missing-data methods in psychosomatic medicine research

• Main Author: Craig K. Enders
• Other Authors: The Pennsylvania State University CiteSeerX Archives
• Format: Text
• Language: English
• Published: 2006
• Subjects: Rubin; DML
• Online Access: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.546.542; http://intl.psychosomaticmedicine.org/content/68/3/427.full.pdf

Abstract: This paper summarizes recent methodologic advances related to missing data and provides an overview of two “modern” analytic options, direct maximum likelihood (DML) estimation and multiple imputation (MI). The paper begins with an overview of missing data theory, as explicated by Rubin. Brief descriptions of traditional missing data techniques are given, and DML and MI are outlined in greater detail; special attention is given to an “inclusive ” analytic strategy that incorporates auxiliary variables into the analytic model. The paper concludes with an illustrative analysis using an artificial quality of life data set. Computer code for all DML and MI analyses is provided, and the inclusion of auxiliary variables is illustrated. Key words: missing data, full information maximum likelihood, direct maximum likelihood, maximum likelihood, multiple imputation, attrition. DML direct maximum likelihood; MI multiple imputation; ML maximum likelihood; LW listwise deletion; AMI arithmetic mean imputation; SRI stochastic regression imputation; DA data augmentation; QOL quality of life; MAR missing at random; MCAR missing completely at random; MNAR missing not at random; LOCF last observation carried forward.