A Comparison of Several Algorithms and Models for Analyzing Multivariate Normal Data with Missing Responses

• Main Authors: Mojtaba Ganjali, H. Ranji
• Other Authors: The Pennsylvania State University CiteSeerX Archives
• Format: Text
• Language: English
• Published: 2007
• Subjects: Maximum Likelihood; Data Augmentation; The EM Algorithm; Multiple imputations; Heckman’s selection model; DML
• Online Access: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.563.9445; http://www.pvamu.edu/Include/Math/AAM/Vol3_No1/Ganjali AAM-R48-MG-071707 Final _5_ 6-12-08.pdf

In this paper we compare some modern algorithms i.e. Direct Maximization of the Likelihood (DML), the EM algorithm, and Multiple Imputation (MI) for analyzing multivariate normal data with missing responses. We also compare two approaches for modeling incomplete data (1) ignoring missing data and (2) joint modeling of response and non-response mechanisms. Several types of Software which can be used to implement the above algorithms are also mentioned. We used these algorithms for a simulation study and to analyze a data set where outliers affect the parameter estimates and final conclusion. As the variance of the estimates cannot be obtained using the available software for some of the algorithms, a bootstrap method is used to find them.