Model-Based Cluster Analysis of Incomplete Continuous Data with Intermittent Measurement Errors

• Activity Number: 186
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Juwon Song*+
• Companies: Korea University/University of California at Los Angeles
• Keywords: Cluster Analysis ; Incomplete Data ; Measurement Error ; Finite Mixture Model

Abstract

Cluster analysis is an analysis technique to classify objects into several groups based on characteristic variables. Model-based cluster analysis assumes that data can be expressed as a finite mixture of distributions. The mixture of the multivariate normal distributions is commonly chosen for model-based clustering. However, when data include intermittent measurement error, the normal distribution may not fit well. Moreover, cluster analysis using standard statistical software assumes that data are fully observed. In this study, we consider model-based cluster analysis of incomplete continuous data with intermittent measurement errors. A finite mixture of multivariate normal distributions is extended to handle both missing values and intermittent errors. An MCEM algorithm is proposed to handle model-based cluster analysis with missing values and intermittent measurement errors by using a measurement error model.