An Application of Stagewise Estimation to Monitor Latent Class Changes Over Survey Periods

• Activity Number: 432 - Contributed Poster Presentations: Section on Statistics in Marketing
• Type: Contributed
• Date/Time: Tuesday, July 31, 2018 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Marketing
• Author(s): Kei Miyazaki* and Takahiro Hoshino and Ulf Bockenholt
• Companies: Kansai University and Keio University and Northwestern University
• Keywords: latent transition analysis; stagewise estimation

Abstract

The purpose of this study is to propose the latent class analysis that can monitor the generation and extinction of latent classes, and to develop a stagewise estimation that allows the latent classes found in past analyses to exist in the present analysis. latent transition analysis (Collins and Wugalter, 1992; Reboussin et al., 1998) is the method for monitoring changes of the composition ratio, the generation and the extinction of latent classes over several survey periods. A standard latent transition analysis, however, we cannot use unless the data from all the time periods is merged while a sequential analysis is often required in practical situations. Based on this practical problem, we improve the standard latent class analysis, and propose a framework for monitoring the changes in latent classes, the generation of new latent classes, and the extinction of existing latent classes over several survey periods. Additionally, we prove the consistency of the estimators of the proposed estimation method. Finally, in order to prove the applicability of our proposed method, we analyze some marketing data set about consumers' purchase behaviors.