General Mixed-Data Model: Extension of General Location and Grouped Continuous Models

• Activity Number: 232
• Type: Topic Contributed
• Date/Time: Tuesday, August 5, 2003 : 10:30 AM to 12:20 PM
• Sponsor: Section on Health Policy Statistics
• Author(s): Alex de Leon*+ and Keumhee Chough Carriere
• Companies: University of Alberta and University of Alberta
• Keywords: latent variable model ; maximum likelihood ; measurement level ; multivariate normal distribution ; polychoric/polyserial correlations ; probit model

Abstract

Health policy decisions are ideally made from recommendations based on sound and defensible data analyses. However, most often such analyses are made ignoring intercorrelations among various dependent health outcomes due to a lack of needed methodology. In this paper, a general model for multivariate data with mixtures of nominal, ordinal, and continuous variables called the general mixed-data model is proposed. The approach adopted in developing the model is motivated by the need to account for the various levels of measurement in the data, which many conventional approaches fail to incorporate in the analysis. The general mixed-data model includes the mixed-data models of Olkin and Tate (1996), Bedrick, et al. (2000), Poon and Lee (1987), and Anderson and Pemberton (1985) as special cases. A full likelihood-based approach that yields maximum likelihood estimates of the model parameters is outlined, and algorithms to implement it are provided. An alternative estimation method based on the pairwise likelihood approach is also presented. Statistical inference is also discussed for testing the model parameters. A real-data example illustrates the utility of the model.