CMStatistics 2021: Start Registration
View Submission - CMStatistics
B1042
Title: A latent multivariate Gaussian process model for longitudinal ageing survey data Authors:  Juhyun Park - ENSIIE (France) [presenting]
Evanthia Koukouli - Lancaster University (United Kingdom)
Andrew Titman - Lancaster University (United Kingdom)
Abstract: The motivation comes from the need of analysing complex survey data coming from the English Longitudinal Study of Ageing, an ongoing biannual longitudinal study which follows up selected individuals aged over 50 years old in England since 2002. This provides valuable resources to study the ageing process from a multi-dimensional perspective and explore how life domains evolve and interrelate with ageing. However, data on individual traits are mostly gathered through questionnaires and tests resulting in a large collection of non-continuous data which are difficult to model directly. Typically, summary scores or other surrogates are derived as continuous variables or limited latent curve modelling techniques are employed, focusing on single traits. In addition, the longitudinal trend is often not examined as a function of age; instead, the regular data collection time is used as the temporal variable and a single estimate is obtained for the age/age group effect. We develop a latent multivariate Gaussian process modelling framework that allows for simultaneous modelling ordinal longitudinal data, which are sampled irregularly and measure multiple life domains. We propose a latent factor structure for the data at a given time point whilst incorporating individual heterogeneity by assuming a multivariate Gaussian process for individuals' latent domain trajectory. We implement our method based on a version of stochastic EM algorithm and present findings from numerical studies.