A0765
Title: Pros and cons of Laplace approximations in mixed hidden Markov models
Authors: Geir Drage Berentsen - University of Bergen (Norway) [presenting]
Abstract: Hidden Markov models (HMMs) are a popular tool for modelling temporal data and has been successfully applied to many problems. Using HMMs for analysing multiple processes, such as longitudinal data, have not yet gained the same popularity. This is partly due to the computational complexity involved in adjusting for heterogeneity in the data, which typically involves numerical or Monte Carlo approximations of multiple integrals. Implications of different model formulations for mixed Hidden Markov models (mHMMs) and pros and cons of using Laplace approximations to obtain the likelihood of the data are discussed. Applications of the method to both real and simulated data will be presented.
