DoctorJohn Holmes

While originally motivated by applications in genomics, specifically as applied to livestock species such as sheep, my research has expanded to include

• Investigating properties of linear mixed models, particularly around random effect prediction in very large datasets.
• Probabilistic approaches to determining latent structure, inspired by classical dimension reduction techniques such as principal components/factor  analysis, for a variety of categorical data types found in the social and health sciences, with examples ranging from arrest and prosecution data, and to simultaneous modelling of cause of death and age of death time series.
• Approximate Bayesian inference, mainly Variational Bayesian methods, with a focus on logistic regression (but can cover negative binomial) type models for a range of priors. This also includes  studies of the properties of logistic transformations of normal random variables, and the possibility of developing more accurate confidence/credible intervals for proportions and odds ratios.