Gilg, Julie A; (2000) Hierarchical Bayesian models for linear and non-linear animal growth curves. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Hierarchical_Bayesian_models_f.pdf Download (4MB) | Preview

Abstract

There are many possible ways to analyse repeated measures such as animal growth data. Recent developments in computational methods mean that the natural approach of modelling the growth of each animal with a parametric curve with the parameters allowed to vary randomly between animals is now practically as well as theoretically feasible. The basic model structure is one level for individuals, one for the population and a third for prior beliefs. This means that the individuals are modelled as being a sample from some population, as indeed they are. We have used Markov chain Monte Carlo methods to fit such models to data for pigs and for cats. For one data set the growth was only recorded over a short time and was approximately linear. For this example we were able to use Gibbs sampling. Over longer time periods animal growth is generally non-linear. We discuss some of the commonly used growth functions for fitting such data. When using these non-linear functions at the first stage of our models we used random walk Metropolis algorithms in order to fit the models. We also include an analysis of some data which included measurements of various body components made after slaughter as well as series of live weights. For this data we were able to use a more sophisticated model which used diphasic functions at the first phase. These functions comprised of the sum of two phases each of which represented a separate group of body components. This approach provided information on the development, or changes in the form of the animals over time, as well as on the overall growth.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as XMLCSVJSON

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item