Dina
Research SchoolDina
Research School
Studies within agricultural research are often characterised by an hierarchical design. Examples are studies based on measurements on plants within fields within farms, or similarly animals within herds within breeds. Hierarchical design arise both in controlled experiments and in epidemiological studies. Statistical analysis of these data are complicated because the design leads to dependency between observations. These dependencies need to be handled in the statistical model. Therefore, parameter estimation becomes very complex in most cases, and specialised numerical methods are necessary. Monte Carlo methods are becoming increasingly popular for this purpose. The workshop will present the principles behind the methods, and their application for analysing data from the hierarchical designs will be demonstrated. Typically, the models include both fixed effects ( treatment) and random components that models the dependency between observations ( a farm effect). Therefore, such models are often called mixed models.
The lectures and case studies will be accompanied by computer exercises where the workshop participants can get a hands on experience with the methodology. The theoretical basis is likelihood based inference for these hierarchical or mixed models, and topics will include both traditional models and so called latent class or mixture models. Different Monte Carlo methods will be presented, e.g. importance sampling and Markov Chain Monte Carlo (MCMC). Models based on both normal (Gaussian) and other probability distributions (generalised linear mixed models) will be covered. The workshop is aiming at
Author:
phd@dina.kvl.dk. Updated: September 2003