Mixed models and multilevel data structures in agriculture
MetadataShow full item record
Permanent link to this item: http://hdl.handle.net/10568/2872
Multilevel data structures can occur in many areas of agricultural research for instance on-farm trials, where there can be information at the village, farm and plot or animal level. Analysis of variance except in balanced or nested designs - has been difficult to apply to data with a multilevel structure. Mixed modelling is becoming a standard approach for analysing these types of data. The mixed model facilities are now available in some of the more powerful statistical packages such as Genstat and SAS. The purpose of this guide book is to review the general concepts of mixed models. The document illustrates by example how to recognise the structure in the data and how to fit and interpret a mixed model analysis. The reader is expected to be familiar with simple analysis of variance methods. Three examples are used in the discussion. Example one, 'fodder production trial', is a fairly traditional agricultural experiment, and is included to show how mixed modelling links to more traditional analysis. Example two, 'concentrate feeding trial', is an on-farm trial with a slightly 'messy' hierarchical structure. It is used to show how the ideas of example one can be extended to other situations, and to demonstrate the benefits of mixed modelling. Example three, 'sheep breeding trial', is a more specialised example of breeding trial. It discusses the implications of formulating models in different ways.