Introduction to Statistical Modelling
Training session with Dr Helen Brown, Senior Statistician, at The Roslin Institute, December 2015.
************************************************
These training sessions were given to staff and research students at the Roslin Institute. The material is also used for the Animal Biosciences MSc course taught at the Institute.
************************************************
*Recommended Youtube playback settings for the best viewing experience: 1080p HD
************************************************
Content:
Mixed (or multilevel) models
-Allows several sources of random variation
---Eg Between and within animals
---Between and within centres in a multi-centre study
---Add Random Effects to the model, eg random effects for animal, centre, etc
-Allows correlation structure for the data
---Eg correlation pattern for repeated measurements
-By contrast GLMs assume random variation at the error level only
---All observations considered independent
---Effects described as Fixed Effects
-Note: Mixed models are similar to GLMs in that they have :
--- One dependent or outcome variable
--- One or more independent variables
Mixed model for repeated measures data
-Repeated measurements on same individual or ‘item’
---Eg repeated measurements on crabs of consumption of different diets
-Crab is fitted as a ‘random’ effect
---Means for crabs are assumed to have a normal distribution
---Account now taken of correlation between multiple measurements on the same animal
-Model is suitable even if some data are missing
-Note: Some packages offer ‘repeated measures ANOVA’, but only suitable for complete data
Potential advantages of mixed models:
-Results take into account variability at several levels
---Eg variability in results between farms
---Eg between and within animals
-Unbalanced or missing data does not cause a problem
-More accurate estimates (particularly if missing data)
- Tags
-