Introduction to Experimental Design
Training session with Dr Helen Brown, Senior Statistician, at The Roslin Institute, January 2016.
***********************************************
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:
Often accepted as sufficient evidence for a result
---Eg Males are statistically significantly heavier than females
---Demonstrates result is reproducible
Sample multiple observations
---How large a sample?
---What to sample
Influenced by amount of variability in data
---Are there design measures that will reduce this?
Which statistical method?
---Appropriate method available given the design to reach desired conclusions?
Need for multiple samples
---Allows us to exclude coincidence
---Allows us to generalise
---Ensure result is reproducible
How many measurements?
-Enough
---Sufficient to have a good chance of achieving statistical significance
---but not too many!
-Too many:
---Likely to draw conclusions but
---Some unnecessary effort, money, animals
-Too few :
---Risk not drawing conclusions
---All effort, money, animals wasted
-Determining sample size – next session
-Practical measures for reducing noise
-Take a ‘pre-treatment’ measurement
---control for it in analysis
-Good experimental practice and technology
-‘Design out’ nuisance factors that may increase variation
---cage, age of animal, etc
-Allocate interventions within individuals
---eg treat different sides of body with different interventions
-Take several measurements on each ‘unit’ and use average
-Less variation in data - More chance of achieving statistical significance
Reduce variability using selection criteria
-Eg Omit animals with, for example :-
---unusual traits
---infection
---over or under specified weights
---etc
-BUT bear in mind effects on resulting interpretation
Reduce noise by taking repeated observations
-Repeat measurement and use average
-Example: Weighing crabs
---Single measures
---Means of 2 measurements per crab
---Means of 5 measurements per crab
- Tags
-