Experimental Design - Parkinson Flashcards
What ARRIVE guidelines relate to reduction?
- Study design
- Sample size
- Allocating animals to experimental groups
- Experimental outcomes
- Statistical methods
- EDA
What are the minefields in experimental design?
- Bias
- False positives, negatives, and power
- Incorrect specification of the experimental unit
- Confounding
How to control against bias?
Randomization, blinding of assay to treatment, blocking
What are false positives?
The change you’ll see a significant result by chance, specified by significance level, p=0.05
What are false negatives?
When we carry out an experiment and don’t see an effect. Power = 1-false negatives = 0.9/0.8
What is the relationship between false positives and negatives
Up power = Up significance e.g. Power = 0.5, sig = 0.05, power = 0.8, sig = 0.01
What is the experimental unit?
The smallest unit which can be independently allocated to a treatment = the replicate
What is confounding?
Differences in your experiment are due to something other than your treatment (asymmetrically = bias, symmetrically = increase error variance and lead to bigger sample size needed?)
What were the problems in the height example?
Sex confounding nationality, only 1/2 population, not significant
How did they fix the problems in the height example?
Blocking gender and using ANOVA
Pros and Cons of using homogeneity to control for confounding?
+ Removes a source of confounding
+ Reduce variability
- Reduce scope
- Reduce sample size
Pros and cons of using block/factoring to control for confounding?
+ Factors out confounding source
+ Reduce variability
+ Maintain scop
+ Maximize sample size
+ Can identify interactions
When do you not integrate sex into research?
- Sex-specific effects E.g. ovarian cancer
-Different M/F models E.g. Lupus (F), kidney damage induced hypertension (M)
How do you integrate sex into research designs
Ensure model works in M and F, and use ANOVA
What are the uses of pilot studies?
Optimize treatment (timing, drug conc.), Obtaining treaetment effect and variability, Streamline procedures
How to control variability?
Make environmental conditions homogenous,
Block/Factor variability (affects sample size = block, affect scope/suspect interaction = factor)
Randomize the rest
How to identify sources of variability?
Raw material, Experimentation, Analysis
What does blocking do?
Systematically removes variation from error thereby increasing power
ONLY PARAMETERS NOT INTERESTED IN E.g. batches of animals
How to use blocking?
Structure as mini experiment w/ at least 1 replicate of each treatment
W/I block keep conditions homogenous E.g Researcher 1/Batch 1/Day 1 and Researcher 2/Batch2/Day2
How to use sequential experiments?
Block and analyze experiment after every block
Advantages and disadvantages of sequential experiments?
+ Highly significant effects detected early
+ Unsuccessful treatments stopped early
+ 20-30% saving animals
-Needs Bonferroni correction = reduce power
- Complicated design => statistician
How does factoring work?
Systematically removes variation from error in parameters you’re interested in = increases power
Quantifies individual treatment effects and interactions
How to use repeated measures?
> 1 treatment on same animal
- less variation
- Non-destructive no carry-over between treatments
- other eye, skin, time series
How do you deal with multiple comparisons?
Minimize number of comparisons => avoid omnibus testing, select comparisons carefully, use ANOVA
