Experimental design

Experimental design

1. Characteristics of a well designed experiment

   • Clearly stated hypothesis
   • Simple
   • Powerful: High statistical power
   • Unbiased: randomisation/blinding
     • Randomisation= the process of making something random
       • Asign numbers to everyone and computer it
     • Blinding= could include the doctor not knowing who is getting what treatment
   • HIGH signal to LOW noise ratio
     • Signal= variance between samples/ treatments
       • This are fixed effect factors (treatment)= variable that can be controlled by the experimentor(signal)
     • Noise= variance within samples/ treatments or between samples

2. Error

   • Type 1 error- reject the null hypothesis when it is true
   • Type 2 error- fail to reject the null hypothesis when it is false

3. Stats tests

   • ANOVA- used when two or more fixed factors affect the measurement parameters
   • Independent sample t-test- No relationship between, compare averages
   • Paired samples t-test- Repeated same measurements with the same individuals and look at the differences, reduce the effect of the variability on samples.
   • all these tests are paremtric tests and require continous data, normally distributed and equal variances between treatments
     • Parametric stats are more powerful than non parametric stats.
     • T= difference between means/ standard error F= between treatment mean sqaures/ error
       • T will be large if we have larger difference between means, low level of variability within our treatment and we have large sample size
         • T= how likely that the difference in means in not due to random chance. bigger T is more likely the difference is significant
           • critical value of T= value of T that must be eqaulied or exceeded for us to have a given level of confidence that the difference is real.

4. Improving the design

   • Increasing the stastical power
     • If statistical power is 50% and there is a signficant effect you only have a 50;50 chance of detecting this effect
       • You want about a 80%= 90% statistcally power
     • Increasing sample size increases statistical power BUT cost animal and money
     • Increasing power can include- larger sample size, increased signal and increased one sided H!.
       • Decreasing power can include- decreased significance level and decreased noise.
   • repeated measures/crossover design e.g ANOVA

5. The experiment design

   • The smallest division of the experimental material such that any two experimental unit can receive
   • Randomised block design- used when you wish to experimentally/statistically control for 1 random factor