I am comparing the variability of the response to thermal stress for different groups. My response variable is percentage of dead tissue and I am looking for a good metric / index for this comparison. Because the response is a percentage, I am not convinced on using the traditional coefficient of variation (CV = SD / Mean). I wonder if just comparing average Standard Deviations (SD) would be ok, or is there any other suitable metrics for this?
Any reference addressing this topic would be very appreciated.
Thanks!
In the past the suggestion was to do an arcsin-square root transformation of the percent data, but that is now strongly discouraged for mathematical reasons. I think CV would still be appropriate; you may also want to try something nonparametric as the interquartile range in order to avoid the most extreme outliers from having any influence.
You could force the scales to have upper and lower limit:
If you are using SPSS try the following transformation:
COMPUTE newscale=((oldscale-minimum scale value)/range)*n
where:
Newscale= score for new scale with defined upper and lower values
Old scale= score on old scale
Minimum scale value= lowest observed value on old scale
Range= range of observed scores on old scale(highest score-lowest score)
n=upper limit for new scale
This way you can define your own desired lower limit and upper limit for your scale.
Try to have a look to the book: Surveys in Social research -Page 193
- Badges
- Science topic
- Similar topics
- Chemistry
- Pharmacology
- Medical Pharmacology
- Clinical Pharmacology