I am a stats rookie and have a relatively basic question regarding the normal distribution of data. If I measure a continuous variable under several different conditions/treatments, I look at the distribution of data points separately under each condition and not the whole data set. What if individual treatments are not normally distributed, but others are? Can I, for instance, transform the values of individual treatments to reach a normal distribution in those or would I have to transform the variable for all treatments? I assume the latter, if I want to do more than a pairwise-comparison with a control group or similar. Thank you very much in advance and sorry in case there is an already existing question that goes in the same direction.
Asked
Active
Viewed 35 times
1 Answers
1
You have to transform all treatments in the same way. It's the same variable after all.
In R, you can use, e.g, https://stat.ethz.ch/R-manual/R-devel/library/MASS/html/boxcox.html to find an optimal power transformation for your data, but this will not solve all possible problems.
If variable transformation to normality does not work, you would usually move to a non-parametric test, or a regression with a different distribution (i.e. GLM).
Florian Hartig
- 6,499
- 22
- 36