Confusion on what level should data follow normality

#1
Hi everyone.

I'm experiencing some confusion as to what level the data should be normal.

For instance, in an unpaired data set, where some vitamin is the main factor (given or not given) applied to a population of mice, you are interested if their blood sugar level is differing between the two treatments across various time points. You want to apply two sample t tests at each time point, so you must make sure the data is parametric. Upon doing some normality tests, you find that the data is not parametric and in need of transformation. Now, at what level does the data need to be parametric? Should different transformations be made at each time point, so that all time points will be parametric? Or, should only one transformation be applied to the whole data set, making it parametric at this level?

Any advice on this would be greatly appreciated!
 
Last edited:
#2
Maybe I'm misunderstanding what you mean by "level" but really whether you believe data is normal or not is a judgement call. T-tests and the like are pretty robust to departures from normality. Can you post some examples of your data or something similiar so we can see how non-normal we are talking? There are obviously other tests you could do that dont rely on normality of data.
 

spunky

Can't make spagetti
#3
For instance, in an unpaired data set, where some vitamin is the main factor (given or not given) applied to a population of mice, you are interested if their blood sugar level is differing between the two treatments across various time points. You want to apply two sample t tests at each time point, so you must make sure the data is parametric.
perhaps you're already aware of this (in which case just disregard this message) but, for the record, you shouldn't do this unless you apply some sort of correction to your type-1 error rate. if you have more than two time periods, you'd be better off doing a repeated-measures ANOVA.