I am performing my experiment which I have three independet experiments with three replicates each (9 samples in total) for each group (infected and mock). For the statistical analysis, should I use all these 9 values or the average value of each experiment (3 values only). I want to aplly non parametric test (mann-whitney U test)

Ex: exp1: 10 exp2: 9,0 exp3: 8,0 or Exp1: 10,2/10,5/10,5 exp2:9,01/9,2/9,0 Exp3: 8,1/8,0/8,2

  • $\begingroup$ The question is whether your "three replicates" are actually valid replicates or whether they are three repeated measurements on the same experimental unit (i.e. "pseudo-replication"). If this is the case, you need to take the average, otherwise go with the full dataset. Statistical power goes up with increasing sample size. So your observation you described in the comment under @Pere 's answer makes sense. $\endgroup$ – Stefan Feb 20 '17 at 16:24

The answer might depend in what you are actually trying to do with your data - that is, what is your research question - but for most analysis you will need your whole data, and for some of them you need at least means and standard deviations of every group.

Since you have 3 groups, if you want to assess differences between groups (not just 2 groups), you might want to use Kruskal-Wallis test instead of Mann-Whitney test. For any of them you need the outcomes of every run of the experiment.

  • $\begingroup$ Hi many thanks! What I actually want to do is to detect differences between the two groups. what I noted is that if I used the mean values of the experiments the Mann whitney shows non significant and when I use all the 9 values the test shows very significant although the difference between the groups are quite small. $\endgroup$ – Julian_DE Feb 20 '17 at 14:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.