To my knowledge typically either parametric or non-parametric methods use mean values for calculations the significance of differences between different groups. In my data I have got, let' s say 2 groups 5 volunteers in each and 20-30 values of some parameter for each volunteer, i.e. the numbers of observations for volunteers are unequal.
Whether exists some well-known technique which allows me to operate not with mean values but with the observations by themselves?