if i use mann-whitney,is it okay if i still use mean (not median) to compare 2 group?

#1
so in my minor thesis, i compare 2 groups (deaf and normal children; n1=n2=15) in their forward and backward digit span scores and word-list memory task (wlmt) scores (there are 4 scores: 1st, 2nd,and 3rd trial, and the mean scores of all three trial).
in the normality tests, distribution is not normal, even when i've transformed it with Log10, Ln, and squareroot. so i use Mann-Whitney to compare.

to present the data, do i really have to use median instead of mean? and if i use mean, will the data still be represented well?

thank you sooo much for ur help :)
 

SiBorg

New Member
#2
Re: if i use mann-whitney,is it okay if i still use mean (not median) to compare 2 gr

Are you comparing the mean scores of all three trials with the Mann-Whitney, or the individual scores from each trial?
 
#3
Re: if i use mann-whitney,is it okay if i still use mean (not median) to compare 2 gr

both. the individual scores (1st, 2nd, 3rd trial) and the mean of all trials' score also.

and one more thing, the score in digit span and word list memory task is numeric data right? 'cos one of my lecturer that i ask about it said that it is ordinal data. and so i'm confused here.

thnk u for ur help!
 

SiBorg

New Member
#4
Re: if i use mann-whitney,is it okay if i still use mean (not median) to compare 2 gr

Hi. I'm no statistical expert so please read all this with a pinch of salt and check with your supervisors...

I'm guessing the forward and backward digit span scores are number of digits remembered forwards and backwards of a predetermined number? I'm guessing, therefore, that that would be ordinary numeric data. The same for the word list memory task. In other words, remembering 4 digits is twice as good as 2 digits and 4 times as good as 1 digit. It would only be ordinal if the scores bear no quantitative relationship to each other.

You should probably use median if you are going to describe non-normal data, as the median should provide a better summary of a skewed dataset (mean and SD only describe a normally distributed dataset well, otherwise they are not so useful). And it's probably best to use non-parametric statistics such as Mann Whitney test to start with.

However, you may be able to use parametric tests if your sample size is large enough, due to the central limit theorem which states that the sample means of infinite samples taken from a population will converge to a normal distribution even if the underlying distribution is not normal. Since t-test, etc, is concerned with means and not the data itself, this can sometimes allow you to use parametric tests on non-normal data. The usual cut-off for this is 30 or so samples (you have n=15) but this cut off is arbitrary. So you could perform random sampling of your dataset to see if the means of the random samples converge to a normal distribution. I'm not sure, however, if this would be a valid thing to do, but it might work. On the other hand, the safe option is just to use the non-parametric stats.
 

Karabiner

TS Contributor
#5
Re: if i use mann-whitney,is it okay if i still use mean (not median) to compare 2 gr

to present the data, do i really have to use median instead of mean? and if i use mean, will the data still be represented well?
Why "either - or" ? I always urge authors to present both, so one has a much better
picture of the data (e.g. if mean = median you know the distribution is symmetric).

Regards

K.
 
#6
Re: if i use mann-whitney,is it okay if i still use mean (not median) to compare 2 gr

thank u so much for the help. i've asked my teacher who tutor me in my minor thesis, and he said it's okay to use median
i just had to change my minor thesis title. my previous title was "the comparison of mean score", and now it's only "the comparison of score between..."
thnk u again :D