Hi,

Here is my problem:
I have performed a study comparing the effect of administration of 2 different doses of morphine on the quality of the recovery form general anaesthesia in horses. The quality of the recovery is evaluated with a composite scoring system whose values can be anything between 4 and infinity. the data distribution for the recovery scores is skewed (lower bound). because the data is not continuous and not normally distributed, I opted for a Mann Whitney U test.

There was no significant difference between the two group using the Mann Whitney U test.

Now I would like to 1/ retrospectively calculate the power of my study and 2/ calculate the sample size required for detecting the observed difference beetween groups (based on the distribution and results of my study) with a 80% power.
I can not find a software that would calculate that. Minitab, which is my usual software apparently does not do power calculation for Mann whitney tests. Neither does R.

I have also tried nQuery Advisor but it will apparently only do the power/sample size calculation based on the assumption that the data is continuous and normal (but not skewed), or that the ordinal and the number of categories can be stated (which I can not because there is an infinity of possibilities).

Would you please be able to help me? Is there a way that I am missing to calculate the Power of my test or the required sample size ?

Alternatively I was thinking of doing this:
The data is skewed and a log(10) transformation yields a normal distribution.
-- Am I allowed to transform this type of data (knowing that the score is not continuous data but a form of ordinal data)
-- If I can do it, is the calculated power/sample size valid fro the Mann whitney U test I did? or should I analyse the transformed data using a 2 sample t test in my study rather than use the Mann Whitney?

As you can see, I am not a statistician, and I would be immensely grateful if someone could help me with this.

Regards,
MAndo.