Hope someone can help me out on this.

I am working on a software for administrating local currency trade systems, for which I want to produce trade overviews with graphs and accompanying statistical tests. The idea is to have a graph with time categories on the x-axis (months in a year), and output (trade activity by members) over the y-axis.

Now I would want to have some test indicating if the variations in output shown in the graph over the months on the x-axis are showing any real differences, or if it is just "random natural variation".

Ordering of output over the months is not to be expected (so no output like "jan < feb < march < ... etc). But a certain month (for example december - christmas) might show much higher output than surrounding months. I would like to be able to test for this.

Data is NOT AT ALL normally distributed.

Questions:

1) is it sound to consider this problem as a test with 12 samples from a population, where each month is a sample?

2) If so, would Kruskal Wallis be allowed? Problem is that the population over the months changes. Some members may be followed in their activity over the months (suggesting paired samples), but often there are big changes, and members come and go. In fact, each sample would consist of a part consisting of the same members as in the previous sample, and an almost equally big part of members which did not show up in the previous sample. So samples are NOT independant, but creating a matrix with pairs would show an enourmous amount of gaps because of unpaired observations.

3) what about ties? Some output data would show a lot of ties, which would limit the use of certain tests (Kruskal Wallis, for example).

So, what would be an appropriate test for this?

Thanks in advance.