Trying to prove significant difference between algal biovolume data

Hello everyone I would like if possible to ask a question regarding a part of the statistical analysis in my MSc project:

Following the sampling of the inlet and outlet of a stormwater pond at different times for different types of phytoplankton

I wish to make comparisons of time same sampling spot at the different dates ( for example Inlet 1 compared to Inlet 2) as well as the difference between inlet and outlet on the same sampling date ( like Inlet 1 with Outlet 1).

So far I have come to the conclusion that the data are very skewed so I likely need to use non-parametric tests.

So far I have tried the following using minitab:

Wilcoxon test and Mann-Whitney tests comparing the pairs above, However I keep getting P values which are over 0.05 despite the what is in my opinion a large variation between the sample data.

I have also tried transforming the data in various ways like ranking and logarithms with no success

Could it be that the tests I am performing are not appropriate for this type of data?

I would be extremely grateful if someone with more experience in statistical tests could help me find if I am just analyzing them wrong or not
Last edited:


TS Contributor
Geia sou Mano.
Why do you have many zeros? The numbers then are going to be very samll, the averages i mean. Logarithms you said? how? The problem is that the sample sie is small and the numbers too low. Any more explanations or suggestions?
I see that is why...
I had started to suspect that this was the case.

This experiment involved counting and identified phytoplankton cells from ponds. The zeros mean that I did not found any algae of that type in that sample or their biovolume was so small that it was not viewable at 6 decimal places.

I will consult my supervisor on this issue.

An acquaintance suggested that I should add a small number on all the data so that I will eliminate the zeros, what do you think? would that be a valid transformation?

Thanks for the reply


TS Contributor
That is an issue, the accuracy of the measurement tool. You know what, if you have abetter accuracy and remove the zeros, try again, ubt the smaple size is still small.
if we put the sample size issue aside, do you think that Wilcoxon test is appropriate for inlet to inlet and inlet to outlet column comparisons?


TS Contributor
no, i prefer the parametric tests, but if you want, yes go ahead with these. Wilcoxn for 2 dependent groups and Mann-Whiteny-Wilcoxon for 2 independent groups. But pairwise tests. If you want all together (3 or more groups), kruskal-wallis for independent groups or the other one for depdent groups whose name I do not rember now.
You probably talk about Friedman test?

Anyways I am not considering parametric tests cause the data are far from being normally distributed ( it happens a lot in biological data I am told). I will keep working on the report and see what I can do about those tests then. Thanks for your input again


TS Contributor
Yes indeed, friedman. Parametric assumptions is not the lkey to decide to use t-test or wilcoxon. There is also non a parametric procedure called bootstrap for the t-test and then the whole t-test is performed i a non-parametric way. But anyway, you do what you think best for you.