I have met some problem with testifying differences between frequencies.

Let's me present it with concrete example.

1. I have 100 sample plots in three habitat types: deciduous (DW), coniferous (CW) and mixed (MW) wood.

2. Each of the plots (DW1,..., DW35, CW1,...,CW35, and MW1,...,MW30) is inhabited by a set of species (different in the plots is term of composition and total number).

3. We are focusing on reproduction stretegies of species in the sample plots. Given species may use one (only) strategy: sexual (S), asexual (A), or mixed (M).

4. So, in each the plots we have data on reproduction strategy frequency. i.e. a number of species with the stretegy S, a number with A, and a number with M.

5. We are interested in the differences in reproduction strategy composition (frequency pattern) between the distingushed wood types.

So, the dataset looks like that:

Plot DW1 DW2 ... DW35 ... CW1 CW2 ... CW35 MW1 MW2 ... MW35

S------10-----5 ... etc

A------11----20 ... etc

M-------2-----5 ... etc

where the numbers represents the number of species with a given strategy.

Question is:

how I can test the differences in frequency patterns of reproduction strategies between the wood types (DW, CW and MW). I could sum up the frequencies within a wood types, of course, and build the 3 (wood types) x 3 (reproduction strategies) table and use a chi-square test or other but it will ignore the differences between the plots of given wood types.

So, is there a test (an ANOVA "analogue") which enables comparing the frequency distribution between the subsets of samples?