I have percent cover data collected over two time periods (immediately after disturbance + 10 years after disturbance) directly at the site of disturbance (central) and around it. Examination of the data indicates that there is a difference in cover at central locations -- only immediately after disturbance and which recovers 10 years later.
Given that the data does not conform to assumptions for parametric tests (skewed distribution), I am looking for non-parametric options. I have used an aligned rank transform ANOVA initially as it is useful for multiple factors including testing for interactions between them. While the overall results is logical (differences are found for the interaction between sampling period and location), the pairwise tests give unhelpful results: the aligned rank of central locations is by far the highest among all positions while the raw rank (or the median) is actually lower in the second sampling period than the other positions. I am not sure that a traditional two-way factorial design and pairwise tests are appropriate when we know are only really interested in a subset of pairs : within positions between years (e.g. Central in Period A vs. Central in Period B) and different positions within years (e.g. Central in Period A vs. Central Period B) but not between positions and years (e.g. Central in Period A vs. Control in Period B is not relevant).
The scientific questions are:
- Is there a difference in cover by position relative to disturbance? (between Central, Peripheral, ...)
- Is there a change in cover from immediately after disturbance to 10 years later (in the plots influenced by the disturbance)?
- Just to be sure, is there a difference in total cover by year? (if there had been a change all the positions over time, it would be attributable to year-effect as opposed to a recovery time effect)