I've got data from eight one-hectare tropical forest plots, with all trees greater than a certain size measured and identified within them. These plots are different forest ages with two replicates per four age classes (e.g. 2 sites of forest which is 40 years old, 2 sites 60 years old etc.). Within these one-ha plots there are a total of ~17000 individual trees. I have several explanatory variables and wish to investigate the effect of age and these variables on species diversity. However, I'm not sure how to proceed with the sample size of 8 plots, and only 2 per age class, seeming incredibly low for any models - despite the large size of these sampling units.

Is there a way I could perhaps randomly subsample within the plots to increase my number of replicates (I would look into spatial autocorrelation) as the data does span a large area? Or would this be a form of pseudoreplication?

  • $\begingroup$ It sounds like your 8 hectares are strata, not units. If you aggregate up all the data you collect within each hectare, it becomes an ecological study (a type of study looking at relationships between averages rather than average relationships). An ecological study would be a fairly deficient design for your hypotheses. $\endgroup$ – AdamO Sep 13 '19 at 14:25
  • $\begingroup$ I'm not entirely sure what you mean by strata, sorry! From what I've googled it looks like I've got 4 strata i.e. the 4 age categories, with 2 plots per stratum. It is an ecological study and the data had already been collected before I started looking into it, I'm trying to make the most of the plots and size of the plots that we have. $\endgroup$ – A.Elsy Sep 13 '19 at 14:46
  • $\begingroup$ Well that is the problem. You refer to "age" but one can only assume you mean age of trees. In what sense does a hectare of trees have an "age"? $\endgroup$ – AdamO Sep 13 '19 at 14:48
  • $\begingroup$ Ah sorry, maybe I wasn't clear - it's the forest age i.e. the amount of time since forest started growing on an area of land. That land could have previously been pasture/farmland etc. I'll edit the question to make that clear. $\endgroup$ – A.Elsy Sep 13 '19 at 15:03
  • $\begingroup$ @ A.Elsy, when @AdamO uses the term "ecological study", that does not mean a study having to do with ecology. This is a technical term used for a particular kind of observational study: en.wikipedia.org/wiki/Ecological_study. This is a little confusing in the context of your study of ecological processes. $\endgroup$ – Jdub Sep 13 '19 at 17:04

Yes, you should explore what you suggest. If you can show that there is no evidence that subsamples are dependent on which larger sample they are in, then you could argue that those subsamples can be considered true replicates.

Depending on your data, one method that might be used for your data would be a Mantel test. I am guessing about your data, but an example would be that you would have two data sets, one with rows that represent subsamples and columns that represent species, another with rows that represent the same subsamples but columns that represent geographical coordinates. You generate a distance matrix for each, then use the Mantel test to determine whether the distance matrices are significantly correlated. There are several things that might be relevant, but that is one example to get you started.

Another option is to simply account for spatial autocorrelation in your analysis. For example, in the context of Mantel tests, you could read:

Crabot, J, Clappe, S, Dray, S, Datry, T. Testing the Mantel statistic with a spatially‐constrained permutation procedure. Methods Ecol Evol. 2019; 10: 532– 540. https://doi.org/10.1111/2041-210X.13141


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.