I have 10 independent locations (separated by at least 5 km) where I have collected abundance data for 17 species of birds. In particular, I want to look for correlations between one particular species and the 16 other species. Each site was sampled multiple times (separated by 2-8 weeks), some only 17 times while others were sampled 50+ times.

I know I can perform Spearman correlations for each location (most data are not normally distributed) and calculate r for each pair. However, as might be expected, not every location shares the same significant relationships and some are even conflicting (some + and some -). Some significant results seem rather spurious (significant based on only 2 observations of one species).

Is there some way I can compile all these different correlation results into a single test to determine whether there is a general trend for the one species to correlate with others? Am I doing this all wrong? I'm floundering and would welcome any advice.


When you have one correlation, or whatever, what do you imagine it will tell you? There is not much point in having a data set rich in texture and then pounding or blending it into one mass with a muddy homogeneity, which in effect is what one numerical summary would imply.

You would be better off seeking some visualizations (I would favour scatter plot matrices) and starting to read about multivariate analysis. You need to ask around according to your mathematical level, but this book covers a lot of ground:


Presumably you are a biologist: I recommend writing "particular species", not "specific species". The last just sounds awkward.

| cite | improve this answer | |
  • $\begingroup$ Thank you for the suggestions. This is one (relatively small) part of a larger study where I'm trying to look for potential effects and relationships between one particular non-native bird species and the larger bird communities. I have used rarefaction and MRPP to look at the communities as a whole and wanted to look at populations of species that might be food competitors. I hoped the correlations could help identify general relationships for more focused future hypothesis testing. If I understand your suggestion correctly, I would be trying to visualize and interpret 160 scatterplots. $\endgroup$ – Christopher Jun 14 '13 at 18:38
  • $\begingroup$ Just one of several possibilities. Correspondence analysis is another. $\endgroup$ – Nick Cox Jun 14 '13 at 23:38
  • $\begingroup$ For now, I'm trying a scatterplot matrix of 16 graphs (focal species with 16 other species) with the data from all 10 sites posted with different symbols. Gets hard to differentiate because the graphs are small and things get clustered in the lower left. I also supplemented this with a table showing r for each species at each site and indicating significant relationships in bold. Gives a feel for the data, but it all seems so subjective. $\endgroup$ – Christopher Jun 19 '13 at 14:22
  • $\begingroup$ Would it be appropriate to do a one-sample t-test for each species based on the 10 sites to determine whether or not the mean correlation coefficient was significantly different from zero? I find it appealing because it seems like an objective way to ascertain a general trend across sites. $\endgroup$ – Christopher Jun 20 '13 at 21:55
  • $\begingroup$ If you have new questions, best to ask new questions with full context. $\endgroup$ – Nick Cox Jun 21 '13 at 8:22

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.