1
$\begingroup$

I have two species community matrices, where each row is a site and each column is a species. The rows and columns are the same for each matrix (i.e., row 1 = site 1, column 1 = species A for both matrices). Although the sites and species are the same, each matrix has species abundance measured in a different way by different people/analyses. One of the matrices has integer abundance counts (0, 1, 2, etc) while the other uses relative abundance calculated from distribution models (so continuous data). This makes comparing between the two tricky, and where I'm not sure how to proceed.

My idea was to compare the integer abundance count matrix to the relative abundance matrix to see how good the relative abundance results are (basically, a good relative abundance matrix should have similar predictions to the actual real world measurements). But I'm having trouble figuring out how to compare the two, since they're essentially completely different types of measurements/predictions. For example, I don't think I can use adonis2() from the vegan R package without somehow getting the two matrices into the same measurement type.

Does anyone have any advice/ideas on the best path forward?

EDIT: Here is some more information about how the relative abundance values were calculated, as requested by commenters! A correlative species distribution model was calculated for each species, using environmental variables as predictors to create a predicted relative abundance raster across a wide area. We did not perform the distribution model ourselves and only have the output (i.e., predicted abundance for each species at locations x, y, z, etc.). However, we have data on species abundance from our own surveys in areas where the model is predicting abundance but does not have any data. So the idea is to validate whether the distribution model were able to accurately predict species abundance in places where they didn't have training data.

The predictions from the models are for specific types of measurement (i.e., 1 person, walking x minutes, at x time). We have point counts, which are also different types of measurements.

Improve this question
$\endgroup$
3
  • $\begingroup$ Could you please say more about how the relative abundance values were calculated? Please provide that information by editing the question, as comments are easy to overlook and can be deleted. $\endgroup$
    – EdM
    Commented Jan 17 at 8:10
  • $\begingroup$ I just edited the post! Let me know if I still should add more clarification $\endgroup$
    – aeiche01
    Commented Jan 17 at 22:22
  • $\begingroup$ Comparing integer values like counts with some continuous variable like a predicted/estimated expectation value is not a problem and has been done at least since the 1900's when Pearson wrote about his chi-squared test. That doesn't seem like a problem to me. Instead, a problem is that your counts might not be independent and following a categorical distribution as is normally assumed. When we are only given a single matrix of observations, it is difficult to say what sort of statistics we are dealing with. $\endgroup$
    – Sextus Empiricus
    Commented Jan 17 at 23:51

2 Answers 2

Reset to default
1
$\begingroup$

It needs more context. Compare in terms of what? What's the expected outcome? If you has wanted to test for some kind of concordance, I'd suggest using a simple Mantel test or even better a PROCRUSTS/PROTEST analysis. These will give you the degree of correlation or concordance between the two matrices. For Mantel, you'd need to compute a distance matrix, but you could use the squared matrices for the PROCRUSTES.

Improve this answer
$\endgroup$
3
  • $\begingroup$ I added more context to the post. Does the PROCRUSTES analysis still work in that context? $\endgroup$
    – aeiche01
    Commented Jan 17 at 22:23
  • $\begingroup$ yes, I'd say that vegan::protest is your best shot $\endgroup$
    – Diogo B Provete
    Commented Jan 18 at 23:36
  • $\begingroup$ Thank you very much! $\endgroup$
    – aeiche01
    Commented Jan 19 at 19:54
1
$\begingroup$

This really much depends on what you mean with 'compare'. There is no heaven given way of comparing two matrices. What you can do is to compare the results of analysis of two matrices. Then it is simple: just analyse the two matrices with the same method and see how the analyses differ. Naturally, with n = 1 it is just descriptive case analysis that you must judge with your intellect: negligible or remarkable difference.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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