I have a dataset with about 80 different species. As usual, some species are very easy to identify with certainty whereas others are more difficult, which means that I am less certain of my identifications for some species than others. I would like to take this uncertainty into account when doing multivariate analyses so that patterns are less affected by potentially erroneous identifications.
I know that it is possible to downweight species with highly variable counts in replicates in PRIMER. But I was wondering if there is a way to do this manually, i.e., specify downweighting factors for each species, based on my level of confidence around the identification?
An alternative would be to lump species in tricky genera together, but then it seems that too much information is lost.
For the species with less certainty around identification, how many other species they could be classified as?
One option would be by splitting the data for those species equally (or proportionally) around these other potential species, so decreasing their weight during the actual analysis. So for example, if one of the species with uncertainty related to identification could be classified as 3 other species, its data value could be multiplied by 0.25, then the other 3 species would be added to the dataset with the same composition values (assuming the uncertainty is proportional among species).
You will end up with a larger dataset and potentially with some species not actually sampled, but their contribution would be small to the final results, while you would have included uncertainty around those taxa.
Does it make sense?
Leonardo Durante Yes I suppose these uncertain species will always come in groups of at least two, and should therefore be treated similarly. Mostly the species are not named species but morphospecies (Genus sp. A, B...). I make drawings of each and keep them to enable me to recognise them later. So, all species in the dataset will be present in the samples. One example of a problem is that a female specimen could potentially belong to a number of morphospecies, which differ only in the structure of male reproductive structures.
Daniel Leduc I see. In this case maybe would be worth it to run different statistical models (including and excluding these females) and compare the output between them. You could try that and varying the weight you give to those species, as mentioned in my first comment. If the results are significantly different between models then you might have a methodological problem, but you can always report the whole set of models, which might present similar main trends anyway. Do you agree with that?
Leonardo Durante Yes, will need to run different analyses to see what the effects are on the results, and hope for the best. But if results vary, I can do what you suggest or be more blunt and lump several (or all) species of a genus together. Thanks
- Badges
- Science method