thomasdonegan
Former amateur ornithologist
Statistics
There was an interesting reference above to a recent paper on Tapaculos by Jorge Avendano and I by RoyN above:
PDF on the journal's website:
http://www.ornitologiacolombiana.org/oc6/doneganyavendano.pdf
I've also looked into variation in Grallaricula antpittas and some other groups, forthcoming:
Unauthorised online PDF of Grallaricula paper:
http://www.scricciolo.com/Nuovo_Neornithes/Grallaricula nana Bull BOC Donegan.pdf
This series of papers examine geographical variation in vocal, biometric and plumage characters in various Andean birds, assessing diagnosability of populations against the following tests:
1. Statistically significant differences (i.e. t-test, or other similar tests which compare averages for non-normally distributed)
2. 50% diagnosability (an old subspecies test: at least 50% of a population can be reliably identified from the other.
3. 75% diagnosability (the traditional subspecies test: at least 75% of a population can be reliably identified from the other).
4. No observed overlap in variables.
5. 97.5% (=close to 100% as anyone would sanely want to get) diagnosability. Phylogenetic species or modern subspecies concept.
The full details of calculations are set out in the paper.
First, it's notable how many times two populations will pass the "statistical significance" test but not even 50% of individuals are diagnosable (see data in the appendices of these two papers). Generally, biologists have been reluctant to describe populations even as subspecies when they show statistically significant but not diagnosable differences. Sure, they are different and this is demonstrated through the wonder of statistics (using the ecologist's love-child, the t-test), but that does not mean that you have to put a name on these populations. Under most species and subspecies concepts, it would be bad practice to do so based just on t-tests.
Secondly, I'm all in favour of birds which are diagnosable only by voice being described as subspecies or species, in groups where voice is innate (e.g. suboscines). See Grallaricula nana hallsi and G. n. nanitaea, in the paper mentioned above, which are mutually fully diagnosable by acoustic frequency of their songs but only show only non-diagnosable-but-significant-and-much-more-so-than-crossbills-differences in biometrics and plumage. (Also, another subspecies G n nana sits in the range in between them.) Where vocal characters studied are likely to be innate due to physiology (e.g. song speed, acoustic frequency range) this should be good enough for oscines too.
That takes me back to the first posting in this chain - and the need for young-rearing experiments to see if any of these differences in crossbill song may be constrained physically, eg. the ability of a bird to trill at certain speeds or reach certain acoustic frequencies. Crossbills, as I see it, reach level 4/5 in song, but that is assumed a learned character until more data is provided. In innate variables (e.g. biometrics) they reach only level 1 at best, and the differences in observed measurements are highly unimpressive.
If you compare the situation in crossbills to our 2008 tapaculos paper (and ignore characters not known/widely accepted to be innate), it would be effectively the equivalent of us having described all of the following as different species in Scytalopus spillmanni:
- west slope Ecuador population
- east slope Ecuador population
- West Andes Colombia population
- Central Andes Colombia population
- East Andes Colombia population.
We also described, but did not name, 4 other populations which are certainly subspecies under all concepts and species under many concepts (Perija population of "griseicollis", Tama population of "griseicollis", a Merida population of Scytalopus cf "spillmanni", Yariguies population of "rodriguezi") in that paper.
The S spillmanni populations all show "level 1" or more differences in voice or biometrics. Various of these are allopatric populations, with statistically significant vocal or biometric differences. But they are not species under a vast majority of concepts, and most of them are not even subspecies under several definitions. As a result, we did not describe any of these populations as even subspecies. (The East Andes Colombia population has much darker plumage, shorter biometrics, lower mass and notable mtDNA differences from other populations but there was a collecting gap in the specimen record which caused us to hesitate there.) This paper was published in a poorly read local journal (Ornitologia Colombiana).
We should have described 6 species in S spillmanni and got the paper published in Condor instead ... well, actually that would not have been a good idea at all.
I often get frustrated that people get so het up about others suggesting any kind of change to taxonomic treatments (e.g. AOU SACC's extreme approach to maintaining Peters and other status quo treatments). I would therefore apologise to the authors of the crossbills paper for getting on their back in this exchange. They have published a really great paper documenting a very interesting instance of geographical variation. However, it is unclear whether the time is right to be describing species-level taxa. (I could buy the new crossbill being a subspecies based on the information they provided although it would not have reached our "level 5" test for innate variables, so I would not have done so myself.) Describing geographical variation is an important exercise. It is best often to leave things at that and exercise restraint at the next step - naming - given that everyone has a different idea of what a species is. That is especially the case in instances like this where followers of many species concepts would not agree with the approach being taken.
There was an interesting reference above to a recent paper on Tapaculos by Jorge Avendano and I by RoyN above:
PDF on the journal's website:
http://www.ornitologiacolombiana.org/oc6/doneganyavendano.pdf
I've also looked into variation in Grallaricula antpittas and some other groups, forthcoming:
Unauthorised online PDF of Grallaricula paper:
http://www.scricciolo.com/Nuovo_Neornithes/Grallaricula nana Bull BOC Donegan.pdf
This series of papers examine geographical variation in vocal, biometric and plumage characters in various Andean birds, assessing diagnosability of populations against the following tests:
1. Statistically significant differences (i.e. t-test, or other similar tests which compare averages for non-normally distributed)
2. 50% diagnosability (an old subspecies test: at least 50% of a population can be reliably identified from the other.
3. 75% diagnosability (the traditional subspecies test: at least 75% of a population can be reliably identified from the other).
4. No observed overlap in variables.
5. 97.5% (=close to 100% as anyone would sanely want to get) diagnosability. Phylogenetic species or modern subspecies concept.
The full details of calculations are set out in the paper.
First, it's notable how many times two populations will pass the "statistical significance" test but not even 50% of individuals are diagnosable (see data in the appendices of these two papers). Generally, biologists have been reluctant to describe populations even as subspecies when they show statistically significant but not diagnosable differences. Sure, they are different and this is demonstrated through the wonder of statistics (using the ecologist's love-child, the t-test), but that does not mean that you have to put a name on these populations. Under most species and subspecies concepts, it would be bad practice to do so based just on t-tests.
Secondly, I'm all in favour of birds which are diagnosable only by voice being described as subspecies or species, in groups where voice is innate (e.g. suboscines). See Grallaricula nana hallsi and G. n. nanitaea, in the paper mentioned above, which are mutually fully diagnosable by acoustic frequency of their songs but only show only non-diagnosable-but-significant-and-much-more-so-than-crossbills-differences in biometrics and plumage. (Also, another subspecies G n nana sits in the range in between them.) Where vocal characters studied are likely to be innate due to physiology (e.g. song speed, acoustic frequency range) this should be good enough for oscines too.
That takes me back to the first posting in this chain - and the need for young-rearing experiments to see if any of these differences in crossbill song may be constrained physically, eg. the ability of a bird to trill at certain speeds or reach certain acoustic frequencies. Crossbills, as I see it, reach level 4/5 in song, but that is assumed a learned character until more data is provided. In innate variables (e.g. biometrics) they reach only level 1 at best, and the differences in observed measurements are highly unimpressive.
If you compare the situation in crossbills to our 2008 tapaculos paper (and ignore characters not known/widely accepted to be innate), it would be effectively the equivalent of us having described all of the following as different species in Scytalopus spillmanni:
- west slope Ecuador population
- east slope Ecuador population
- West Andes Colombia population
- Central Andes Colombia population
- East Andes Colombia population.
We also described, but did not name, 4 other populations which are certainly subspecies under all concepts and species under many concepts (Perija population of "griseicollis", Tama population of "griseicollis", a Merida population of Scytalopus cf "spillmanni", Yariguies population of "rodriguezi") in that paper.
The S spillmanni populations all show "level 1" or more differences in voice or biometrics. Various of these are allopatric populations, with statistically significant vocal or biometric differences. But they are not species under a vast majority of concepts, and most of them are not even subspecies under several definitions. As a result, we did not describe any of these populations as even subspecies. (The East Andes Colombia population has much darker plumage, shorter biometrics, lower mass and notable mtDNA differences from other populations but there was a collecting gap in the specimen record which caused us to hesitate there.) This paper was published in a poorly read local journal (Ornitologia Colombiana).
We should have described 6 species in S spillmanni and got the paper published in Condor instead ... well, actually that would not have been a good idea at all.
I often get frustrated that people get so het up about others suggesting any kind of change to taxonomic treatments (e.g. AOU SACC's extreme approach to maintaining Peters and other status quo treatments). I would therefore apologise to the authors of the crossbills paper for getting on their back in this exchange. They have published a really great paper documenting a very interesting instance of geographical variation. However, it is unclear whether the time is right to be describing species-level taxa. (I could buy the new crossbill being a subspecies based on the information they provided although it would not have reached our "level 5" test for innate variables, so I would not have done so myself.) Describing geographical variation is an important exercise. It is best often to leave things at that and exercise restraint at the next step - naming - given that everyone has a different idea of what a species is. That is especially the case in instances like this where followers of many species concepts would not agree with the approach being taken.