Dear Antonio,

I agree with Thomas. I see no options to try and do what you seem to propose. Species descriptions are simply not drawn up in any standardized way. Authors may use variation from within the area their publication deals with, but on the other hand may also include material from outside that area, when the first is scarce. Major revisions will try to include all global museum/herbarium collections available.

Thus, only new observations, acquired in a standardized way, will probably give you the answers you are looking for, but then, that sort of work has been performed exhaustively in the 1980's and 1990's, and so the methods are fairly old. Nowadays, I would suggest applying DNA barcoding or other molecular methods that have become available (microsatellites and the like).

with best regards,

Marc

Antonio,

With certain forms of molecular data (e.g., AFLPs), people often use ordination techniques (especially non-metric multi-dimensional scaling) to analyze their data and look - qualitatively - for clustering of multiple individuals into species clusters.

I'm not sure, however, what your question really is.  If you are seeking to tell whether published species descriptions are adequate for separating species, there really is no way to use the species descriptions themselves as a statistical test, in my opinion at least, because the description will have collapsed the covariance structure among the variables and thus prevent you from seeing if there is non-overlap among species in multi-dimensional space even if there is overlap among those species (let's assume) in each individual character.  You'd need a  population of samples from each species, and then apply such approaches as univariate and multivariate ANOVAs and discriminant function analyses (e.g., see Sánchez-Ruiz & Sanmartín 2000, Eur J Entomology).

Hope this helps!

Cheers, Tom

Dear Antonio,

I agree with Thomas. I see no options to try and do what you seem to propose. Species descriptions are simply not drawn up in any standardized way. Authors may use variation from within the area their publication deals with, but on the other hand may also include material from outside that area, when the first is scarce. Major revisions will try to include all global museum/herbarium collections available.

Thus, only new observations, acquired in a standardized way, will probably give you the answers you are looking for, but then, that sort of work has been performed exhaustively in the 1980's and 1990's, and so the methods are fairly old. Nowadays, I would suggest applying DNA barcoding or other molecular methods that have become available (microsatellites and the like).

with best regards,

Marc

I agree with Thomas Givnish opinion, that when dealing with description data represented as minimum-maximum range, or as means of each variable, it seems impossible to include such a data to statistical multivariate analysis. We simply con not assume what is the distribution type and variability of each variable. The distribution frequently can be not Gaussian, and without data from direct measuring it will be impossible to deal with such variables.

The only possibility, for my opinion, is when you have some variables with not overlapping minimum-maximum ranges between species, that will allow some comparison at least in graphical form. The separation of species is potentially possible when you have several variables and each species have at least one variable (more than one - better) that have not overlapping range with the rest of data. But this analysis will be in any way not directly comparable with multivariable ordination techniques.

In general, it is not possible to make comparison tests if only min-max ranges are provided, additionally a dispersion statistic is commonly needed. Moreover, if the distribution of the variable is not normal (as is the case in many morphometric parameters), even average values are useless.

If all you want is a graphical display of differences amongst species (i.e. not statistical testing) then you could try various  multivariate methods.  A problem you have not mentioned is how to deal with qualitative characters (usually very important for distinguishing species) such as colour, dispersal, fruit type etc.

Hill & Smith (1976) have the solution to that particular problem.  See attached paper

I suggest that you examine and measure herbarium specimens from across their geographic range for each of the species that you are interested in and then use the data to discriminate between the species using multivariate approaches. 

Dear José Antonio,

Strona and coworkers have recently proposed a multivariate method for parasite identification on the basis linear measures (ranges). This method has been applied to parasite worms, but can be conceptually used for any taxon. So, I think you can try it for plants. Attached is Strona et al.'s paper.

All the best,

Simone

Tom,

Thank you for your answer.

A more precise question is How to use species descriptions including a combination of qualitative and quantitative (ranges for each character) data to assess morphological similarity among species (numerically or graphically) without statistical testing?

In describing a new species we often confront the need to determine more objectively which of the other species is morphologically the most similar. A graphical display of the clusters of individuals would be valuable, however, for now I only have a combination of qualitative data and ranges of characters for each Magnolia species, many of them often from one or two populations (same locality) and very few individuals, each in different stage (sterile, flower bud, open flower, with styles at anthesis or afterward, immature fruit, dehisced fruit, with seeds).

Is there an acceptable way to transform range data to binary data? For instance breaking into lower (0) and higher values (1) where clear disjunctions (non overlapping ranges) exist.

Alternatively, Is it acceptable, as a measure of dissimilarity, to use a semi-matrix (species x species) with the numbers of none overlapping characters between species?

Cordially,

Antonio

Viacheslav,

Thank you for your answer.

Certainly,  each of the Magnolia species has from 6 to 12 non-overalving ranges with the rest of the species. What sort of graphical display could be obtained?

Cordially,

Antonio

M D Swaine,

Thank you for your answer,

The reference will be very valuable to me.

Cordially,

Antonio

Simone Fattorini,

Thank you for your answer,

The reference will be very valuable to me.

Cordially,

Antonio

Jose: Check these publications where we used multivariate statistics to separate groups of individuals. Scott

Gardner, S.L.; Luedders, B. (2014) “Hymenolepis robertrauschi n. sp. from grasshopper mice Onychomys spp. in New Mexico and Nebraska, U.S.A.” Occasional Papers, Museum of Texas Tech University.

 Gardner, S.L.; Schmidt, G.D. 1988. “Cestodes of the genus Hymenolepis Weinland, 1858 sensu stricto from pocket gophers Geomys and Thomomys spp. [Rodentia: Geomyidae] in Colorado and Oregon, with a discriminant analysis of four species of Hymenolepis.” Canadian Journal of Zoology. 66: 896-903. 

Thank you everyone for your imput

Please send a cpoy in pdf ob this paper:

how can I use multivariate techniques (ordination and classification) on data from species descriptions?

Thanks

Dear Yoandri: You can use these techniques only if you have access to the measurements of the individuals of the specimens used in the species descriptions. If you cannot access the actual measurements, you will have to take them yourself from the specimens from deposited museum specimens. Scott

Antonio:

Thinking on the graphical representation of character variability ranges:

I believe that the bivariate plot for characters with error bars could be a good possibility. The character variability for each specie with source observation data available could be represented as points with error bars in two dimensions (representing confidence interval, minimum-maximum range excluding outliers or any other variability measure).

Look Fig. 4 in Bodbyl Roels & Kelly 2011 for the example of such a plot (http://onlinelibrary.wiley.com/doi/10.1111/j.1558-5646.2011.01326.x/pdf); the same figure is available here:  http://beacon-center.org/wp-content/uploads/2013/12/Bodbyl_Figure_2.jpg )

In such a bivariate plot you can represent boxes of minimum-maximum ranges for species known only from description. If the box of given species do not overlaps with the observed variability for another species, that will be a good criteria to consider it morphologically distinctive.

Regarding the issue of combining quantitative and qualitative (binary or other) charaters there is the possibility to construct dendrograms using e.g. Gower's distance. This standardizes characters by their range and will let you combine the two kinds. Is there are problems with multistate categorical features these could be recoded as series of additive binary characters. For a first approach you could used the package 'PAST' which is free and easy to use.

Cheers

If you have species descriptions, then the species have already been delimited. Why not use the morphological data in the descriptions to produce a phylogenetic tree

Dear Antonio,

This paper can help you:

Carrijo et al., 2013 - Pollen morphology of some related genera of Vernonieae (Asteraceae) and its taxonomic significance - Plant Systematics and Evolution

Karlo

http://link.springer.com/article/10.1007%2Fs00606-013-0795-9