I would say that within each taxonomic group, if you are not a specialist,you should ask advice of specialists whether the trait is usually varibale within genus or family...

But obviously, this is a problem to assign a trait when you have no data. I would let missing data but it seems that BTA do not like well missing data (I do not know, but I remarkeed people working on this usually infer that within a genus, it is the same).... An ni:portant thing then, if you assigned unknown traits based on congeneric species traits, do not then analyse phylogenetic diversity and conclude that it is well related to functional biodiversity( that based on traits) (a mistake I recently observed by ,someone who had a good experience on the topic of biodiversity analyses and traits).

The problem gets worse at microbiological Level. DNA ecologists use just sequences to assign traits to OTUs, whithout knowing their ecological role (not even whether sequences correspond to organisms). Their traits are inferred by OTUs position in a phylogenetic tree AT CLASS OR ORDER LEVEL! Can one get any ecological trends based on OTUs? I would love to hear your opinion.

About microbes (although I am not a microbiologist....) I think a precision is necessary:

There are cases where DNA sequences are used to infer functions (ecological role) which may be more justified than others: in metagenomic studies, it is not only 16S genes (barcoding genes) but also many genes with well known biochemical functions that are characterized ... I do not know whether Maria Paz's previous message thought of such studiesor not.

For the ecological studies I am refering to, only 16S genes are analyzed

It might be dangerous to assume traits are always shared within a genus or family. Although in general there is quite some evidence for ecological niche conservatism within clades, it also seems to vary over clades which characteristics are changing and which are not. An example from botany might be the cases of two quite comparable usually rain forest dwellers: Begonia and Impatiens. In one (Begonia) flowers do not vary a lot, but there is a large variation in life forms (epiphytic shrubs, climbers, stemless ground dwellers, near succulents, etc), especially leaf shape and characteristics, which has an influence on their trait. In Impatiens on the other hand their is hardly any difference in vegetative organs (a little on leaf shape and indumentum, but not that much), but there is a very large difference in floral characters, resulting in very different pollinators. So for traits related to pollination, Begonia is more or less all the same accros the genus, while in Impatiens it would be a big mistake to assume one could copy floral traits from one species to the other. For vegetative traits it the other way round, in Impatiens one might get away with it, but not in Begonia.

A general danger lies in island species, where the availability of niches made it easy to jump from one trait to another, classic example being the Darwin finches of the Galapagos. So if your study contains islands: be more cautious.

I think important points have been raised, with good examples.  To try to generalise a bit:

(Note: I use the term 'gap filling', which refers to the replacement of missing trait data in a species x traits matrix with estimated values.)

*Whether to exclude missing trait data or use gap-filled data depends on the question being asked.  Anne gave a good example of this.  Another might be that you should not use gap-filled trait data to investigate conservatism of those traits.  On the other hand, if you are calculating trait means for communities, then excluding missing trait data can introduce more bias than using gap-filled data.

*Gap filling complicates analysis because the independence of different observations is often compromised - i.e. pseudoreplication problems.

*I strongly agree with the responses above that point out (paraphrasing) that it is more reasonable to gap-fill some traits (particularly those not under selective pressure) than others (particularly those under strong selective pressure).

*As Jan pointed out, some clades lend themselves to gap filling more than others.

*Some methods of gap filling work better than others. If gap filling is being undertaken (after careful thought first!) then, as suggested above, expert knowledge is a good method (but be warned that some referees do not appear to like this, preferring inferior but more repeatable methods).

*If gap-filling is done with a large dataset, a variety of algorithms exist to automate the process.  I do not recommend simply filling each missing value with mean from the lowest available clade (genus, family or whatever) because of problems like those already mentioned.  The best algorithm I know of is hierarchical probabilistic matrix factorisation (HPMF) - see Shan et al (2012) Gap Filling in the Plant Kingdom—Trait Prediction Using Hierarchical Probabilistic Matrix Factorization. Proceedings of the 29th International Conference on Machine Learning, Edinburgh, Scotland, UK, 2012. (http://icml.cc/2012/papers/652.pdf).

I hope that helps!

Dear All,

Thanks for the valuable comments and propositions.

Obvious, is an open issue for the scientific community but in the era of budjet cut in the research and especially in the field of the ecology of species, fuzzy approaches and loans from close species/genus/families is the only that we can rely on.

For my group, marine molluscs in the Mediterranean, i have to follow the borrowing method from congeneric species as well as to apply a fuzzy coding as one species can have more that one feeding type, but only one larval type, as an example.

Best Regards to all and hope to come back with ideas on this subject