None of these methods are absolutely better than the others, philosophically, but different researchers, different fields, and more importantly different journals/reviewers will have their idiosyncrasies. If you're working in paleontology or with datasets that are missing a lot of data points, as is the case with a lot of paleontological phylogenetics, bootstrap/jackknife values are not expected to be as high as those with more complete (i.e. genetic) datasets. A bootstrap of 95% is always strong support, but depending on how rigorous you want to be/how much you hate polytomies it can be advisable to retain all clades with >50% support.

None of these methods are absolutely better than the others, philosophically, but different researchers, different fields, and more importantly different journals/reviewers will have their idiosyncrasies. If you're working in paleontology or with datasets that are missing a lot of data points, as is the case with a lot of paleontological phylogenetics, bootstrap/jackknife values are not expected to be as high as those with more complete (i.e. genetic) datasets. A bootstrap of 95% is always strong support, but depending on how rigorous you want to be/how much you hate polytomies it can be advisable to retain all clades with >50% support.

Thanks Isaac, that is also what I think about this issue. it is very different among researches.

Alberto  -

As mentioned, different researcher place different emphasis on values for bootstrap.  Bayesian inference has better statistical foundations and thus strictly speaking  posterior probabilities should be treated as significant (≥0.95) or not (

Hi Ken,

Thanks so much for your answer. I am very interested about the Approximately Unbiased (AU) test. Frankly, I didn´t know about that. Do you know  what software is the best for making  this test? I use to run my analysis with PAUP.

Thanks so much in advance,

Best regards,

Alberto

If you run a parsimony analysis, you can end up with clades that are actually unsupported by the data (meaning that there are no specific characters supporting these relationships). You can then collaps these branches into polytomies, based on different rules (I usually use "rule 3" in TNT, but it seems that also PAUP has it implemented, see Coddington & Scharff 1994). After doing so, all of your branches in the collapsed most parsimonious trees have some support from your data. That said, bootstrap and bremer values can still be quite low, so you might still apply some more testing, especially to placements that seem dubious. If you have some evidence or previous suggestions that another placement of a particular taxon could be more parsimonious, you could run constrained searches and see how much longer the MPTs of the constrained searches are (the idea behind this is similar to the one from the Bremer value), or you could run a Templeton test to actually check if your data is strong enough to rule out different positions of dubious taxa.

Coddington, J., & Scharff, N. (1994). Problems with zero‐length branches. Cladistics, 10(4), 415-423.

Thanks so much Emanuel, That seems very interesting. It reminds me how you obtain the Bremmer value in PAUP.  I will investigate about your suggestions.