Please always cc the mailing list.
I took a look at the datasets you sent to me; meta.csv does not contain a
variable called 'field' and the 'species' variable is actually called
'species1', but I could work around that.
Based on your example, I noticed an issue with rma.mv() which caused some
checks on the 'R' argument not being run as intended. That has been fixed
in the 'devel' version; see here:
http://www.metafor-project.org/doku.php/installation#development_version
Please install that and try again. When you try to fit the model, you will
then get the error:
There are levels in 'species1' for which there are no rows/columns in the
corresponding 'R' matrix.
As the error will tell you, h1$species contains some levels that are not
represented in the 'phylo' matrix. I did a bit of checking and this appears
to due to these species:
[1] Zygiella_x-notata_ott504327
[2] Neolamprologus_pulcher_--Daffodil--_ott311695
[3] Dugesia_sp_06_MR-2009_ott859610
Make sure that these are also contained in 'phylo' and that the
row/columnnames are spelled exactly in the same way. rma.mv() doesn't try
to do some kind of fuzzy matching, but expects 1-1 matches between the
levels of the variable specified via the 'random' argument and the
dimension names of the corresponding matrix specified via 'R'.
By the way, 'phylo' appears to be a distance matrix (at least, the
diagonal elements are all 0's, so this strongly suggests that this is what
it is). But 'R' should be used to specify covariance/correlation matrices.
As for your other question: As I mentioned, the second argument is for
specifying the sampling variances, not the standard errors. The sampling
variances are the square of the standard errors. I don't know what you mean
by 'May you provide your calculation of standard error?' Each outcome
measure has a different formula for the standard errors (and hence sampling
variance), so I don't know what you are asking here.
Best,
Wolfgang
-----Original Message-----
From: Rafael Rios [mailto:biorafaelrm at gmail.com]
Sent: Thursday, 28 September, 2017 4:22
To: Viechtbauer Wolfgang (SP)
Subject: Re: Questions about phylogenetic meta-analisys
ATTACHMENT(S) REMOVED: phylo.csv | meta.csv
Dear Wolfgang,
Thank you for the contributions. My data follows on attached. "meta.csv"
is my dataset, and "phylo.csv" is the phylogenetic distance matrix. Both
files were saved as CSV (MS-DOS) format.
I have another doubt. You said that I should use rma.mv(zf, sezf^2, ...).
However, my advisor said that most meta-analists frequently confound
standard error with variance. So, we must use rma.mv(zf, sezf, ...). Is
that true? May you provide your calculation of standard error? Thank you!
Best wishes,
Rafael R. Moura.
scientia amabilis
Doutorando da P?s-gradua??o em Ecologia e Conserva??o de Recursos Naturais
Universidade Federal de Uberl?ndia, Uberl?ndia, MG, Brasil
ORCID: http://orcid.org/0000-0002-7911-4734
Curr?culo Lattes: http://buscatextual.cnpq.br/buscatextual/
visualizacv.do?id=K4244908A8
Research Gate: https://www.researchgate.net/profile/Rafael_Rios_Moura2
2017-09-27 18:40 GMT-03:00 Viechtbauer Wolfgang (SP) <
wolfgang.viechtbauer at maastrichtuniversity.nl>:
Dear Rafael,
First of all, 'sezf' sounds like these are standard errors. The second
argument of rma.mv() (and rma()/rma.uni()) is for the *sampling
variances* (the square of the standard errors). So, you should use: rma.mv(zf,
sezf^2, ...).
This aside, I cannot diagnose what the problem is without a reproducible
example or at least more information. Could you at least post:
unique(h1$species)
class(phylo)
dimnames(phylo)
Best,
Wolfgang
-----Original Message-----
From: Rafael Rios [mailto:biorafaelrm at gmail.com]
Sent: Wednesday, 27 September, 2017 20:49
To: r-sig-meta-analysis at r-project.org; Viechtbauer Wolfgang (SP)
Subject: Questions about phylogenetic meta-analisys
Dear Dr. Wolfgang Viechtbauer and other meta-analysists,
I am trying to conduct a phylogenetic meta-analysis using function rma.mv
with the following format:
meta1=rma.mv(zf, sezf, mods= ~phylum,
random = list (~1|study, ~1|species, ~1|field),
R=list(species=phylo),
data = h1)
Where I have a moderator (phylum) and three random variables (study,
species and if it was an observational or experimental study). We included
phylogenetic correlation matrix (phylo) in argument R. However, I found the
following error message: "Error in R[[j]][s.levels[[j]], s.levels[[j]]] :
subscript out of bounds". I also tryied to use a matrix with phylogenetic
distances, but I found the same error. I hope that this e-mail finds you
well.
Best wishes,
Rafael R. Moura.
scientia amabilis
Doutorando da P?s-gradua??o em Ecologia e Conserva??o de Recursos Naturais
Universidade Federal de Uberl?ndia, Uberl?ndia, MG, Brasil
ORCID: http://orcid.org/0000-0002-7911-4734
Curr?culo Lattes: http://buscatextual.cnpq.br/buscatextual/
visualizacv.do?id=K4244908A8
Research Gate: https://www.researchgate.net/profile/Rafael_Rios_Moura2