Hi Gopal,


If your data are really zero-inflated, then my impression has been the best 
thing to do with this situation is to break up your dependent data points to 
test two hypotheses: 1 a logistic model for what determines whether the outcome 
is a zero or >0, 2 a linear or other appropriate model for what determines the 
continuous value of non-zero outcomes. That's been my impression at least that 
with zero inflation there is no longer any single test or distribution that is 
very applicable. There are some approaches that do all this in a single model, 
but when you look under the hood it still is breaking the data into two parts 
with two models

http://stats.idre.ucla.edu/r/dae/zip/



To make this phylogenetic, I think you want to run the appropriate phylogenetic 
code for the logistic outcome with the 0/1 coding of the variable, and then 
pgls for the continuous value of all the >0 outcomes. MCMCglmm could do this, 
but also there is a phylogenetic logistic regression that Ives and Garland 
(2010) produced with Matlab code. I'm not sure if that routine was ever moved 
into R code. Maybe someone else on this list knows? For the pgls, you could use 
MCMCglmm, ape, or caper.



Here's the Ives and Garland ref:
Ives, A.R., & Garland, T. (2010). Phylogenetic logistic regression for binary 
dependent variables. Systematic Biology, 59, 9_26.



This is all assuming the data really are zero-inflated, which I wouldn't 
consider as quite the same as right-skewed. A zero-inflated distribution after 
the arcsine transform would have a big spike of zeros and then with a 
normal-ish looking distribution of positive values.



Best

Luke



Luke J Matthews

Behavioral and Social Scientist

RAND Corporation





Message: 1

Date: Wed, 15 Mar 2017 11:51:21 +0000

From: Manabu Sakamoto 
<manabu.sakam...@gmail.com<mailto:manabu.sakam...@gmail.com>>

To: Gopal Murali <goopaalmur...@gmail.com<mailto:goopaalmur...@gmail.com>>

Cc: R phylo mailing list mailing list 
<r-sig-phylo@r-project.org<mailto:r-sig-phylo@r-project.org>>

Subject: Re: [R-sig-phylo] Zero inflated model accounting for

                phylogeny

Message-ID:

                
<caerhmt3chufva2ek_aidprwktjvk5qelhuzkyrvvxknzjdo...@mail.gmail.com<mailto:caerhmt3chufva2ek_aidprwktjvk5qelhuzkyrvvxknzjdo...@mail.gmail.com>>

Content-Type: text/plain; charset="UTF-8"





Dear Gopal,



Maybe you can look into MCMCglmm?

MCMCglmm can take a lot of error distributions into account.



thanks,

Manabu



On 15 March 2017 at 10:24, Gopal Murali 
<goopaalmur...@gmail.com<mailto:goopaalmur...@gmail.com>> wrote:



> Dear list,

> I am trying to fit a regression model using a dataset containing

> dependent variable as a percentage (arcsine transformed) and a

> categorical independent variable controlled for phylogeny using PGLS.

> Following the post below, I found the normality of the residuals to be

> non-normal as the data is right skewed (zero-inflated). Is there any

> way to overcome this issue?

> http://blog.phytools.org/2013/02/a-comment-on-distribution-

> of-residuals.html

>

> Thanks in advance,

> Gopal Murali

>

>         [[alternative HTML version deleted]]

>

> _______________________________________________

> R-sig-phylo mailing list - 
> R-sig-phylo@r-project.org<mailto:R-sig-phylo@r-project.org>

> https://stat.ethz.ch/mailman/listinfo/r-sig-phylo

> Searchable archive at http://www.mail-archive.com/r-

> sig-ph...@r-project.org/<mailto:sig-ph...@r-project.org/>

>






__________________________________________________________________________

This email message is for the sole use of the intended r...{{dropped:9}}

_______________________________________________
R-sig-phylo mailing list - R-sig-phylo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/

Reply via email to