Skip to content

zero one inflated beta mixed model

2 messages · jonnations, Morgane Brachet

Original
jonnations
# 
Hi Morgane,

Like Ben said, brms is a good option for zero one inflated beta models.
I have used proportional data in the past and, when learning about mixed
models, was surprised to find out that there is no obvious distribution
family for these data. In my own experience zero one inflated models
required more data than I had to make good parameter estimates. Depending
on the distribution of your data, an ordinal model could be a good choice!
You could set lots of thresholds if you want. There is a great
manuscript&tutorial on ordinal models in brms available here
https://psyarxiv.com/x8swp/  I think there is a formatted published version
out there for free as well.


Message: 2
Date: Wed, 12 Feb 2020 09:56:44 -0500
From: Ben Bolker <bbolker at gmail.com>
To: Morgane Brachet <morgane.brachet at hotmail.com>
Cc: "r-sig-mixed-models at r-project.org"
        <r-sig-mixed-models at r-project.org>
Subject: Re: [R-sig-ME] zero one inflated beta mixed model
Message-ID:
        <CABghstQfprvqDm+=oVOjuiF2UKP_KaC+EKq2YRBAYnup+K_bGw at mail.gmail.com>
Content-Type: text/plain; charset="utf-8"

    At present glmmTMB doesn't do zero-one-inflated betas, only
zero-inflated betas. As far as I know your options are (1) use brms,
(2) squish your 1 values to something slightly less than 1, or (3) do
the hurdle model manually (i.e. fit two separate models, one for the
probability that the response== 1, and another (conditional) model for
the zero-inflated beta distribution applied only to the responses <1).

  Others on the list may have other suggestions ... (e.g. does INLA
does zero-one-inflated betas?)

On Wed, Feb 12, 2020 at 8:42 AM Morgane Brachet
<morgane.brachet at hotmail.com> wrote:
https://github.com/glmmTMB/glmmTMB/issues/355). I am trying to fit
proportion data with lots of 0s and a few 1s into a hurdle model using
glmmTMB. Is this possible? Would you have any example code please?
https://github.com/glmmTMB/glmmTMB/issues/355>
? GitHub<https://github.com/glmmTMB/glmmTMB/issues/355>
have any example code? will it still have the same issue? Actually i
thought using the ziformula option is the way to fit hurdle model, but
since it shows such errors of inappropriate values, i guess i may misse
some options for hurdle model.

  
    

      
      
    

  


  


    

    

      
      

        


  

        

      Morgane Brachet
    

    

      
      #
    

  


  

      
Thank you very much for your help, I will look into it!

Morgane
________________________________
De : jonnations <jonnations at gmail.com>
Envoy? : mercredi 12 f?vrier 2020 21:37
? : morgane.brachet at hotmail.com <morgane.brachet at hotmail.com>
Cc : r-sig-mixed-models at r-project.org <r-sig-mixed-models at r-project.org>
Objet : Re: [R-sig-ME] zero one inflated beta mixed model

Hi Morgane,

Like Ben said, brms is a good option for zero one inflated beta models.
I have used proportional data in the past and, when learning about mixed models, was surprised to find out that there is no obvious distribution family for these data. In my own experience zero one inflated models required more data than I had to make good parameter estimates. Depending on the distribution of your data, an ordinal model could be a good choice! You could set lots of thresholds if you want. There is a great manuscript&tutorial on ordinal models in brms available here https://psyarxiv.com/x8swp/  I think there is a formatted published version out there for free as well.


Message: 2
Date: Wed, 12 Feb 2020 09:56:44 -0500
From: Ben Bolker <bbolker at gmail.com<mailto:bbolker at gmail.com>>
To: Morgane Brachet <morgane.brachet at hotmail.com<mailto:morgane.brachet at hotmail.com>>
Cc: "r-sig-mixed-models at r-project.org<mailto:r-sig-mixed-models at r-project.org>"
        <r-sig-mixed-models at r-project.org<mailto:r-sig-mixed-models at r-project.org>>
Subject: Re: [R-sig-ME] zero one inflated beta mixed model
Message-ID:
        <CABghstQfprvqDm+=oVOjuiF2UKP_KaC+EKq2YRBAYnup+K_bGw at mail.gmail.com<mailto:oVOjuiF2UKP_KaC%2BEKq2YRBAYnup%2BK_bGw at mail.gmail.com>>
Content-Type: text/plain; charset="utf-8"

    At present glmmTMB doesn't do zero-one-inflated betas, only
zero-inflated betas. As far as I know your options are (1) use brms,
(2) squish your 1 values to something slightly less than 1, or (3) do
the hurdle model manually (i.e. fit two separate models, one for the
probability that the response== 1, and another (conditional) model for
the zero-inflated beta distribution applied only to the responses <1).

  Others on the list may have other suggestions ... (e.g. does INLA
does zero-one-inflated betas?)

On Wed, Feb 12, 2020 at 8:42 AM Morgane Brachet

        
<morgane.brachet at hotmail.com<mailto:morgane.brachet at hotmail.com>> wrote:

            

      
      
      
    

        

      
      
    

        
--
Jonathan A. Nations
PhD Candidate
Esselstyn Lab<https://esselstyn.github.io/>
Museum of Natural Sciences<https://www.lsu.edu/mns/>
Louisiana State University
jonnynations.com<https://jonnynations.com/>