[R-sig-ME] Bradley Terry GLMM in R ?

[Shira Mitchell]

Thanks so much, Thierry ! This is great.

This works except that I cannot subtract because:
f(home, model = "iid")) - f(away, copy = "home")

just drops the second term. Apologies that I'm not super familiar with INLA
syntax yet.



On Fri, Oct 7, 2022 at 10:19 AM Thierry Onkelinx <thierry.onkelinx using inbo.be>
wrote:

> Hi Shira,
>
> I fit such models with the INLA package (https://www.r-inla.org/). The
> trick is to define two random effects but force their parameter estimates
> to be identical.
>
> The code would contain something like f(home, model = "iid")) + f(away,
> copy = "home"). Meaning home ~ N(0, sigma_beta_i) and home[i] = away[i]
>
> Best regards,
>
> ir. Thierry Onkelinx
> Statisticus / Statistician
>
> Vlaamse Overheid / Government of Flanders
> INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE AND
> FOREST
> Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
> thierry.onkelinx using inbo.be
> Havenlaan 88 bus 73, 1000 Brussel
> www.inbo.be
>
>
> ///////////////////////////////////////////////////////////////////////////////////////////
> To call in the statistician after the experiment is done may be no more
> than asking him to perform a post-mortem examination: he may be able to say
> what the experiment died of. ~ Sir Ronald Aylmer Fisher
> The plural of anecdote is not data. ~ Roger Brinner
> The combination of some data and an aching desire for an answer does not
> ensure that a reasonable answer can be extracted from a given body of data.
> ~ John Tukey
>
> ///////////////////////////////////////////////////////////////////////////////////////////
>
> <https://www.inbo.be>
>
>
> Op vr 7 okt. 2022 om 15:00 schreef Shira Mitchell <shiraqotj using gmail.com>:
>
>> We want to fit Bradley-Terry-style GLMM models in R. We looked into:
>>
>>
>> https://cran.r-project.org/web/packages/BradleyTerry2/vignettes/BradleyTerry.pdf
>> and
>> http://bbolker.github.io/mixedmodels-misc/glmmFAQ.html
>>
>> We have voter-specific variables x that influence which political message
>> (i vs j) wins for them:
>>
>> logit[pr(i beats j | person with covariate x)] = lambda_i - lambda_j +
>> (beta_i - beta_j) x
>>
>> We then model parameters as random effects:
>> lambda_i ~ N(0, sigma_lambda)
>> beta_i ~ N(0, sigma_beta)
>>
>> Is there a way to do this in R ? We do this in TensorFlow in Python by
>> directly specifying design matrices with the 0,-1,1 or 0,-x,x entries.
>> However, I do not see how to do this in R using lme4, BradleyTerry2, mgcv,
>> etc.
>>
>> Thanks so much,
>> Shira
>>
>>         [[alternative HTML version deleted]]
>>
>> _______________________________________________
>> R-sig-mixed-models using r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
>

	[[alternative HTML version deleted]]