Request: log link function for beta family

[mpirikL]

Can I request that the log link function be added to the beta family. I realize that this may seem like a strange request, the log link function cannot ensure that the mean is between 0 and 1. However, I have used it with rstanarm's betareg function, INLA's fitting function. And, there is some theory to back up my use of the log link function in my setting, at least in my econometric setting.

[paul-buerkner]

Yeah, I think we can do that.

[paul-buerkner]

Done.

[jenmunoz]

Hi, I am wondering if this could also by implemented for the link_phi. The reason is that when I fit a Beta model, phi= "identity" even when I specify it to be "log" in the family. e.g ( family=brmsfamily("Beta", link="logit", link_phi = "log"), phi seems to be on the identity scale and is very large! and does not actually use the log link for the phi parameter. -----------
This is the output : Family: beta
Links: mu = logit; phi = identity
Formula: mean_attentiveness_nest ~ scale(degree) + scale(elevation) + scale(mass_female_or_general) + (1 | species) + (1 | gr(species_jetz, cov = phy_cov))

[lunafazio]

Hi @jenmunoz, it doesn't seem like you're specifying any regression on the phi parameter, so the default behavior here is to always use the identity link. If you nonetheless want to look at the parameter in the log scale, you can specify an intercept-only regression and here it will use the log link by default.

Example:

library(brms)
fit1 <- brm(bf(qsec ~ cyl), family = brmsfamily("Beta"), data = mtcars/25)
fit2 <- brm(bf(qsec ~ cyl, phi ~ 1), family = brmsfamily("Beta"), data = mtcars/25)

[jenmunoz]

Thanks so much, that was very helpful.

…

On Sat, Oct 7, 2023 at 1:18 AM Luna Fazio ***@***.***> wrote: Hi @jenmunoz <https://github.com/jenmunoz>, it doesn't seem like you're specifying any regression on the phi parameter, so the default behavior here is to always use the identity link. If you nonetheless want to look at the parameter in the log scale, you can specify an intercept-only regression and here it will use the log link by default. Example: library(brms) fit1 <- brm(bf(qsec ~ cyl), family = brmsfamily("Beta"), data = mtcars/25) fit2 <- brm(bf(qsec ~ cyl, phi ~ 1), family = brmsfamily("Beta"), data = mtcars/25) — Reply to this email directly, view it on GitHub <#1316 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEB7CLYYYIQQ3Q5MCUKWJFLX6EF33AVCNFSM5QSX4DF2U5DIOJSWCZC7NNSXTN2JONZXKZKDN5WW2ZLOOQ5TCNZVGE3DINZXGUZQ> . You are receiving this because you were mentioned.Message ID: ***@***.***>

-- *JENNY M MUÑOZ- ZAPATA* Graduate student University of British Columbia Biodiversity Research Centre Department of Zoology 4270 University Boulevard Vancouver, BC Canada V6T 1Z4