Flexible distribution for below LOD

[Jonas]

Hi,

I am currently setting up a joint model with INLABRU, where one of the likelihoods is left-censored due to a limit of detection. I already looked into a lognormal.surv and loglogisticsurv, but I need more distributional flexibility to capture the shape of my data.

Is it possible to use a 3-parameter weibull or a mixture of distributions (lognormalsurv & lognormal) within INLABRU? Or other flexible options to consider?

Kind regards,

Jonas Meijerink

[Finn Lindgren]

Hi,

essentially all inla models can be run via the inlabru interface, so it just depends on if it has been implemented as an inla() model.

I see a lognormalsurv in INLA::inla.doc("lognormalsurv") so looks like it should work.

Finn

--
You received this message because you are subscribed to the Google Groups "R-inla discussion group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to r-inla-discussion...@googlegroups.com.
To view this discussion, visit https://groups.google.com/d/msgid/r-inla-discussion-group/bc6ff3f8-a250-40df-8801-3f741b4ac3f9n%40googlegroups.com.

--

Finn Lindgren
email: finn.l...@gmail.com

[MEIJERINK Jonas]

Hi Finn,

Thank you for the quick answer! I looked into it but couldn't find a 3-parameter Weibull distribution, nor any built-in options for mixture models with continuous distributions, in INLA/inlabru. Is there a creative workaround or specific implementation trick to get this working within the INLA/inlabru framework? Or is this simply not feasible given the latent Gaussian structure it relies on?

Kind regards,

Jonas

---

Van: r-inla-disc...@googlegroups.com <r-inla-disc...@googlegroups.com> namens Finn Lindgren <finn.l...@gmail.com>
Verzonden: woensdag 22 juli 2026 18:40
Aan: MEIJERINK Jonas <jonas.m...@uhasselt.be>
CC: R-inla discussion group <r-inla-disc...@googlegroups.com>
Onderwerp: Re: [r-inla] Flexible distribution for below LOD

 

U ontvangt niet vaak e-mail van finn.l...@gmail.com. Ontdek waarom dit belangrijk is

To view this discussion, visit https://groups.google.com/d/msgid/r-inla-discussion-group/CAHC4gzbdGQfXMZwiJzer3ojR%2B5050fBYpcJw-WV5ddyYgSYQ9A%40mail.gmail.com.

[Finn Lindgren]

Hi,

I missed the “mixture” mention, sorry! Mixture distributions aren’t generally supported, but I think one can implement specific ones in principle, e.g. via cloglink that allows you to code your own observation likelihood. Not sure what a sensible parameterisation of such a model would be though, as it can only involve one linear predictor value. (There’s ongoing research into implementing multi predictor observation models, but nothing that’s useable yet, and it’s not clear if it will help for mixtures.) One option is to use the “poisson trick” to convert a general density model into a point process likelihood, but that can be expensive, and I’m not sure if/how to combine that with the survival/censoring models.

Finn

On 24 Jul 2026, at 08:05, MEIJERINK Jonas <jonas.m...@uhasselt.be> wrote:



[Helpdesk (Haavard Rue)]

...we should be able to do mixtures with known `number of components', like with
a mix of two know likelihoods. this will then be 'hard-coded' into one new
likelihood.

@Jonas, if you're able to do C coding, you can check 'inla.doc("cloglike")'.
otherwise, we can discuss it again when I'm back at KAUST 4th Aug

On Fri, 2026-07-24 at 08:27 +0100, Finn Lindgren wrote:
> Hi,
> I missed the “mixture” mention, sorry! Mixture distributions aren’t generally
> supported, but I think one can implement specific ones in principle, e.g. via
> cloglink that allows you to code your own observation likelihood. Not sure
> what a sensible parameterisation of such a model would be though, as it can
> only involve one linear predictor value. (There’s ongoing research into
> implementing multi predictor observation models, but nothing that’s useable
> yet, and it’s not clear if it will help for mixtures.) One option is to use
> the “poisson trick” to convert a general density model into a point process
> likelihood, but that can be expensive, and I’m not sure if/how to combine that
> with the survival/censoring models.
> Finn
>
> > On 24 Jul 2026, at 08:05, MEIJERINK Jonas <jonas.m...@uhasselt.be>
> > wrote:
> >
> > 
> > Hi Finn,
> >
> >
> > Thank you for the quick answer! I looked into it but couldn't find a 3-
> > parameter Weibull distribution, nor any built-in options for mixture models
> > with continuous distributions, in INLA/inlabru. Is there a creative
> > workaround or specific implementation trick to get this working within the
> > INLA/inlabru framework? Or is this simply not feasible given the latent
> > Gaussian structure it relies on?
> >
> >
> > Kind regards,
> > Jonas
> >
> >
> >
> >

> > > email tor-inla-discussio...@googlegroups.com.

> > > To view this discussion, visit
> > > https://groups.google.com/d/msgid/r-inla-discussion-group/bc6ff3f8-a250-40df-8801-3f741b4ac3f9n%40googlegroups.com
> > > .
> >
> >
> > --
> > Finn Lindgren
> > email: finn.l...@gmail.com

--

Håvard Rue
he...@r-inla.org