mixed lognormal hurdle model with multiple grouping factors
For a hurdle model for repeated measurements data, the dichotomous outcome I(diversity > 0) is also a repeated measurements outcome. Hence, in the logistic regression model for this dichotomous outcome you will need to include random effects to account for the correlations. And it is logical to assume that the random effects from this logistic regression model will be correlated with the random effects of the linear mixed model for only the positive responses. In this case the likelihood of the two parts does not split in two functionally independent parts that can be separately maximized. If this is indeed the case, then fitting the two parts separately may cause bias and loss of efficiency. Best, Dimitris ?? Dimitris Rizopoulos Professor of Biostatistics Erasmus University Medical Center The Netherlands
From: R-sig-mixed-models <r-sig-mixed-models-bounces at r-project.org> on behalf of Mollie Brooks <mollieebrooks at gmail.com>
Sent: Wednesday, December 18, 2019 3:08 PM
To: Help Mixed Models; Guillaume Adeux
Subject: Re: [R-sig-ME] mixed lognormal hurdle model with multiple grouping factors
Sent: Wednesday, December 18, 2019 3:08 PM
To: Help Mixed Models; Guillaume Adeux
Subject: Re: [R-sig-ME] mixed lognormal hurdle model with multiple grouping factors
Hi Guillaume, I don?t think the hurdle lognormal can be fit in a single function call to glmmTMB since the model for the non-zero response requires log-transforming the response. Other types of hurdle models could be fit in glmmTMB using the zero-inflation model. I don?t think you gain much information in hurdle models by modeling the two parts (zeros and non-zeros) in one function call. The only potential benefit to fitting a hurdle in a single function call is that you get likelihood and AIC for the entire data set, but I don?t know if those are produced by brms. You could just fit a binomial model for the zero-non-zero process (i.e. monoculture) like mod_binom=lmer((diversity>0)~block+syst+(1|plot)+(1|year)+(1|plot:year), data=density, family=binomial) and then fit a model to the log of the positive data mod_gaus=lmer(log(diversity)~block+syst+(1|plot)+(1|year)+(1|plot:year), data=subset(density, diversity>0)) Or, given that the outcome is non-negative and continuous, it might make sense to try a Tweedie distribution, but I?m not sure I?ve seen this applied to diversity indices in the literature. Has anyone else seen this done? mod_twe = glmmTMB(diversity~block+syst+(1|plot)+(1|year)+(1|plot:year), data=density, family=tweedie) Cheers, Mollie > On 18Dec 2019, at 14:45, Cesko Voeten <c.c.voeten at hum.leidenuniv.nl> wrote: > > Hi Guillaume, > > If you're not afraid to go Bayesian, brms can do it. Alternatively, you may be able to use glmmTMB and treat the hurdle part as zero inflation, but this is conceptually not the same thing as a hurdle model so you would need to judge whether that would make sense at all for your application. > > HTH, > Cesko > > Op 18-12-2019 om 13:48 schreef Guillaume Adeux: >> Hi everyone, >> I am looking for a package which can handle "hurdle.lognormal" distribution >> family and multiple grouping factors. >> GLMMadaptive seemed as the way to go but unfortunately, to the best of my >> knowledge, it does not handle multiple grouping factors (random effects). >> You may ask why? I am analyzing plant diversity and one of the treatments >> led to plots which were dominated by one species. Hence, certain diversity >> indices are estimated as zero in these plots, and produces a mass at zero. >> All other values are positive and continuous. >> Anyone have an idea of a package/function which can handle this? Or any >> alternative approach? >> In lmer syntax, the model is the following: >> mod=lmer(diversity~block+syst+(1|plot)+(1|year)+(1|plot:year),data=density,REML=F) >> Thank you for your time and help. >> Sincerely, >> Guillaume ADEUX >> [[alternative HTML version deleted]] >> _______________________________________________ >> R-sig-mixed-models at r-project.org mailing list >> https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fstat.ethz.ch%2Fmailman%2Flistinfo%2Fr-sig-mixed-models&data=02%7C01%7Cd.rizopoulos%40erasmusmc.nl%7C0f8f22d15f244f0190f208d783c3ca4b%7C526638ba6af34b0fa532a1a511f4ac80%7C0%7C0%7C637122749245106064&sdata=XWAMTSylM6j81Lf%2BLb8qikDEho3QIo8l%2Bh%2BnxSpwjHM%3D&reserved=0 >> > > _______________________________________________ > R-sig-mixed-models at r-project.org mailing list > https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fstat.ethz.ch%2Fmailman%2Flistinfo%2Fr-sig-mixed-models&data=02%7C01%7Cd.rizopoulos%40erasmusmc.nl%7C0f8f22d15f244f0190f208d783c3ca4b%7C526638ba6af34b0fa532a1a511f4ac80%7C0%7C0%7C637122749245106064&sdata=XWAMTSylM6j81Lf%2BLb8qikDEho3QIo8l%2Bh%2BnxSpwjHM%3D&reserved=0 _______________________________________________ R-sig-mixed-models at r-project.org mailing list https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fstat.ethz.ch%2Fmailman%2Flistinfo%2Fr-sig-mixed-models&data=02%7C01%7Cd.rizopoulos%40erasmusmc.nl%7C0f8f22d15f244f0190f208d783c3ca4b%7C526638ba6af34b0fa532a1a511f4ac80%7C0%7C0%7C637122749245106064&sdata=XWAMTSylM6j81Lf%2BLb8qikDEho3QIo8l%2Bh%2BnxSpwjHM%3D&reserved=0