Truncated variables in jags

[Allan Clark]

Good morning all

I have a really simple question! How does one specify the following priors in jags?

tau1 ~ dgamma(a, b)

and now for some positive value 'k'

tau2 ~ dgamma(a, b) truncated on the interval tau1/k < tau2 < tau1

I've tried the following: 

method 1

--------------

tau1 ~ dgamma(a, b)

tau2 ~ dgamma(a, b) T(tau1/K2, tau1)

This does not work since T() has to have deterministic values for the lower and upper bound entries of T()

method 2

--------------

Given the above comment. I them transform from tau2 to tau2t where tau2t = tau2/tau1. In this case tau2t ~G(a, b*tau1) truncated on the interval 1/k < tau2t < 1.

tau1~dgamma(a,b)

hi<- b*tau1

tau2t~dgamma(a,hi)T(1/K2, 1)      #I've also tried tau2t~dgamma(a, b*tau1)T(0.02, 1)

tau2<-tau1*tau2t

This does not work either! I get the following error:

"Error in update.jags(object, n.iter, ...) : Error in node y[1]

Failure to calculate log density"

Does anyone have any ideas? 

regards

The complete code is included below.

Its a really simple model:

Basically a linear model where some of the data points have inflated error variance!

#The data

hills<- DAAG::hills

X<-cbind(1, hills$dist, hills$climb)

y<-matrix(hills$time, ncol=1)

#the jags code

writeLines("

model{

  #prior specifications

  b0~dnorm(0, inv.v)

  b1~dnorm(0, inv.v)

  b2~dnorm(0, inv.v)

           

  tau1~dgamma(a,b)

  tau2t~dgamma(a,b*tau1)T(1/K2, 1)

  tau2<-tau1*tau2t

           

  #the likelihood of the data

  for (i in 1:n1){

    mu[i]<- b0 + b1*x1[i] + b2*x2[i] 

    y[i]~ dnorm(mu[i], tau1)

  }

  for (i in (n1+1):n){

    mu[i]<- b0 + b1*x1[i] + b2*x2[i] 

    y[i]~ dnorm(mu[i], tau2)

  }

           

}

", con = "model2.txt")

#---------------------------------------------------------------------

#inputs for the Bayesian method

niter <- 50000

nburn <- 20000

nthin <- 1

nchains <- 1

nkeep<-(niter-nburn)*nchains

# Input data 

index<- c(1, 18, 33)

Data <- list(x1=c(X[-index,2],X[index,2]), x2=c(X[-index,3],X[index,3]),

             y=c(y[-index],y[index]), n1=32, n=35 , a=0.001, b=0.001, inv.v=1/1000, K2=50)

# Initial values

inits <- function()

{

  list("b0"=rnorm(1),"b1"=rnorm(1),"b2"=rnorm(1), "tau1"=10, "tau2t"=0.5)

}

inits1=inits(); #inits2=inits(); inits3=inits()

Inits=list(inits1)#,inits2,inits3)

# Paramaters to be monitored       

parameters <- c("b0", "b1", "b2", "tau1","tau2") 

modfile = "model2.txt"

rm(Runs2)

Runs2 <- jags(data=Data, inits=Inits, parameters.to.save=parameters, 

             model.file=modfile,

             n.iter = niter, n.burnin = nburn, n.thin = nthin, 

             n.chains = nchains, DIC = FALSE, verbose=FALSE, parallel=TRUE) 

Runs2

[Perry de Valpine]

Truncation with model nodes as truncation bounds should work in NIMBLE, which uses (nearly) the same model syntax as JAGS.

Truncation is described in section 5.2.7 of the NIMBLE User Manual, available at http://r-nimble.org

-Perry

--
*** Three hierarchical modeling email lists ***
(1) unmarked: for questions specific to the R package unmarked
(2) SCR: for design and Bayesian or non-bayesian analysis of spatial capture-recapture
(3) HMecology (this list): for everything else, especially material covered in the books by Royle & Dorazio (2008), Kéry & Schaub (2012), Kéry & Royle (2015)
---
You received this message because you are subscribed to the Google Groups "hmecology: Hierarchical Modeling in Ecology" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hmecology+unsubscribe@googlegroups.com.
To post to this group, send email to hmec...@googlegroups.com.
Visit this group at https://groups.google.com/group/hmecology.
To view this discussion on the web visit https://groups.google.com/d/msgid/hmecology/8a77708d-9efb-4e71-a3b9-3d94f651eae3%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Allan Clark]

Hello Perry

Thank you for the suggestion!

Regards

To unsubscribe from this group and stop receiving emails from it, send an email to hmecology+...@googlegroups.com.