EXAMPLE 1

model { z[1] <- 1; z[1] ~ dbern(p[1]); p[1] <- 1/th[1]; 
             th[1] ~ dexp(1) I(x[1],); th1 <- th[1]-x[1]
             x[2] ~ dexp(th[2]);  th[2] ~ dexp(1)
  # Likelihood terms for models   
 LL[1] <-  log(1/th[1]);  LL[2]  <-  log(th[2])-th[2]*x[2]
# prior ordinates for models    
 pr.ord[1]  <- -th[1];  pr.ord[2]<- -th[2]
# pseudo-prior ordinates for models
g.ord[1] <- (r[1]-1)*log(th[1]-x[1])-s[1]*(th[1]-x[1])+r[1]*log(s[1])-loggam(r[1])
g.ord[2] <- (r[2]-1)*log(th[2])-s[2]*th[2]+r[2]*log(s[2])-loggam(r[2]) # Combining likelihoods, priors and importance samples    
   L[1]<-LL[1]+ pr.ord[1]-g.ord[1]+log(0.5); 
   L[2]<-LL[2]+ pr.ord[2]-g.ord[2]+log(0.5)
# Scaling against largest likelihood
 maxL<-ranked(L[],2);  SL[1]<-  L[1]-maxL;    SL[2]<-  L[2]-maxL
# exponentiating and normalizing    
    expSL[1] <- exp(SL[1]); expSL[2]<- exp(SL[2]);    
# model weights    
   w[1]<-  expSL[1]/sum(expSL[]); w[2]<-expSL[2]/sum(expSL[])
   B12 <- w[1]/w[2]}

Data (x=0.2)
# r[1] and s[1] based on sampled th1
list(x=c(0.2,0.2),r=c(0.631,2),s=c(1.356,1.2))

Data for x=0.9 
list(x=c(0.9,0.9),r=c(0.83,2),s=c(1.27,1.9))

Inits
list(th=c(2,2))

EXAMPLE 2

   model {x[1]~dbin(p[1],10); p[1]~dbeta(1,1) 
               x[2]~dbin(p[2],10); p[2]~dbeta(30,30)  
  # Likelihood terms for models   
 LL[1]<-logfact(10)-logfact(10-x[1])
             -logfact(x[1])+x[1]*log(p[1])+(10-x[1])*log(1-p[1])    
LL[2]<-logfact(10)-logfact(10-x[2])
                -logfact(x[2])+x[2]*log(p[2])+(10-x[2])*log(1-p[2])    
# prior ordinates for models    
 LP[1]<-loggam(2)-loggam(1)-loggam(1)+0*log(p[1])+0*log(1
         -p[1])  
LP[2]<-loggam(60)-loggam(30)-loggam(30)+29*log(p[2])
         +29*log(1-p[2])    
# importance sample ordinates for models    
IMP[1]<-loggam(12.1)-loggam(2)-loggam(10.1)+1*log(p[1])+9.1*log(1-p[1])    
IMP[2]<-loggam(70.8)-loggam(31.3)-loggam(39.5)+30.3*log(p[2])+38.5*log(1-p[2])    
# Combining likelihoods, priors and importance samples    
L[1]<-LL[1]+LP[1]+log(0.5)-IMP[1]; L[2]<-LL[2]+LP[2]+log(0.5)-IMP[2]    
# Scaling with the largest likelihood, and protecting for underflow    maxL<-ranked(L[],2); SL[1]<-max(L[1]-maxL,-700); SL[2]<-max(L[2]-maxL,-700)    
# exponentiating and normalizing    expSL[1]<- exp(SL[1]); expSL[2]<- exp(SL[2]);    
# model weights    
w[1]<-expSL[1]/sum(expSL[]); 
w[2]<-expSL[2]/sum(expSL[])}

Data
list(x=c(1,1)).