Difference between revisions of "POPs in Baltic herring"

library(OpasnetUtils)
library(ggplot2)
library(rjags)

dat <- opbase.data("Op_en3104", subset = "POPs")
dat <- dat[dat$Fish_species == "Baltic herring" , ]

congeners <- levels(dat$POP) #names of different congeners in data

Y <- length(congeners) #number of congeners and j's in for loop

#resultsall <-c('congener', 'meanorig', 'sdorig', 'meanlog', 'meanpost') #comined results table structure to store results of individual congener runs

compdat <- dat[dat$POP %in% congeners[1:3] , ] #data for current congener
Compound <- log(compdat$Result+1E-2) #+1E-2 because zero concentrations are not allowed

mo <- textConnection("model{
    for (j in 1 : 3 ) { # for testing reduced to 3 congeners, real loop amount is Y (=79)
      tau1[j] ~ dunif(0.001, 1000)
      muOfCompound[j] ~ dnorm(0, 0.001)
    }
    for( i in 1 : N ) {
      Compound[i] ~ dnorm(muOfCompound[POP[i]], tau2[i])
      tau2[i] <- tau1[POP[i]]*sqrt(n[i])
    }
  }
")

#  shap <- shapiro.test(compdat$Result) #test of normality

#filename <- paste("shap", j, sep="") #save shapiro test result for each congener, just for fun

#assign(filename, shap)


#  if(shap$p.value < 0.05) { #if skewed, logtransformation, if not continue with original data. Or should the logtransformation be done in any case?
#Compound <- log(compdat$Result+1E-2) #+1E-2 because zero concentrations are not allowed
#  } else {
#    Compound <- compdat$Result
#  }

dataList = list(
  Compound = Compound,
  POP = as.numeric(compdat$POP),
  N = length(Compound),
  n = compdat$N_individuals #number of fishes in sample
)

jags <- jags.model(mo, data = dataList, n.chains = 4, n.adapt = 1000)

close(mo)

#update(jags, 1000)
out <- coda.samples(jags, c('tau1', 'muOfCompound'), 500) # Stores a posterior sample
plot(out)

#filename <- paste("out", j, sep="") #save posterior for each congener
#assign(filename, out)


resultsall <- data.frame(
  Meanorig = aggregate(compdat$Result, compdat["POP"], mean), #calculate mean for original data
  Sdorig = aggregate(compdat$Result, compdat["POP"], sd)$x, #calcaulte sd of original data
  Meanlog = aggregate(Compound, compdat["POP"], mean)$x, #calculate mean for logdata
  Sdlog = aggregate(Compound, compdat["POP"], sd)$x, #calculate sd for logdata
  Meanpost = c(mean((out[[4]])[,1]), #calculate mean of mu for posterior (test 4)
               mean((out[[4]])[,2]),
               mean((out[[4]])[,3])
    ),
    Sdpost =  c(sqrt(1/(mean((out[[4]])[,4]))), #calculate mean of mu for posterior (test 4)
                sqrt(1/(mean((out[[4]])[,5]))),
                sqrt(1/(mean((out[[4]])[,6])))
    )
)

oprint(resultsall)

ggplot(dat, aes(x = Result, colour = Catch_location))+geom_density()+scale_x_log10()