eggCounts (version 2.4)

fecr_stanExtra: Model the reduction of faecal egg count using custom models

Description

Models the reduction in faecal egg counts with custom model formulation using Stan modelling language (for advanced users).

Usage

fecr_stanExtra(preFEC, postFEC, rawCounts = FALSE, preCF = 50, postCF = preCF, 
  modelName = NULL, modelCode = NULL, modelFile = NULL, modelData = NULL,
  nsamples = 2000, nburnin = 1000, thinning = 1, nchain = 2, 
  ncore = 1, adaptDelta = 0.95, verbose = FALSE)

Value

Prints out the posterior summary of FECR as the reduction, meanEPG.untreated as the mean pre-treatment epg, and meanEPG.treated as the mean after-treatment epg. The posterior summary contains the mean, standard deviation (sd), 2.5%, 50% and 97.5% percentiles, the 95% highest posterior density interval (HPDLow95 and HPDHigh95) and the posterior mode.

The returned value is a list that consists of:

stan.model

an object of class stanmodel-class that was used

stan.samples

an object of S4 class stanfit representing the fitted results

posterior.summary

a data.frame that is the same as the printed posterior summary. Not available for custom models.

Arguments

preFEC

numeric vector. Pre-treatment faecal egg counts. Not required if modelData is supplied.

postFEC

numeric vector. Post-treatment faecal egg counts. Not required if modelData is supplied.

rawCounts

logical. If TRUE, preFEC and postFEC correspond to raw counts (as counted on equipment). Otherwise they correspond to calculated epgs (raw counts times correction factor). Defaults to FALSE. Not required if modelCode or modelFile is supplied.

preCF

a positive integer or a vector of positive integers. Pre-treatment correction factor(s). Not required if modelCode or modelFile is supplied.

postCF

a positive integer or a vector of positive integers. Post-treatment correction factor(s). Not required if modelCode or modelFile is supplied.

modelName

string. One of four availale models ("Po", "UPo", "ZIPo", "ZIUPo") from eggCountsExtra package, which corresponds to outlier-adjusted version of paired, unpaired, paired with zero inflation and unpaired with zero inflation models. Not required if modelCode or modelFile is supplied.

modelCode

stan model code. Not required when modelName or modelFile is supplied.

modelFile

stan model file with file extension ‘*.stan’. Not required when modelName or modelCode is supplied.

modelData

stan data list. A named list or environment providing the data for the model, or a character vector for all the names of objects in the current enviroment used as data. Not required when modelName is supplied.

nsamples

a positive integer. Number of samples for each chain (including burn-in samples).

nburnin

a positive integer. Number of burn-in samples.

thinning

a positive integer. Thinning parameter, i.e. the period for saving samples.

nchain

a positive integer. Number of chains.

ncore

a positive integer. Number of cores to use when executing the chains in parallel.

adaptDelta

numeric. The target acceptance rate, a numeric value between 0 and 1.

verbose

logical. If TRUE, prints progress and debugging information.

Author

Craig Wang

Details

If modelName is one of c("Po", "UPo", "ZIPo", "ZIUPo"), then outlier-adjusted models are used.

• In paired models, outliers are those counts with postFEC > preFEC. Outlier weights are assigned as the inverse of postFEC/preFEC,

• In unpaired models, outliers are those counts with postFEC greater than the 95th percentile of a Poisson distribution, where the Poisson mean is computed based on the mean of postFEC excluding postFEC > Q3 + 1.5*IQR. Q3 is the 75th percentile and IQR is the interquartile range. The lowest outlier weight is assigned as 0.01, and other outliers assigned proportionally.

• In both cases, non-outliers are assigned with outlier weight = 1.

The first time each model is applied, it can take up to 20 seconds for Stan to compile the model.

The default number of samples per chain is 2000, with 1000 burn-in samples. Normally this is sufficient in Stan. If the chains do not converge, one should tune the MCMC parameters until convergence is reached to ensure reliable results.

Examples

if (FALSE) {
library(eggCountsExtra)
data(epgs) ## load sample data

## apply paired model with outliers 
model1 <- fecr_stanExtra(epgs$before, epgs$after, rawCounts=FALSE, 
         preCF=10, modelName = "Po")
samples <- stan2mcmc(model1$stan.samples)
fecr_probs(model1$stan.samples, threshold = 0.99)

## apply a simple custom model
code <- "data{
  int J; // number of animals
  int y_before[J]; // after treatment McMaster count
  int y_after[J]; // before treatment McMaster count
}
parameters{
  real mu;
  real delta;
}
model{
  mu ~ gamma(1,0.001);
  delta ~ beta(1,1);
  y_before ~ poisson(mu);
  y_after ~ poisson(mu*delta);
}"

dat <- list(J = nrow(epgs), y_before = epgs$before,
            y_after = epgs$after)
model2 <- fecr_stanExtra(modelCode = code, modelData = dat)
}