template_huiwalter {runjags}

Create a Hui-Walter model based on paired test data for an arbitrary number of tests and populations

Description

Create a Hui-Walter model based on paired test data for an arbitrary number of tests and populations

Usage

template_huiwalter(
  testdata,
  outfile = "huiwalter_model.txt",
  covariance = data.frame(Test_A = character(0), Test_B = character(0), Active =
    logical(0)),
  se_priors = "dbeta(1,1)",
  sp_priors = "dbeta(1,1)",
  prev_priors = "dbeta(1,1)",
  cov_as_cor = FALSE,
  specify_populations = FALSE,
  outcome_check = TRUE,
  check_min_obs = 20L
)

Arguments

testdata

the input paired test data, where each column name corresponds to a test result - except possibly "ID" which is ignored, and "Population" indicating a population identifier for that row. Each row must represent test results from the same individual either as logical or a factor with two levels (and where the first level indicates a negative test result). Data may be missing at random (except for Population).

outfile

the name of the text file to save the model representation

covariance

a data frame specifying which conditional depdendence terms should be included (either activated or deactivated) with columns Test_A, Test_B, Active_Se and Active_Sp. A single logical FALSE is allowed for back-compatibility, and a single logical TRUE is also currently allowed but is deprecated.

se_priors

the priors to use for sensitivity parameters (can be adjusted in the model once it is generated)

sp_priors

the priors to use for specificity parameters (can be adjusted in the model once it is generated)

prev_priors

the priors to use for prevalence parameters (can be adjusted in the model once it is generated)

cov_as_cor

option for the prior for covariance terms to be put on the correlation rather than covariance directly (deprecated; currently ignored with a warning)

specify_populations

option for the active populations to be retrieved from a PopulationsUsing vector in the R environment - this facilitates sensitivity analysis by excluding subsets of populations without re-generating the model

outcome_check

option to facilitate comparing the observed tallies to the predicted tallies in order to assess model fit

check_min_obs

the minimum number of total observations required before an outcome_check is generated (this prevents e.g. outcome checks being generated for partially missing data)

Examples

N <- 600
status <- rbinom(N, 1, rep(c(0.25,0.5,0.75), each=N/3))
testdata <- data.frame(Population = rep(1:3, each=N/3),
    FirstTest = rbinom(N, 1, status*0.95 + (1-status)*0.05),
    SecondTest = rbinom(N, 1, status*0.75 + (1-status)*0.02),
    ThirdTest = rbinom(N, 1, status*0.5 + (1-status)*0.01)
)
template_huiwalter(testdata, outfile="huiwalter_model.txt",
    covariance=data.frame(
      Test_A="FirstTest",
      Test_B="SecondTest",
      Active_Se=TRUE, Active_Sp=FALSE
    )
)

## Then examine and verify the code manually!
cat(readLines("huiwalter_model.txt"), sep="\n")

## Before running the model:
## Not run: 
  results <- run.jags("huiwalter_model.txt")
  results

## End(Not run)

## Cleanup:
unlink("huiwalter_model.txt")