nplcm  R Documentation 
Uses JAGS
(OSX or Windows) operating system for Bayesian posterior inference
(see README
file for an instruction to install JAGS
). If running JAGS
on windows,
please go to control panel to add the directory to JAGS
into ENVIRONMENTAL VARIABLE.
nplcm(data_nplcm, model_options, mcmc_options)
data_nplcm 
Cases are on top of controls in the rows of diagnostic
test results and the covariate matrix. This is assumed by

model_options 
A list of model options: likelihood and prior.

mcmc_options 
A list of Markov chain Monte Carlo (MCMC) options.

A JAGS
output result, fitted by function R2jags::jags2()
from R2jags
.
It is an object of class nplcm
and bugs
.
Current implemented models follow the hierarchy below:
no regression: Fitted by at low level by nplcm_fit_NoReg
regression: Given disease class (control or a class of cases with the same subset of causative agents):
local independence model for BrS measures: Fitted at lower level by
nplcm_fit_Reg_NoNest deals with the setting with two sets of covariates, one for CSCF regression and the other for FPR regression. The two sets of covariates may be identical, overlapping or nonoverlapping. This function is called when there exists one or more than one discrete covariate among the union of the two covariate sets. The method implemented by this function directly lets FPR depend upon covariates. This is different from Wu and Chen (2020+), which let the subclass weights depend upon covariates. We implemented this function for methods comparison.
nplcm_fit_Reg_discrete_predictor_NoNest deals with the setting with all discrete covariates for FPRs and CSCFs. The strata defined by the two sets of covariates need not be identical, e.g., as a result of distinct sets of covariates. Again, this is directly to let FPR be stratified by covariates, hence different from Wu and Chen (2020+) We implemented this function for methods comparison.
local dependence model for BrS measures: Fitted at lower level by nplcm_fit_Reg_Nest: This is the method introduced in Wu and Chen (2020+): CSCF regression + case/control subclass weight regression. It does not provide a specialized function for the setting with all discrete covariates.
data(data_nplcm_noreg) cause_list < LETTERS[1:6] J.BrS < 6 model_options_no_reg < list( likelihood = list( cause_list = cause_list, k_subclass = 2, Eti_formula = ~1, # no covariate for the etiology regression FPR_formula = list( MBS1 = ~1) # no covariate for the subclass weight regression ), use_measurements = c("BrS"), # use bronzestandard data only for model estimation. prior= list( Eti_prior = overall_uniform(1,cause_list), # Dirichlet(1,...,1) prior for the etiology. TPR_prior = list(BrS = list( info = "informative", # informative prior for TPRs input = "match_range", # specify the informative prior for TPRs by specifying a plausible range. val = list(MBS1 = list(up = list(rep(0.99,J.BrS)), # upper ranges: matched to 97.5% quantile of a Beta prior low = list(rep(0.55,J.BrS)))) # lower ranges: matched to 2.5% quantile of a Beta prior ) ) ) ) set.seed(1) # include stratification information in file name: thedir < paste0(tempdir(),"_no_reg") # create folders to store the model results dir.create(thedir, showWarnings = FALSE) result_folder_no_reg < file.path(thedir,paste("results",collapse="_")) thedir < result_folder_no_reg dir.create(thedir, showWarnings = FALSE) # options for MCMC chains: mcmc_options_no_reg < list( debugstatus = TRUE, n.chains = 1, n.itermcmc = as.integer(200), n.burnin = as.integer(100), n.thin = 1, individual.pred = TRUE, # < must set to TRUE! < NOTE! ppd = FALSE, result.folder = thedir, bugsmodel.dir = thedir ) BrS_object_1 < make_meas_object(patho = LETTERS[1:6], specimen = "MBS", test = "1", quality = "BrS", cause_list = cause_list) clean_options < list(BrS_objects = make_list(BrS_object_1)) # place the nplcm data and cleaning options into the results folder dput(data_nplcm_noreg,file.path(thedir,"data_nplcm.txt")) dput(clean_options, file.path(thedir, "data_clean_options.txt")) rjags::load.module("glm") nplcm_noreg < nplcm(data_nplcm_noreg,model_options_no_reg,mcmc_options_no_reg)
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.