bhlm: Bayesian Hierarchical Latent Mixture Model

In arcuo/bachelors_meta: Bayesian Hierarchical Latent Mixture model meta analysis and plots.

Description

Run a BHLM model with custom number of latent variables according to specified number of outcome options. The model sets priors for each outcome, group mean (theta) and estimate variance.

Run following for model specification (Graphical notation):

plot(0:1,0:1,type="n",ann=FALSE,axes=FALSE) rasterImage(bhlm_model_spec,0,0,1,1)

Usage

bhlm(dataframe, grouping_factor_cols, estimate_col, outcome_options_col,
  outcome_options, outcome_priors, lambda_prior, theta_prior,
  field_theta_precision = NULL, identifier_col = NULL,
  bayes_method = "jags", jags_init = NULL, jags_chains = 3,
  jags_iter = 2000, jags_burnin = floor(jags_iter/2), jags_thin = max(1,
  floor(jags_chains * (jags_iter - jags_burnin)/1000)), jags_DIC = TRUE,
  save_model = NULL)

Arguments

dataframe	A data frame with studies, two grouping factor cols (e.g. studies*outcomes), estimates, and optional ID column.
grouping_factor_cols	A data vector of two grouping factor columns of the data frame (e.g. c("Study", "Outcome")). These columns must be integer vectors.
estimate_col	Column of estimates in the data frame.
outcome_options_col	Column containing the different outcome names Outcomes : Factor w/ 3 levels "Physical","Psychological",..: 2 2 2 2 2 2 2 2 1 1 ...
outcome_options	Chosen outcomes. You can choose a subset of outcomes.
outcome_priors	Priors for outcomes. Takes character vector or matrix. Define all outcome priors to one distribution set prior (throws Warning) as character: "dnorm(0, 1)" Priors defined as character strings are checked with regex: grepl("^[a-zA-Z]+\(((\d*\.)*\d+,\s?)*(\d*\.)*\d+\)", .) or as data vector c("dnorm", 0, 1) Define all outcome priors manually with matrix. Notice: byrow = TRUE matrix(c("dnorm", 0, 1, "dnorm", 0, 1.1), nrow = 'number of outcomes', ncol = 'maximum amount of dist parameters', byrow = TRUE) or as character vector of characters c("dnorm(0,1)", "dnorm(0,1.1)") Notice: priors set manually, are set in the same order as outcome_options in character vector or in matrix row order (top to bottom).
lambda_prior	Prior for Lambda as character or data vector.
theta_prior	Prior for Theta as character or data vector.
identifier_col	ID column. Chosen column is added to the used data for easier overview.
bayes_method	Use JAGS or STAN (currently only JAGS is implemented).
save_model	Input FALSE to not save the model file, NULL to save as 'model_file.txt' in working directory, or filepath location with name to save the model file as .txt (Separated with '\' or '/').
field_theta	Optional: Include a another hierarchical level in the form of a field-wide mean distribution. Use a string vector with the same distribution as theta_prior and precision (no mean). The field-wide mean prior, is set to theta_prior. This solution is currently a fix-up, and should only be used on a case by case basis.
jags	Parameters for R2jags::jags: init (NULL set as default), chains, iter (iterations per chain), burning (length of burn), thin (sim thinning rate), and DIC (compute deviance, pD and DIC).

Author(s)

Hugh Benjamin Zachariae

Examples

# INCOMING

arcuo/bachelors_meta documentation built on May 5, 2019, 2:41 a.m.