mtc.nodesplit: Node-splitting analysis of inconsistency

Description

Generate and run an ensemble of node-splitting models, results of which can be jointly summarized and plotted.

Usage

mtc.nodesplit(network, comparisons=mtc.nodesplit.comparisons(network), ...)
mtc.nodesplit.comparisons(network)

Arguments

network

An object of S3 class mtc.network.

comparisons

Data frame specifying the comparisons to be split. The frame has two columns: 't1' and 't2'.

...

Arguments to be passed to mtc.run or mtc.model. This can be used to set the likelihood/link or the number of iterations, for example.

Value

For mtc.nodesplit: an object of class mtc.nodesplit. This is a list with the following elements:

d.X.Y

For each comparison (t1=X, t2=Y), the MCMC results

consistency

The consistency model results

For summary: an object of class mtc.nodesplit.summary. This is a list with the following elements:

dir.effect

Summary of direct effects for each split comparison

ind.effect

Summary of indirect effects for each split comparison

cons.effect

Summary of consistency model effects for each split comparison

p.value

Inconsistency p-values for each split comparison

cons.model

The generated consistency model

Details

mtc.nodesplit returns the MCMC results for all relevant node-splitting models [van Valkenhoef et al. 2015]. To get appropriate summary statistics, call summary() on the results object. The summary can be plotted. See mtc.model for details on how the node-splitting models are generated.

To control parameters of the MCMC estimation, see mtc.run. To specify the likelihood/link or to control other model parameters, see mtc.model. The ... arguments are first matched against mtc.run, and those that do not match are passed to mtc.model.

mtc.nodesplit.comparisons returns a data frame enumerating all comparisons that can reasonably be split (i.e. have independent indirect evidence).

Examples

Run this code

# NOT RUN {
# Run all relevant node-splitting models
# }
# NOT RUN {
 result.ns <- mtc.nodesplit(parkinson, thin=50) 
# }
# NOT RUN {
# (read results from file instead of running:)
result.ns <- readRDS(system.file('extdata/parkinson.ns.rds', package='gemtc'))

# List the individual models 
names(result.ns)

# Time series plots and convergence diagnostics for d.A.C model
plot(result.ns$d.A.C)
gelman.diag(result.ns$d.A.C, multivariate=FALSE)

# Overall summary and plot
summary.ns <- summary(result.ns)
print(summary.ns)
plot(summary.ns)
# }