outbreaker (version 1.1-6)

select MCMC chains: Select 'good' runs from independent MCMC chains

Description

The function selectChains is used to discard 'bad' MCMC chains from outbreaker's ouptput (functions outbreaker and outbreaker.parallel). This is useful whenever several chains were run and converged towards different posterior modes or distributions. This can happen for instance when imported cases are hard to disentangle, resulting in different runs identifying different imports and therefore having different likelihood.

Three modes are available, depending on the argument select (see also arguments below):

• visual: (default) interactive mode plotting the log-posterior values for the different chains and asking the user to identify runs to be discarded.
• auto: an automatic procedure is used to discard 'bad' runs; see details.
• [numbers]: numbers indicating the runs to be discarded.

Usage

selectChains(x, select="visual", alpha=0.001, ...)

Arguments

x

the output of outbreaker or outbreaker.parallel.

select

a character string matching visual or auto, or a vector of integers indicating the runs to be discarded.

alpha

the alpha threshold to be used to the automatic procedure (see details)

...

further arguments to be passed to plotChains.

Value

• These functions similar objects to the inputs, from which 'bad' runs have been discarded.

Details

The automatic procedure relies on the following recursive process:

• 1. Make the ANOVA of the log-posterior values as a function of the run identifier.
• 2a. If the P-value is greater than alpha (non-significant), exit.
• 2b. Otherwise, discard the run with the lowest mean log-posterior value, and go back to 1.