jagshelper (version 0.4.0)

plotRhats: Plotting all Rhat values

Description

Plotting all values of Rhat (or alternately n.eff) from an output object returned by jagsUI, or perhaps a subset of parameters. This function is intended as a quick graphical check of which parameters have adequately converged.

Rhat (Gelman-Rubin Convergence Diagnostic, or Potential Scale Reduction Factor) is calculated within 'JAGS', and is commonly used as a measure of convergence for a given parameter node. Values close to 1 are seen as evidence of adequate convergence. n.eff is also calculated within 'JAGS', and may be interpreted as a crude measure of effective sample size for a given parameter node.

Usage

plotRhats(
  x,
  p = NULL,
  n.eff = FALSE,
  fence = NULL,
  plotsequence = FALSE,
  splitarr = FALSE,
  margin = NULL,
  ...
)

Value

NULL

Arguments

x

Output object returned from jagsUI

p

Optional vector of parameters to subset

n.eff

Optionally, whether to plot n.eff instead of Rhat. Defaults to FALSE.

fence

Value of horizontal lines to overlay as reference. Accepting the default value (NULL) will give fence values of 1.1 (a commonly used value) and 1.01 for Rhat, or 100 and 500 for n.eff.

plotsequence

Whether to plot parameter vectors (or matrices) in a sequence, running left to right, which may be useful for time series models, etc. If the default (FALSE) is used, a vertical cluster will be plotted for each parameter, resulting in a simpler plot if there are many parameters. Note that the Rhat values will still be plotted in sequence if the default (FALSE) is used.

splitarr

Whether to split 2+ dimensional parameter arrays by a given dimension, rather than plotting the full array in one vertical cluster or continuous sequence. This may be recommended in the case of large arrays. Defaults to FALSE.

margin

If splitarr= is set to TRUE, which array margin to split by. In the case of a 2-dimensional array, setting margin=2 will separate the array by column. If the default (NULL) is accepted, the function will split by the smallest dimension, therefore splitting into the fewest groups.

...

additional plotting arguments

Author

Matt Tyers

References

Gelman, A., & Rubin, D. B. (1992). Inference from Iterative Simulation Using Multiple Sequences. Statistical Science, 7(4), 457–472. http://www.jstor.org/stable/2246093

See Also

traceworstRhat, check_Rhat, qq_postpred, ts_postpred, plot_postpred

Examples

## plotting everything
plotRhats(SS_out)
str(SS_out$Rhat)  # the associated values

plotRhats(SS_out, n.eff=TRUE)
str(SS_out$n.eff)  # the associated values

## behavior of splitarr and margin are shown
plotRhats(SS_out)
plotRhats(SS_out, splitarr=TRUE)
str(SS_out$Rhat) # the associated values

## plotsequence may be useful in the case of a sequence of values
plotRhats(SS_out, p=c("trend", "cycle_s"), splitarr=TRUE, plotsequence=TRUE)