extract.sienaBayes: Extraction of posterior samples or posterior means from...
In RSienaTest: Siena - Simulation Investigation for Empirical Network Analysis

Description Usage Arguments Details Value Author(s) References See Also Examples

The first function extracts posterior samples from a list of sienaBayesFit object to be used, e.g., for assessing convergence.
The second function extracts posterior means and standard deviations per group from a sienaBayesFit object.

extract.sienaBayes(zlist, nfirst=zlist[[1]]$nwarm+1, extracted,
                   sdLog=TRUE)
extract.posteriorMeans(z, nfirst=z$nwarm+1, pmonly=1, excludeRates=FALSE,
                   verbose=TRUE)

`zlist`	A list of `sienaBayesFit` objects, further called 'chains', resulting from calls to `sienaBayes` with a common data set and model specification.
`z`	A `sienaBayesFit` object.
`nfirst`	Integer: the first element for the first MCMC chain used for calculating properties of the chain.
`extracted`	The parameters for which posterior samples are to be extracted: `"all"`: all parameters; `"rates"`: all groupwise rate parameters; `"varying"`: all varying non-rate parameters: global means and standard deviations; `"non-varying"`: all estimated non-varying (and therefore, non-rate) parameters; `"objective"`: all non-rate parameters.
`sdLog`	Logical: should the logarithms be taken of the posterior standard deviations (applies only to varying parameters).
`pmonly`	if `pmonly=0` the posterior means and standard deviations are extracted, if `pmonly=1` only the posterior means, if `pmonly=2` only the posterior standard deviations.
`excludeRates`	Logical: whether to exclude the rate parameters.
`verbose`	Logical: gives some console output to show ongoing activity (may be reassuring for large number of groups).

extract.sienaBayes produces a 3-dimensional array, iterations by chains by parameters, suitable for use, e.g., in function monitor() of package rstan. The 'iterations' are draws from the posterior distribution (provided that convergence was achieved), after the thinning in sienaBayes implied by its parameters nrunMHBatches, nSampVarying, nSampConst and nSampRate.
The number of iterations should be the same for all chains.
It is not checked that the specifications of the chains are identical. For the parameter names, the shortNames in the first chain are used.

extract.posteriorMeans produces the posterior means of the groupwise varying parameters. The resulting matrix has the groups in the rows. The effects are in the columns: for each effect comes first the posterior mean, then the posterior standard deviation.

For extract.sienaBayes a 3-dimensional array, iterations by chains by parameters, the third dimension having the names of the parameters.
For extract.posteriorMeans a matrix with the groups in the rows and all effects in the columns, with for each effect two columns: first the posterior mean (effect name preceded by "p.m.") and then the posterior standard deviation (effect name preceded by "psd.").
If some names are duplicated (e.g., because of the presence of evaluation and creation effects), these can be changed by the user after creation of the array.

Tom Snijders

See the manual and http://www.stats.ox.ac.uk/~snijders/siena/

sienaBayes, plotPostMeansMDS

  Group1 <- sienaDependent(array(c(N3401, HN3401), dim=c(45, 45, 2)))
  Group3 <- sienaDependent(array(c(N3403, HN3403), dim=c(37, 37, 2)))
  Group4 <- sienaDependent(array(c(N3404, HN3404), dim=c(33, 33, 2)))
  Group6 <- sienaDependent(array(c(N3406, HN3406), dim=c(36, 36, 2)))
  dataset.1 <- sienaDataCreate(Friends = Group1)
  dataset.3 <- sienaDataCreate(Friends = Group3)
  dataset.4 <- sienaDataCreate(Friends = Group4)
  dataset.6 <- sienaDataCreate(Friends = Group6)
  FourGroups <- sienaGroupCreate(
        list(dataset.1, dataset.3, dataset.4, dataset.6))
  FourEffects <- getEffects(FourGroups)
  FourEffects <- includeEffects(FourEffects, transTrip)
  FourEffects <- setEffect(FourEffects, density, random=TRUE)
  FourEffects <- setEffect(FourEffects, recip, random=TRUE)
  print(FourEffects, includeRandoms=TRUE)
  # Note this also shows the "randomEffects" column.
  FourAlgo <- sienaAlgorithmCreate(projname = "FourGroups", maxlike=TRUE)
## Not run: 
  bayes.models <- replicate(4, sienaBayes(FourAlgo, data = FourGroups,
        effects = FourEffects, nwarm=10, nmain=25, nrunMHBatches=10),
        simplify=FALSE)
  bayes.extracted <- extract.sienaBayes(bayes.models, extracted="all")
  dim(bayes.extracted)
  dimnames(bayes.extracted)
  extract.posteriorMeans(bayes.models[[1]])

## End(Not run)