
Posterior parameter summaries
summary.magComp.RdSummarizes posterior draws. summary(fit, ...) calls
samples(fit, ...) and then summarizes that sample object. The default
reports scalar parameters; use type = "effects" or
type = "combined" for individual effects or combined coefficients. Multiple-chain fits include the classical unsplit Gelman-Rubin
\(\widehat{R}\) diagnostic; single-chain fits report NA for
\(\widehat{R}\). Effective sample size is computed from within-chain
autocorrelation and summed across chains, without a between-chain correction.
The R-hat calculation is bounded below by one and does not split or rank-normalize
chains. See the technical guide's Diagnostics section in magComp-package
for the formulas.
Arguments
- object
A fitted
magCompobject or an object returned bysamples().- probs
Nonempty numeric vector of posterior quantile probabilities strictly between zero and one.
- ...
For a fitted model, arguments passed to
samples(), includingtype,start,thin,chains,combine_chains, andgroup. Unused for an already extracted sample object.
Value
A data frame of class magCompSummary, identifying component, block, term,
parameter, index, and manuscript symbol, followed by
posterior means, standard deviations, requested quantiles,
\(\widehat{R}\), effective sample size, retained samples per chain, and
number of chains.
Details
Parameter names match the public sample entries: coefficients,
size, variance, loadings, and covariance parameter
names such as phi, lambda, and nu. The
index column identifies the relevant design column, stick/covariate
pair, stick/factor pair, or factor. Component, block, and term identities stay
the same as in magCompSettings and samples.
Symbols use portable ASCII notation. Fixed size and covariance values are excluded;
positive anchor-diagonal loadings are included as estimated parameters.
Fixed upper-anchor zeros are omitted from the scalar summary. Latent effects
and scores are available through summary(fit, type = "effects").
Combined coefficients also identify their fitted unit or group, coefficient,
allocation row, and parent context. R-hat and ESS use the selected draws
within their original chains, even when storage has been pooled. One-chain
selections report NA for R-hat.