ResIN 2.3.1
Major changes
- Correlation engine overhaul. ResIN now uses a
unified, more efficient correlation backend that supports:
cor_method = "pearson" (fully supported across
weighted/unweighted settings, pairwise/listwise missingness, and
optional p-value pruning)cor_method = "polychoric" (currently supported for
unweighted correlation estimation; unsupported option combinations
transparently fall back to Pearson correlation)
- New missingness control for correlations. Added
missing_cor = c("pairwise","listwise") to explicitly
control missing-data handling during correlation estimation (default:
"pairwise").
Weighted correlations
- Improved weights interface. Survey
weights can now be supplied either as:
- a numeric vector of length
nrow(df), or
- a character scalar naming a weights column in
df (e.g.,
weights = "psweights").
When weights is supplied as a column name and
node_vars = NULL, the weights column is automatically
excluded from the response-node variables used for estimation.
Edge pruning
- New multiple-testing option for pruning. Added
remove_nonsignificant_method, enabling Benjamini–Hochberg
FDR control ("fdr") as an alternative to raw p-value
pruning ("default"). sign_threshold is
interpreted as the raw p-value cutoff under "default" and
as the target FDR level under "fdr".
- EBICglasso retired.
EBICglasso and
EBICglasso_arglist are retained for backwards compatibility
but are deprecated and ignored as of 2.3.0. This
feature is not statistically appropriate for the rank-deficient one-hot
encoded ResIN correlation matrix.
Bootstrapping
workflow (S3 methods & robustness)
- Added/expanded S3 methods and documentation for ResIN bootstrap
workflow objects:
ResIN_boots_prepped and
ResIN_boots_executed now have clearer print/summary
behavior.ResIN_boots_execute() is more robust (handles failed
iterations gracefully) and improves progress-bar formatting.
ResIN_boots_extract() now supports multi-quantity
extraction and improved summarizing behavior, including consistent
data-frame output for summaries.
Conversion
of ResIN objects for interoperability and manipulation in other software
environments
- Added conventional S3 conversions for ResIN objects:
as.igraph.ResIN(), as.qgraph.ResIN(),
as.gephi.ResIN()as.network.ResIN() (statnet/network ecosystem), with
as.network() re-exported for convenience
- Added lightweight cross-language export helpers:
as.networkx.ResIN() exports NetworkX compatible edge
& node CSV tables for Python workflows (loadable via
pandas + networkx).as.graphsjl.ResIN() exports Graphs.jl compatible edge
& node CSV tables for Julia workflows (with integer vertex IDs
suitable for Graphs.jl), while preserving node/edge
metadata as table columns.
- Legacy conversion functions (e.g.,
ResIN_to_igraph(),
ResIN_to_qgraph(), ResIN_to_gephi()) are
retained for compatibility and delegate to the new conversion methods
where appropriate.
Improved
Multimodal network estimation and plotting features
- Enhanced the optional multimodal output
(
multimodal = TRUE) to return tidy, analysis-ready tables
in addition to graph objects:
ResIN_edgelist_multimodal: participant–response edge
list (one row per observed response node selection).ResIN_nodeframe_multimodal: vertex table containing
node identifiers, node type (participant vs response node), layout
coordinates, and available metadata.
- Improved multimodal plotting:
- Response nodes are now rendered larger than participant nodes for
improved visual separation.
- multimodal plots now reuse key analytic settings from the main ResIN
workflow where applicable (e.g., covariate/cluster-based colouring via
plot_whichstat).
- When
plot_whichstat = "cluster", the legend title
includes the clustering algorithm and reports participant/response
counts per detected cluster.
- Added flexible edge overlay controls for multimodal
plots via
multimodal_edge_overlay:
"none": no edges drawn"multimodal": draw participant–response edges only"ResIN": draw the original ResIN (response–response)
edges as an overlay, with line width optionally mapped via
plot_edgestat"both": draw ResIN edges as a base layer and multimodal
edges on top ResIN edge overlays are mapped to the multimodal layout
coordinates to ensure visual alignment with the multimodal node
geometry.
- Added adjecency matrix retaining negative edges as
part of
aux_objects
Bug fix for cluster
Leiden function call
Documentation
- Improved help page titles and vignette metadata (more informative
package-level index display).
- Added/updated documentation for new S3 classes/methods, conversions,
and pruning options.
ResIN 2.3.0