Skip to contents

Compare multiple modality networks on one panel

Source: R/compare_modality_graphs.R
compare_modality_graphs.Rd

Visualises two or more catmodgraph objects side-by-side with a shared node layout, for exploring how category-level marginal association structure differs across populations, sites, or time points.

Usage

compare_modality_graphs(
  x,
  restrict = c("common", "union"),
  pruning = c("individual", "none"),
  min_weight = 0.1,
  max_p = 0.05,
  signed = FALSE,
  layout_fn = igraph::layout_with_fr,
  vertex_size = 14,
  edge_scale = 6,
  ...
)

Arguments

x

A named list of catmodgraph objects. The list names become panel titles. A minimum of two graphs is required.

restrict

Character. How to handle modalities that appear in only some of the input graphs. One of:

"common" (default)

Restrict every panel to modalities present in all input graphs. Safest for comparison.

"union"

Use the union of modalities across all inputs. Modalities absent from a given graph appear as isolated vertices in that panel.

pruning

Character. How to filter edges before plotting. One of "individual" (default), "none". Pooled pruning is not supported for modality networks because it requires pooling the underlying row data and re-estimating, which may change the modality set non-trivially. Use "individual" with a shared min_weight to ensure comparable per-panel thresholds.

min_weight

Numeric. Minimum phi threshold for "individual" pruning. Default 0.10.

max_p

Numeric. Maximum p-value for "individual" pruning. Default 0.05.

signed

Logical. If TRUE, edges are coloured by sign of the stored std_resid attribute (green = attraction, red = repulsion). Default FALSE.

layout_fn

Function. An igraph layout function applied to the union graph. Default igraph::layout_with_fr.

vertex_size

Numeric. Vertex size. Default 14.

edge_scale

Numeric. Multiplier for edge widths. Default 6.

...

Further arguments passed to plot.igraph.

Value

Invisibly returns the reference union graph.

Details

Modality graphs are inherently harder to compare than variable-level graphs because modality sets may differ across groups (e.g., a response category endorsed in population A but not population B). The restrict = "common" default avoids this by reducing all panels to a shared vocabulary; "union" preserves all nodes but panels will differ in vertex presence.

Formal testing

For inferential comparison of modality networks, see test_modality_graph_equality (omnibus) and test_modality_edge_differences (edge-wise post-hoc).

See also

build_modality_graph, compare_catgraphs

Examples

# Build two modality graphs from subsets of HairEyeColor
df <- expand_table(HairEyeColor)
mg_f <- build_modality_graph(df[df$Sex == "Female", c("Hair", "Eye")])
mg_m <- build_modality_graph(df[df$Sex == "Male",   c("Hair", "Eye")])

compare_modality_graphs(list(Female = mg_f, Male = mg_m))