In most projects, the data contains many co-terminal links: multiple coded claims with the same cause and the same effect. We call these bundles of links.

This extension is best thought of as two pieces:

1. A filter (data transformation)
2. An interpretation rule (how to read a “link” on a map)

1) The filter: bundle the links table#

The filter operates on the current links table (one row per coded claim / citation) and produces a derived table with fewer rows by grouping on the (current, possibly transformed) factor labels:

• Bundle key: \((\text{cause label}, \text{effect label})\)
• One output row = one bundle (one unique cause→effect pair)

The bundled output adds aggregate columns such as:

• bundle: a readable key like cause >> effect
• citation_count: number of underlying link rows in the bundle (how many claims / coded citations)
• source_count: number of distinct sources contributing at least one link to the bundle (how many sources said it)

Optionally, further bundle-level summaries can be computed from the underlying rows, for example:

• mean_sentiment: mean effect sentiment across the bundle (if sentiment is present)
• per-tag or per-group counts (e.g. “how many links in this bundle have tag X?”), when you want enriched breakdowns

Important: bundling uses the current filtered labels, i.e. after any upstream label transforms (collapse, zoom hierarchies, bracket removal, soft recode, combine opposites, etc.). So the bundle definition reflects the conceptual normalisations you have chosen.

Most causal mapping approaches which have recorded data from more than one source or context has done this.

2) The interpretation rule: what “a link” means on a map#

Strictly, maps show bundles, not individual citations. But in practice we often still say “link” to mean:

the bundle representing “many similar claims that \(x\) influences \(y\)”.

So a map statement like:

“\(n\) sources claimed that \(x\) influences \(y\) (in \(m\) citations)”

is shorthand for: there are \(m\) underlying coded link rows whose (possibly transformed) labels share the same cause and effect, contributed by \(n\) unique sources.

Showing bundle statistics on maps#

Bundling also supports a family of conventions for displaying bundle-level statistics visually, for example:

• edge width proportional to citation_count
• link label showing source_count or citation_count
• link label showing bundle summaries like mean sentiment (where available)

These are display choices layered on top of the same underlying bundle definition.