Summary#

This extension is about summarising a causal map as a network using simple statistics.

It can be useful for evaluation work because it helps you move from:

• “here is the picture” → to
• “here are a few headline properties of the picture”.

When to use it#

• Reporting: you want a short paragraph or slide that characterises the map (dense vs sparse, highly central factors, etc.).
• Comparing views: you want to compare two filtered views (e.g. before/after simplification, or one group vs another) using the same set of stats.
• Sanity checks: you want to spot oddities (e.g. one “hub” factor connected to everything because of a label issue).

What kinds of stats are most interpretable#

Global (“whole map”) summaries#

• Number of factors / links: basic size of the current view.
• Density / sparsity: how connected the map is relative to how many factors it contains.
• Share of self-loops (if present): how much “plain coding” or A→A evidence is in view.

Local (“per factor”) summaries#

• In-degree / out-degree: how many incoming vs outgoing links a factor has (as outcome vs as driver).
• Centrality (use carefully): identifies “connector” factors that sit on many paths.

How to interpret them (practical cautions)#

• These are properties of the current filtered view (sources selection + analysis filters). Change the pipeline and the stats change.
• Network stats measure structure, not “truth” and not causal effect size.
• Centrality can be inflated by label choices (e.g. a very broad parent label used as a bucket).
• For group comparisons, it’s usually better to compare like-for-like pipelines (same transforms, same simplification rules), otherwise the stats conflate analytic choices with group differences.

Formal notes (optional)#

If you want the more formal network picture:

• Build a directed graph \(G = (V,E)\) from the current links table (after bundling / transforms as appropriate).
• \(V\) are factor labels; \(E\) are directed edges (cause→effect).
• Common global stats include \(|V|\), \(|E|\), and density (e.g. \(|E|/(|V|(|V|-1))\) if you treat it as a simple directed graph without self-loops).
• Common local stats include in-degree, out-degree, and centrality measures (betweenness, closeness, eigenvector/PageRank variants). These can be computed on the simple graph or on a weighted graph (e.g. weighting edges by source count or citation count).