Hierarchical Graph (2D)
Updated: Mar 20, 2026
What This Chart Shows
This chart shows your causal influence network as a 2D diagram: nodes are entities (e.g., channels, touchpoints, or events) and lines are directed influences between them. Nodes are arranged in tiers from left-to-right or top-to-bottom. You will see how marketing influence flows from drivers through intermediate steps to final outcomes, and which nodes sit at which stage of the journey.
Key Questions This Chart Helps Answer
- How does influence flow from sources to outcomes in my marketing system?
- Which nodes are pure sources, which are outcomes, and which are in between?
- Where do the main causal chains sit in the graph?
- How is the causal structure organized by stage (early / mid / late)?
Axes, Metrics, and Units
Element | Description |
|---|---|
| X-axis | Layout position (not labeled). Horizontal position is determined by tier and by node index within each tier. |
| Y-axis | Layout position (not labeled). Vertical position is tier-based: higher tier value = lower Y (e.g., Sources at top, Outcomes at bottom). |
| Nodes | Graph vertices. Each node is one entity in the causal graph (e.g., a series or conceptual event). |
| Node color | Tier: Sources (tier 0), Early (1), Mid (2), Late (3), Outcomes (4). Color palette is fixed per tier. |
| Edges (lines) | Directed links from source node to target node. Gray lines; up to 200 edges shown (strongest by absolute weight if the graph is large). |
Positions are layout coordinates (tier-based placement). Edge presence and subsampling use model-derived weights from the CSR graph.
Control Options Reference
This chart has no user-facing control options in the code; it renders the current graph from the bridge as-is.
How to Interpret the Results
- Tier 0 (Sources): Nodes with no incoming edges—entry points or root drivers (e.g., paid media, owned touchpoints).
- Tiers 1–3 (Early / Mid / Late): Intermediate nodes; tier is based on in-degree vs out-degree ratio, so you see where they sit in the flow.
- Tier 4 (Outcomes): Nodes with no outgoing edges—endpoints (e.g., conversions, revenue).
- Edges: Direction is source → target. Many edges into a node = many drivers; many edges out = many downstream effects.
- Dense vs sparse: Dense sections show tightly connected parts of the journey; sparse sections show fewer modeled links.
- Only top 200 edges: In large graphs, only the 200 edges with largest absolute weight are drawn; the rest are hidden, so the chart emphasizes the strongest influences.
Practical Applications for Marketers
Application | How to use this chart |
|---|---|
| Journey mapping | See how touchpoints and channels connect from first touch to conversion. |
| Driver identification | Identify source nodes and trace which outcomes they can influence. |
| Bottleneck discovery | Look for nodes with many incoming and few outgoing edges (or the reverse) as potential bottlenecks or hubs. |
| Model transparency | Explain that attribution uses a causal graph and show its high-level shape. |
| Planning and storytelling | Use the tier structure (sources → outcomes) to align with funnel or stage-based planning. |
Common Mistakes & Misinterpretations
Mistake | Why it is a problem | How to avoid |
|---|---|---|
| Treating every line as "causes in the real world" | Edges are model-inferred influences; strength and even presence depend on data and model. | Use the graph as the model’s view of structure, and validate important paths with tests or other analyses. |
| Ignoring edge subsampling | Only up to 200 edges are shown; weaker links are omitted. | Don’t assume "no line = no relationship"; treat the chart as showing the strongest links. |
| Reading exact position as importance | X/Y are layout positions for clarity, not a direct importance or time axis. | Rely on tier (color/legend) and connectivity, not pixel position. |
| Assuming nodes are channels only | Nodes are graph vertices (e.g., series or conceptual events); they may be channels, events, or other entities depending on the graph build. | Check what the graph is built from (e.g., conceptual events, series) in your setup. |
| Over-interpreting a single dense cluster | One dense area may reflect data density or model structure, not necessarily the "most important" part of the journey. | Combine with attribution and outcome metrics to judge importance. |
Caveats & Considerations
- No edges: If the graph has nodes but zero edges, the chart shows: "Graph has V nodes but no edges."
- Edge limit: At most 200 edges are displayed; edges are chosen by largest absolute weight. Very large graphs are simplified.
- Fixed layout: Positions are tier-based and algorithmic (in/out degree); not force-directed. The same graph will look the same each time.
- No node labels in main view: Nodes are labeled generically (e.g., "Node i") in hover; detailed naming depends on whether the app adds labels from metadata.
- Tier definition: Sources and Outcomes are defined by in/out degree only; "Early/Mid/Late" are derived from degree ratios and may not match your exact funnel stages.