What This Chart Shows
This chart shows how your causal influence network changes over time. It’s a 3D network with animation by time window: edges appear and disappear depending on when each relationship is valid, and node size reflects current influence (sum of active edge weights) in that window. You can play, pause, change speed, and scrub by date to see when causal links form, persist, or fade—useful for understanding seasonality, campaign timing, and structural breaks.
Key Questions This Chart Helps Answer
- When do causal relationships between touchpoints and outcomes form or disappear?
- How does the strength of influence at each node change over time?
- How does the causal structure align with campaign or market events?
- Where and when do new or fading connections appear?
Axes, Metrics, and Units
Element | Description |
|---|---|
| X-axis | Layout position (hidden). Horizontal position from tier-based layout. |
| Y-axis | Layout position (hidden). Horizontal position from tier-based layout. |
| Z-axis | Causal Tier: 0 = Sources, 1 = Early, 2 = Mid, 3 = Late, 4 = Outcomes (Outcomes at top). |
| Nodes | Graph vertices (e.g., touchpoints, outcomes). Fixed position across frames; size varies by window. |
| Node size | Current influence in that window. Larger = more active influence. |
| Node color | Tier (Sources, Early, Mid, Late, Outcomes)—fixed per node. |
| Edges (lines) | Directed links that vary by frame: only edges valid in the TCEG validity window for that frame are shown. Gray; up to 200 edges per window. |
| Tier bands | Semi-transparent horizontal planes at each Z level for visual grouping. |
| Frame | One time window. Each frame has a date label (e.g., start date or date range of the window). |
| Playback | Play/Pause, Speed (0.5x, 1x, 2x, 4x), and a slider to jump to a window by date. |
Positions are layout coordinates (tier-based). Edge presence is model-derived from TCEG validity; node activity is derived from edge weights in that window.
Control Options Reference
Control | Meaning |
|---|---|
| Play | Start animation; frames advance automatically. |
| Pause | Stop animation at the current frame. |
| Speed (0.5x / 1x / 2x / 4x) | Animation speed (frame duration). |
| Slider (Date) | Scrub to a specific time window; label shows the window’s date or date range. |
How to Interpret the Results
- Stable layout: Node positions don’t change over time; only edges and node sizes change. Use this to track the same nodes across windows.
- Edges appearing: A new line in a later window = a causal link that became valid in that period (e.g., new campaign or data).
- Edges disappearing: A link that was present earlier but not in a later window = relationship no longer valid in that window (e.g., seasonality or structural change).
- Node size pulsing: Larger nodes = more incident edge weight in that window (more influence); smaller = less. Use this to see when nodes are “hot” or “cold.”
- Scrubbing: Use the slider to align big changes in the network with known events (launches, policy changes, etc.).
- Stable core vs periphery: Nodes and edges that stay across many windows = stable structure; those that come and go = transient.
Practical Applications for Marketers
| Application | How to use this chart |
|---|---|
| Campaign timing | See when causal links to outcomes appear or strengthen and align with campaign start/end. |
| Structural breaks | Spot windows where many edges appear or disappear (e.g., data or market shifts). |
| Journey stability | Identify touchpoints and paths that are consistently present vs occasional. |
| Event correlation | Pause on a window and compare with external events to explain why the network looks different. |
| Planning and storytelling | Use “when do relationships form/fade?” to communicate time-varying attribution and influence. |
Common Mistakes & Misinterpretations
| Mistake | Why it is a problem | How to avoid |
|---|---|---|
| Treating “edge appears” as proven new causation | Validity is from the TCEG model (data + windows); it’s when the model says the link is valid, not necessarily when the true cause started. | Use animation as a signal; validate with tests or other analyses. |
| Ignoring window length | Each frame is an aggregate over a window (e.g., 7 days). Short-lived effects may be smoothed or missed. | Check window size and consider whether you need finer or coarser time. |
| Assuming all changes are marketing-driven | Edges can change due to data quality, model updates, or non-marketing factors. | Correlate with known events; don’t assume every change is campaign-related. |
| Over-reading node size in one frame | Size = activity in that window only. One big node in one frame may be noise or a spike. | Look at size over several frames to see persistent vs one-off influence. |
| Comparing different windows as if same scale | Activity (and thus size) is window-relative; total activity can differ by window. | Focus on relative size within a frame and on patterns over time. |
Caveats & Considerations
- Minimum windows: Needs at least 2 time windows for animation.
- Edge limit: At most 200 edges per window are shown (by weight or kernel logic). Dense graphs are simplified per frame.
- TCEG semantics: Edge presence is driven by TCEG validity windows (when the model treats the link as valid). Window length and indexing depend on how TCEG was built.
- Precomputed frames: All frames are precomputed on the GPU; playback uses that precomputed data. Changing underlying data requires recomputing.
- Performance: Heavier 3D animation and many frames may be slower on some devices or browsers.
- Edge color: Edges are currently drawn in a single gray style. Documentation may describe age-based coloring (e.g., new=green, fading=orange); if your version supports it, use it to read “new” vs “fading” links.