Distinguishing between what you can observe and what you can’t directly measure — like true loyalty vs. transactional convenience.

Structural Equation Modeling (SEM)

SEM distinguishes between what you can observe (manifest variables) and what you can’t directly measure (latent variables) like true loyalty or satisfaction.

! Why This Matters

A customer who appears “loyal” based on transaction volume alone might actually be transactionally loyal — driven by convenience or lack of competition — and may defect as soon as a competitor opens nearby.

  • Emotional Loyalty: Brand affinity, genuine preference, resistant to competitor offers
  • Transactional Loyalty: Convenience-driven, price-sensitive, high defection risk
  • SEM uses simultaneous equations to model how different factors cause loyalty types
  • Path analysis shows causal relationships that simple models miss

Path Workbench

Draw a causal model, then fit it to data. Coefficients are estimated by maximum likelihood, and a χ² test checks whether the drawn structure can reproduce the observed covariances.

1 · Draw the causal DAG

Draw an arrow with the From → To bar, or single-click one node then another on the canvas, or tick targets in a node’s editor (double-click the node). Reverse (⇆) or delete (×) arrows from their chips; cycles are rejected. Double-click a node to rename it (names must match your data’s column names), delete it, or mark it as the target outcome. Drag nodes to arrange the diagram.

Add a directed path: →
Click “+ Add Variable” to create your first variable.

2 · Load the data

Upload a CSV (first row = column names, numeric columns) or load the built-in dataset: 400 customers with satisfaction, convenience, loyalty, and repurchase, generated by a known causal process. N is the number of rows. Click + on a column chip to add it to the canvas.

No data loaded yet.

3 · Fit the model

Try it in 60 seconds

Press Load Default Dataset (400 customers), then click + on all four column chips to put them on the canvas. Draw satisfaction → loyalty, convenience → loyalty, loyalty → repurchase, and convenience → repurchase. Double-click repurchase and tick Is Target. Press Fit Model: the arrows light up with β ≈ 0.50, 0.35, 0.71, 0.18, and the test reads χ²(1) ≈ 1.3, p ≈ .26 — the structure is consistent with the data. Satisfaction’s total effect on repurchase (≈ 0.35) flows entirely through loyalty. Now delete convenience → repurchase and refit: χ²(2) ≈ 30, p < .001 — the data reject that one missing arrow.

🎯
Strategic Implication

Different loyalty types require different interventions. Transactionally loyal customers need barriers to switching; emotionally loyal customers need brand reinforcement — not discounts that might cheapen the relationship.