Creating Interaction Terms
Multiply Variables creates interaction terms by multiplying two variables together, allowing you to test whether marketing channels work synergistically.
What Are Interaction Terms?
Beyond Additive Effects
Standard Linear Model (Additive):
Copy Sales = β₀ + β₁×TV + β₂×Digital Assumes: TV and Digital effects are independent and simply add together
Interaction Model:
Copy Sales = β₀ + β₁×TV + β₂×Digital + β₃×(TV × Digital) Tests: Do TV and Digital work better (or worse) together than alone?
When to Use Interaction Terms
Synergy Testing
Scenario: TV builds awareness, Digital captures intent
Hypothesis: TV + Digital together more effective than sum of each
Solution:
If β₃ > 0: Positive synergy (amplification effect) If β₃ < 0: Negative synergy (cannibalization)
Cross-Channel Effects
Test combinations:
TV × Radio (similar messaging across channels)
Brand × Performance (awareness + conversion)
Online × Offline (digital-physical synergy)
Creating Multiplication Variables
In Variable Workshop
Step 1: Select first variable (e.g., TV_Spend)
Step 2: Choose "Multiply Variables"
Step 3: Select second variable (e.g., Digital_Spend)
Step 4: Preview multiplication result
Step 5: Name new variable: TV_x_Digital
Step 6: Create
Positive Coefficient = Synergy
Example Model:
At low spend:
Interaction: $10K × $5K × 0.001 = $50
At high spend:
Interaction: $50K × $30K × 0.001 = $1,500
Insight: Channels amplify each other at higher investment levels
Negative Coefficient = Cannibalization
Example:
Interpretation: Channels competing for same audience, diminishing combined effect
Common Use Cases
Geographic Overlap
Seasonal Interactions
✅ Use with transformed variables - Apply adstock/saturation first, then multiply
✅ Test one interaction at a time - Don't add many interactions simultaneously
✅ Require sufficient data - Interactions need more observations to estimate reliably
✅ Business hypothesis driven - Only test theoretically meaningful interactions
❌ Don't multiply everything - Most interactions are non-significant
❌ Watch for multicollinearity - Interactions correlated with main effects
Key Points:
✖️ Multiply = Interaction term - tests synergy/cannibalization
🤝 Positive coefficient = synergy - channels amplify each other
⚔️ Negative coefficient = cannibalization - channels compete
📊 Requires more data - interactions harder to estimate than main effects
🎯 Hypothesis-driven - test specific, meaningful combinations
Powerful tool for understanding how marketing channels work together!
Last updated 4 months ago