Split by Date

Isolating Specific Time Periods

Split by Date creates separate variables for different time periods, allowing you to model how effectiveness changes between periods or isolate specific campaigns.

---

What is Split by Date?

Creating Time-Specific Variables

Takes one variable and splits it into two (or more):

Original:

TV_Spend: Values for all 104 weeks

After Split:

TV_Spend_2023: Values for 2023 only (zeros elsewhere)
TV_Spend_2024: Values for 2024 only (zeros elsewhere)

Use Case: Test if TV effectiveness changed between years

---

When to Use Split by Date

Scenario 1: Campaign Period Isolation

Problem: Want to measure specific campaign separately from baseline

Example:

TV_Spend_Holiday_Campaign (Nov-Dec only)
TV_Spend_Baseline (all other months)

Model:

Sales ~ TV_Holiday + TV_Baseline

Result: Compare holiday vs. baseline TV ROI

---

Scenario 2: Structural Changes

Problem: Business changed significantly mid-period

Examples:

• Product reformulation

• Market expansion

• New competitor entry

• Pricing strategy shift

Solution:

Digital_Pre_Change (before event)
Digital_Post_Change (after event)

Model: Test if digital effectiveness changed

---

Scenario 3: Test vs. Control Periods

Problem: Ran experiment during specific period

Example:

TV_Test_Period (Weeks 20-32)
TV_Control_Period (all other weeks)

Analysis: Did test period show different effectiveness?

---

Scenario 4: Seasonal Effectiveness

Problem: Channel may work differently by season

Example:

Radio_Summer
Radio_Winter

Model: Compare seasonal ROI

---

Creating Split Variables

In Variable Workshop

Step 1: Select base variable to split

Step 2: Choose "Split by Date"

Step 3: Define split criteria:

Option A: Date Range

Start Date: 2023-11-01
End Date: 2023-12-31
Name: TV_Spend_Holiday_2023

Option B: Custom Conditions

Include weeks where: Holiday = 1
Name: TV_Spend_Holiday_Weeks

Step 4: Preview split (see which periods included)

Step 5: Create variable

---

Multiple Splits

Create complementary variables:

1. First split: TV_Q4 (Oct-Dec)

2. Second split: TV_NonQ4 (Jan-Sep)

Together: Cover entire time period

Model: Compare coefficients

---

Naming Convention

Recommended formats:

{Variable}_{Period}
{Variable}_{Condition}

Examples:
TV_Spend_Q4
TV_Spend_2024
Digital_Holiday_Campaign
Radio_Summer_Months
TV_Post_Rebrand

---

Interpretation

Separate Coefficients

Model:

Sales = β₀ + β₁×TV_Q4 + β₂×TV_NonQ4

Coefficients:

• β₁ = 0.8 → TV in Q4 generates $0.80 per $1 spent

• β₂ = 0.5 → TV in other quarters generates $0.50 per $1

Insight: Q4 TV is 60% more effective (seasonal boost)

---

Common Patterns

Before/After Analysis

Test structural change impact:

TV_Before_Rebrand
TV_After_Rebrand

If β_after > β_before: Rebrand improved TV effectiveness

---

Seasonal Quarters

Create 4 variables:

TV_Q1, TV_Q2, TV_Q3, TV_Q4

Model all simultaneously: See which quarter has highest ROI

---

Campaign Flight Isolation

Isolate 6-week campaign:

TV_Campaign_Flight (Weeks 15-20)
TV_Baseline (all other weeks)

Compare ROI: Was campaign more effective than baseline?

---

Best Practices

✅ Clear rationale for split - don't split arbitrarily

✅ Meaningful periods - business-driven boundaries (campaigns, seasons, events)

✅ Complementary splits - ensure full coverage (Period A + Period B = Total)

✅ Sufficient data in each split - at least 10-15 observations per period

✅ Test one split at a time - don't over-complicate initially

❌ Don't split too granularly - avoid 1-2 week periods (too sparse)

❌ Don't create too many splits - limit to 2-4 per variable

---

Combining with Other Transformations

Split THEN Transform

Example: Holiday TV with Adstock

1. Split: TV_Holiday (Nov-Dec only)
2. Apply adstock: TV_Holiday_ads60
3. Use in model

Captures: Holiday campaign with proper carryover, isolated from baseline

---

Summary

Key Points:

📅 Splits variable by time period - creates period-specific variables

🎯 Use for campaigns, seasons, structural changes - test effectiveness differences

✂️ Complementary splits - Period A + Period B = complete coverage

📊 Separate coefficients - compare ROI across periods

💡 Business-driven boundaries - align with real campaigns or events

Powerful tool for understanding how marketing effectiveness varies across time!