Add ITS lift testing notebook for MMM calibration

[drbenvincent]

Adds a new notebook demonstrating how to conduct lift tests using Interrupted Time Series analysis when control groups are unavailable (e.g., national-level campaigns).

Key features:

• Simulates realistic marketing data with adstock and saturation effects
• Shows how to run ITS analysis to measure campaign lift
• Extracts mean and standard deviation outputs for MMM calibration
• Calculates absolute lift, percentage lift, and ROI with uncertainty
• Includes guidance on formatting results for PyMC-Marketing integration

This bridges the gap between experimental lift testing and MMM calibration, providing the workflow for generating experimental evidence that can improve attribution models.

---

📚 Documentation preview 📚: https://causalpy--547.org.readthedocs.build/en/547/

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.50%. Comparing base (828ba2e) to head (dc21e9c).
⚠️ Report is 39 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #547      +/-   ##
==========================================
+ Coverage   95.59%   96.50%   +0.90%     
==========================================
  Files          29       31       +2     
  Lines        2681     4173    +1492     
==========================================
+ Hits         2563     4027    +1464     
- Misses        118      146      +28     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[drbenvincent]

@cetagostini I think I should be adding in the channel spend variables into the model formula, right?

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

cetagostini commented on 2025-11-05T07:17:35Z
----------------------------------------------------------------

All good, but two important points:

1. The spend or response should not be scaled. So, you can safely pass everything in the original scale, the model will handle everything under the hood. Doc here: https://github.com/pymc-labs/pymc-marketing/blob/main/pymc_marketing/mmm/multidimensional.py#L1828
2. The impact input should be MEAN not cumulative. Basically, this same values divide by the N length of the experiment.

drbenvincent commented on 2025-11-13T14:11:05Z
----------------------------------------------------------------

Removed mention of normalising the results in cb64c19

drbenvincent commented on 2025-11-13T14:28:55Z
----------------------------------------------------------------

remaining stuff about the mean (lift/week of the experiment) should be done in dc21e9c

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

cetagostini commented on 2025-11-05T07:17:36Z
----------------------------------------------------------------

Good, I'll make a PR to make things clear there as well.

drbenvincent commented on 2025-11-06T09:47:21Z
----------------------------------------------------------------

I've already given that a go in a PR that I think you've seen pymc-labs/pymc-marketing#2061

[drbenvincent]

I've already given that a go in a PR that I think you've seen pymc-labs/pymc-marketing#2061

---

View entire conversation on ReviewNB

[drbenvincent]

Removed mention of normalising the results in cb64c19

---

View entire conversation on ReviewNB

[drbenvincent]

remaining stuff about the mean (lift/week of the experiment) should be done in dc21e9c

---

View entire conversation on ReviewNB