How to extract longitudinal trends of a brick-and-mortar brand's foot traffic
summary
Built a Macro Model that strips technology and panel-change noise from app-derived foot-traffic, using a proprietary panel of 2 million users and ~600 million visits/year. Method computes month-to-month changes inside stable cohorts (apps / OS / SDK), weights them by cohort size, and cumulatively reconstructs a longitudinal series. That approach removes Simpson effects (simpson's paradox), turning a spurious 23% YoY jump into a plausible 7% growth and produces materially more stable year‑over‑year trends.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 2:22Foot traffic
- 2:50Why longitudinal trends are important
- 3:44Our panel
- 4:47Apps
- 5:24Privacy
- 7:40The Core Problem
- 10:03Simpsons Paradox
- 11:20Example
- 16:01The model
- 18:03Delta Y
- 21:53Most time consuming
- 25:10Technology impact
- 27:03Summary
- 29:23Questions
- 31:38iOS vs Android
related talks
34mSuperstar Session: The State of Mobile Gaming 2020
38mContinuous Marketing Effectiveness Measurement
29m5 Simple Steps to Create Meaningful Features from Clickstream Data
39mBuilding Causal Measurement
32mInternet Trends
23mGrowth Hacking - Product Analytics at Scale using R and RStudio
29mUser Motivation: Refining Customer Segments with Location
26mThe Need for People Based Measurement