From Research to ROI: Case Study - Geographic Boundaries as Regression Discontinuities
- Maria Alice Maia
- Dec 23, 2024
- 2 min read
You launched a bold new sales incentive program in specific cities, but not in their surrounding suburbs. A quarter later, sales in the cities are up 15%.
Is the new program a brilliant success? Or are you just measuring the fact that city markets are different from suburban ones?
This is a classic "Doing Data Wrong" scenario for a Sales Department. A simple comparison between a "treated" city and an "untreated" suburb is hopelessly confounded. The customer bases are different, economic activity is different, and team history is different. You're not measuring your program's impact; you're measuring the difference between two markets.
But what about the sales territories that are right on the border? The territory just inside the city line versus the one just outside.
This is where a powerful causal method is hiding in plain sight: the Geographic Regression Discontinuity (GRD) Design.
As outlined in research by Keele & Titiunik, a geographic boundary can act like the sharp cutoff in an RDD. It creates a "natural experiment" by splitting otherwise similar units into treated and control groups based on their location. By comparing the performance of sales reps in a very narrow band just inside and just outside the city limits, we can isolate the true causal effect of the new incentive, stripping away the broader market differences.
However, geography is tricky. Applying this method requires more than just running a standard RDD. It demands a higher level of strategic thinking.
Here are two critical challenges you must address:
The Sorting Problem: Unlike a test score, people can precisely choose which side of a boundary to live or work on. Did your top-performing reps maneuver to get the "treated" city territories? A GRD is only valid if the units on either side of the border are comparable. You must test this by checking for discontinuities in pre-program covariates (like past performance, team tenure, etc.) along the boundary.
The Compound Treatment Problem: What else changes at that city border? Tax laws? Local regulations? Competing store density? If your new incentive isn't the only thing that changes at the boundary, you can't isolate its effect. The best GRD analyses find segments of a border where only the treatment of interest changes, holding all other factors constant.
As a leader who has designed sales territories and go-to-market strategies, I know that business boundaries are rarely random. But by rigorously testing our assumptions and carefully selecting the right segments of a boundary, we can transform these operational lines into powerful sources of causal insight.
My mission is to translate these powerful frameworks from the academic frontier to help you solve your most complex business challenges. This knowledge isn't mine to keep.
If you’re ready to move beyond simple geographic comparisons and learn to find the natural experiments hidden in your business, join my movement. Subscribe to my email list.
And if you’re trying to measure the impact of a location-based strategy, book a 20-minute, no-nonsense consultation with me. Let’s map out a rigorous approach together.
