Core Concept: Regression Discontinuity Designs (RDD) - Intuition & Sharp RDD
- Maria Alice Maia
- Dec 9, 2024
- 2 min read
Your 'Gold Tier' loyalty program is full of high-spending customers. Did the program make them high-spenders, or did you just give perks to the people who already were?
This is a fundamental ROI question for any Consumer Goods company, and it’s where a very common “Kindergarten Data” mistake happens.
The Wrong Way (Naive Comparison): To get Gold Tier status, a customer must spend $1,000 in a year. Your team compares the average spending of all Gold members next year to all non-Gold members. The Gold members spend far more. You celebrate.
You've just celebrated a meaningless correlation. Customers who spend over $1,000 are, by definition, your best customers. They were already going to spend more. This analysis is hopelessly confounded by selection bias.
The Right Way (Finding the Natural Experiment): But what about the customer who spent $1,001 versus the one who spent $999?
These two people are almost identical in their motivation, loyalty, and income. The only meaningful difference is that one just barely qualified for the treatment, and the other just barely missed it.
This is the magic of Regression Discontinuity Design (RDD), one of the most powerful quasi-experimental methods in our toolkit. The spending threshold ($1,000) isn't just a rule; it creates a local randomized experiment hiding in your data. The customer at $999 is the perfect counterfactual for the customer at $1,001.
A Sharp RDD analysis, where the rule is deterministic, doesn't compare all treated vs. all control. Instead, it plots the outcome (next year's spending) against the 'running variable' (last year's spending) and looks for a clean jump, or discontinuity, right at the cutoff point. That jump is the true, local average causal effect of your program.
But this only works if customers can't perfectly manipulate their spending to land just over the threshold. A critical first step in any RDD analysis is to test for this by checking the distribution of customers around the cutoff. A suspicious "bunching" of people at $1,000 invalidates the design.
As an executive and consultant, I’m always hunting for these natural experiments created by business rules. Sales quotas, eligibility criteria, pricing tiers—they are everywhere. Learning to spot and analyze them is a massive strategic advantage. It allows you to find causality in the wild, without the cost and complexity of a full-blown RCT.
My passion is translating these powerful research designs, laid out in seminal works like those of Lee & Lemieux, from academia to the real world of business. This knowledge isn't mine to keep.
If you’re ready to stop being fooled by selection bias and learn how to find the hidden experiments in your own data, join my movement. Subscribe to my email list.
And if you think you have a business rule that could be an RDD, book a 20-minute, no-nonsense consultation with me. Let’s see if we can measure its true impact.
