Case Study: Conditional Cash Transfers & Children's Outcomes using an RCT
- Maria Alice Maia
- Mar 10
- 3 min read
You’re spending millions on programs, but can you prove they actually work?
I’m not talking about pretty dashboards showing a “before and after.” I’m talking about undeniable, causal proof. Most companies can't provide it. They’re stuck in what I call "Kindergarten Data": looking at two different groups and drawing dangerously wrong conclusions.
Let's be direct. This isn't just bad science; it's a colossal waste of money and opportunity.
The "Doing Data Wrong" Scenario: The Education Group Fiasco
Imagine an education company launches a scholarship program for high school students with a grade average of 89% or lower, hoping to boost their performance. A year later, they compare the grades of students who got the scholarship (the <=89% group) with those who didn't (the >89% group).
They find the scholarship group's grades are still lower. The program is declared a failure and funding is cut.
What went wrong? Everything. This was never a fair fight. You were comparing a group that was already struggling with a group of high-achievers. You didn’t measure the program's impact; you just confirmed that students with lower grades have... lower grades. This is correlation, not causation. And it’s the kind of basic error that kills promising initiatives.
The "Right Way": Finding Truth at the Tipping Point
So how do we fix this? We stop comparing apples to oranges and get surgically precise. We can use a powerful causal inference method called Regression Discontinuity Design (RDD).
Instead of looking at everyone, RDD focuses intensely on the cutoff, the tipping point. In our case, the 89% grade threshold.
Think about it: a student with an 88.9% and one with a 89.1% are virtually identical in terms of background, talent, and motivation. The only significant difference between them is that one just barely got the scholarship and the other just barely missed out.
By comparing the future success of ONLY these students right at the edge of the cutoff, we can isolate the true, causal effect of the scholarship. We’re not guessing anymore. We’re measuring with rigor.
This isn't just a textbook theory. Cutting-edge academic research uses this exact method to answer tough questions. A brilliant 2024 study in the American Political Science Review used RDD in Brazilian municipal elections. By comparing mayors who won by a razor-thin margin to those who barely lost, the researchers proved that female mayors—despite having fewer resources—are far more effective at building inclusive, resilient political parties by reducing the gender gap in membership. They found the cause, not just a correlation.
What to Do Now:
Managers: Stop accepting kindergarten data. Start asking your teams: "Where is the cutoff rule in this program? Can we use a Regression Discontinuity Design to prove it caused this outcome?" Demand causal answers. Your budget depends on it.
Tech & Data Professionals: This is your moment to be a hero. Move beyond building dashboards. Find the natural experiments in your business—eligibility thresholds, sales quotas, discount tiers—and propose an RDD analysis. You won’t just be reporting data; you’ll be delivering the ground truth that shapes strategy and creates real value.
My entire career, from founding a tech company to leading growth at giants like Itaú and Stone, has been about one thing: generating tangible results. The insights I've gathered from FGV, Berkeley, HEC, and Harvard aren't mine to keep. My purpose is to share them so we can all stop "doing data wrong."
If you’re tired of data theater and want to be part of a community dedicated to unlocking real value with no-nonsense, research-backed methods, join my private email list. Let’s fix this, together.
