From Research to ROI: New Method -Longitudinal Data and the Estimation of Dynamic Treatment Effects
- Maria Alice Maia
- Feb 17
- 2 min read
Your new loyalty program launched six months ago. You know the one-month effect on customer spending, but what about the six-month effect? Is the impact growing as customers engage more, or did the initial novelty wear off?
Answering this question—understanding the dynamic effects of your programs over time—is critical for strategy. Yet, the standard methods many teams use are often flawed, a subtle but serious form of "Doing Data Wrong."
Let's take a Consumer Goods company evaluating its new loyalty program.
The Wrong Way (The Contaminated Event Study): The analytics team wants to see the program's impact quarter by quarter. The go-to method is often a two-way fixed effects (TWFE) regression with leads and lags to create an "event study" plot. However, as we've discussed, if the program had a staggered rollout, these dynamic estimates can be contaminated by "forbidden comparisons," leading to a distorted picture of the program's true impact curve. You can't trust the story the model is telling you.
The Right Way (A Simpler, More Robust Approach): So, how do we get a trustworthy picture of how an effect evolves? Recent research from economists like Dube, Jordà, and Taylor offers a powerful and refreshingly direct alternative: the Local Projections approach to Difference-in-Differences (LP-DiD).
Instead of one complex model that tries to estimate the entire dynamic path at once, the LP-DiD method asks a series of simple questions, one for each time horizon:
What is the effect one quarter after a customer joins? (Run a simple DiD for that horizon).
What is the effect two quarters after a customer joins? (Run another simple DiD for that horizon).
What is the effect three quarters after joining? (And so on...).
By running a separate, clean regression for each time horizon, this approach avoids the negative weighting and contamination issues of the standard TWFE event study. When you plot these individual estimates together, you get a clean, reliable "impulse response function"—the true story of your program's impact over time.
As a leader responsible for growth, the "average effect" of a program is far less important to me than its trajectory. At Alura, understanding the long-term engagement curve of new corporate clients was everything. It told us whether we had a sustainable model or a leaky bucket. When I founded NaHora.com, we lived and died by our ability to model the dynamic effects of pricing changes. A single average would have hidden the truth and killed the business.
The LP-DiD method is a perfect example of the research frontier providing simpler, more robust, and more interpretable tools. My mission is to champion these methods that favor clarity over needless complexity. This knowledge is not mine to keep.
If you’re ready to accurately map the long-term ROI of your initiatives, join my movement. Subscribe to my email list for more no-nonsense, research-backed playbooks.
And if you’re struggling to understand how the impact of your programs evolves over time, book a 20-minute, no-nonsense consultation with me. Let's chart the true course of your impact together.
