Sometimes you can’t run a clean A/B test, you just have a billboard that’s already up and a question to answer. Correlation Isn’t Enough investigates whether a regional billboard campaign in Porto Alegre actually caused customer deposits to rise, in a setting where randomized testing wasn’t an option because the intervention was offline and city-wide.
The causal question: Did the campaign in Porto Alegre increase deposits compared to what would have happened without it? Answering that means estimating a counterfactual, the deposits that never happened, rather than just eyeballing a before-and-after chart.
The data: Panel data across two Brazilian cities and two time periods. Porto Alegre is the treatment group that received the campaign, Florianopolis is the control group that didn’t, with May as the pre-intervention period and July as the post-intervention one.
The method: Difference-in-Differences is the core strategy, comparing trends across both groups and time to isolate the campaign’s effect. That’s backed by OLS regression with time, treatment, and interaction terms to attach real statistical inference, standard errors, p-values, and confidence intervals, to the estimate. Naive before-after and treated-versus-control comparisons are included too, mostly to show how badly they mislead you.
Built with Python.