Why the treatment window is a design choice
The treatment window is the period during which treatment is applied and outcomes are measured. Longer windows increase statistical power but also raise the risk that other shocks contaminate the comparison. Short windows reduce confounding but may be underpowered.
Seasonality is the main confounder
Marketing outcomes are seasonal: holidays, weather, and competitive cycles create predictable swings in sales and ad effectiveness. If treatment and control windows fall in different seasonal phases, the estimated effect can be biased.
Design-first fixes for seasonality
The strongest protection comes from aligning design, not from modeling.
- Align seasonal phases: compare treated and control windows from the same seasonal period (e.g., holiday-to-holiday across years, or concurrent regions).
- Span multiple seasons: longer windows can average out seasonal fluctuations when timing is not aligned.
When modeling helps
Seasonal fixed effects or detrending with historical patterns can improve precision, but they introduce functional-form assumptions. Use these tools after you have aligned the design, not as a substitute for alignment.
Practical guidance
- Define the window length based on the expected effect horizon and exposure to other shocks.
- If treatment targets a seasonal event, anchor the control period to the same event window.
- Document the seasonal alignment decision in the design protocol.
Takeaway
Treatment windows are not a technical afterthought. They are part of the causal design. The most credible studies align treatment and control on seasonality before turning to modeling adjustments.
References
- Shaw, C. (2025). Causal Inference in Marketing: Panel Data and Machine Learning Methods (Community Review Edition), Section 3.3.2.