Experimental Design for GEODifference-in-Differences, Time Series, and Synthetic Controls

How matched changes over time, intervention discontinuities, and weighted controls create stronger field evidence when randomization is unavailable.

Difference-in-Differences compares changes, not levels

DiD subtracts the control group’s before-to-after change from the treated group’s change. It can remove shared platform or seasonal movement when both groups would otherwise have followed comparable trends.

Parallel trends is the key assumption

Before treatment, treated and control units should move similarly. If their trajectories were already diverging, the post-treatment difference may continue an existing pattern rather than identify the intervention.

Interrupted time series estimates level and slope changes

With enough repeated periods before and after a clearly timed intervention, an interrupted time series can test whether the outcome shifts immediately, changes trajectory, or both. Engine releases and other simultaneous events remain important confounders.

Synthetic controls combine several untreated comparators

When no single control matches the treated unit, a weighted combination of untreated pages, markets, themes, or brands can approximate its pre-treatment trajectory. The weights and donor pool must be disclosed, and placebo-unit tests should assess whether the apparent effect is unusual.