Constructing a meaningful never-deplatformed control group for causal analysis

Determine whether a control group consisting of influencers who have never been deplatformed can be identified and justified as a meaningful comparison group for difference-in-differences estimation of deplatforming’s impact on online attention, and develop a concrete procedure to obtain such influencers while ensuring comparability to treated units under the parallel-trends assumption.

Background

To estimate the causal effect of deplatforming, the paper employs a stacked difference-in-differences design that compares deplatformed influencers to yet-to-be-deplatformed influencers, avoiding biases from variation in treatment timing.

The authors explicitly avoid using never-deplatformed influencers as controls due to uncertainty regarding their suitability and the practical challenge of identifying them. Resolving this uncertainty would enable broader causal designs by introducing never-treated controls where appropriate.