Treatment Group

A treatment group is the share of users who, in an experiment, see the changed version or the new feature, and whose behavior is compared against the control group.

The treatment group is the active half of the experiment: the change the hypothesis claims is applied only here, and the difference between this group's behavior and the control's is what is called the effect. If the experiment has several versions, each version gets its own treatment group and all are judged against one shared control. The main risk is 'contamination': if a treatment-group user also sees the control version through a shared device or cache, the boundary blurs and the effect is diluted. The treatment group must also be large enough that, if an effect exists, it rises above the noise.