Control Group

A control group is the share of users who, in an experiment, do not see the new change and experience the current version, so they serve as the baseline for comparison.

Without a control group there is no experiment, only a change and a number after it; the control group measures 'what would have happened without the change' at the same time, so the real effect is separated from natural and seasonal variation. The key point is that assignment must be random; if control and treatment come from two different sources or two time windows, their difference can be from that initial gap rather than the change. The common mistake is comparing 'before and after' instead of concurrent 'control and treatment', because a time comparison charges seasonality and outside events to the change.