Subgroup effects

Split the affected population into named, identifiable groups and read each group's result separately. An average can stay strong while one group's outcome reverses entirely. You are looking for where the direction of the effect flips, not just where it weakens.

Split the affected population into named, identifiable groups and read each group's result separately. An average can stay strong while one group's outcome reverses entirely. You are looking for where the direction of the effect flips, not just where it weakens.

Example

You cut the course's 82% completion rate by device, prior experience, language, and assistive-tech use. Four groups sit between 79% and 88%. Screen-reader users complete at 0%, all nine of them out by week one.