Grouping hides within-group differences and needs an expand path

What it is

When a group shows only its representative, differences among members—price, condition, chapter, in-stock or not—fold into that one row. An expand path lets people pull every option into view without dismantling the grouping. Without that path, grouping stops being “fewer rows” and becomes “the system already chose.” Search engines’ “omitted similar results,” if they cannot be opened, are deletions.

Expand is about in-group visibility. It is not whether to group, and not whether identity merge should have happened. The representative still holds the list slot; expand is navigation inside the group.

Why it happens

The representative is chosen as the centre of the cluster, not as the carrier of every member’s decision attributes. People see the centre and extrapolate its attributes to the whole group: representative out of stock, whole group out of stock; representative is hardback, paperback is no longer expected. Extrapolation saves time and creates a blind spot. Expand turns extrapolation into a falsifiable move: only on opening does the attribute distribution become visible, and only then does working memory allow “there may be exceptions in the group.”

The path has to be affordable. Buried on a second page after clicking the representative, or present only as an icon with no “N more options” count, discoverability is near zero. People will not hunt a hidden control to test a hypothesis that was never raised. A count and “see all” write “unread options remain in this group” as a visible property of the current row, and expand becomes a possible next step.

Studying it

Put the target on a non-representative member and compare a discoverable expand path, an undiscoverable one, and default full expansion.

• Paradigms: one member of the cluster carries the attribute the task needs, the representative does not; three UIs. Variant picking in retail, “similar coverage” in news, “other versions” in scholarly search are usable settings.
• Independent variables: form of the path (count-as-link / icon only / expanded by default), whether the count is accurate, whether expand keeps the original query context.
• Dependent variables: discovery rate of the non-representative member, rate of believing the group lacks the attribute, return rate to the representative after expand.
• Methodological note: if people are told “click this row, there are more,” the study measures compliance, not discoverability. The path has to be found unprompted. Inflated or deflated counts change the decision to expand; manipulate count accuracy separately.

Where it stops holding

When a group has one or two members and the difference already sits in the representative snippet, a mandatory expand path is noise. When membership runs to hundreds (a global seller list), expand cannot be a flat page; it needs in-group filtering, or the path opens a new overload. Assistive technology users must be able to learn “this row is a group and can be expanded”; otherwise the count is visual-only. Expanding every group by default returns to side-by-side listing. The path is on demand; it is not grouping undone.

Applying it

• Write “N more options,” or an equivalent verb, beside each representative. A click should expand in context, not jump to a new page that drops the query.
• N must match the number of members actually viewable. Do not count items removed by permission or region.
• After expand, collapsing must still be possible, the representative stays in place, and the list must not lose scroll position.
• Verify with a target that exists only as a non-representative member of some group. Unprompted, people should reach it from the count or link. If they cannot, or expand drops the query, or N says 8 while 2 exist, the path has not given in-group differences back.

Related

• Within the group: G3.15.1 Multiple sources of the same entity should collapse to one representative · G3.15.2 Near-identical but distinct results should be grouped, not listed side by side · G3.15.4 Dedup rules must be intelligible so known items do not silently vanish · G3.15.5 Group by the dimension the user is deciding on, not by a database field
• Adjacent: G3.06 Result snippets · G3.08 Filters · G4.03 State persistence
• Search terms: omitted similar results · expand cluster · result group