Analysis & results

Open a study and go to its results to explore how participants sorted. All analysis is computed live in your browser, so sliders and filters update instantly.

Recruitment funnel

The Results tab has five views: Overview, Analysis, Placements, Participants and Report. The five chart views (similarity matrix, dendrogram, 3D clusters, suggested groups and standardization) all sit under Analysis, as a second row of buttons. The Include and Segment filters sit under the tab row and apply to whichever view you are on. Every study type's Results tab has this same shape: Overview, Analysis with its charts as a second row, the type's own data view, Participants and Report, with a Download row (Excel and raw data) just under the tabs. The view you are on is kept in the page address, so a reload or a shared link lands on the same view.

When your study has a screener, the Overview view shows a recruitment funnel: how many people started, how many were screened out, how many hit a quota full, how many passed the screener, and how many completed. It also breaks down screen-outs by which question sent people away, so you can see whether a particular qualifying question is turning away more people than you expected.

Overview & filters

The top of the results view shows: how many participants are in the current view, how many were set aside for quality, the median sort time, and the average number of groups per participant.

Two filters control everything below them:

Include flags: toggle ok and rejected responses in or out of the analysis, so a low-effort response can be dropped with a click. There is no middle "suspect" rating: a response is either fine or it is set aside, and softer signals appear on the Participants list as reasons to take a second look rather than as a verdict.
Segment: narrow analysis to people who chose a particular screener answer (single/multiple-choice questions), to compare segments.

Filters are live-only: they reset on reload and don't change your data.

Similarity matrix

A card-by-card heatmap. The darker a cell, the more often those two cards were placed in the same group. Cards are ordered by clustering so related cards sit together. Hover any cell for the exact pairing and percentage. Export it as CSV.

Dendrogram (clustering tree)

Shows how cards merge into groups, from individual cards up to the whole set.

Cluster slider: one slider sets how many groups to split into, from 1 (everything together) up to one group per card. As you drag it, a live average within-group agreement % updates so you can see how tightly the cards cohere at each number of groups (it climbs as you add more, smaller groups). This count is shared with the 3D view and Suggested groups.
Linkage method: Average agreement (the default; balanced) or Strongest pair (single-linkage; useful with smaller samples).
A legend lists each cluster's color, your saved name, and its cards.

3D cluster view

The same clustering shown as an interactive 3D scatter of the cards: drag to rotate, scroll to zoom, click a point to focus a cluster. A sidebar lists clusters by name and card count, and you can name groups with Graham here too.

Suggested groups

The dendrogram cut into your chosen number of groups, shown as editable cards. Each group shows its cards, a within-group agreement % (green ≥60%, amber 30 to 60%, red <30%), and the labels participants actually used for it. Rename any group inline. Names are saved by the set of cards in the group, so a group keeps its name as you adjust the slider, and the name is shared across the dendrogram, 3D view, and report. Click Have Graham name groups and in one pass he gives every group a short, distinct name and a one-line description of what it is about, both saved with the group and shown beneath its name. Each group gets its own name, so you never see two groups labelled the same. On paid plans you can also download the groups as CSV from here. These named, described groups are what you can port straight into a pairwise study (see the pairwise article).

Popular placements

For closed/hybrid sorts (and standardized open sorts): a card × category table showing how many participants put each card in each category, with a "times sorted" total per card. Darker means more agreement. Up to 14 categories shown; export as CSV.

Standardizing created names (open/hybrid)

Participants invent their own group names, so you'll see many variations of the same idea. The standardization grid lists every raw name with how often it was used and its quality flag, and lets you map them to a single standardized name. Standardizing rolls those responses together everywhere else (popular placements, clustering, report). The grid shows names only from the responses currently in your analysis, so setting a junk response aside or filtering it out clears its group names off this screen instead of leaving a wall of them to wade through.

Participant management

See each response individually: effort flag and reasons, duration, and the groups they made.

A card sort response is judged by seven checks, and any one of them sets it aside:

Pace: placements made faster than a person could read the cards.
Too many cards too fast: a large share of individual cards placed faster than they could be read.
Block dumping: the groups mirror contiguous blocks of the shuffled order, with little sign of decision-making.
Too few groups: a large deck sorted into only one or two groups, on a study that asked for more.
Giant pile: most of the cards placed in a single group.
Over-splitting: nearly every card left alone in its own single-card group, which is clicking through rather than sorting (the opposite of the giant pile). Only on open/hybrid sorts, where participants could make their own groups.
Under the minimum time: finished well under the minimum the study set.

The few-groups, giant-pile, over-splitting and minimum-time checks are driven by the study's Response quality settings. Group names are judged separately by a category-quality flag, which sets a response aside for gibberish or foul language.

Two signals flag without setting anything aside: somebody who went far faster than the study's own median, and (on studies recorded before the over-splitting check existed) somebody who left nearly every card in its own group. Either moves up the list with that reason against them, and nothing happens to their data or their payment.

You can also set a response aside yourself, by hand, from the Participants list. It takes that response out of the analysis, the standardize screen and the report, but leaves its data and any panel payment untouched and keeps it on the list, marked "Set aside", so one click Restores it. This is your own judgement call, separate from the automatic quality checks, and always reversible.

Replay a sort as a timeline (where movement data was captured).
Export one participant's data as JSON (for data-subject requests).
Delete a response: permanent, and it's removed from all charts and counts immediately.