PADI · Information Architecture Research

Open Card Sort — Potential Divers

01 Executive summary

02 Method & analytical approach

This was an open card sort (participants invented and named their own groups) of 39 PADI content cards, run with a panel of self-identified potential divers. Each participant sorted the same 39 cards into as many or as few groups as they wished and labelled each group.

How the numbers are built

Standardisation of category labels

03 Participants & data quality

04 Agreement matrix

Each row is a card; each column is a standardised category. The number is the percentage of participants who filed that card under that category. The solid cell in each row is the winning (most popular) placement.

How to read it. Read one row at a time. A row with a single strong solid cell and little else is a confidently-placed card — people agree where it goes. A row with colour spread across several columns is a contested card whose label needs work. Rows are grouped by their winning category, so the dominant themes read top-to-bottom. Hover any cell for the exact count; click a card name to isolate its row; use the controls to sort or filter.
Agreement matrix
Card × standardised category · solid = most popular placement

What it shows

05 Category mapping (full audit)

How to read it. Labels are grouped under the standardised category they were mapped to. The number beside each label is how many participants used that exact wording. The category totals here match the column totals in the agreement matrix above.

Why this matters

The agreement matrix is the only view that depends on label wording. Publishing the mapping in full means the client can verify or challenge any single classification — the similarity matrix, dendrogram and clusters are computed purely from which cards were grouped together and need no such standardisation.

06 Similarity matrix

A staircase of every one of the 741 card pairs. Cards are ordered so the most-related sit next to each other, so the natural groups appear as dark blocks running down the steps.

How to read it. Each square is the percentage of participants who placed those two cards in the same group; darker means stronger. Follow the right-edge labels downward — where consecutive rows stay dark you are inside a cluster (the coloured chip marks which of the six). Pale runs between dark blocks are the cluster boundaries. Hover a square for the exact pair and value; click a card label to highlight all of its relationships.
Similarity matrix — full lower triangle (741 pairs)
Card × card co-occurrence · ordered by dendrogram · chip colour = cluster
LowerHigher Clusters:

07 Dendrogram

The hierarchical clustering tree. Cards that merge further to the left were grouped together by more participants; the six coloured branches are the groups used throughout this report.

How to read it. Each card starts as a leaf on the left and merges rightward into larger groups. The horizontal position of a join = the distance at which it happens, where distance 0 means "always grouped together" and 1 means "never". A vertical cut through the tree defines your number of groups. Hover a card name to trace its full path up the tree; hover a join to see the agreement level at which it forms.
Dendrogram — average linkage · cophenetic r = 0.83
Distance = 1 − co-occurrence · the dashed line marks the six-cluster cut

08 The six groups, in full

Every card assigned to each cluster, with the cluster's internal agreement (average co-occurrence among its members).

09 Card-level statistics

Every card with its winning category and placement confidence, how many distinct categories it landed in (a measure of ambiguity), and its single strongest partner card.

Click any column header to sort. Cards appearing in many categories with low confidence are the ones needing attention.

10 All card pairs

The complete pairwise co-occurrence table — all 741 pairs, sortable and searchable — so every relationship in the analysis is auditable.

11 Ambiguous & homeless cards

The cards most likely to cause hesitation in live navigation — their strongest single relationship with any other card is weak, and the agreement matrix shows their placement spread across several categories.

12 Recommended information architecture

*Items marked with an asterisk are ambiguous in the data and should be confirmed by tree testing before launch.

13 Conclusions & next steps

14 Appendix — raw participant data