Myroslav spent a Wednesday processing card sort results from a moderated study he ran for a student IA project. The interface was a municipal services portal. The working assumption was that users would group services by department, because that is how the municipality organizes them internally. The data said otherwise with polite, consistent force.
What Card Sorting Data Looks Like
Card sorting generates similarity matrices: a grid showing how often participants grouped two items together across all sessions. With 24 participants and 40 cards, the output is not intuitive to read at first. Myroslav used OptimalSort to generate a dendrogram, a tree diagram that clusters items by grouping frequency.
The clearest finding: 83% of participants grouped waste collection, pothole reporting, and water outage notifications into a single category they labeled something like problems or urgent issues. The current site placed these items across three separate departments with no cross-linking. A user reporting a water leak had to know which department handled it to find the form.
The Mental Model Gap
This is the central problem information architecture research exposes. According to research published by the Nielsen Norman Group, navigation labels that reflect internal organizational structure instead of user task orientation consistently produce higher error rates and longer completion times.
Users do not know how your organization is structured. They know what they need to do.
By Thursday, Myroslav had drafted three proposed navigation structures based on the clustering data, each reflecting task-based groupings rather than departmental ones. The next step is tree testing: giving participants a simplified version of the proposed structure and asking them to locate specific items without visual design cues. The card sort told him what users group together. The tree test will tell him whether the structure actually works.