INF2170H Chap.4 Labels, Categories and Card Sorting
Labels, Categories and Card Sorting
Test boundaries and language
Category criteria define what belongs, while labels communicate the boundary. This chapter compares open and closed card sorting, shows how card wording and granularity can cue a result, and combines final piles with hesitation and think aloud evidence.
It treats difficult placements as diagnostic evidence rather than noise.
Compute without overclaiming
Original sort data support step by step similarity and distance calculations. The matrix and cluster tree summarise co occurrence, but they do not discover a true taxonomy or explain participant reasoning.
Design judgment still turns the evidence into criteria, labels and a next test.
Prepare cards that do not reveal the answer
State the item level with a sentence such as “Each card represents one resource.” Review the whole set for mixed granularity, inconsistent information density and proposed category words.
Participants sort the representation you created, so an image, title or description can manufacture agreement. Give enough context to identify the object while withholding language that instructs the intended pile. Pilot with someone outside the design work and revise both confusing and leading cards.
Open and closed sorts answer different design questions.
In an open sort, participants create and name groups, producing candidate relationships and vocabulary. In a closed sort, the designer supplies categories, allowing the study to test proposed boundaries and labels. Neither method reproduces a real navigation task.
If the decision concerns finding a destination through a hierarchy, build the candidate structure and use tree testing afterward.
Interpret clusters with the session evidence
For a pair of cards, similarity equals together placements divided by valid sorts, and distance equals one minus similarity. Print the numerator, denominator and exclusion rule. A missing comparison should not silently become separation.
Once the matrix is calculated, inspect the session record. Co occurrence can arise from a shared property, a wording cue or a task that participants understood differently.
Final piles are only part of the evidence. Hesitation, requests for duplicate placement, rejected labels and think aloud explanations reveal difficult boundaries.
A recurring group with unstable names may support the relationship while leaving vocabulary unresolved. Define the candidate criteria, test labels separately and preserve alternate access when an item genuinely serves several tasks. The final design remains a reasoned hypothesis, not a taxonomy discovered automatically by the software.
What this chapter covers
- 01
Category criteria
- 02
Label quality
- 03
Open card sort
- 04
Closed card sort
- 05
Representation bias
- 06
Similarity and distance
Worked example · free
Calculate card pair distance
- 2Set the valid denominator.
- 2Compute similarity.
- 2Compute distance.
- 2State the evidence boundary.
Key terms
- Open Sort
- Open Sort lets participants create and label their own groups.
- Closed Sort
- Closed Sort asks participants to place cards into supplied categories.
- Pair Similarity
- Pair Similarity divides together placements by valid sorts for two cards.
- Pair Distance
- Pair Distance subtracts pair similarity from one.
Labels, Categories and Card Sorting FAQ
When is an open card sort appropriate?
Use it to explore candidate groupings and participant vocabulary when the category model is still uncertain. Keep the proposed labels off the cards, use comparable item representations and collect explanations. The output is generative evidence, not a finished menu.
What can a closed card sort test?
A closed sort can test whether participants associate items with proposed categories and whether labels communicate their boundaries. It does not reproduce a real navigation task or evaluate visual presentation. Use tree testing later when the question concerns paths through a hierarchy.
Why can a dendrogram be misleading?
A dendrogram summarises distances derived from co occurrence, but it does not reveal a true taxonomy, explain why a participant grouped items or settle a label. Inspect the printed sorts, denominator, card wording and qualitative session notes before recommending a structure.
Assessment move
Prepare a small original card universe, audit every description for cues and keep the item level consistent. Compute selected pair similarities and distances by hand, then use session explanations to qualify the pattern. Compare open, closed and tree testing claims. Start every exercise with a sentence that states what each card represents. Remove proposed category terms from titles and keep description length comparable.
Pilot the cards for comprehension before treating any placement as evidence. During analysis, calculate a few pair similarities from the raw sorts, write the valid denominator and convert each result to distance. Check omitted cards rather than treating missing comparisons as separation. Place the numerical pattern beside participants' labels, hesitations and requests for duplicate placement.
Draft category criteria from the combined evidence, then test whether the proposed label communicates the same boundary. Rehearse method choice by changing the question: use open sorting to generate relationships, closed sorting to examine proposed groups and tree testing to evaluate routes. End with a limited claim tied to the participant group and card universe.
A recurring pile is a candidate structure, not proof of a universal mental model, and a cluster diagram never replaces interpretation.
