Description
Present items briefly or degraded, count which are mistaken for which, symmetrise the confusion counts and cluster them to find families of confusable items.
How often people mistake one stimulus for another is a direct behavioural measure of perceived similarity. Items such as icons, symbols, sounds or digits are presented repeatedly under difficult conditions (brief exposure, small size, noise) and responses are tabulated in a confusion matrix. The matrix is asymmetric, so it is made symmetric, for example by combining the i-for-j and j-for-i counts, and then clustered. The resulting tree shows families of confusable items; Levine's digits split into curved and straight-line families, with 0 and 8, and 2 and 7, most confused. The method was used to design signals and controls operators must not mistake. Unlike the machine-learning confusion matrix used to evaluate classifiers, this one measures human perception, and it needs no rating task, only identification.
Attributes
- Discipline
- Cognitive Psychology & Perception
- Loss type
- Extinct
- Stage
- Design, Evaluation
- Data
- Quantitative
- Effort
- Days
Steps
- Choose the stimulus set (icons, symbols, audio cues or labels).
- Present each stimulus repeatedly under difficult conditions, such as 4-20 ms exposures, small size or noise, and record the identification response.
- Tabulate the full confusion matrix of stimulus by response.
- Symmetrise it, checking first that the direction of confusion does not matter for the goal.
- Cluster the similarity matrix (for example with complete linkage) and draw the tree.
- Redesign members of the most confusable clusters and retest.
Sources
- Cluster Analysis for Researchers (Romesburg, 1984)
Also known as
- Clustering stimulus confusions
- Confusability analysis
- Symmetrised confusion matrix
Categories
- User Research
- Cognitive Psychology & Perception