Description
Finds hidden user types from patterns of categorical survey answers, giving probabilistic, statistically tested segments.
Cluster analysis on survey data is popular for personas but has no statistical test for the number of groups and handles categorical answers poorly. Latent class analysis assumes an unobserved categorical variable (the user type) explains patterns in responses, and estimates both how many classes exist and each person's probability of belonging to each. Model fit statistics (BIC, entropy) guide the number of classes, and class profiles describe how each type answers. Sauro and Lewis position it as the advanced route to persona development, above k-means clustering in rigour.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Unmigrated
- Goal
- Segment users
- Stage
- Definition
- Data
- Quantitative
- Effort
- Weeks
Steps
- Survey a large sample (several hundred) on categorical behaviours, needs and attitudes.
- Fit LCA models with 1 to k classes.
- Choose the number of classes using BIC, bootstrap likelihood tests and interpretability.
- Profile each class by its response probabilities and name it.
- Validate classes against behaviour data and build a classification tool (for example with discriminant analysis) to type new users.
Sources
- Factor and Discriminant Analysis (Sauro & Lewis, MeasuringU training)
Also known as
- LCA
- Latent class segmentation
Categories
- Analytics & Behaviour
- Statistics & Modelling