Description
Builds personas from dimension reduction of user survey data, using component loadings or correspondence-analysis clusters as the statistical core of each archetype.
Sinha argued that persona practice favours precision (rich detail) over accuracy (representative users). Her remedy surveys users on the importance of many dimensions of a domain, runs principal components analysis, and gives designers the rotated loadings as the motivational core of each persona; demographics flesh them out and targeted interviews verify them. Kim and Wiggins applied the same logic to service importance ratings, assigning each respondent to a factor by factor score and cross-tabulating with demographics. Where data are categorical or coded from open answers, multiple correspondence analysis (MCA) replaces PCA: cases, variables and categories are plotted in one space, discrimination measures show which variables drive each dimension, and visible clusters of cases become segment hypotheses, with demographics projected as supplementary variables. In every variant the statistics propose segments; designers and further qualitative research turn them into personas.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Latent
- Stage
- Discovery, Definition
- Data
- Mixed
- Effort
- Weeks
Steps
- List the dimensions of the domain (Sinha used 32 for a restaurant finder) from competitor review and a few exploratory interviews, and add demographic and behaviour items.
- Field the questionnaire with importance or satisfaction ratings on a 5-point scale, aiming for about ten responses per variable (Sinha had 63 respondents, Kim and Wiggins 435).
- Check factorability with the Kaiser-Meyer-Olkin measure and Bartlett's test of sphericity; for nominal or coded data, build a cases-by-categories matrix for MCA instead.
- Extract components and compare four- to seven-component solutions with varimax or equamax rotation, keeping the most interpretable (Sinha kept five explaining 53% of variance); for MCA choose dimensions from inertia and Cronbach's alpha.
- Save component scores or MCA object scores, assign each respondent to their highest-scoring component or visible cluster, and cross-tabulate with demographics.
- Give designers the loadings or discrimination measures with a short explanation; they build each persona around primary and secondary motivations, merging components that are not distinct.
- Test stability by removing outlier clusters or re-running with different variable sets, and watch for implausible clusters that signal coding errors or social-desirability bias.
- Recruit high scorers on each component for interviews and observation to verify and enrich the personas.
Sources
- Persona Development for Information-rich Domains (Sinha, 2003)
- A Factor Analysis Approach to Persona Development using Survey Data (Kim & Wiggins, 2016)
- Using Correspondence Analysis to Monitor the Persona Segmentation Process (Laporte et al., 2012)
Also known as
- Factor-analytic persona development from information-need ratings
- Factor-analytic persona development from service rating surveys
- PCA-based persona development
- Survey plus PCA personas
- Needs-based statistical personas (Sinha)
- Factor analysis personas
- Multiple correspondence analysis for persona segmentation
- MCA persona segmentation
- Correspondence analysis personas
- Homogeneity analysis for personas
Categories
- Analytics & Behaviour
- Statistics & Modelling