Description
Models the probability that an interaction fails from design and context variables, giving a curve that shows the size or time needed for an acceptable error rate.
Instead of reporting a single error rate, logistic regression models log(p/(1-p)) = b0 + b1x1 + ..., where p is the probability of error and the predictors are target size, distance, glove use, time pressure, user experience and so on. The fitted curve gives the predicted error probability at any design value, with confidence intervals, and identifies which factors matter once others are controlled. Mixed-effects logistic models add random intercepts for participants and items, which is essential because repeated trials from one user are not independent. The design team can then read off the target size required for, say, P(error) below 1%. Rigour comes from sufficient events (at least 10 errors per predictor), checking calibration, and avoiding extrapolation beyond tested ranges.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Latent
- Stage
- Design, Evaluation
- Data
- Quantitative
- Effort
- Weeks
Steps
- Specify the binary outcome (error or not) and candidate predictors (size, distance, condition, user group, fatigue).
- Collect trial-level data across a range of predictor values; ensure at least 10 error events per predictor parameter.
- Fit a mixed-effects logistic regression with random intercepts for participant (and item if relevant).
- Check model fit and calibration (for example Hosmer-Lemeshow or calibration plot) and report odds ratios with 95% confidence intervals.
- Invert the curve to find the design value at which predicted P(error) meets the risk threshold (for example 1% for routine, 0.1% for critical actions).
- Validate predictions on a held-out sample or follow-up study.
Sources
- Clark, J., Designing for Touch (2015)
- Shneiderman, B. & Plaisant, C., Designing the User Interface
- Touch IT
- Gopinath, V., Methods Atlas
- Designing for Touch (Josh Clark)
- Shneiderman & Plaisant
- Methods Atlas (Gopinath)
- The Inquiry Series, Ch 3 — The Ballistics of the Human Body — Fitts' Law & Motor Control
Also known as
- Error probability modelling
- Logistic error model
- Hit-probability regression
Categories
- Analytics & Behaviour
- Statistics & Modelling