Description
Models a continuous outcome as a weighted sum of several predictors, estimating each predictor's unique contribution while holding the others constant.
Multiple regression fits an outcome such as task time, error count or a SUS score as a linear combination of predictors, for example interface complexity, user expertise and device type. Each coefficient estimates the expected change in the outcome per unit change in that predictor, conditional on the others, which is what separates it from a pile of bivariate correlations. Rigour comes from pre-specifying the model, checking assumptions (linearity, independent errors, homoscedasticity, normal residuals, low collinearity), reporting effect sizes with confidence intervals rather than p-values alone, and validating predictive accuracy on held-out data. In UX it is the workhorse behind driver analysis, benchmarking and predicting error rates from design attributes, but it is descriptive unless the causal structure justifies the covariates chosen.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Vestigial
- Stage
- Evaluation, Post-launch
- Data
- Quantitative
- Effort
- Days
Steps
- Write the outcome, the candidate predictors and the reason each is included before looking at the data, ideally with a causal diagram.
- Plan sample size: a common rule is at least 10-20 observations per predictor, or run a power analysis (e.g. G*Power, f2 = 0.15 for a medium effect).
- Clean and code predictors: centre continuous variables, dummy-code categories with a clear reference level, and inspect distributions.
- Fit the model and inspect diagnostics: residual plots, Q-Q plot, VIF (investigate above 5, act above 10) and Cook's distance (flag above 4/n).
- Report standardised and unstandardised coefficients with 95% confidence intervals, plus R-squared and adjusted R-squared.
- Validate with k-fold cross-validation (k = 5 or 10) and report out-of-sample RMSE or R-squared, not only in-sample fit.
- Translate coefficients into design language, e.g. each additional form field adds 4.2 s (95% CI 3.1-5.3) to completion time.
Sources
- Applied Predictive Modeling (Max Kuhn & Kjell Johnson)
- R for Marketing Research and Analytics (Chris Chapman & Elea McDonnell Feit)
- Applied Linear Statistical Models (Kutner, Nachtsheim, Neter & Li)
- Applied Predictive Modeling (Kuhn & Johnson)
- R for Marketing Research (Chapman & Feit)
- The Inquiry Series, Ch 43 — Regression, Moderation & Mediation for UX Researchers
- Veal, A. J. & Darcy, S., Research Methods in Sport Studies and Sport Management (2014)
Also known as
- Multiple linear regression
- OLS regression
- Linear model
Categories
- Analytics & Behaviour
- Statistics & Modelling