Description
Enter each person's mean of a time-varying predictor and each moment's deviation from it as separate predictors, to separate between-person differences from within-person associations.
A correlation across pooled repeated measures mixes two questions: do people who are more stressed on average feel worse, and does a person feel worse when more stressed than usual? A model with the raw predictor gives an uninterpretable blend. Adding the person mean and the person-mean centred value separates the two; they can differ in size and even sign. In the ESM handbook example the between-person slope was over three times the within-person slope, and the within estimate was about 60 times more precise. Bolger and Laurenceau's daily process model (daily conflict predicting same-day intimacy) adds control for time trend and AR(1) errors. They argue diary designs give better causal leverage than one-off surveys because measurement follows causes and effects as they change, while still ranking far below experiments. The split extends to lagged predictors and group differences at each level. Person means from few observations are unreliable.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Unmigrated
- Stage
- Discovery, Post-launch
- Data
- Quantitative
- Effort
- Days
Steps
- Compute each person's mean of the time-varying predictor.
- Compute person-mean centred values (each moment minus the person's own mean).
- Fit a mixed model with both as predictors, a random intercept and a random slope for the centred term.
- Control for time trend and allow autocorrelated (AR(1)) residuals in daily or momentary data.
- Interpret the centred coefficient as the within-person association and the mean coefficient as the between-person association.
- Add group interactions at each level if needed, and report both effects; never infer within-person effects from cross-sectional data.
Sources
- The Open Handbook of Experience Sampling Methodology (Myin-Germeys & Kuppens, 2022)
- Intensive Longitudinal Methods (Bolger & Laurenceau, 2013)
Also known as
- Person-mean centring
- Contextual effect model
- Disaggregating within and between effects
- Within-subject causal process model with person-mean centring
- Daily process model
- Within-person causal process analysis
Categories
- Analytics & Behaviour
- Statistics & Modelling