Description
Analyse diary data from both members of a pair in one model, estimating each partner's process separately and how the partners' daily residuals and person-level effects correlate.
When both members of a pair keep a diary on the same days, their data are not independent: the pair is the sampled unit. Analysing each partner separately avoids the problem but discards the most interesting questions. The model treats occasions as nested within pairs and writes two parallel equations, one per distinguishable member (partner A and B, patient and carer, buyer and seller), using role dummies with no common intercept so each member has their own intercept and slopes. At level 1 the partners' same-day residuals may correlate and each series may be autocorrelated. At level 2 their random intercepts and slopes may correlate, so you can ask whether a strongly reactive member tends to have a strongly reactive partner. Predictors are split into person means and deviations for each member. A three-level model is not identifiable for distinguishable pairs, and complex error structures may not converge in all software.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Unmigrated
- Stage
- Discovery, Post-launch
- Data
- Quantitative
- Effort
- Days
Steps
- Confirm members are distinguishable by a meaningful role; otherwise use an indistinguishable-dyad approach.
- Stack the data so each pair has one block of rows per member, with pair id, person id, day, one role dummy per member and the outcome.
- Grand-mean centre each predictor, then split it into person means and person-mean deviations separately for each member; centre and rescale time.
- Plot each pair's two series in panels to inspect co-movement and trends.
- Fit a model with no global intercept: each role dummy and its interactions with within-person and between-person predictors and time.
- Add random intercepts and within-person slopes for each member, with their covariances, at the pair level.
- Specify level-1 errors that allow same-day correlation between partners and autocorrelation over days.
- Report fixed effects per member, slope variances and cross-partner covariances, with a spaghetti plot of fitted lines and ellipses of random effects.
Sources
- Intensive Longitudinal Methods (Bolger & Laurenceau, 2013)
Also known as
- Dyadic multilevel model
- Two-intercept dyadic model
- Multivariate multilevel model for couples
- Dyadic intensive longitudinal analysis
Categories
- Analytics & Behaviour
- Statistics & Modelling