Description
Fitting a vector autoregressive model to repeated self-reports or behaviour logs to obtain a directed temporal network (what predicts what at the next moment) and an undirected contemporaneous network (what co-occurs within the same moment).
When people answer the same short questionnaire many times a day, or when behaviour is logged at regular intervals, the data can show how states influence each other over time. Graphical vector autoregression regresses each variable at one time point on all variables at the previous time point, with regularisation, and models the remaining same-moment associations as a partial correlation network. The result is a temporal network with arrows, read as predictive (Granger-type) relations, and a contemporaneous network. Versions exist for a single person, for many people with multilevel estimation, and for panel data. The book lists practical threats: these models need many reliable, stationary measurements; the sampling rate must be fast enough for the process studied; equal spacing between measurements is assumed unless continuous-time models are used; and predictive arrows are not proof of causation, which ultimately requires intervention. For UX, the method can show whether frustration tends to precede abandonment, whether stress predicts later use of a wellbeing feature, or how motivation, effort and use feed each other during habit formation.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Siloed
- Stage
- Discovery, Post-launch
- Data
- Quantitative
- Effort
- Weeks
Steps
- Choose a small set of states or behaviours to track and an interval that matches how fast they change.
- Collect intensive repeated measures (experience sampling prompts or regular log aggregates) over days or weeks.
- Check and handle trends, missing prompts and unequal spacing; detrend or model time as needed.
- Fit a single-person graphical VAR or a multilevel VAR across participants.
- Inspect the temporal, contemporaneous and, for multilevel models, between-person networks.
- Interpret arrows as predictive relations, then test promising links with an intervention study.
Sources
- van der Maas, H., Complex Systems Research in Psychology (2024)
Also known as
- Graphical VAR
- GVAR
- Temporal network analysis
- Multilevel VAR networks
- mlVAR
Categories
- Analytics & Behaviour
- Statistics & Modelling