vector autoregressive model
Definition
Vector autoregressive model is a variation of multiple regression in which the independent variables are lagged forms of the dependent variables, with as many equations in the model as there are dependent variables, and each dependent variable modeled as a function of its own prior value and the prior values of all other variables in the system. Applied to idiographic psychological networks, the model captures temporal dynamics between variables such as symptoms, emotions, and cognition through these lagged relationships, representing the predicted associations between a variable at one time point and variables at the next. Under standard formulations, the connections between variables and their mean levels are assumed to be time-invariant, which limits the model's capacity to represent clinical change. The fixed moderated time series model extends this by allowing all parameters, including the innovation structure, to be moderated by contextual factors such as social interaction or sleep quality, and by using the state space framework to estimate changes in mean levels directly.
Sources: Bringmann et al. (2024)
Related Terms
- networks (1 shared article)
- personalized models; moderation analysis (1 shared article)
- time series (1 shared article)
Applications
Vector Autoregressive Model and Idiographic Psychological Networks
Idiographic psychological networks, employed increasingly in clinical practice to depict associations among a person's symptoms, emotions, and behaviors over time, are grounded in the vector autoregressive model. The network structure itself represents the lagged relationships estimated by the model, showing how one variable at a given time point predicts another at the next. Standard VAR-based networks assume stationarity, meaning the connections and mean levels do not change, a constraint that moderated extensions of the model are designed to relax.
Sources: Bringmann et al. (2024)
Vector Autoregressive Model and Moderation Analysis
Moderation analysis in the context of VAR-based networks examines whether contextual factors influence the model's parameters, including lagged connections and the innovation structure. Earlier VAR formulations based on linear or additive regression permitted moderation of the lagged connections but not the innovation variance, leaving part of the network unexamined. The fixed moderated time series model addresses this gap by allowing all parameters of the VAR model to be moderated, as demonstrated with depression patients whose emotion networks were tested against binary and continuous moderators.
Sources: Bringmann et al. (2024)
