VAR(1) based models do not always outpredict AR(1) models in typical psychological applications.

Area of Science:

• Psychological research methods
• Quantitative psychology
• Dynamical systems modeling

Background:

• Modeling within-person dynamics is crucial in psychology.
• Vector autoregressive (VAR(1)) models are popular but complex.
• Model complexity raises concerns about overfitting and predictive accuracy.

Purpose of the Study:

• To evaluate if VAR(1) models outperform simpler models in psychological applications.
• To compare the predictive accuracy of various cross-validation (CV) techniques.
• To assess the generalizability of VAR(1) models to new, unseen data.

Main Methods:

• A simulation study compared five CV techniques (LOOCV, K-fold, blocked CV, hv-block CV, accumulated prediction errors).
• Blocked CV demonstrated superior performance in simulations mimicking psychological data.
• Blocked CV was then applied to three real psychological datasets to compare VAR(1) with AR(1) models.

Main Results:

• Blocked cross-validation proved most effective for time-dependent psychological data.
• In three psychological applications, VAR(1) models did not show superior predictive accuracy compared to simpler AR(1) models.
• The findings suggest VAR(1) models may be unnecessarily complex for some psychological data.

Conclusions:

• Simpler autoregressive (AR(1)) models may be sufficient and more parsimonious than VAR(1) models in certain psychological contexts.
• Researchers should carefully consider model complexity and predictive accuracy when analyzing dynamical psychological processes.
• The choice of cross-validation technique significantly impacts the assessment of model performance, with blocked CV recommended for time-series data.