Choosing between AR(1) and VAR(1) models in typical psychological applications

Area of Science:

• Psychological research methodology
• Quantitative psychology
• Time series analysis

Background:

• Individual subject time series data are increasingly common in psychological research.
• Vector Autoregressive (VAR) models are popular for analyzing these data but require substantial observations for reliable coefficient estimation.
• Small sample sizes in psychology can question the reliability of VAR model coefficients, suggesting simpler Autoregressive (AR) models might be more appropriate.

Purpose of the Study:

• To investigate the relative performance of AR and VAR models in psychological research with small sample sizes.
• To determine how model performance depends on sample size (n) and true model characteristics.
• To quantify selection uncertainty and evaluate model selection strategies for applied researchers.

Main Methods:

• A simulation study was conducted to directly compare AR and VAR model performance.
• The study analyzed the impact of varying sample sizes and underlying model structures.
• Different model selection strategies were assessed for their effectiveness.

Main Results:

• The simulation directly assessed the performance of AR and VAR models under typical psychological data conditions.
• Findings illustrate the dependency of relative model performance on sample size and true model properties.
• The study quantifies the uncertainty associated with choosing between AR and VAR models.

Conclusions:

• Provides a comprehensive guide for applied researchers on the appropriate use of VAR models in psychological research.
• Offers insights into selecting between AR and VAR models based on data characteristics and sample size.
• Enhances the understanding of model selection strategies for time series data in psychology.