Identifying Model Misspecification in VAR(1) Models: A Predictive Accuracy Analysis Approach

Eiling, Ward (2023) Identifying Model Misspecification in VAR(1) Models: A Predictive Accuracy Analysis Approach. Bachelor thesis, Psychology.

Text BachelorThesis_Eiling_s3983919.pdf Restricted to Repository staff only until 20 July 2025. Download (611kB)

Abstract

The rising popularity of the network model of psychopathology, which captures the dynamic interplay between symptoms over time, has led to the widespread use of the lag-1 vector autoregressive (VAR(1)) model. Networks based on the VAR(1) model have the potential to provide valuable insights for clinicians in understanding and treating mental disorders. However, to establish whether meaningful inferences can be drawn from these networks, the quality of the model must be ensured. To this end, predictive accuracy analysis (PAA) may be used to evaluate a model's effectiveness to capture dynamics and generalize to unseen data. While simulation-based PAA may identify overfitting, it may overlook violations of model assumptions. This study investigates this limitation by conducting empirical and simulation-based analyses to shed light on the potential consequences of employing large datasets in a psychopathological context. Specifically, it suggests that such datasets may be prone to violate the stationarity assumption, resulting in model misspecification. Therefore, when utilizing the VAR(1) model, researchers are advised to carefully balance sample size, ensuring power and preventing overfitting, while also avoiding model misspecification. In line with the hypothesis, the simulation-based PAA yielded overoptimistic results and failed to identify violations of the model's assumptions. To avoid misinterpretation of inaccurate networks, the study recommends the use of empirical-based cross-validation procedures to evaluate generalizability and predictive accuracy in real-world applications. Future research should address pressing questions regarding predictive accuracy metrics and explore the relationship between VAR assumption violations, model misspecification, and predictive accuracy outcomes.

Item Type:	Thesis (Bachelor)
Supervisor name:	Bringmann, L.F. and Albers, C.J.
Degree programme:	Psychology
Differentiation route:	None [Bachelor Psychology]
Date Deposited:	21 Jul 2023 09:04
Last Modified:	21 Jul 2023 09:04
URI:	http://gmwpublic.studenttheses.ub.rug.nl/id/eprint/2513

Actions (login required)

View Item