A Bayesian Vector Autoregressive Model with Nonignorable Missingness in Dependent Variables and Covariates:...

Area of Science:

• Psychometrics
• Statistics
• Longitudinal Data Analysis

Background:

• Intensive longitudinal designs with repeated assessments are prone to nonignorable attrition and selective response omission.
• Standard time series models like vector autoregressive (VAR) models often overlook nonignorable missing data, potentially biasing results.

Purpose of the Study:

• To introduce a novel Bayesian model for analyzing multivariate, multiple-subject time series data.
• To simultaneously address over-time dependencies using a VAR model and handle both ignorable and nonignorable missing data.
• To provide practical tools and comparisons for data analysis.

Main Methods:

• Developed a Bayesian model integrating a vector autoregressive (VAR) model with a missing data framework.
• The model accounts for complex temporal dependencies and various missingness mechanisms.
• Software code for implementation and simulation studies were provided.

Main Results:

• The joint Bayesian approach demonstrated effectiveness in handling nonignorable missingness within time series data.
• Simulation results provided insights into the comparative performance of the joint model versus traditional two-step multiple imputation.
• The study highlights the strengths and weaknesses of different methods in practical scenarios.

Conclusions:

• The proposed Bayesian model offers a more comprehensive approach to analyzing intensive longitudinal data with nonignorable missingness.
• This method enhances the accuracy and reliability of time series analyses in the presence of complex data issues.
• The findings guide researchers in selecting appropriate analytical strategies for challenging longitudinal datasets.