A Note on Ising Network Analysis with Missing Data

Area of Science:

• Psychometrics
• Statistical Modeling
• Network Analysis

Background:

• The Ising model is widely used for item response data analysis.
• Standard likelihood methods are computationally expensive for many variables.
• Pseudo-likelihood methods are hindered by missing data, and listwise deletion can cause bias.

Purpose of the Study:

• To propose a conditional Bayesian framework for Ising network analysis that effectively handles missing data.
• To address the limitations of existing methods in the presence of missing values.
• To provide a statistically sound and computationally efficient approach for Ising model inference.

Main Methods:

• Integration of a pseudo-likelihood approach with iterative data imputation.
• Development of an asymptotic theory for the proposed method.
• Implementation of a Pólya-Gamma data augmentation procedure for efficient parameter sampling.

Main Results:

• The proposed conditional Bayesian framework successfully handles missing data in Ising network analysis.
• Asymptotic theory supports the validity and consistency of the method.
• Simulations and a real-world application demonstrate the method's performance and efficiency.

Conclusions:

• The conditional Bayesian framework offers a robust solution for Ising model analysis with missing data.
• The method provides unbiased estimations and reliable interpretations, overcoming limitations of traditional approaches.
• This framework is applicable to complex psychological and epidemiological datasets, such as those from NESARC.