Embedding Network Autoregression for Time Series Analysis and Causal Peer Effect Inference

2025-12-31 · Source: JMLR · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

Jae Ho Chang and Subhadeep Paul introduce the Embedding Network Autoregressive Model (ENAR) for multivariate networked longitudinal data, published in 2026. This model unifies two problems: predicting multivariate networked time series and estimating causal peer influence with confounding from finite-time longitudinal data. ENAR assumes the network is generated from a latent variable model, incorporating these unobserved variables into a structural peer effect or time series network autoregressive framework. The estimation strategy involves estimating latent variables from the observed network, then applying least squares to the network autoregressive model. The authors prove consistency and asymptotic normality for momentum and peer effect parameters as network vertices ($N$) grow, covering both growing and finite time points ($T$). They also develop a selection criterion for an unknown number of latent vectors ($K$) that prevents under-selection, with theoretical guarantees. The method was applied to analyze peer effects in conflict and school climate perception using data from over 7000 students across 23 schools.

Key takeaway

For research scientists or data scientists analyzing social networks or complex systems, this Embedding Network Autoregressive Model offers a robust framework. If you are grappling with confounding homophily in peer effect studies or predicting networked time series, consider applying ENAR. Its proven consistency for large networks and method for selecting latent variables can improve the accuracy of your causal inferences and predictions. You should explore its application to your specific longitudinal network datasets.

Key insights

The Embedding Network Autoregressive Model unifies networked time series prediction and causal peer effect inference via latent variables.

Principles

• Latent variables explain network structure.
• Parameter consistency holds for large networks.
• K selection criteria prevent under-selection.

Method

Estimate latent variables from the observed network, then use least squares for the network autoregressive model. A selection criterion for \$K\$ is developed.

In practice

• Infer causal peer effects in groups.
• Model school climate perceptions.
• Predict multivariate networked time series.

Topics

• Embedding Network Autoregression
• Time Series Analysis
• Causal Peer Effects
• Latent Variable Models
• Network Analysis
• Longitudinal Data

Best for: AI Scientist, Research Scientist, Data Scientist

Related on AIssential

• Multi-Level Causal Embeddings — Causal embeddings generalize abstraction, mapping multiple detailed models into a coarser causal model's sub-systems.
• Estimating Mutual Information between Time Series and Temporal Event Sequences Across Diverse Analysis Tasks — A nonparametric mutual information estimator quantifies dependence between heterogeneous time series and event sequences without transformation or discretization.
• Knowledge-Embedded Latent Projection for Robust Representation Learning — Leveraging semantic side information improves latent space model estimation in high-dimensional, imbalanced data.
• Causal Effect Estimation with Learned Instrument Representations — ZNet learns latent instrumental variables from observed data, enabling robust causal inference without explicit instruments.
• Identifying Weight-Variant Latent Causal Models — Identifying latent causal variables requires addressing transitivity, permutation, and scaling indeterminacies, especially via modulated linear-Gaussian models.
• Faithful Embeddings of Irregular and Asynchronous Data for Online Log-NCDEs — Explicit data reconstruction is unnecessary for continuous-time models when using a continuous and injective embedding.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by JMLR.