Accounting for Edge Uncertainty in Stochastic Actor-Oriented Models for Dynamic Network Analysis.

Area of Science:

• Network Science
• Computational Statistics
• Neuroscience

Background:

• Stochastic Actor-Oriented Models (SAOMs) are standard for analyzing dynamic social networks.
• The assumption of error-free network data in SAOMs is often unrealistic.
• Real-world network data frequently contains false positive and false negative edges.

Purpose of the Study:

• To propose a Hidden Markov Model (HMM) extension to SAOMs to handle noisy network observations.
• To develop an expectation-maximization algorithm for parameter estimation in the proposed HMM-SAOM framework.
• To enhance the accuracy of network dynamic estimation in the presence of observation noise.

Main Methods:

• Developed a two-component model: a latent Markov process for true network evolution and a measurement model for observed networks.
• Employed an expectation-maximization algorithm for parameter estimation.
• Utilized the missing information principle and particle filtering to manage computational challenges of large state spaces.

Main Results:

• Simulation studies demonstrated improved estimation accuracy compared to standard SAOMs when data is noisy.
• The HMM-SAOM approach revealed larger effect sizes in functional brain networks derived from EEG data.
• The proposed method outperforms the naive application of standard SAOMs on noisy network data.

Conclusions:

• The HMM extension to SAOMs provides a more robust framework for analyzing dynamic networks with imperfect observations.
• This method offers significant improvements in accuracy for network inference, particularly in fields like neuroscience.
• Accurate modeling of network dynamics, even with noise, is crucial for reliable scientific discovery.