Analytic Strategies for Longitudinal Networks with Missing Data.

Area of Science:

• Social network analysis
• Statistical modeling

Background:

• Missing data pose challenges in longitudinal social network analysis.
• Stochastic actor-based models are commonly used but can be sensitive to missing data.
• Current practices often exclude partially observed participants, limiting representativeness.

Purpose of the Study:

• To propose and evaluate an inclusive approach for analyzing longitudinal social network data with missing observations.
• To compare the proposed method against standard practices for handling missing data.
• To assess the impact on data retention, sample representativeness, and model bias.

Main Methods:

• Review of existing approaches for missing data in longitudinal network analysis.
• Development and application of an alternative sub-setting and analysis strategy.
• Utilized data from a school friendship network with four waves of observation (N=694).
• Comparison of the proposed method with standard practices using model convergence and estimation bias as metrics.

Main Results:

• The proposed inclusive approach retained more information from partially observed individuals.
• The analytic sample generated by the new method was more representative of the original network.
• Model estimates derived from the inclusive approach exhibited less bias in the case study.
• The method facilitated model convergence more effectively than standard practices.

Conclusions:

• An inclusive approach to sub-setting and analyzing longitudinal social network data with missing observations is effective.
• This method offers advantages over standard practices by retaining more data and reducing bias.
• The findings have significant implications for improving the analysis of complex longitudinal network data.