A Network Science Approach to Understanding Child–Parent Discrepancies in Anxiety Symptoms

Link to the Github Project Link to the Analysis

• Big idea — Traditional psychological scoring methods collapse symptoms into a single number, losing the relationships between symptoms and the discrepancies between child and parent reports.
• Small idea — Network analysis treats each symptom as a node and the relationships between symptoms as edges, allowing us to detect central, influential, or bridging symptoms in child–parent pairs.
• Bird’s eye view of the idea — By modeling the SCARED questionnaire responses as a symptom network, we can quantify hubs, clusters, and pathways of anxiety — and compare how children and parents perceive them differently.
• Technical details — This project uses Graph Theory, Centrality Metrics (degree, betweenness, closeness, eigenvector), community detection, and arc diagram visualization to map 41 symptoms and 451 edges from a sample of 31 child–parent dyads.

• What’s next — These networks can guide future clinical interpretation, automated discrepancy detection, and targeted interventions focused on the most influential symptoms.

📊 Network Metrics Summary

🧠 Snapshot of Findings

• 41 nodes representing SCARED anxiety symptoms
• 451 edges representing correlations between symptom responses
• Top hub symptoms appear consistently in both child and parent networks but with notable magnitude differences
• Eigenvector centrality highlights influential “core anxiety” symptoms
• Arc diagram shows strong clustering in Generalized Anxiety and Social Anxiety items
• Parent reports show stronger symptom clustering than child reports
• Child reports show more variability and weaker adjacency patterns

🌀 Preview (Mock Arc Diagram)

Arc diagram preview placeholder — visualization shows symptom clusters by subscale, with children displaying more dispersed connections than parents.

Arc diagram mock-up of symptom relationships across the SCARED survey

🎯 Objective

• Understand the structure of anxiety symptoms in children using a network science lens
• Compare child vs. parent symptom networks to identify discrepancy patterns
• Use network metrics to identify core, influential, or bridging symptoms
• Build a framework for quantifying disagreement in mental health assessments

💪 Challenge

• Parents and children often report anxiety symptoms very differently
• Traditional scoring does not capture inter-symptom connectivity
• Small sample sizes (e.g., 31 child–parent dyads) challenge robustness
• Symptom-level analysis requires nuanced statistical and network modeling

🧪 Solution

• Construct correlation networks for child reports and parent reports
• Use centrality measures to identify important symptoms
• Compute discrepancy metrics for each symptom pair
• Visualize relationships using arc diagrams, force-directed graphs, and heatmaps
• Evaluate how network topology changes across reporting sources

📁 Data Source

• SCARED (Screen for Child Anxiety Related Emotional Disorders)
• 41-item questionnaire completed by children (n=31) and parents (n=31)
• Responses standardized, cleaned, and merged into dyadic networks

📐 Metrics

- **Centrality Metrics**
  - Degree
  - Betweenness
  - Closeness
  - Eigenvector

- **Discrepancy Metrics**
  - Absolute score difference per symptom
  - Rank-order difference
  - Network distance difference

- **Community Metrics**
  - Subscale clustering
  - Modularity score