Friendship Formation Across Life Stages: A Comparative Study

A comprehensive social network analysis examining how friendships form across different educational stages, from primary school through university.

Project Overview

This research project investigates the dynamics of friendship formation across four distinct educational stages:

• Primary School (ages 6-12)
• Middle School (ages 11-13)
• High School (ages 15-18)
• University (ages 18-22)

Research Questions

• How do different factors (age, gender, behavior) influence friendship formation at various life stages?
• What are the differences between wireless sensor data and survey-based data collection methods?
• Do consistent patterns emerge across different educational contexts?

Team

University of Bologna, Alma MAter Studiorum - Artificial Intelligence MSc

• Guglielmo Biagini
• Elisa Castagnari
• Matteo Fusconi
• Luca Trambaiollo

Datasets

We analyzed four publicly available datasets, each representing a different educational stage:

Stage	Source	Method	Nodes	Edges	Metadata
Primary School	French school (2009)	RFID sensors	21	45	Sex
Middle School	Dutch class (2003-2004)	Survey	26	33-117*	Sex, Age, Religion, Alcohol, Delinquency
High School	French classes préparatoires	RFID sensors	28	73	Sex
University	Groningen Sociology (1996-1997)	Survey	23	36-147*	Sex, Age, Smoking, Drugs, Religion, Associations

*Undirected-Directed graph variations

Data Collection Methods Compared

RFID/Wireless Sensors:

• Objective, unsupervised proximity detection
• Large-scale data collection
• Limited contextual information
• Requires threshold selection (we used 5-minute minimum contact duration)

Questionnaire Surveys:

• Rich metadata (behaviors, preferences, characteristics)
• Captures meaningful relationships
• Potential recall bias
• Smaller sample sizes

Methodology & Measures

Network Analysis Techniques

We implemented comprehensive social network analysis using NetworkX (Python), examining:

1. Homophily Analysis

Measured tendency for individuals to befriend similar others using assortativity coefficient:

• Sexual homophily: 0.62-0.63 (Middle School & University) - strong same-gender clustering
• Age homophily: 0.83 (University) - students befriend same-age peers
• Behavioral factors (smoking, drugs): ~0 - minimal influence

2. Centrality Measures

• Degree Centrality: Identified most popular individuals
• In-Degree/Out-Degree: Analyzed friendship reciprocity (directed graphs)

3. Transitivity (Clustering Coefficient)

• Primary: 0.39
• Middle School: 0.38
• High School: 0.40
• University: 0.46 (higher clustering, more cohesive groups)

4. K-Cores Analysis

Identified tightly-knit subgroups within each network.

5. Reciprocity (Directed Graphs)

• Middle School: 0.76
• University: 0.49 - approximately 50% of friendships are non-reciprocal

Key Findings

1. Gender is the strongest predictor of friendship formation across all life stages

   • Consistent same-gender clustering (homophily 0.62-0.63)
   • Less pronounced in wireless datasets due to higher connectivity

2. Age matters more in university than earlier stages

   • University students cluster by age groups (homophily 0.83)
   • Life experiences and shared schedules drive connections

3. Behavioral factors (smoking, drugs) have minimal influence

   • Low assortativity (~0) for substance use
   • School association involvement helps create connections but with lower reciprocity

4. Transitivity increases with age

   • University shows highest clustering (0.46)
   • More dynamic, resourceful social behavior in older students

5. Friendship reciprocity varies significantly

   • Only ~50% of university friendships are mutual
   • Could reflect survey bias OR genuine asymmetric relationships

6. Data collection method matters

   • Wireless: More connections, less structure, fewer metadata
   • Surveys: Richer context, potential recall bias

Technologies Used

• Python 3.x - Primary programming language
• NetworkX - Social network analysis and graph algorithms
• NumPy/Pandas - Data manipulation
• Matplotlib - Visualization
• LaTeX - Academic paper formatting

Implications & Future Work

• Educational Policy: Understanding social dynamics can inform classroom organization and intervention strategies
• Disease Modeling: RFID contact networks applicable to epidemic spread analysis
• Social Psychology: Insights into developmental changes in friendship formation

Limitations & Future Directions

• Longitudinal analysis: Track how friendships evolve over academic years
• Cross-cultural comparison: Datasets from diverse geographical/cultural contexts
• Personality traits: Incorporate psychological assessments (cheerfulness, resourcefulness)
• Standardized metadata: Develop consistent feature collection across all life stages
• Generational shifts: Compare historical data with contemporary patterns (e.g., earlier smoking initiation)