# Wings of Unrest: Flight Connectivity and Social Instability Research

A quantitative research project investigating the relationship between global air connectivity and social unrest across 2,620 cities worldwide.

## Overview

This study explores whether cities with greater flight connectivity experience systematically different levels of protest activity compared to less connected urban areas. Using a Negative Binomial regression model, the analysis reveals that increased air connectivity is significantly associated with higher frequencies of unrest events (p < 0.001), explaining approximately 32% of the deviance.

## Key Findings

- **Air connectivity** positively correlates with protest frequency (each additional flight → 0.41% increase in expected events)
- **Population served** by airports shows strong positive association with unrest
- **Unemployment** and **freedom of expression** are positively associated with protest intensity
- **HDI** and **land area** exhibit negative associations with unrest
- Model pseudo R² = 0.323, suggesting moderate explanatory power

## Dataset

The analysis combines multiple authoritative data sources:

- **Air connectivity**: OpenFlights (2014 direct flight data)
- **Social unrest**: ACLED (Armed Conflict Location & Event Data Project, June 2020 - April 2025)
- **Economic indicators**: World Bank (GDP per capita, unemployment, land area)
- **Development metrics**: UNDP (HDI, life expectancy, education)
- **Political measures**: V-Dem (freedom of expression, civil society indices)

**Final dataset**: 2,620 cities from 191 countries

## Methodology

### Data Processing Pipeline

1. **Spatial matching**: Cities linked to airports within 50km radius using Haversine distance
2. **Service score calculation**: `score = departures / (distance + 1)` to weight airport assignment
3. **Event geocoding**: Protest events mapped to nearest city with same country constraint
4. **Country standardization**: Harmonized country names across all datasets

### Statistical Model

Negative Binomial regression (handles overdispersion in count data):

```
log(E[Number_of_Events]) = β₀ + β₁(Flights) + β₂(Population) + β₃(Land_area)
                           + β₄(Unemployment) + β₅(Freedom_Expression) + β₆(HDI)
```

## Project Structure

```
FlightUnrestResearch/
├── Data/                      # Raw datasets (airports, routes, ACLED, V-Dem, etc.)
├── clean_data.csv            # Final processed dataset for analysis
├── clean_data.ipynb          # Data cleaning and preparation pipeline
├── haversine.py              # Geospatial distance calculations (PyTorch-accelerated)
├── mappings.py               # Country name standardization mappings
├── FINAL_r_reg_final.ipynb   # Statistical analysis (Negative Binomial regression)
├── final_paper.pdf           # Complete research paper with literature review
└── final_paper.odt           # Source document
```

## Core Components

### `haversine.py`
PyTorch-accelerated geospatial functions for large-scale spatial assignments:
- `calculate_served_population()`: Assigns city populations to airports using weighted scoring
- `assign_events_to_cities()`: Maps protest events to nearest cities via Haversine distance

### `clean_data.ipynb`
Complete data pipeline:
- Merges flight routes with airport locations
- Calculates served populations within 50km radius
- Geocodes 428K+ ACLED events to cities
- Integrates World Bank, V-Dem, and HDI indicators
- Handles missing data and country name harmonization

### `FINAL_r_reg_final.ipynb`
Statistical analysis in R:
- Negative Binomial regression model
- Multicollinearity diagnostics (VIF analysis)
- Model fit evaluation (pseudo R², deviance)
- Coefficient interpretation and significance testing

## Requirements

### Python
```
pandas
numpy
torch
openpyxl
tqdm
```

### R
```
MASS (for glm.nb)
```

## Usage

### 1. Data Preparation
```bash
jupyter notebook clean_data.ipynb
```
Outputs: `clean_data.csv` with all variables merged and geocoded

### 2. Statistical Analysis
```bash
jupyter notebook FINAL_r_reg_final.ipynb
```
Runs regression model and produces coefficient tables

## Key Variables

| Variable | Description | Source |
|----------|-------------|--------|
| `Number_of_Flights` | Total direct flights at city airports | OpenFlights |
| `Number_of_Events` | Protest/riot incidents (Jun 2020-Apr 2025) | ACLED |
| `Served_Population` | Population within 50km of airports | Cities database + calculation |
| `GDP_per_capita` | National GDP per capita (2019) | World Bank |
| `Unemployment` | National unemployment rate (2019) | World Bank |
| `HDI` | Human Development Index | UNDP |
| `Freedom_of_Expression` | Civil liberties index (2021) | V-Dem |
| `Civil_Society_Index` | Civil society strength (2021) | V-Dem |

## Limitations

- **Cross-sectional design**: Cannot establish causality
- **Temporal mismatch**: Flight data (2014) vs. protest data (2020-2025)
- **Measurement error**: Potential ACLED under-reporting in authoritarian contexts
- **National-level controls**: HDI, unemployment applied uniformly within countries
- **Spatial uncertainty**: 50km radius may misrepresent complex metro areas
- **Data access**: High-resolution aviation data remains proprietary

## Citation

Bitton, R. (2025). *Wings of Unrest: The Relationship Between Global Flight Connectivity and Social Instability*.

## Related Research

This work contributes to literature on:
- Globalization and domestic political contention
- Infrastructure networks and protest diffusion
- Urban political dynamics and global integration
- Spatial determinants of collective action

## Contact

Raphael Bitton
rbitton@uchicago.edu

## License

Research data sourced from publicly available datasets. Analysis code available for academic use.