🔬 physics

Coalition Structure and Polarization in Parliamentary Voting Networks: Evidence from the Italian Parliament

This paper introduces a network-based framework analyzing over 4 million Italian parliamentary votes to demonstrate that legislative polarization is driven by coalition structure and opposition cohesion rather than size, revealing hidden voting patterns and concentrated mediator roles through quantitative metrics like modularity and betweenness centrality.

Original authors: Francesca Collu, Antonio Scala, Emilia La Nave

Published 2026-02-05

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Francesca Collu, Antonio Scala, Emilia La Nave

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Idea: Who Actually Votes With Whom?

Imagine a high school cafeteria. You have the official "seating chart" posted on the wall that says who sits with whom based on their clubs or teams. But if you actually watch the lunch period, you might see that the "soccer team" is actually sitting with the "band," and the "debate club" is arguing with the "art club."

This paper is about building a tool to watch the Italian Parliament's "lunch period" (their voting records) to see who actually sits together, rather than just trusting the official seating chart (party labels). The authors used a method called Network Analysis, which treats every politician like a person in a social network and every vote like a handshake.

The Toolkit: How They Measured the "Vibe"

The researchers didn't just count votes; they looked at the shape of the relationships. They used three main tools:

The "Clump" Detector (Modularity): Imagine looking at a crowd. Are there two distinct groups huddled in opposite corners, or is everyone mixed together in one big, messy circle? This tool measures how "polarized" or separated the parliament is.
The "Loyalty" Meter (Coherence): This checks how tight-knit a specific party is. Do all the members of the "Red Team" vote the exact same way every time? Or do some of them wander off to vote with the "Blue Team"?
The "Bridge" Finder (Betweenness): In a room full of people shouting at each other, who is the one person quietly talking to both sides to pass a message? This tool finds the politicians who act as bridges between opposing groups.

The Experiment: Three Different Governments

The authors studied the Italian Parliament during a very chaotic time (2018–2021) when the country had three different governments. Think of this as watching the same group of students under three different principals:

Government 1 (Conte I): A partnership between two populist parties (M5S and Lega).
Government 2 (Conte II): A new partnership between M5S and the center-left (PD).
Government 3 (Draghi): A "Grand Coalition" or "Technical Government" where almost every party joined forces to fix the economy, leaving only one small party in opposition.

The Surprising Discoveries

Here is what the "social network" revealed that the news headlines missed:

1. The "Big Tent" Paradox
You would think that when almost everyone joins the government (like in the Draghi cabinet), the parliament would be very united and calm.

The Reality: The network showed the opposite. The "Grand Coalition" was actually the least united. Even though they were all officially on the same team, they voted in many different, scattered ways.
The Analogy: It's like a massive family reunion where everyone is technically "family," but the cousins are arguing over the music, the aunts are in the kitchen, and the uncles are outside. Just because they are under one roof doesn't mean they are acting as one unit.
The Contrast: The smaller, partisan governments (Conte I) actually had clearer "us vs. them" lines, making the network look more structured and separated.

2. The "Bridge" is Rare
In a polarized parliament, you might expect many politicians to act as mediators.

The Reality: Only about 2% of the politicians acted as bridges.
The Analogy: Imagine a bridge over a deep canyon. Most people are stuck on one side or the other. Only a tiny handful of people (usually veterans or those switching parties) know how to walk across the canyon to talk to the other side. The rest of the parliament is just shouting across the gap.

3. The "Immigration" Twist
The authors tested this on a hot-button issue: immigration.

The Reality: The level of fighting (polarization) didn't depend on how united the government was. It depended on how united the opposition was.
The Analogy: If the government is a solid wall, but the people trying to break it down are arguing with each other, the "fight" looks messy and less intense. But if the government is shaky, and the opposition is a single, disciplined team, the "fight" looks very sharp and clear. The network tool could see this difference, while traditional methods might just look at who won the vote.

4. The "Switchers" are the Key
The people who acted as bridges were often the ones who had recently switched parties or were about to switch.

The Analogy: If you are a soccer player who just transferred from Team A to Team B, you still know the players on Team A. You are the only one who can pass a ball between the two teams. The network showed that these "transitional" players held all the power to connect the divided groups.

The Bottom Line

This paper argues that we shouldn't just look at the official party labels to understand how a parliament works. By using a "social network" map of votes, we can see:

That big, all-inclusive governments can actually be very messy and divided.
That true unity comes from how the opposition is organized, not just the government.
That the power to connect divided groups is held by a very small, specific group of politicians.

The authors built a tool that uses only public voting records (no secret meetings or private texts needed) to show us the real, hidden structure of political power. It's like having X-ray vision for democracy.

Technical Summary: Coalition Structure and Polarization in Parliamentary Voting Networks

Problem Statement
Ensuring legislative accountability in multi-party systems requires quantitative tools that reveal actual voting behavior beyond formal party affiliations. Traditional legislative analysis often fails to capture the discrepancies between declared political alliances and actual legislative behavior. While network science has been applied to specific parliamentary case studies (e.g., US Congress, European Parliament), existing approaches are largely descriptive and lack a general, transferable framework that operates simultaneously across multiple analytical scales (from individual MPs to party systems). There is a need for a reproducible methodology that integrates standard network measures to capture coalition dynamics, group coherence, and individual influence using exclusively public data.

Methodology
The authors propose a network-based framework applied to the Italian Parliament during the 18th legislature (2018–2021), covering three distinct government formations: the Conte I (M5S-Lega), Conte II (M5S-PD-LEU-IV), and Draghi (technocratic "grand coalition") cabinets. The study analyzes over 4 million individual vote expressions from both the Chamber of Deputies and the Senate.

The methodology consists of four core components:

Data Collection: Utilizing open-access SPARQL endpoints from the Italian Parliament, the authors compiled a relational database tracking MP identifiers, vote expressions (yes/no/abstain/absent), bill identifiers, and party affiliations. The analysis focuses on the first 50 days following each cabinet formation to capture coalition stabilization dynamics.
Network Construction: Co-voting networks are constructed where nodes represent MPs and weighted edges represent the frequency of identical voting behavior. The adjacency matrix is symmetric with zero diagonals.
Metrics and Algorithms:
- Network Modularity ( $Q$ ): Used as a system-wide measure of voting polarization emerging from pairwise relations, rather than a direct proxy for government-opposition separation. Communities are detected using Greedy modularity optimization (NetworkX).
- Voting Distances and Coherence: A voting distance metric is defined based on the Hamming distance of vote vectors, normalized to [0, 1]. This allows for the calculation of intra-group coherence ( $c(G, G)$ ) and inter-group coherence ( $c(A, B)$ ). Individual loyalty is measured via a coefficient of location ( $c_p$ ).
- Betweenness Centrality: Calculated to identify MPs acting as structural bridges between voting communities.
Reproducibility: The framework relies solely on public roll-call data, requiring no text analysis, media data, or proprietary information, making it applicable to any legislature with transparent voting records.

Key Results
The application of this framework to the Italian Parliament yielded several non-intuitive findings:

Coalition Structure Paradox: System-level polarization (modularity) varies systematically with coalition structure rather than coalition size. The Draghi cabinet, despite having the broadest formal support (a "grand coalition"), exhibited the lowest network modularity ( $Q \approx 0.02-0.08$ ). In contrast, the narrower partisan coalitions (Conte I) showed higher modularity ( $Q \approx 0.20-0.27$ ). This indicates that broad formal support does not translate to unified voting behavior; instead, parties in the Draghi government maintained distinct identities through selective defection.
Implicit Coalitions and Defection: Community detection revealed that declared alliances often diverge from actual voting blocs.
- In the Conte II cabinet, 15% of M5S members and 7% of PD members voted more frequently with the opposition than with their formal government coalition.
- In the Draghi cabinet, the Senate showed a third community composed of 73% of the LSP-PSDAZ group, which voted consistently with the opposition despite formal government membership.
Party Coherence Dynamics: Party discipline is not static. Groups transitioning between government and opposition roles (e.g., Lega, M5S) showed sharp declines in internal coherence. Under the Draghi government, all parties except the sole opposition party (Fratelli d'Italia) experienced a decline in coherence, suggesting that political ambiguity undermines collective discipline.
Concentrated Mediator Power: Betweenness centrality analysis revealed that bridge-building power is highly concentrated. Only approximately 2.1% of MPs (21 individuals) exhibited significant betweenness centrality. These actors were primarily party switchers or veteran politicians with cross-party relationships, rather than formal party leaders.
Issue-Specific Polarization (Immigration Case Study): On polarizing issues like immigration, network polarization depends on the structure of the opposition rather than just government cohesion.
- During Conte I, a cohesive government faced a fragmented opposition, resulting in lower modularity.
- During Conte II, a more divided government faced a unified opposition, resulting in higher modularity ( $Q \approx 0.45$ ).
- This demonstrates that modularity captures system-wide interaction structures, not just government-opposition separation.

Significance and Claims
The authors claim that their primary contribution is methodological: the demonstration of a framework capable of revealing latent coalition structures and polarization patterns that are invisible to traditional roll-call analysis.

Beyond Formal Labels: The framework proves that network modularity and community detection can uncover "implicit coalitions" and "selective defection" that formal party labels miss.
System-Level Insights: The study argues that polarization is an emergent property of the interaction structure of the entire system, influenced heavily by opposition fragmentation, rather than a simple function of government size or cohesion.
Generalizability: By relying exclusively on public roll-call data, the framework provides a reproducible infrastructure for comparative analysis across different legislatures and time periods. It enables the systematic assessment of legislative behavior without requiring institutional parameters specific to any single country.
Democratic Accountability: The methodology offers a tool for citizens, journalists, and researchers to monitor whether representatives vote consistently with their declared positions and to identify key intermediaries in legislative decision-making.

The paper concludes that while the observed patterns are specific to the Italian context of the 18th legislature, the framework provides a common analytical language for investigating structural regularities in multi-party legislatures globally.