Introduction

Committee work is at the heart of legislative activity in the European Parliament (EP). One of the European Union’s (EU) most influential committees is the Committee on International Trade (INTA). With exclusive competence over trade policy, the EU makes critical decisions that affect global trade by negotiating international agreements with partner countries or regulating which products are allowed to enter European markets. Despite its prominence, understanding the inner workings of the INTA committee remains the purview of a limited number of policy specialists.

Understanding the work of parliamentary committees requires not only profound knowledge of official documents and legal processes. Behind the scenes, relationships among Members of the European Parliament (MEPs) shape legislative initiatives. This article uses amendment co-authorships as an approximation of relationships among MEPs to gain insights about collaborative patterns in the INTA committee. Interpreting these co-authorship relationships as a social network, the article uses visual and analytical tools from network analysis to answer relevant questions: what different types of engagement in collaborative work do MEPs use to gain influence and who are the most central players in the network?

This analysis builds on work I've done for the Borderlex Datahub, specifically the Amendment Tracker. This tool enables users to search, filter, and read amendments with improved clarity compared to official PDF documents. Beyond organizing information, this article presents a compelling use case for how the Borderlex Datahub can deliver policy monitoring and intelligence services from data collection to actionable insights.

In the first section of the article, I will present the data used to construct the network of amendment co-authorships in the INTA committee. Second, I will present an interactive visualization of the network and explore its overall structure. The next section is about identifying key actors and different types of engagement in the network. Finally, I will summarize key findings, discuss limitations and ideas for future analysis.

Data Sources and Network Construction

This article examines amendment co-authorship patterns during the 9th European Parliamentary term (July 2019 to June 2024) through the lens of network analysis. As is common in network (or graph) analysis, I will use the terms nodes and edges throughout the text. You can just imagine nodes as points and edges as lines connecting these points.

MEP data: Using the European Parliament's official API, I collected comprehensive data on all MEPs with any connection to INTA committee work. This includes not only formal committee members but also external MEPs who contributed amendments in the INTA committee. The 9th parliamentary term included the exit of the United Kingdom from the EU. As a consequence, the data contains a large number of incoming and outgoing MEPs.

Amendment documents: The INTA committee’s archive of amendment documents published during the 9th parliamentary period is available here. With the Amendment Tracker of the Borderlex Datahub, we have developed a reliable strategy for parsing these documents and converting them into structured data.

MEPs as nodes

Each MEP becomes a node in the network with the following attributes:

• Country represented and political group affiliation
• INTA committee role (full member, substitute, chair/vice-chair, or external)
• Total amendment count (all amendments) and collaborative amendment count (amendments co-authored with at least one other MEP)

Political group affiliations and committee roles can change during a parliamentary term. To maintain analytical consistency, I simplified these dynamic attributes by selecting the political group with which each MEP initially acquired their mandate, and the committee role held for the longest duration during the term.

Amendment co-authorships as edges

Amendment co-authorships become network edges connecting MEP nodes. For amendments with multiple authors, I create edges between all possible pairs of co-authors. Each edge is weighted by the frequency of collaboration, i.e. the total number of amendments two MEPs have co-authored together. Single-author amendments do not generate edges.

This simple version of the analysis treats all amendments equally regardless of type or content, focusing on collaboration patterns. The analysis examines the complete network at term's end, providing a comprehensive view of collaborative relationships that developed over the full parliamentary cycle.

Network Visualization and Exploratory Analysis

The resulting network for the 9th parliamentary term contains 253 nodes (MEPs with INTA involvement) and 77,719 edges (unique collaboration pairs). The edges are undirected and weighted representing the intensity of bidirectional collaborative relationships. Only half of all amendments (5,684 out of 11,365) during this term actually have multiple authors. The next visualization shows the resulting network.

Visual Encoding

• Node color: Political group affiliation
• Outer node size: Total amendment authorship (both individual and collaborative combined)
• Inner node size: Total number of collaborative amendments only
• Edge thickness: Frequency of co-authorship between MEP pairs
• Spatial positioning: MEPs who frequently co-author amendments are positioned closer together. The exact position of the nodes can vary as it is to some extent random and also depends on factors like screen size.
• Interactivity: Hovering over nodes reveals each MEP's name and highlights their direct collaborative partners.

Network Structure

The network has a low density, only 4.8% of all theoretically possible edges between nodes exist. This confirms that collaboration is selective rather than broadly distributed across all possible MEP pairs.

As a result, the network is not fully connected, i.e. there is no guaranteed path between all MEPs. Many nodes are completely isolated with no collaborative amendment co-authorships (43). Two additional components are triplets of MEPs that have co-authored amendments to a limited extent. The large, remaining component contains the overwhelming majority of amendment co-authorships.

Multi-author amendments have almost exclusively been prepared by MEPs belonging to the same group. Group clusters are clearly visible in the visualization. MEPs that work together more often are positioned closer to each other. The colors of the nodes help identify group affiliations as a dominant determinant of amendment co-authorships. Intensive intra-group collaboration can be especially seen in the great number of bold edges in the EPP, Renew and S&D groups that clearly form the center of the largest component in the network.

Multi-author amendments crossing political group boundaries are rare. Only 107 amendments (i.e. 1.9% of all collaborative amendments, and 0.9% of all amendments in total) fall into this category. A straightforward conclusion would be to consider political groups as the defining communities in the network.

Co-authorship Communities

To examine how well political groups capture observed collaborative patterns, we can look at the modularity of the network. Modularity measures how well a network divides into distinct communities by evaluating whether connections within communities are much denser than connections between different communities. A high modularity score (closer to 1) indicates strong community structure where nodes within groups are tightly connected to each other while having few connections to nodes in other groups. Negative values (up to -1) indicate that the proposed community division performs poorly at capturing natural groupings in the network.

In the case of the INTA amendment co-authorship network, the modularity resulting from defining the 8 political groups as the communities of the network is 0.679. This is a high score, especially given that we included independent MEPs (NI) as a group.

We can compare this score to the modularity of communities found by algorithmic approaches for community detection. Greedy modularity communities and Louvain communities are two common detection algorithms. Both algorithms start with each node as its own community, gradually merging nodes into larger communities using different strategies to optimize modularity. In the case of the INTA amendment collaboration network, the differences between the two algorithms are less important, as the communities detected by both algorithms achieve a similar modularity of around 0.713. This is slightly better than the modularity based on formal political groups.

Excluding isolated nodes, both community detection algorithms identify communities that correspond mostly to formal political groups. However, each community also contains one or two additional MEPs belonging to a different group. For instance, both algorithms identify Danti (S&D) and Zullo (NI) as part of the Renew community, and Fiocchi (ECR) as part of the ID community.

The small improvement in the modularity score of 0.034 compared to the modularity of formal political groups confirms what we can observe in the visualization. Although dominated by intra-group collaboration, the presence of a small set of inter-group edges contributes to the emergence of a large component inside the network. Without these rare cross-party collaborations, the network would fragment into separate political group clusters.

Co-authorships between different groups, although an exception, appear most often among MEPs from the central groups (EPP, Renew, S&D). This is consistent with the experience that the process of European Integration has until recently been carried largely by grand coalitions of centre-right and centre-left political groups. However, the network structure indicates that the 9th parliamentary term has also seen a shift towards broader coalitions. The dominant connected component extends to the Greens and the Left. Co-authorships linking the ECR group with the EPP and Renew extend amendment collaboration to the right. The integration of the far-right ID group hinges on just a single connection via a node of the ECR group.

Finally, an interesting question regarding communities is also the effect of nationality on collaborative behavior. Although trade policy is an exclusive EU competence, national interests play a major role in trade agreements, especially when key national industries are affected by changes in trade policy. However, the modularity score for communities defined as 28 national groups (before Brexit) is very low (0.015). Nationality is not a major determinant for collaborative work in the INTA committee.

Key Actors and Collaboration Types

Having developed a basic understanding of the network structure, we can move on to ask more practical questions about the INTA committee: who are the key actors that drive amendment co-authorships and what different behavioral patterns exist? The network structure provides us with important clues.

Collaboration breadth and intensity

Based on the network structure, we can observe key indicators of node centrality, especially the degree of nodes (the number of direct co-authors a MEP is related to) and the weight of edges (the frequency of co-authorships). The simple degree of a node can be interpreted as collaboration breadth; the weights can be interpreted as collaboration intensity. Adding the number of amendments (total and collaborative), we get a comprehensive picture of different forms of collaborative engagement in the INTA committee.

The maximum degree of a node in the network is 55, and the maximum weighted degree is 5188. Both degree distributions follow a power law distribution as is typical for social networks. A few nodes have many edges, while most nodes have only a few or none.

The next visualization shows the network from the perspective of how degrees and weighted degrees are distributed across the nodes. It shows collaborative breadth (number of partners) on the horizontal axis and collaborative intensity (weighted degree) on the vertical axis. The size of hollow outer circles represents how many amendments each MEP has authored in total. The filled inner circle shows how many amendments a MEP has co-authored with at least one colleague. The positioning of the circles in the scatterplot reveals distinct behavioral patterns that correspond to different influence strategies within INTA.

Collaborative Leaders occupy the upper-right quadrant, combining extensive partnership networks with intensive collaboration and typically high amendment productivity. These MEPs represent the collaborative elite who coordinate extensively across the committee. Hovering over the largest nodes in this region reveals MEPs like Mato (EPP) and Vedrenne (Renew).

Independent Producers appear as large nodes in the lower-left region of the visualization, achieving high amendment output primarily through individual work rather than collaborative partnerships. Notable examples include Puigdemont i Casamajó (NI) in the lower-left area with 704 amendments but no collaborative relationships, and Maurel (The Left) with 666 amendments and a limited co-authorship network.

Intensive Partners are those MEPs that have a high weighted degree relative to the number of co-authors. In the visualization, their nodes get pulled upwards on the left and middle part of the horizontal axis. They focus on deep collaborative relationships with fewer partners. For instance, Gallée (The Greens) has a small intensive co-authorship network, clearly confirmed by the small number of thick edges in the network visualization.

In contrast to the previous category, Broad Networkers position themselves at the lower region of the visualization, maintaining many collaborative relationships with lower intensity. MEPs in this category work with a large number of colleagues. However, they typically work on fewer amendments with the same partner. This pattern suggests that these MEPs perform roles as connectors or coordinators. Jongerius (S&D), who co-authored 167 amendments with 21 colleagues from four different groups, is an example. However, broad networking can also be shallow if it only concerns a small number of amendments compared to the total amendment activity of a MEP. For instance, a MEP can co-author a single amendment with 20 other colleagues, and draft 100 more individually. This MEP would still have a degree of 20.

That is why it is always useful to put collaborative measures like degrees into perspective by looking at the difference between the number of individual and collaborative amendments. Large nodes with small inner circles indicate Occasional Collaborators. Scholz (The Left) is an example of a MEP with a broad network, but also a strong focus on individual amendment productivity.

Different measures of node centrality

We can complement the exploratory analysis of collaborative styles by applying measures of network centrality. Centrality is about identifying the most influential MEPs based on their position within the network. In the context of amendment co-authorship, centrality captures different dimensions of political influence and collaborative importance. A highly central MEP may shape legislative outcomes not just through their individual contributions, but through their ability to coordinate with others, build coalitions, and facilitate broader collaborative efforts.

Some common measures of network centrality are not directly useful for this analysis. For instance, classic degree centrality (i.e. the total number of partners), has clear drawbacks as we have seen with the category of occasional collaborators. Even MEPs with many edges might still author most of their amendments individually without being a consistently active collaborator.

Combining degree centrality with edge weights (based on Opsahl) yields a centrality measure often used to analyze social networks. This measure recognizes that frequent, repeated partnerships carry more weight than occasional co-authorships. MEPs with high scores on this measure are both broadly collaborative (working with many different colleagues) and intensively collaborative (engaging in repeated partnerships).

The top five MEPs according to this type of centrality are:

1. MATO, Gabriel (ESP), EPP, Member
2. VEDRENNE, Marie-Pierre (FRA), Renew, Vice Chair
3. HAHN, Svenja (DEU), Renew, Substitute
4. WINKLER, Iuliu (ROU), EPP, Vice Chair
5. RAFAELA, Samira (NLD), Renew, Member

This list confirms the observations made in the upper right quadrant of the scatterplot. These five MEPs can safely be regarded as collaborative leaders as defined in our typology of behavioral patterns.

Another important measure of centrality is eigenvector centrality (also sometimes referred to as “prestige score“). It identifies who is best connected with other important nodes in the network. In other words, eigenvector centrality does not only look at the number of partners, but also their quality as central figures in the network. The top five MEPs according to this measure are

1. VEDRENNE, Marie-Pierre (FRA), Renew Vice Chair
2. HAHN, Svenja (DEU), Renew, Substitute
3. DECERLE, Jéremy (FRA), Renew, Substitute
4. ANDREWS, Barry (IRL), Renew, Member
5. SCHREINEMACHER, Liese (NLD), Renew, Member

This ranking shows some overlap with the previous list. However, we also learn something new: this list is exclusively composed of MEPs from the Renew group. The result underlines Renew’s position at the centre of collaborative work in the INTA committee during the 9th parliamentary term. MEPs from Renew had connections not only with their own leaders, but also with influential figures from other groups as can be checked with the interactive network visualization above.

Next, betweenness centrality indicates which nodes are most often on the shortest paths between any two other nodes. It is not directly applicable to the mere writing of amendments. But it could be interpreted as the capacity to activate communication and working relationships flowing from direct co-authorship partners to other MEPs. MEPs with a high betweenness centrality can be seen as bridge builders or coalition shapers. The list of top five MEPs has been calculated using the inverse weight of edges as distances, i.e. MEPs who collaborate more often together have closer links.

1. LOISEAU, Nathalie (FRA), Renew, External
2. RODRIGUEZ-PINERO, Inma (ESP), S&D, Member
3. DANTI, Nicola (ITA), S&D, Substitute
4. DANJEAN, Arnaud (FRA), EPP, Member
5. MATO, Gabriel (ESP), EPP, Member

Surprisingly, Loiseau (Renew) tops the list as an external contributor to the INTA committee. Her high rank corresponds to the collaborative pattern as a broad networker observed in the data (84 amendments published with 34 other MEPs). This is also the case for Danjean (EPP) who published 227 collaborative amendments with 44 colleagues. Danti (S&D) positions himself as a bridge builder between the S&D and Renew, while Rodríguez-Piñero (S&D) is at the centre of the S&D cluster with links reaching from Renew to the Greens and the Left. Mato (EPP), already identified as a major collaborative leader, also appears as a coalition shaper within the EPP as well as in relation to MEPs from S&D and Renew.

Finally, closeness centrality measures how close a node is to all other nodes. This measure identifies which MEPs have the best position to reach all other nodes in the network. Again, this measure is calculated using the inverse weight as the distance between nodes.

1. DANJEAN, Arnaud (FRA), EPP, Member
2. MATO, Gabriel (ESP), EPP, Member
3. LOISEAU, Nathalie (FRA), Renew, External
4. WINKLER, Iuliu (ROU), EPP, Vice Chair
5. WARBORN, Jörgen (SWE), EPP, Member

The list of top five MEPs shows some overlap with betweenness centrality. With Mato (EPP) and Winkler (EPP), two of the above identified collaborative leaders also appear on this list. Warborn (EPP) is the only new name. He focuses primarily on collaboration within his own group, with some links to the ECR group.

Conclusion

Amendment co-authorship relationships in the INTA committee share many structural features with social networks. Network analysis offers a view behind the scenes of EU trade policy making, focusing on the structure of collaboration and relationships in the INTA committee. This type of analysis holds valuable insights for professionals who want to keep track of the key players and how they engage in the legislative process.

This article has only scratched the surface of what can be done with these and other data from the Borderlex Datahub. Some key findings are:

Political groups are highly effective frameworks for collaboration. The modularity analysis demonstrates that existing political group boundaries align well with observed collaborative communities. In contrast, cross-party amendment co-authorships were exceptional. Moreover, nationality does not seem to be a major factor in selecting partners for amendment co-authorships. This validates the expectation that the EP should form a transnational counterweight to the intergovernmental dynamics playing out in the Council of ministers. Coordination within groups is driving collaborative work in the INTA committee.

MEPs follow a variety of collaboration strategies based on their position in the network. MEPs have used different pathways to exert influence on the amendments produced in the committee. The article has suggested a typology of collaborative engagement distinguishing collaborative leaders, intensive partners, broad networkers, occasional collaborators and individual contributors.

Using different measures of centrality as well as visual analysis, we can identify individual MEPs as central actors based on their position in the network. Vedrenne (Renew), Hahn (Renew), Winkler (EPP) and Rafaela (Renew) appear as the collaborative leaders that have shaped amendment co-authorships in the INTA committee during the 9th parliamentary term. Moreover, Renew stands out as the best connected group in the network. Other measures of centrality are well adapted to identify bridge builders in the network. MEPs such as Danti (S&D) and Danjean (EPP), who have achieved high betweenness centrality. They establish rare, but strategically relevant inter-group connections.

MEPs with formal roles, such as vice chairs, appear in the top five lists for different measures of centrality. However, the results indicate that network leadership does not depend on having a formal leadership role. Other MEPs, who were normal members, substitutes or even external members throughout most of the term, can occupy important positions in the network. This suggests that collaborative influence operates through both official and informal channels.

Limitations and Future Directions

While this analysis provides valuable insights into collaborative patterns, several important limitations suggest directions for future research:

Amendment success and actual influence. This analysis treats all amendments equally, regardless of whether they were adopted or rejected. Incorporating data on amendment adoption rates would enable analysis of which collaborative strategies prove most effective in achieving legislative outcomes. The effectiveness of amendment strategies also relates to understanding intentions behind amendments. Intentions can vary widely from contributing specialized knowledge, breaking political deadlock by formulating compromises, political signaling, and appealing to core constituents at home.

Semantic network analysis. Amendment co-authorship represents only one dimension of political collaboration. MEPs may coordinate through informal division of labor, submit separately authored amendments with similar content to amplify specific positions, or engage in behind-the-scenes negotiations that don't appear in formal documents. Analyzing amendment text similarity could reveal coordinated campaigns and thematic alliances that transcend formal co-authorship.

Temporal evolution. This analysis examines the complete 9th term as a single period, potentially missing important developments in collaborative relationships over time. Legislative priorities shift, political crises emerge, and relationships develop. All these factors could significantly alter collaboration patterns during a five-year parliamentary term. Comparing collaboration patterns across different periods within the same term or in subsequent terms could reveal how collaborative patterns evolve in response to changing circumstances and legislative priorities.

Predictive political analysis. The comprehensive network data creates opportunities for machine learning approaches to predict various political outcomes. Amendment success prediction could incorporate author centrality scores, cross-party collaboration patterns, and network positioning to forecast adoption likelihood. Beyond individual amendments, predictive models could identify MEPs likely to form future coalitions, anticipate cross-group cooperation opportunities, and forecast influence trajectories for newer members based on previous collaboration patterns.