Abstract
The European Union legal system is one of the most complex and sophisticated in the world. This article models the Acquis Communautaire (i.e. the corpus of European Union law) as a network and introduces the Evolution of European Union Law data set, which tracks connections between European Union primary law, European Union secondary law, European Union case law, national case law that applies European Union law, and national law that implements European Union law. It is the largest, most comprehensive data set on European Union law to date. It covers the entire history of the European Union (1951–2015), contains over 365,000 documents, and records over 900,000 connections between them. Legislative and judicial scholars can use this data set to study legislative override of the Court of Justice of the European Union, the implementation of European Union law, and other important topics. As an illustration, I use the data set to provide empirical evidence consistent with legislative override.
The European Union (EU) legal system is one of the most complex and sophisticated in the world. The Acquis Communautaire (i.e. the corpus of EU law) encompasses multiple international treaties, well over 100,000 legislative acts, and tens of thousands of court rulings. Unlike other international legal orders, the EU legal system is deeply integrated with the domestic legal systems of member states. National courts interpret and apply EU law in domestic litigation, and national parliaments enact legislation to transpose EU law into national law.
This article models, for the first time, the vast network of legal instruments that constitute the Acquis Communautaire and introduces the Evolution of European Union Law (EvoEU) data set, which records the structure of the network. The data set covers the entire history of the EU (1951–2015). It tracks over 900,000 connections between over 365,000 EU legal instruments across five domains of EU law: EU primary law (e.g. treaty articles), EU secondary law (e.g. directives, regulations, and decisions), EU case law (e.g. judgments by the Court of Justice, judgments by the General Court, and advocate general (AG) opinions), national case law that refers to EU law, and national law that implements EU law. These connections capture legislative and judicial outcomes that scholars are interested in, such as amendments to secondary laws, decisions regarding documents in court rulings, actions by national governments and courts to implement and apply EU law, and citations between legal instruments.
Legislative and judicial scholars will see many uses for this data. Due to space constraints, I focus on how scholars can use the data to study the relationship between the legislative and judicial processes. I present a brief research application that uses the EvoEU data set to show that member states are more likely to amend laws that have been interpreted by the Court of Justice of the European Union (CJEU), which indicates a feedback loop between the legislative and judicial processes consistent with legislative override—attempts by member states to constrain the Court by amending secondary laws that the Court has interpreted (Carrubba et al., 2008; Larsson and Naurin, 2016; Martinsen, 2015).
The primary empirical challenge to studying big-picture questions about the development of EU law is a lack of comprehensive data on the network of EU legal instruments that scholars can use to establish stylized facts and test theory. Recently, there have been a number of important data-collection efforts related to EU law, but these efforts, while overlapping in some cases, are largely uncoordinated (Carrubba and Gabel, 2015; Derlén and Lindholm, 2014; Frankenreiter, 2018; Häge, 2011; Hübner, 2016; Larsson and Naurin, 2016). This article addresses this challenge by introducing the EvoEU data set.
The data set contributes to the empirical study of EU law (a) by providing scholars with a comprehensive picture of how the many facets of EU law relate to each other and (b) by providing a framework for future data collection. Successful scientific cumulation in the study of EU law will depend on the public availability of comprehensive, compatible data sets that scholars can easily use and build on (Gabel et al., 2002). The EvoEU data set draws on raw metadata from EUR-Lex, the EU’s official archive of legal instruments. It contains the vast majority of the documents in EUR-Lex and uses the same document identifiers, so it provides a useful framework for the development of an ecosystem of compatible data sets. Existing data sets that include EUR-Lex identifiers (e.g. Carrubba et al., in press; Fjelstul, in press) will fit into this framework.
Theoretical questions
This section reviews open theoretical questions related to the relationship between the legislative and judicial processes, discusses the main data challenges related to theory-testing, and explains how the EvoEU data set addresses these challenges. Due to space constraints, I focus on recent large-N empirical work. Of course, there are many other important theoretical questions that scholars can research using the EvoEU data set, including how member states transpose EU directives into national law and how national courts interpret and apply EU law. I briefly point out these opportunities at the end of this section, but a review of these expansive literatures is beyond the scope of this study.
Scholars are interested in the strategic interaction between the Court and member states (Carrubba et al., in press; Carrubba and Gabel, 2015; Carrubba et al., 2008; Larsson and Naurin, 2016; Larsson et al., 2017; Martinsen, 2015). There is a long-standing debate in the literature about the degree to which member states can use the legislative process to constrain or empower the Court (Blauberger and Schmidt, 2017).
Secondary laws are incomplete contracts; they cannot fully specify the legal obligations of member states under every possible set of circumstances. When there are disputes about the obligations of member states under secondary law, the Court can facilitate cooperation by completing the contract—by declaring what constitutes a violation and what does not (Carrubba and Gabel, 2017). When the contract is sufficiently incomplete, national courts can refer questions about how to interpret a secondary law to the Court (under Article 267 TFEU). The Court issues a preliminary ruling that answers the national court’s questions and remands the case back down to the national court for a final ruling.
How do member states react when the Court attempts to complete the contract? Do they ever go back and update the contract by amending the secondary laws that the Court has interpreted? If the Court interprets a secondary law in a preliminary ruling, and enough member states disagree with how the Court has completed the contract, there are two basic mechanisms by which member states can try to undermine the Court’s ruling (Blauberger and Schmidt, 2017). First, member states could refuse to comply with the ruling. If the Court cares about compliance, we might expect it to make concessions to member states by not issuing adverse rulings that are too costly for member states to comply with (Carrubba and Gabel, 2015; Carrubba et al., 2008). Second, member states could amend a secondary law to override adverse rulings that interpret that law. Member states could also revise the treaties to curb the Court—a more extreme version of legislative override (Davies, 2016; Larsson and Naurin, 2016; Martinsen, 2015).
There is a growing empirical literature on legislative override (Blauberger and Schmidt, 2017). Larsson and Naurin (2016) find that the Court will be more sensitive to the third-party briefs filed by member states when it is easier for the Council to override the Court’s decision, such as in areas with a qualified majority voting (QMV) rule. Larsson et al. (2017) look at whether the Court is more likely to embed their judgments in case law, making them more persuasive, when they are likely to be politically controversial (e.g. when the Court disagrees with member states, the Commission, or the balance of member state briefs). They find that the Court uses precedent as political cover when issuing rulings that run counter to the preferences of the member states. Martinsen (2015) presents qualitative evidence of legislative override, looking at a small sample of directives.
But there is debate about whether legislative override is even possible (Blauberger and Schmidt, 2017). Davies (2016: 847), for example, argues that the notion of legislative override “rests on a misunderstanding of the distinctive features of the EU structure” and claims that it does not happen. Member states can only override the Court by amending secondary law when the Court interprets secondary law. Davies (2016) argues that most cases deal with the interpretation of the Treaties, so even if legislative override is possible in the subset of cases that do concern secondary law, the threat of override cannot significantly constrain the Court. I use the EvoEU data set to show that, in references for preliminary rulings, there are about twice as many instances of the Court interpreting secondary law than there are of the Court interpreting the Treaties, which casts doubt on this critique.
There are significant empirical challenges to studying this mechanism. As Blauberger and Schmidt (2017) point out, Larsson and Naurin (2016) argue that the Court is sensitive to the threat of legislative override, but they do not provide direct evidence that legislative override actually occurs. If it does not, threats of override are unlikely to be credible. To determine whether override actually occurs in practice, we need to know whether member states are more likely to amend secondary laws that have been the subject of court rulings. This requires (a) data on which secondary laws are affected by each court ruling and (b) data on amendments to all secondary laws, including those that have not been interpreted by the Court (to serve as a counterfactual).
Existing data sets of secondary laws (e.g. Häge, 2011; König et al., 2006) do not systematically track amendments. Existing data sets of court rulings (e.g. Carrubba and Gabel, 2015; Frankenreiter, 2018; Larsson and Naurin, 2016) do not include information on which secondary laws a case relates to. The EvoEU data set, in contrast, includes comprehensive data on which secondary laws have been interpreted by the Court (or treaty articles, for scholars who are interested in treaty revisions in response to court rulings) and on amendments to secondary laws. This will allow scholars to estimate the effect of court rulings on the probability that a law will be amended. The EvoEU data set also includes comprehensive data on citations in court rulings to existing case law, secondary law, and treaty articles, which will help scholars evaluate how the Court uses citations to deter noncompliance and override (Larsson et al., 2017).
Beyond this topic, scholars can also use the EvoEU data set to study how national courts apply EU law (Alter, 2000; Conant, 2002; Hübner, 2016; Tallberg, 2002). Empirical studies on the role of national courts have focused on preliminary rulings (Carrubba and Murrah, 2005; Stone Sweet and Brunell, 1998), but national courts frequently apply EU law without referring questions to the CJEU. Without data on the application of EU law in national court cases in which there was not a reference for a preliminary ruling, scholars have focused on variation in counts of preliminary rulings without knowing anything about the population of cases that generate them (e.g. Carrubba and Murrah, 2005; Stone Sweet and Brunell, 1998). Hübner (2016) introduces a database on national court cases that refer to EU law, but researchers have to extract the records for individual cases to code variables. The EvoEU data set extracts much of this information from the original source material (now part of EUR-Lex), including citations and whether the national court referred a question for a preliminary ruling, and makes it directly accessible to researchers.
Finally, scholars can use the EvoEU data set to study the legislative strategies that member states use to transpose directives (Angelova et al., 2012; Finke and Dannwolf, 2015; Franchino and Høyland, 2009; König and Luetgert, 2009; König and Luig, 2014; Steunenberg and Toshkov, 2009; Thomson, 2010; Toshkov, 2010; Zhelyazkova, 2013). A number of recent empirical studies rely on EUR-Lex metadata on national implementing measures (NIMs), which are national laws that member states use to transpose directives, in order to code transposition performance. For example, Toshkov (2008) study a random sample of 119 directives (1999–2005). Thomson et al. (2007) analyze a sample of 24 directives (1999–2000). Luetgert and Dannwolf (2009) look at a more representative sample of 1192 directives (1986–2003). However, none of these samples include data on the 13 member states that have joined since 2004. The EvoEU data set extracts the raw metadata from EUR-Lex (through 2015) and makes it available to researchers.
The EvoEU data set
To help scholars answer these questions about the evolution of EU law, I use network analysis to model the Acquis Communautaire and introduce the EvoEU data set, which records the structure of this network. The data set covers five categories of EU law: EU primary law (treaty articles), EU secondary law (directives, regulations, and decisions), EU case law (Court of Justice judgments, General Court judgments, and AG opinions), national case law (national court rulings that apply EU law), and national laws that member states pass to transpose directives (NIMs). It covers the period from 18 April 1951 (the date the Treaty of Paris, which established the European Coal and Steel Community, was signed) to 31 December 2015. Scholars can filter the EvoEU data set to recreate this network as it was at any point in time.
The Acquis Communautaire network consists of nodes and edges. Nodes represent documents and edges represent connections between those documents. As such, the EvoEU data set has two components: a nodes data set and an edges data set. Edges capture legislative and judicial outcomes scholars are interested in, including citations between documents, amendments to secondary laws, decisions with respect to documents in court rulings, and actions by member states to implement EU law. In total, I code 13 different types of edges, which I discuss below (see Table 1 for a summary). See the Online appendix for details about how I extract data from EUR-Lex and structure that information into the nodes and edges data sets (including information on reproducibility).
Summary of edge types.
Note: Secondary laws refer to directives, regulations, and decisions. CJEU judgments refer to Court of Justice judgments and General Court judgments.
The nodes data set includes approximately 365,000 legal instruments. I identify all documents by their Communitatis Europeae Lex (CELEX) number, which is a unique ID number assigned by EUR-Lex. The nodes data set includes the CELEX number of the document, the type of the document (e.g. Court of Justice judgment, AG opinion, directive, etc.), the date that the document was published, and the author of the document (e.g. the Court of Justice, the Commission, etc.). See the Online appendix for a codebook. Figure 1 summarizes the contents of the nodes data set.
Number of observations in the nodes data set by type of node.
The edges data set records all connections between all documents in the nodes network that are recorded in EUR-Lex metadata. Each connection is represented by one edge, which is one observation in the edges data set. Documents from the nodes data set only appear in the edges data set when they are connected to at least one other document by at least one edge. The document creating the connection is the outgoing document and the document receiving the connection is the incoming document.
I provide two versions of the edges data set: a document-to-document version and a document-to-clause version. In many cases, a document is not just connected to another document generally, but to one or more specific clauses of that other document. If that is the case, then, in the document-to-clause version, I code multiple edges, one for each separate clause (if applicable). The document-to-document version aggregates the edges in the document-to-clause version such that there is exactly one edge per pair of documents per type of edge. See the Online appendix for an in-depth explanation with examples. All of the descriptive statistics that I present are based on the document-to-document version of the data set. Figure 2 summarizes the contents of the edges data set.
Number of observations in the edges data set by type of edge.
The document-to-clause version of the edges data set includes the CELEX numbers of the two documents that are connected (i.e. the outgoing document and the incoming document), the types of the two documents (as recorded in the nodes data set), the type of the edge, and the relevant clause, if applicable. The document-to-document version of the data set omits the last of these. The document-to-clause version contains approximately 1.1 million observations, and the document-to-document version contains approximately 900,000 observations. Table 1 summarizes the different types of edges and indicates which types of documents can be connected by them.
Researchers can use the edges data set to construct quantitative variables. For example, they can filter the edges data set by document type or edge type. Then, they can group edges by the outgoing document to see all of the (older) documents that a document creates connections to or by the incoming document to see all of the (newer) documents that create connections to it. From there, they can count or aggregate edges.
Limitations and validity
Scholars interested in using the EvoEU data set should be aware of its limitations. Because it relies on EUR-Lex metadata, the EvoEU data set is only as complete and as accurate as EUR-Lex. As such, scholars using the EvoEU data set should carefully consider whether the contents of EUR-Lex could be biased in a way that affects their analysis.
EUR-Lex is administered by the EU Publications Office (which is part of the Commission) and is primarily based on the Official Journal of the EU. EUR-Lex contains all documents that appear in the Official Journal, which includes treaty articles, CJEU judgments, AG opinions, and secondary law (regulations, directives, and decisions). As such, the EvoEU data set should have complete coverage of these instruments. EUR-Lex’s coverage of documents that do not appear in the Official Journal, which includes NIMs and national court cases, is almost certainly incomplete. Moreover, scholars have rightly pointed out that these samples are probably biased.
With respect to national court cases that apply EU law, EUR-Lex only includes cases that the Commission has been made aware of. The database that Hübner (2016) introduces has been incorporated into EUR-Lex. Thus, the EvoEU data set has effectively the same coverage as Hübner (2016), but the data are available as a ready-to-use data set, rather than a database where users have to query individual documents to extract information. Hübner (2016) provides an excellent discussion about potential bias in EUR-Lex’s coverage of national court cases. With respect to NIMs, EUR-Lex only contains measures that are notified to the Commission by member state governments, which may not always have the capacity or incentives to accurately report information in a timely manner. Scholars do not know exactly how complete the EUR-Lex metadata is for NIMs, especially for earlier years (König and Luetgert, 2009).
The EvoEU data set does not solve the potential missing data problems that these scholars point out, but it does provide the most complete sample of national court cases and NIMs to date. And all of the data are in a ready-to-use format, with no need to query individual documents. Users should carefully review these studies to determine if the data are appropriate for their needs.
Coding connections between documents
Next, I describe the 13 types of edges that I code. Due to the large number of nodes and edges in the network, we cannot visualize the network directly. Instead, I use a circular dendrogram to visualize the structure of the network (see Figure 3).
Circular dendrogram of the EvoEU data set. Note: This figure visualizes the structure of the edges data set using a circular dendrogram. The inner ring includes the types of outgoing documents, the middle ring includes the types of edges that each outgoing document can create, and the outer ring includes the types of incoming documents that can be connected to each type of outgoing document via each type of edge. The size of the circles indicates the number of observations (based on the document-to-document version of the edges data set) in each branch of the tree (including any subsequent branches).
The first type of edge records citations between documents (“E1: Cites all or part of”), which is the most common (over 478,000 observations). Unlike existing data sets, which only include case law citations in court rulings (e.g. Derlén and Lindholm, 2014; Larsson et al., 2017), the EvoEU data set includes all citations made in case law, including AG opinions, and secondary law.
Figure 4 uses an alluvial plot to visualize the subnetwork for citations. The majority of citations in the network are made by court rulings and AG opinions. AG opinions tend to be more extensively researched than court rulings. Secondary laws can cite the treaties, other secondary laws, and sometimes even EU case law (citations to case law do not necessarily imply that a secondary law is overriding a court ruling). 1 Citations in national court decisions indicate how national courts are applying EU law. National court decisions rarely cite EU case law, but cite the Treaties more frequently than CJEU rulings. They are by no means hesitant to directly apply the EU Treaties.
Alluvial plot showing citations between documents. Note: The left axis indicates types of outgoing documents and the right axis indicates types of incoming documents. The flow lines indicate which types of outgoing documents cite which types of incoming documents. The width of the flow lines indicates the number of edges.
It is notable that citations patterns in AG opinions differ from citations patterns in court rulings. I calculate the Jaccard similarity (a standard measure of the overlap between two sets) between citations in AG opinions and citations in court rulings for cases that have an AG opinion. The AG and the Court often agree about which secondary laws to cite, but not necessarily about which treaty articles or case law to cite (see Figure 5). This indicates significant differences in legal reasoning. If we believe that the AG opinion reflects the legal merits of the case, as Carrubba and Gabel (2015) argue, this could indicate that the Court is playing politics, rather than ruling on the legal merits, and using citations strategically to deter noncompliance and override (consistent with Larsson et al., 2017).
Similarity between citations in Court of Justice judgments and AG opinions. Note: This figure shows the distribution of the Jaccard similarity between citations in Court of Justice judgments and citations in AG opinions (calculated on a case-by-case basis) for all cases with an AG opinion. Values closer to one indicate greater agreement about which legal instruments to cite.
I code six types of edges created by secondary laws, five of which relate to amendments to existing secondary law. A new secondary law can partially change the text of a clause of an existing law (“E2: Changes text in clause of”), or it can replace the text of the clause entirely (“E3: Replaces clause of”). It can add text to an existing clause (“E4: Adds text to clause of”), or it can insert an entirely new clause (“E5: Inserts new clause in”). It can also repeal one or more clauses, including the whole law (“E6: Repeals all or part of”). Figure 6 visualizes the subnetwork for amendments. The Commission reviews the text of each document, codes these relationships, and makes that information available as metadata in EUR-Lex. 2 See Table 1 for a detailed summary.
Alluvial plot showing amendments to secondary law. Note: The left axis indicates types of outgoing documents, the middle axis indicates types of edges, and the right axis indicates types of incoming documents. The flow lines indicate the ways in which different types of secondary laws amend each other. The width of the flow lines indicates the number of observations.
I also code a type of edge that records the legal basis of secondary laws (“E7: Has legal basis in”). Laws can have a legal basis in treaty articles or other laws. The legal basis of a law is stated at the beginning of the preamble, just before the recitals. Figure 7 shows the number of edges between each type of secondary law. It is common for secondary laws to have a legal basis in multiple legal instruments. Directives usually have a legal basis in a treaty article or another directive. Regulations usually have a legal basis in another regulation. Decisions, in contrast, rarely have a legal basis in other decisions.
Alluvial plot showing the legal basis of secondary laws. Note: The left axis indicates types of outgoing documents and the right axis indicates types of incoming documents. The flow lines indicate which types of secondary laws have a legal basis in which types of incoming documents. The width of the flow lines indicates the number of observations.
I code five types of edges created by court rulings. Court rulings can interpret documents (“E8: Interprets all or part of”), respond to references for preliminary rulings (“E9: Answers question referred by”), determine compliance by member states with documents (“E10: Determines compliance with”), uphold documents (“E11: Upholds all or part of”), or overturn documents (“E12: Overturns all or part of”). Figure 8 visualizes this subnetwork. Figure 9 uses the “E9: Answers question referred by” edge type to calculate the proportion of national court cases that refer a question to the CJEU. Again, the Commission codes all of this information and makes it available as metadata in EUR-Lex. The types of edges that a given ruling can create and the types of documents that a given ruling can be connected to depend on the legal procedure of the case. See the Online appendix for a discussion of the major legal procedures and the types of edges each can create. Table 2 summarizes this discussion.
Alluvial plot showing the effects of CJEU judgments. Note: The left axis indicates types of outgoing documents, the middle axis indicates types of edges, and the right axis indicates types of incoming documents. The flow lines indicate which types of CJEU judgments create which types of edges to which types of incoming documents. The width of the flow lines indicates the number of observations.
Proportion of national court cases that refer a question to the CJEU. Note: This figure shows the proportion of national court cases in each member state that refer questions to the CJEU via the reference for a preliminary ruling procedure.
Summary of edge types for CJEU judgments by legal procedure.
Note: This table omits the “E9: Answers question referred by” edge type, which connects references for preliminary rulings heard by the Court of Justice to national court decisions.
This data will allow scholars to study mechanisms by which member states can constrain the Court—like legislative override and treaty revision—by indicating which secondary laws and treaty articles the Court most frequently interprets. Scholars can use this data to assess how likely those secondary laws are to be amended, compared to secondary laws that the Court has not interpreted. The same can be done for treaty articles.
As previously mentioned, Davies (2016) argues that legislative override cannot happen in the EU because preliminary rulings almost always interpret treaty articles (usually the Treaty on European Union or the Treaty on the Functioning of the European Union). Contrary to this claim, the data show that there are about twice as many instances of the Court interpreting secondary laws than there are of the Court interpreting treaty articles. There are approximately 7900 instances of the Court interpreting a legal instrument (the Court can interpret multiple documents in a single preliminary ruling) and approximately 67% of these involve a secondary law, while only 33% involve a treaty article. Thus, there are plenty of opportunities for member states to override the Court.
Finally, I code one type of edge created by NIMs. I code an edge between each NIM and every secondary law that it at least partially implements (“E13: Transposes all or part of”). Figure 10 provides summary statistics. Member states use an average of approximately three NIMs to transpose a directive. Each NIM transposes an average of approximately 1.5 directives. There is significant variation across member states. Note that the distributions in Figure 10 both have very long right tails (not shown). Some directives take hundreds of measures to implement and some NIMs transpose hundreds of directives. 3
Summary of member state implementation activities. Note: This figure shows the distribution of the number of NIMs that member states use to transpose each directive and the distribution of the number of directives that each NIM transposes. It also shows the average number of NIMs per directive and directives per NIM by member state.
In sum, the EvoEU data set tracks connections between EU legal instruments that capture political outcomes including citations between documents, amendments to secondary laws, decisions with respect to legal instruments in court rulings, and actions by member states to implement EU law. Scholars can filter the edges data set by date to reconstruct the network of EU law at any point in time. See the Online appendix for descriptive statistics on how the structure of the network has changed over time.
Research application
In this section, I present a brief research application to illustrate how scholars can use the EvoEU data set to study feedback between the legislative and judicial processes. I present empirical evidence that member states are more likely to go back and amend laws that the Court has interpreted. If legislative override is happening, the primary observable implication is that member states should be more likely to amend laws that the Court has interpreted. Such evidence is necessary but not sufficient to demonstrate override because member states could be trying to codify court rulings into secondary law, as opposed to overriding them (Martinsen, 2015). 4 Using event history analysis, I find empirical evidence consistent with this implication. My findings highlight a need for more theoretical and large-N empirical work on legislative mechanisms by which member states can constrain the Court in response to adverse rulings.
I use event history analysis to assess whether member states are more likely to amend laws that the Court has interpreted. The objective of event history analysis is to estimate the hazard rate of failure. The hazard rate is the probability of failure at a specific point in time conditional on it not having failed up to that time. In this case, failure is the amendment of a secondary law. I focus on amendments to directives, which are more politically salient than regulations. There can be multiple failures because directives can be amended multiple times. These failures are ordered in the sense that a directive cannot be amended a second time until it is amended the first time.
I estimate the hazard rate of member states amending a directive as a function of the natural log of the cumulative number of preliminary rulings in which the Court has interpreted that directive. 5 I use the document-to-document version of the EvoEU data set to identify every instance of member states amending a directive and every instance of the Court interpreting a directive in a preliminary ruling. The data include, according to EUR-Lex, the full universe of such events over the entire history of the EU (1951–2015).
I reshape the data into the standard format for event history data. Each observation in the event history data set is an interval of time. I observe every directive every day starting on the day it is adopted until the day it is no longer in force or until the last day I observe them, which is day the EvoEU data set ends (31 December 2015). 6 If a directive is no longer in force, it exits the risk set, which is the set of directives that are at risk of being amended. Every time member states amend a directive or the Court interprets a directive, there is a new observation. The data are right-censored because every directive is still at risk of being amended on the final day of observation, unless it is no longer in force. The sample includes 3986 directives. 7 There are 11,718 time intervals (observations). The shortest time-to-amendment (or right-censoring) is three days, while the longest is 20,025. The directives in the sample are cumulatively at risk of amendment for a total of 21.8 million days. There are 5421 failures (amendments).
I estimate the hazard rate using a Cox proportional hazards model. I use the standard Anderson–Gill approach to defining the risk set (Andersen and Gill, 1982), which assumes that the risk of amendment is the same regardless of how many times the directive has previously been amended. 8 I cluster the standard errors by directive. Using metadata from EUR-Lex, I control for several observable characteristics of the directive, including whether the directive was enacted by the Commission (as opposed to the Council and Parliament), whether the legal basis of the directive was one of the EU Treaties (as opposed to another secondary law), and whether the directive was addressed to a specific subset of member states (as opposed to all member states).
Figure 11 presents the results. See the Online appendix for a regression table, tests of the proportional hazards assumption, and other robustness tests and alternative specifications. I find that a one-unit increase in the log of the number of cases is associated with a 44.27% increase in the hazard that the directive will be amended. The standard deviation is 1.18, so that a two standard deviations change approximately doubles the estimated hazard rate (a 104.48% increase). This result indicates that member states are legislatively responsive to court rulings: they go back and revise the contract in response to efforts by the Court to complete it. This finding implies that, in some situations, neither the Court nor member states achieve their first-best outcome. The Court does not always get its preferred interpretation of the law (member states will not always amend laws in ways that are fully consistent with the Court’s preferences), and member states do not always get to keep the policy they enacted, as originally written.
Coefficient plot for Cox proportional hazard model. Note: This figure shows the results of a Cox proportional hazards model. The x-axis shows the hazard ratio with asymmetric confidence intervals. As the log of the cumulative number of cases increases, the hazard of amendment increases. This is consistent with legislative override.
In sum, the data indicate a feedback loop between the legislative and judicial processes. Since laws are incomplete contracts, national courts often ask for clarification (via the preliminary reference procedure) about how to apply laws. The Court helps national courts by attempting to complete the contract. This prompts member states to revise the contract. Future research should focus on discriminating between legislative override and codification—that is, on determining how member states are changing the contract and whether those changes are consistent with the preferences of the Court.
Conclusion
This article models the Acquis Communautaire as a network of legal instruments and introduces the EvoEU data set, which tracks connections between them across the entire history of the EU (1951–2015). The data set contains over 365,000 legal instruments and tracks 13 different types of connections between them that capture legislative and judicial outcomes scholars are interested in. To illustrate how scholars can use the data set to study legislative override, I provide a brief research application. I find empirical evidence that member states are more likely to amend directives that have been interpreted by the Court, which indicates a feedback loop between the legislative and judicial processes.
Researchers can easily merge the EvoEU data with other data sets that use CELEX numbers. For example, it can be merged with Fjelstul and Carrubba (2018), which contains information on Commission infringement proceedings. It can also be merged with data from Fjelstul (in press) and Carrubba et al. (in press), which both contain extensive information on Court of Justice cases. For scholars interested in the legislative process, it can be combined with data from the European Union Policy-Making (EUPOL) data set (Häge, 2011).
Scholars can build on the EvoEU data set by developing new, compatible data sets that focus on particular types of documents or particular types of relationships between documents. Since it includes a comprehensive list of legal instruments (with CELEX numbers), the EvoEU data provides scholars with a framework on which to build. Scholars can use this list as a starting point for collecting or coding additional variables of interest, such as information about the preferences of relevant actors, or the rules of relevant institutions. I hope this will encourage more collaborative data collection in the study of EU law (Gabel et al., 2002: 495).
Supplemental Material
Supplemental material for The evolution of European Union law: A new data set on the Acquis Communautaire
Supplemental material for The evolution of European Union law: A new data set on the Acquis Communautaire by Joshua C Fjelstul in European Union Politics
Footnotes
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
