Legislative amendments and informal politics in the European Union: A text reuse approach

Abstract

The ability to amend legislative proposals introduced by the Commission is central to legislative process in the European Union. Despite this, very few attempts have been made to capture and explain such amendments. This study addresses this gap in the literature by considering the changes between the Commission’s proposals and the final legislative outcome passed by the European Union. It does so by implementing minimum edit distance algorithms to measure changes between legislative proposals and outcomes. The findings suggest that legislative amendments are determined by the formal and informal institutional structures in which negotiations take place and characteristics of the proposal itself. Our conclusions contribute to the ongoing debate on the nature and distribution of legislative powers in the European Union.

Keywords

Bargaining co-decision informal politics legislative studies quantitative text analysis

Introduction

Since its inception, the European Union (EU) has witnessed significant debate over the legislative proposal powers awarded to the European Commission by the Member States (Rasmussen, 2007a). The European Commission has been allocated the sole right to propose legislation (Pollack, 1997).¹ Once a proposal has been introduced, the Council of Ministers and the European Parliament have formal powers to reshape the legislation by proposing and approving amendments. These amendment powers are shaped by the legislative procedures and voting thresholds governing the negotiation process.

In parallel to the formal institutional rules shaping the legislative process, informal decision-making processes and settings including ‘early agreements’ and ‘conciliation committees’ have become more common since their introduction with the Treaty of Amsterdam. Some argue they represent an attempt by the Council and Parliament to overcome formal procedures and thus reduce the agenda-setting power of the Commission (Reh et al., 2013). Whether or not this empowers the Council and Parliament to further amend Commission proposals is not known.

Despite the clear link between formal and informal rules and amendment abilities, the degree to which they impact upon legislative proposals in practice remains an open empirical question. In this study we directly assess the impact of formal and informal institutional settings on amendment capabilities. In order to achieve this goal, we introduce a new set of automated text analysis methods based upon the concept of minimum edit distances, and show how these methods are capable of extracting amendments from legislative texts in a systematic, replicable and consistent manner. We examine the usefulness of these methods for capturing legislative amendments using a new dataset of legislative proposals decided upon between 1994 and the 2013. The findings presented demonstrate that amendments are influenced by variation in decision-making rules under different legislative procedures, as well as by informal arrangements such as early agreements, and conflict within and between the EU institutions about what a piece of legislation should look like. Our findings have implications for the study of agenda setting and the influence of formal and informal institutional rules and norms over policy outcomes. Furthermore, the method we introduce for automatically tracking amendments to versions of political text are applicable to alternative settings and have the potential to significantly reduce the time and costs associated with detecting policy change.

Literature review

Agenda setting

A fundamental characteristic of the EU legislative process is the fact that formal proposal powers have been delegated to the Commission (Art. 225 Treaty on the Functioning of the European Union (TFEU)). The impact of this Treaty provision on the agenda-setting power of the Commission has been much debated (see Kreppel and Oztas, 2016). Some maintain that the Commission is the dominant EU agenda-setter due to its ability to frame issues when making proposals (Pollack, 1997), while others argue that the Commission must account for the views of other actors when setting the agenda (Crombez and Hix, 2011). Agenda-setting power is thought to derive from the Commission’s ability to structure the agenda, to exclude certain items and to frame issues in a certain way (Daviter, 2007). This ability to set the agenda is shaped by leadership factors and Commissioners’ nationality (Wonka, 2008), party identification (Hix, 2008), the influence of interest groups (Varela, 2009), treaty change (Crombez and Hix, 2011) and the gradual evolution of the legislative process over time (Höing and Wessels, 2013).

As Kreppel and Oztas (2016) point out, much of the empirical evidence rests on indirect measures of agenda-setting power based on case studies or on the degree to which the Commission successfully introduces legislation in line with the policy priorities set out in its Annual Work Programmes (Osnabrügge, 2015). This study takes a more direct approach to capturing the ability of the Commission to affect policy outcomes by considering the degree to which its proposals are amended.

Formal legislative procedures

The manner in which formal rules empower and constrain different actors at different stages in the EU legislative process has been explored using formal models (Crombez, 2001; Tsebelis and Garrett, 2000).² Under the consultation procedure the Commission proposes as agenda-setter, while Parliament provides an opinion and the Council decides (Crombez, 1996), while under the co-decision procedure, the Parliament has a significantly strengthened role (Tsebelis and Garrett, 2000). The most well developed attempt to assess the predictive accuracy of these formal models is the Decision Making in the EU project (Thomson, 2011; Thomson et al., 2006, 2012). This project compared the predictive accuracy of a large selection of formal models of the legislative process. The main finding was that models that explain legislative outcomes as a result of the distribution of power, policy positions and relative saliency performed better than those that relied solely on the formal rules of procedure. This suggests that one must go beyond formal rules for a more complete explanation of legislative outcomes.

Informal bargaining processes

A second set of literature focusing on the informalisation of EU decision-making has also recently developed which argues that the introduction of early agreement under the Treaty of Amsterdam, as well as the use of the conciliation committee, has lead to the development of informal bargaining processes. In both these settings, the role of formal decision-making rules is much reduced, resulting in a redistribution of power away from the Commission.

Early agreements involve the Commission, Council and Parliament reaching an informal compromise on legislation at the first or early second reading stage of co-decision (Rasmussen and Reh, 2013). Farrell and Héritier (2004) argue that early agreements empower relais actors such as the Rapporteur in the Parliament and the sitting Council President. This is because early agreement negotiations are: (a) held in closed-door settings; (b) based on decision-making processes occurring simultaneously between institutions rather than sequentially; (c) governed by informal decision-making processes; and (d) subject to lower voting thresholds where formal rules are in play (Hagemann and Høyland, 2010). Interestingly, Rasmussen and Reh (2013) found that none of these factors seem to affect the relative bargaining success of the Council and Parliament. This is surprising given the rise of early agreements as a mode of decision-making and the theoretical expectations outlined above. We aim to reexamine the idea that early agreements impact upon actors’ ability to amend the Commission proposal.

Conciliation occurs if all other decision procedures under co-decision have failed to produce a text acceptable to the Council and Parliament. The conciliation committee brings together representatives of the Council, Parliament and Commission, who are delegated power to find a compromise (Rasmussen, 2008). In this setting informal bargaining processes are thought to dominate. Franchino and Mariotto (2013) examined the relative success of the Council, Commission and Parliament at getting their internally agreed-upon text reflected in the final legislation decided upon. They show that on average, the increased utilisation of co-decision over time strengthens the Parliament's hand vis-a-vis the Commission and Council. This rather encouraging finding from the viewpoint of the Parliament is however tempered by the fact that the Parliament is at a disadvantage when negotiations reach the conciliation committee, as the structure of conciliation negotiations is biased in favour of the Council. Based on extensive interviews, König et al. (2007) showed that the Parliament wins most conflicts, but that the Council is more successful in multi-dimensional disputes. They further provide evidence that the Commission has an impact on conciliation outcomes, despite not being a party to agreements, by aligning itself with one of the other institutions. Rasmussen (2008) furthermore showed that while there is potential for conciliation committees to act as runaway agents, given the amendment powers delegated to them, in practice this rarely happens. In general, the delegations from each institution share the same goals as their parent institutions, and information asymmetries between the parent institutions and their representatives on the conciliation committee tend not to be too serious.

Theory

We first consider formal institutional rules and their expected impact upon the ability of the Council and Parliament to amend Commission proposals. Two formal legislative procedures shall be examined; the consultation procedure and the co-decision procedure. The consultation procedure³ is the simpler of the two so we begin with that.

The consultation procedure begins when the Commission introduces a proposal. The Parliament must then examine this proposal and provide a non-binding opinion to the Council. The Council then considers this opinion, and any amendments that Member States wish to put forward and then votes on the text. Any amendments approved of by the Council must be adopted under the unanimity voting rule. If no amendments are adopted, the Council can assent to a proposal by qualified majority. The broader implications of this institutional setup are that amending Commission proposals is difficult due to the unanimity requirement in the Council, and the minimal formal role of the Parliament.

Under the co-decision procedure,⁴ the Parliament holds much greater formal power to influence the decision-making process. This procedure is initiated when the Commission submits a proposal to the Council and Parliament for negotiation and decision. Discussions proceed on a parallel basis within both the Council and Parliament until the Parliament introduces its first reading opinion to the Council, decided upon by a simple majority of Members of the European Parliament (MEPs) present in plenary. At this point, the Commission can decide to withdraw a proposal if there is not enough support evident. If this does not happen and the Council approves of the Parliaments position by qualified majority, then the act can be adopted without further debate. If the Council wishes to change the positions presented by the Parliament the rules for doing so depends on the position of the Commission. If the Commission approves of the proposed amendments it adopts its own positions by qualified majority, if the Commission does not approve unanimity in the Council is required. The proposal is then returned to the Parliament for a second reading.⁵ At this stage, the act is adopted if the Parliament approves of the position submitted by the Council by simple majority, while any re-introduced amendments or rejection of Council amendments must be approved by absolute majority. Should the Parliament reject the Council text, then the law fails to be adopted. If it modifies the Council text, the Commission provides an opinion and the modified text is then passed back to the Council. At this stage, the Council can either approve the text as it stands by qualified majority or convene a conciliation committee that brings together Council and Parliamentary representatives who attempt to agree upon a compromise text. If such a text is agreed upon, it must be approved by both the broader Council and Parliament plenary before it can become law. If such an agreement cannot be reached, the proposal can be further negotiated, shelved by the Council President, or withdrawn by the Commission.

The first theoretical expectation that emerges from the differences between these two legislative procedures is that the Commission proposal will see more amendments under the co-decision procedure than under the consultation procedure. This is due to the argument that amendments are significantly more difficult to agree upon under consultation as they always need to be agreed upon unanimously in the Council.⁶

H1: Proposals will see more amendments under the co-decision procedure.

The second set of theoretical expectations to be explored is that the number of amendments observed under the co-decision procedure will vary significantly depending upon the informal decision-making processes in play at each stage of negotiations. By informal decision-making processes we mean, in line with Reh et al. (2013), decision-making rules that are non-codified and enforced outside official channels, which operate in a setting characterised by restricted participation and seclusion, and where the outcome requires subsequent formalisation. These informal processes are most prevalent during the first and third reading stages of co-decision, where early agreements and conciliation committees come into play.

Early agreements have become more and more frequent since their introduction with the Treaty of Amsterdam (Reh et al., 2013). They represent an attempt to fast-track the legislative process through informal compromise between the Council and Parliament in ‘trilogues’, before either body issues its opinion or common position. When early agreements take place, the Council and EP reduce the formal agenda-setting power of the Commission by agreeing on amendments that are integrated into the EP text and then voted through by the Council at the first reading stage. The EP-amended text essentially takes account of both the Council and EP amendment demands in one step outside the formal negotiation settings, thus avoiding some of the more stringent majority requirements at the second and third reading stages of co-decision. In practice this should allow significantly more amendments to a Commission proposal than would otherwise be the case, as such amendments are easier to push through.

H2a: Proposals will see more amendments if early agreements are used under co-decision.

An important subtlety of the co-decision procedure at the third reading stage is that the Parliament and Council can collude against the Commission and force negotiations into a conciliation committee where formal voting rules and procedures are less of a constraint on the two institutions (Tsebelis and Garrett, 2000). Conciliation committees are made up of an equal number of representatives from the Council (one from each member state) and the Parliament (assigned by quotas reflecting the composition of the Parliament), alongside a team of Commission officials. Within conciliation committees, a compromise text is negotiated informally, and a qualified majority of the Council representatives and a simple majority of the Parliamentary representatives must approve of amendments therein. The compromise text is then voted on by the Council with a qualified majority and by the Parliament with a simple majority. Importantly, once a conciliation compromise has been agreed, no further amendments are possible and the text is either accepted or rejected as is by the broader institutions (Rasmussen, 2008). The informal bargaining processes involved in creating a compromise text, the reduced majority demands for approving this text and the reduced role of the Commission in conciliation committees, suggest that there should be more successful amendments to a Commission proposal in this context.⁷ Actor preferences also play a role, as it could be the case that the Council could agree with the Commission and be opposed to suggested Parliamentary amendments (or vice versa). The fact that a third reading has been reached indicates significant differences between the policy demands of the actors, which is a second reason to expect more amendments at this stage of negotiations.

H2b: Proposals will see more amendments if the conciliation committee stage is reached under co-decision.

Successful amendment attempts can also be based upon an institution demonstrating that internal controversy exists over what a proposal should look like. Such controversy is usually resolved by relevant parties trading support for amendments they care less about in return for support on amendments they care more about (Arregui, 2008). The net effect of such position trading will be increased amendments to the original proposal. Furthermore, internal conflict within an institution can increase the uncertainty surrounding its ‘true’ position, and can thus lead to Commission mistakes in presenting an acceptable proposal to other actors in the first place. We use as a proxy the length of time it takes each institution to reach internal agreement on a proposal to capture internal controversies. The intuition here is that the time it takes for each institution to reach internal agreement reflects the level of disagreement that exists within that institution over the Commission proposal. The result of this disagreement is that the institution in question will present amendments to the Commission proposal, which are likely to be accepted in order to placate dissenting internal actors.

H3a: Proposals will see more amendments as the time it takes for the Council to reach a common position increases.

H3b: Proposals will see more amendments as the time it takes for the Council to reach a reasoned opinion (under consultation) or as the time it takes the Parliament to supply a first reading agreement (under co-decision) increases.

Beyond our major hypotheses of interest, we also control for other confounding factors that may influence whether or not a proposal is amended. We control for the type of legislation under consideration as this relates to the amount of discretion member states are likely to enjoy when implementing a piece of legislation (Thomson and Torenvlied, 2010). Our dataset contains a selection of EU decisions, directives and regulations. Regulations must be directly transposed into national law as they appear in the text agreed upon, directives in general allow for more discretion than regulations, and decisions are only relevant to those to whom they are addressed. As a result, we expect that regulations will be subject to the most amendments, directives will be subject to somewhat less amendments and decisions will be the least amended of the legislation analysed.

We also control for treaty changes (from Amsterdam to Nice to Lisbon) which have empowered the Parliament vis-a-vis the other institutional actors (Kreppel, 2002; Rasmussen, 2007b), and introduced new decision-making procedures like early agreements (Farrell and Héritier, 2004). Such treaty changes represent a form of external shock to the legislative process (Bølstad and Cross, 2016), which can be expected to affect the amount of successful amendments to Commission proposals. We control for the enlargement round in 2004, as adding new member states represents another form of external shock to the legislative process that broadens the variety of interests represented, and is thus likely to increase the number of attempted amendments (Aleskerov et al., 2002). We control for an annual time trend in order to capture the gradual evolution of the legislative process over time. Finally, we control for proposal characteristics including the length of a proposal and whether or not a proposal was withdrawn and amended by the Commission previously.

Capturing amendments using minimum edit distances

Amendment tracking

The goal of understanding the legislative process has driven existing amendment-tracking efforts in the literature. Tsebelis et al. (2001) track legislative amendments throughout negotiations by examining nearly 5000 separate amendments to a selection of 231 examples of EU legislation negotiated under both co-decision (79 cases) and co-operation (152 cases) between 1988 and 1997. These amendments were tracked over the course of negotiations in order to capture the degree to which they were adopted. The main finding of this study was that while the co-decision procedure empowers the Parliament relative to the other institutional actors, the Parliament also has important conditional agenda-setting power under the co-operation procedure.

Franchino and Mariotto (2013) made progress in automating the process for detecting amendment success by using Slapin and Proksch (2008)’s WORDFISH algorithm. WORDFISH takes raw political texts and estimates policy positions based on a ‘bag of words’ approach and a statistical scaling model of word frequencies in documents. This method has been successfully been applied to problems as diverse as estimating MEP policy positions from Parliamentary speeches (Slapin and Proksch, 2010), and estimating Japanese political party positions on the basis of election pledges (Proksch et al., 2011). The major advantage of utilising such a method is that it minimises coder-reliability problems and allows for easy replication of results. However, a major disadvantage of this particular application is that amendment success is not directly captured, but instead estimated at the proposal level based on similarities in word frequencies across versions of a proposal. The document-level measures produced also hide significant variation in amendment success across versions of legislation and are weakly correlated with direct measures of amendment success.

In this study we introduce a new automated text analysis method to the political science literature, which is based on the idea of text reuse. We capture successful proposal amendments based upon the minimum edit distances between the initial proposal and the final policy outcome (Levenshtein, 1966). This method has been developed in bio-informatics, computer science and natural language processing to measure the number of edit operations (insertions, deletions, or substitutions) required to change one string of characters or words into another (Wagner, 1974). Minimum edit distances have successfully been applied to problems as diverse as creating accurate spell checkers (Wagner, 1974; Wagner and Fischer, 1974), assessing differences between different dialects in computational linguistics (Kessler, 1995; Nerbonne and Heeringa, 1997) and assessing genetic alignments in computational biology (Fitch and Margoliash, 1967; Henikoff and Henikoff, 1992). This method of capturing the change between two text strings is suitable for detecting amendments, as the basic structure of the problem of is essentially the same as in the previous applications mentioned.

The most similar work that can be found in the political science literature is that of Wilkerson et al. (2015). In that work the authors utilise text-reuse methods to track the promulgation of legislative text across sets of laws emerging from the US Congress. This is achieved by taking account of the sequencing of words in a text using a ‘n-gram’ approach to capture similarities between sub-sequences of text across documents (Smith and Waterman, 1981). Their results demonstrate that many bills in the US Congress are in fact a mixture of text (and thus policy ideas) that first appeared in earlier legislative proposals. The most fundamental difference between the approach of Wilkerson et al. (2015) and the one taken in this study is that their work is geared towards detecting similarities between texts whilst ours is geared towards detecting differences. While the underlying logic of the algorithms is very similar, the application in practice is very much different.

A number of explicit assumptions must be made when employing minimum edit distances to measure changes to legislation. The first is that changes to a legislative text have substantive meaning in terms of policy outcomes. There are a number of arguments justifying this assumption. Legislative texts most often consist of definitions of concepts, or the rights and responsibilities of one party to another. Over time, and in order to reduce legal ambiguities, the structure, style, vocabulary and grammar of these definitions become subject to very strong norms and best practices within a polity.⁸ This formalistic and precision-centred nature of legislative texts means that there are very few, if any, alternative ways of expressing the same legislative message. Precision is the guiding principle of legislative drafting. Ambiguities may of course arise by accident, or be unavoidable, but recent research shows that they often represent a conscious attempt by the drafter to refer the interpretation to relevant courts (see for example Wallace, 2012). In other words, even when ambiguities do arise in legislation they are usually carefully planned and deliberate. In addition, and in contrast to other written or verbal messages, legislative texts (when entering into force) have a unique impact on the real world. Any single change may have substantive consequences for those subject to the law, in terms of their rights and responsibilities. Because of these singular characteristics of legislative texts, political conflict in legislative bodies is almost always focused on the exact wording of laws. The legislative rules involved in the writing of laws further ensures that making changes to legislative texts is difficult, meaning that spurious or non-salient changes are unlikely to be successful.

Since ‘every word matters’, substantively, legally and politically, we argue that the number of words that have been edited between versions of a legislative text is a credible indicator of the number of substantive changes that have been made to the document. But exactly how accurate is such an indicator? It is difficult to argue strongly, for example, that three edits represent a three times larger change than a single edit, at least without knowing anything more about the specific words or context. This is especially true for longer documents, given a certain probability for unplanned ambiguity, misspellings and grammatical errors per word. However, if for example one hundred words have been edited, in contrast to three, it can reasonably be assumed that this represents a larger, substantive change in the consequences of the law. In our analysis, the measure ranges between 0 and 37,934 edits between the European Commission’s proposals and the final legislative acts decided upon. In other words, the practical range of the indicator gives confidence that we are indeed observing large differences in the amount of substantive change to the Commission’s proposals demanded by the Council and Parliament.

Finally, there are a number of advantages to using this methodology to measure changes to legislation over the course of negotiations. The first advantage is that it gives an easily replicable measure that is comparable across different legislative decisions. The second advantage is a gain in efficiency, in that it can be applied to a much larger selection of texts than would be possible with hand coding alone. This enables us to examine all legislative proposals between 1994 and the present, rather than a sample as has been the case in the literature to date.

DocuToads: Introducing a new minimum edit distance algorithm

This section describes the implementation of a new minimum edit distance algorithm, which we refer to as DocuToads.⁹ We include an extensive set of appendices explaining how this algorithm functions, and validating algorithm outputs through replication of existing studies of amendment tracking found in the literature. Before outlining how this algorithm functions, we briefly discuss an existing algorithm found in the literature and identify an important weakness of the algorithm that makes it unsuitable for studying the evolution of legislative texts. Both the existing and new algorithms are described in terms of edit operation on text strings, with each edit operation operating on a single element of a string. Here and in the analyses that follow the basic element of a string is considered to be an individual word.

The classic minimum edit distance algorithm that appears in the existing literature is the Levenshtein distance (Levenshtein, 1966). The Levenshtein distance is calculated as the minimum number of edit operations (insertions, deletions or substitutions) required to change one string into another. These edit operations essentially define amendments to a text, and by tracking the number of edit operations required to convert one version of a text into another we get a detailed account of proposal amendments.

One important weakness of the Levenshtein distance is that when applied to two versions of a text, it places a particularly heavy weight (penalty) on situations where sections of text have moved location within the document, i.e. when a text transposition has occurred. This is problematic when tracking amendments to legislative texts, as in the amendment process, it is quite common for whole articles to move position within a larger text. In such cases, the text in question remains in the proposal, but has been moved to a different location. For the minimum edit distance application applied here, such changes need to be discounted, as they do not entail any policy changes, but instead represent a reorganisation of a legislative text.

The Levenshtein distance cannot account for these text transpositions, as it focuses on dynamically determining how to edit (delete, insert or substitute) a document locally on a word by word basis. In order to avoid the heavy penalisation associated with moving large sections of a string using the standard Levenshtein distance algorithm, while retaining accuracy and computational efficiency, we propose the DocuToads edit distance algorithm, which has been developed to account for such transpositions.

The first step of the DocuToads algorithm is very similar to that of the Levenshtein distance algorithm. A matrix is created with the width and height of the longest of the two strings (in number of words), with an additional row of zeroes at the top and an additional column of zeroes at the far left to represent the null string of each. Each of the cells in the matrix, termed D(i, j), represents a comparison between one word in the first text, S₁(i), and one word in the second text S₂(j). The algorithm then proceeds to fill the empty cells using a dynamic-programming approach, starting from the top left and working row by row until all elements of each text have been compared. If the two words being compared in a given cell do not match, the value of that cell is set to zero; D(i,j) = 0. If the two words do match, the value is set to the value of the cell diagonally up to the left plus one; $D (i, j) = D (i - 1, j - 1) + 1$ .

D (i, j) = {0; if S_{1} (i) \neq S_{2} (j) D (i - 1, j - 1) + 1; if S_{1} (i) = S_{2} (j)

(1)

In the second step, a vector is constructed of the maximum values in each column. These maximum values are found in the cells of the matrix representing the best match for every word in the second text (i.e. the part of the longest unbroken sequence of words in the first text). It can be easily shown that an unbroken sequence of matching words will result in an unbroken rising sequence of values in the vector of the maximum values in each column. Furthermore, any additions, deletions, replacements, copy-paste or cut-paste edits will cause breaks to that sequence of rising values, and these breaks can be detected and accounted for by the algorithm to yield the minimum edit distance. In particular, every zero in the vector of rising values (except the very first one) corresponds to a word that exists in one of the texts, but not the other. In addition, every positive value in the vector that does not equal the previous value plus one corresponds to a copy-paste or cut-paste edit of the text. Following these two simple rules, the algorithm counts the instances of zeroes or broken progression of values to find the minimum number of edits needed to transform text one into text two. It is also possible to record not just the number of edits needed, but also which exact combination of edits (e.g. words added, deleted, substituted or transposed) most efficiently transforms the first text into the second. This provides the possibility to detect where in a text changes occurred and to perform secondary analysis of for example topics as well as human validation of the algorithm. A step-by-step walkthrough of the DocuToads algorithm can be found in the Online appendix.

By reducing the string matching problem to the analysis of one vector, the algorithm only requires (if the length of the longest text is n) n²+ 2n computations, very much comparable to the fastest minimum edit distance algorithm available (Levenshtein, 1966), which requires (if the length of the texts are n and m, respectively) n × m computations. This represents a large improvement on suffix trees, which are another existing method for detecting cut–paste edits, but which requires at least $(n \times \log (n) + m \times \log (m)) + n + m$ computations. Our approach therefore allows analysis of much larger corpora of text.

Replication exercise

To validate the algorithm, and to demonstrate that automated methods can perform just as well as and sometimes better than hand-coding efforts, we replicated the hand-coded amendment tracking efforts performed by Franchino and Mariotto (2013). The Online appendix provides a full account of this replication, so here we focus only on the substantive results of the replication process. Franchino and Mariotto (2013) encode substantive differences between the three draft versions of a sample of 20 EU legislative acts that were resolved by Conciliation Committee. The authors categorize 525 substantive amendments in the recitals and articles of these texts, and associate them with particular legislative articles. As a first step, we simply compare the number of human-coded amendments with the DocuToads minimum edit distance for the 60 available document pairs. Figure 1(a) demonstrates the relationship between the two measurement, suggesting a strong linear correlation (R²= 0.84). Second, we compare the automated and manual approaches at the article level. We sum all human-coded amendments on an article-by-article basis, and compare this to the article-specific minimum edit distances produced by the respective algorithms. There is a total 2328 such articles for comparison. Once again, DocuToads and hand-coding are highly correlated (R-squared of 0.76) in a simple linear model. The article-level relationship between the DocuToads results and hand-coding of amendments is demonstrated in Figure 1(b).

Figure 1.

Replication of Franchino and Mariotto (2013) at document and article level. (a) Document-level correlation; (b) Article-level correlation.

Data

In order to detect successful amendments to Commission proposals, we need the proposal and outcome texts, which are available online from the legislative observatory of the European Parliament. Once the relevant texts have been scraped from this source, they are cleaned of extraneous content not directly related to the policy content of the proposal itself. They are then broken down into their constituent parts (preambles, articles, and appendices) using simple regular expressions (text features that reliably distinguish between text sections), and these are then matched with the original proposal and final outcome. Here we focus solely on the articles in a proposal as they contain the main content of the legislation of interest. It is worth noting that no additional text pre-processing (removing stopwords or stemming, etc.) is needed, as we are interested in the raw ordered legislative texts rather than some representation thereof. The final dataset utilised to examine the hypothesis presented above contains 1578 proposal-outcome text pairs decided upon between 1994 and 2013.

The independent variables of interest are measured as follows. The legislative procedure is accounted for as a dummy variable coded 0 for consultation and 1 for co-decision. A dummy variable for early agreements is taken from Reh et al. (2013), while a dummy variable for conciliation takes on a value of 1 for proposals reaching the third reading stage and 0 otherwise. The length of time it took for the Council and Parliament to reach a common position or first reading opinion respectively was taken from Prelex. This database contains information on the dates that the Commission introduced a proposal and the dates of the respective agreements within each of the other institutions. The variable used here is simply the number of days between the introduction of a proposal and the date a decision was reached in each institution.

For the control variables, the type of legislation is a categorical variable coded 0 for a decision, 1 for a directive and 2 for a regulation. Treaty changes are captured by a categorical variable for the dates each Treaty was in force. The enlargement variable is a dummy capturing the pre-2004 period and post-2004 periods. The gradual evolution of the legislative process over time is captured by a year indicator. We also control for policy area with a variable indicating the subject area of the proposal as found on the PRELEX database. Finally, we control for the length of a proposal by using a word count, and include an indicator for proposals that were officially withdrawn and amended by the Commission before being reintroduced, as these proposals are likely to be subject to more changes.

Analysis

Our DocuToads measure clearly demonstrates for the first time the extent to which Commission proposals are amended. Figure 2 shows the frequency of edit operations detected in the proposals considered. As can be seen, the large majority of proposals have relatively few successful edit operations carried out on them. This suggests that in general, the Commission proposals avoid large-scale revision by the Council and Parliament, but that there remains a significant level of amendments to be explained.

Figure 2.

Frequency of DocuToads algorithm output.

We can explore the relationship between successful amendments and the institutional environment in which negotiations take place based on this new measure. We capture such effects using a robust negative binomial regression, as the data are of a count nature and over-dispersed. The coefficients are exponentiated so as to represent incidence rate ratios. The study describes three distinct models, summarised in Table 1. Model 1 considers the effects of legislative procedure alongside controls, while Models 2 and 3 look at the role of informal politics under co-decision and the role of internal conflict within the Parliament and Council, respectively. The dependent variable in all models is the minimum edit distance between Commission proposal and final outcome, meaning that a larger coefficient indicates more successful amendments to the Commission proposal. Figure 3 plots the substantive size of the effects under consideration.

Table 1.

Determinants of amendment success.

		Model		Model		Model
	Hypothesis	1		2		3
Procedure	+	1.679***	(0.000)
Early agreement	+			1.332**	(0.007)	1.503***	(0.000)
Conciliation	+			1.789***	(0.000)	1.873***	(0.000)
Length in EP	+			1.002***	(0.000)
Length in Council	+					1.001***	(0.000)
Nice		1.347*	(0.014)	1.138	(0.392)	1.073	(0.642)
Lisbon		2.726***	(0.000)	2.351***	(0.001)	2.392***	(0.001)
Directive		1.835***	(0.000)	1.419**	(0.007)	1.380*	(0.012)
Regulation		1.325**	(0.006)	1.336*	(0.023)	1.350*	(0.017)
2004 enlargement		1.151	(0.223)	1.127	(0.342)	1.024	(0.874)
Year		0.932***	(0.001)	0.933**	(0.008)	0.959	(0.159)
Length of proposal		1.000***	(0.000)	1.000***	(0.000)	1.000***	(0.000)
Amended proposal		1.536***	(0.000)	1.274**	(0.005)	1.255*	(0.012)
Constant		5.92789e + 62***	(0.000)	4.89294e + 62**	(0.005)	4.54018e + 38	(0.135)
lnalpha
Constant		1.347***	(0.000)	0.889*	(0.026)	0.920	(0.109)
Observations		1605		703		699

EP: European Parliament.

Exponentiated coefficients; p values in parentheses.

p < 0.05, **p < 0.01, ***p < 0.001.

Figure 3.

Substantive effects of significant variables. Note: (a) Procedure, (b) Early Agreement, (c) Conciliation and (d) Treaty.

As can be seen, our expectations about the manner in which the legislative procedure affects the likelihood of successful amendments to the Commission proposal finds support. The amount of successful amendments to co-decision proposals increases by a factor of 1.68 compared to consultation proposals in Model 1. This effect is significant at the p < 0.001 level. This result is in line with our expectation that the more forgiving amendment rules under co-decision and the addition of the Parliament as co-legislator increases successful amendments, and reduces the ability of the Commission to put forward amendment-proof proposals.

When we start to unpack the subtleties of the co-decision procedure in terms of the variation in the formality of decision rules at different stages of the negotiation process, we arrived at two testable hypotheses relating to the propensity for successful amendments to Commission proposals. We first expected there would be an increase in the number of successful amendments to Commission proposals when a proposal is decided by early agreement (H2a). This hypothesis finds strong support in our analysis, suggesting that the introduction of early agreements has had an important impact on legislative outcomes as expected. Looking at Models 2 and 3, respectively, it can be seen that early agreements are associated with a 1.33 and 1.5 factor increase in proposal amendments. This informal setting is thus allowing the Council and Parliament to collude to amend Commission proposals and to bypass the more stringent amendment requirements in more formalised settings.

If and when negotiations reach conciliation, we also see a significant increase in amendments to proposals. Successful amendments to proposals that went through conciliation increase by a factor of 1.79 according to Model 2 and by a factor of 1.87 according to Model 3, with both coefficients significant at the p < 0.001 level. In line with H2b, this finding suggests that conciliation empowers the Council and Parliament and allows them to introduce amendments to solve impasses in negotiations (Tsebelis and Garrett, 2000).

We next consider the effect of internal conflict within the Council or Parliament, as captured by the amount of time it took of each institution respectively to reach internal agreement. The amount of time it takes for the Parliament to make a decision is found to be significant and in the expected direction, lending support to H3a. Each extra day that it takes for the Parliament to reach a decision leads to an increase in the number of successful amendments to a Commission proposal by a factor of 1.002. This suggests that internal conflict within the Parliament leads to significant differences between the Commission proposal and the final outcome. Similarly, the amount of time it takes the Council to reach a decision is also found to have a significant effect on the number of successful edit operations on a Commission proposal (Table 1). Each extra day that it takes for the Council to reach a decision leads to an increase in the number of successful edits to a Commission proposal by a factor of 1.001, lending support to H3b.

Of the control variables considered, both of the treaty changes occurring during the period under consideration have an effect on Commission agenda-setting failure, with Lisbon in particular being associated with increased amendment success. The type of legislative proposal is also somewhat important in determining the amount of successful amendments observed. Decisions were significantly different to Directives and Regulations in terms of their propensity to be amended. This can be attributed to the fact that these forms of legislation usually have further reaching consequences, and so are likely to be subject to more critical evaluation. Finally, enlargement does not seem to have an impact, while longer legislative proposals are subject to more amendments (not surprisingly), as are proposals that were officially withdrawn and amended by the Commission.

Conclusions

The analyses presented above have provided a number of interesting insights into the determinants of amendment success in the legislative process of the EU. The first finding relates to the impact of formal legislative procedures that structure the relationship between legislative actors during negotiations. We demonstrated that there are significantly more successful amendments to a Commission proposal under co-decision compared to consultation. It appears that the empowerment of the Parliament under co-decision is having a significant impact on the ability of the Commission to get proposals through unamended. The substantive implications of this finding are important to consider in light of the debates about democratic legitimacy in the EU and the increased use of co-decision that we observe over time. It appears that increasing the use of co-decision can go some way to redressing the balance of power between institutions and addressing the criticism that the Commission has too much power. Under co-decision, the other institutions have more ability to amend Commission proposals and are using these abilities to do exactly that.

A second aspect of legislative politics in the EU that we have considered is the increasing use of informal procedures and practices during negotiations. Our findings suggest that early agreements significantly affect the probability of successful amendments to a Commission proposal. This finding is of particular interest, given the increased use of this informal procedure that we are observing in practice, and previous assertions that early agreements do not impact on the decision-making process (Reh et al., 2013). It appears that the reduced formal constraints in informal trialogues are providing the Council and Parliament with the opportunity to collude on amendments and push them through early in the negotiation process. In terms of democratic legitimacy, one would have to ask whether this is a desirable attribute of the decision-making process and whether it undermines the seeming gains in democratic inputs due to the increased use of co-decision, given the intransparent nature of early agreements and the associated negotiations.

We also observe significantly more successful amendments when negotiations reach conciliation. In general, this stage will only be reached when there is significant disagreement between the institutions as to what a piece of legislation should look like. It appears that the Council and Parliament can leverage the informal setting of conciliation to significantly alter the proposal in front of them. The Commission has reduced formal powers to resist such amendments in the conciliation committee by design, as Tsebelis and Garrett (2000) have pointed out. This check on Commission proposal power seems to be functioning as planned, and this in line with the argument that informal institutions often develop as a check on formal distributions of powers (Farrell and Héritier, 2003).

Internal disagreement within the Council or Parliament also has an important impact upon the Commission’s ability to avoid amendments. Internal disagreement leads to concessions to accommodate recalcitrant parties. It appears that amendment demands emanating from the Council and Parliament are often accommodated into the final legislative outcome, which is a desirable attribute of a political system of checks and balances. It once again appears that the worries that the Commission enjoys too much unchecked power to drive the European integration project beyond the preferences of the other institutions have been over-stated.

All of these findings taken together suggest that the EU legislative process is a dynamic game between institutional actors with competing policy demands. The ability of these actors to have their policy demands reflected in policy outcomes is shaped and constrained by the mixture of formal and informal institutional constraints that are in place.

Taking a step back from our examination of EU legislative politics, we have shown that minimum edit distance measures can provide scholars with a very fine-grained picture of the negotiation processes, when different drafts of legislation are available. Further investigations could make use of the intermediate documents, e.g. the Council common positions and Parliament opinions, to follow the amendment process step by step. There is significant potential to analyse negotiations on an edit-by-edit basis, rather than at the proposal or article level, as is currently the trend in the literature. Furthermore, our approach is more efficient and replicable than the hand-coding methods that are currently the gold standard in the field. Utilising our method provides the option to focus upon large-scale analysis of trends in amendment success as we have demonstrated here, but also allows one to delve into individual articles in proposals for a very fine-grained analysis should one choose to do so, as we aim to demonstrate in future work. It is envisaged that further development and refinement of the measures proposed here can provide new ways to test existing theories of the legislative process at the micro level and give the impetus for further theoretical developments based upon the nuanced picture of legislative negotiations that such measure can provide.

Footnotes

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship and/or publication of this article: The research underpinning this article was paid for in part by an Individual Postdoc Grant, 4003-00071B, from the Danish Free Research Council.

Supplemental material

Supplementary material is available for this article online.

Notes

References

Aleskerov

Avci

Iakouba

et al. (2002) European Union enlargement: Power distribution implications of the new institutional arrangements. European Journal of Political Research 41(3): 379–394.

Arregui

(2008) Shifting policy positions in the European Union. European Journal of Political Research 47(6): 852–875.

Bølstad

Cross

(2016) Not all treaties are created equal: The effects of treaty changes on legislative efficiency in the EU. Journal of Common Market Studies 54(4): 793–808.

Crombez

(1996) Legislative procedures in the European community. British Journal of Political Science 26(02): 199–228.

Crombez C (2001) The Treaty of Amsterdam and the co-decision procedure. In: Schneider G and Aspinwall MD (eds) The Rules of Integration. Institutional Approaches to the Study of Europe. Manchester: Manchester University Press, pp. 101–122.

Crombez

Hix

(2011) Treaty reform and the commissions appointment and policy-making role in the European Union. European Union Politics 12(3): 291–314.

Crombez

Vangerven

(2014) Procedural models of European Union politics: Contributions and suggestions for improvement. European Union Politics 15(2): 289–308.

Daviter

(2007) Policy framing in the European Union. Journal of European Public Policy 14(4): 654–666.

Farrell

Héritier

(2003) Formal and informal institutions under codecision: Continuous constitution building in Europe. Governance 16(4): 577–600.

10.

Farrell

Héritier

(2004) Interorganizational negotiation and intraorganizational power in shared decision making: Early agreements under codecision and their impact on the European Parliament and Council. Comparative Political Studies 37(10): 1184–1212.

11.

Fitch

Margoliash

(1967) Construction of phylogenetic trees. Science 155(760): 279–284.

12.

Franchino

Mariotto

(2013) Explaining negotiations in the Conciliation Committee. European Union Politics 14(3): 345–365.

13.

Hagemann

Høyland

(2010) Bicameral politics in the European Union. Journal of Common Market Studies 48(4): 811–833.

14.

Henikoff

(1992) Amino acid substitution matrices from protein blocks. Proceedings of the National Academy of Sciences 89(22): 10915–10919.

15.

Hix

(2008) Towards a Partisan Theory of EU politics. Journal of European Public Policy 15(8): 1254–1265.

16.

Höing O and Wessels W (2013) The European Commission's position in the post-Lisbon institutional balance: Secretariat or partner to the European Council. In: Chang M and Monar J (eds) The European Commission in the Post-Lisbon Era of Crisis. Brussels: Peter Lang, pp. 123–146.

17.

Kessler B (1995) Computational dialectology in Irish gaelic. In: Proceedings of the seventh conference on European chapter of the Association for Computational Linguistics, Dublin, Ireland, 27–31 March 1995, pp. 60–66. Morgan Kaufmann Publishers Inc.

18.

König

Lindberg

Lechner

et al. (2007) Bicameral conflict resolution in the European Union: An empirical analysis of conciliation committee bargains. British Journal of Political Science 37(02): 281–312.

19.

Kreppel

(2002) Moving beyond procedure: An empirical analysis of European Parliament Legislative influence. Comparative Political Studies 35(7): 784–813.

20.

Kreppel A and Oztas B (2016) Leading the band or just playing the tune? Reassessing the agenda-setting powers of the European Commission. Comparative Political Studies 50(8), 1–33.

21.

Levenshtein VI (1966) Binary codes capable of correcting deletions, insertions, and reversals. Soviet Physics Doklady 10: 707–710.

22.

Nerbonne J and Heeringa W (1997) Measuring dialect distance phonetically. In: Proceedings of the third meeting of the ACL special interest group in computational phonology (SIGPHON-97), Madrid, 12 July 1997.

23.

Osnabrügge

(2015) The European Commission and the implementation of its legislative programme. European Union Politics 16(2): 241–261.

24.

Pollack

(1997) Delegation, agency, and agenda setting in the European Community. International Organization 51(1): 99–134.

25.

Proksch

Slapin

Thies

(2011) Party system dynamics in post-war Japan: A quantitative content analysis of electoral pledges. Electoral Studies 30(1): 114–124.

26.

Rasmussen

(2007a) Challenging the Commission’s Right of Initiative? Conditions for institutional change and stability. West European Politics 30(2): 244–264.

27.

Rasmussen

(2007b) Twenty years of co-decision since Maastricht: Inter- and intrainstitutional implications. West European Politics 30(2): 244–264.

28.

Rasmussen

(2008) The EU Conciliation Committee: One or several principals? European Union Politics 9(1): 87–113.

29.

Rasmussen

Reh

(2013) The consequences of concluding codecision early: Trilogues and intra-institutional bargaining success. Journal of European Public Policy 20(7): 1006–1024.

30.

Reh

Héritier

Bressanelli

et al. (2013) The informal politics of legislation explaining secluded decision making in the European Union. Comparative Political Studies 46(9): 1112–1142.

31.

Slapin

Proksch

(2008) A scaling model for estimating time series party positions from texts. American Journal of Political Science 52(3): 705–722.

32.

Slapin

Proksch

(2010) Look who’s talking: Parliamentary debate in the European Union. European Union Politics 11(3): 333–357.

33.

Smith

Waterman

(1981) Identification of common molecular subsequences. Journal of Molecular Biology 147(1): 195–197.

34.

Thomson R (2011) Resolving Controversy in the European Union. Cambridge University Press.

35.

Thomson

Arregui

Leuffen

et al. (2012) A new dataset on Decision-making in the European Union Before and After the 2004 and 2007 Enlargements (DEUII). Journal of European Public Policy 19(4): 604–622.

36.

Thomson R, Stokman FN, Achen CH, et al. (2006) The European Union Decides. Cambridge University Press.

37.

Thomson

Torenvlied

(2010) Information, commitment and consensus: A comparison of three perspectives on delegation in the European Union. British Journal of Political Science 41(01): 139–159.

38.

Tsebelis

Garrett

(2000) Legislative politics in the European Union. European Union Politics 1(1): 9–36.

39.

Tsebelis

Jensen

Kalandrakis

et al. (2001) Legislative procedures in the European Union: An empirical analysis. British Journal of Political Science 31(4): 573–599.

40.

Varela

(2009) Just a lobbyist?: The European Parliament and the consultation procedure. European Union Politics 10(1): 7–34.

41.

Wagner

(1974) Order-n correction for regular languages. Communications of the ACM 17(5): 265–268.

42.

Wagner

Fischer

(1974) The string-to-string correction problem. Journal of the ACM 21(1): 168–173.

43.

Wallace

(2012) Bringing people with us: Legislative writing as political rhetoric. Text and Talk 32(1): 83–101.

44.

Wilkerson

Smith

Stramp

(2015) Tracing the flow of policy ideas in legislatures: A text reuse approach. American Journal of Political Science 59(4): 943–956.

45.

Wonka

(2008) Decision-making dynamics in the European Commission: Partisan, national or sectoral? Journal of European Public Policy 15(8): 1145–1163.

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.

0.03 MB

0.00 MB

0.36 MB