European Journal of Law and Economics

, Volume 39, Issue 2, pp 287–311

Analyzing preliminary references as the powerbase of the European Court of Justice

Article

DOI: 10.1007/s10657-014-9449-9

Cite this article as:
Hornuf, L. & Voigt, S. Eur J Law Econ (2015) 39: 287. doi:10.1007/s10657-014-9449-9

Abstract

The European Court of Justice (ECJ) is a very powerful court compared to other international courts and even national courts of last resort. Observers almost unanimously agree that it is the preliminary references procedure that made the ECJ the powerful court it is today. In this article, we analyze the factors that lead national courts to use the procedure. We add to previous studies by constructing a comprehensive panel dataset (1982–2008) and identify the economic structure, familiarity with EU law, and tenure of democracy as new determinants.

Keywords

European Court of Justice Court behavior Preliminary reference procedure 

JEL Classification

H77 K33 

1 Introduction

The European Court of Justice (ECJ) is evaluated by many observers as an extremely powerful court. For example, Nugent (1999, p. 277f) writes: ‘In exercising their responsibilities the Courts, and especially the ECJ, sometimes not only interpret law but also make it. Of course judges everywhere help to shape the law, but this is especially so in the EU where the Courts have much more manoeuvrability available to them than is customary within states.’

On its path to power, the court’s preliminary reference procedure played an important role, both quantitatively and qualitatively. Today, a majority of all cases brought to the ECJ draw on this procedure. Many of the ECJ’s decisions that resulted in far-reaching implicit constitutional change were based on the preliminary reference procedure:1 these include Van Gend en Loos (1963), in which the ECJ proclaimed that European Legislation takes ‘direct effect’ in the EU member states without the necessity of national parliaments passing corresponding laws; Costa v. Enel (1964), in which the ECJ decided that European law takes precedence in the case of conflict between it and national legislation; and Francovich (1991), in which the ECJ established that member states were responsible for damages attributable to their failure to enact (or incorrect enactment of) legislation mandated by EU directives. All three of these examples involve procedural issues, but the court has also used the preliminary reference procedure to strengthen or extend EU policy. Barber (1990), in which the ECJ ruled that occupational pensions are part of an employee’s pay and must therefore comply with the Treaty article stipulating equal pay for women and men, is one prominent example of how the court extended EU policy competence with regard to social security entitlements.

Tridimas and Tridimas (2004, p. 128) write: ‘The preliminary reference system has led, in effect, to transfer of powers at three levels, namely (a) from the governments of the Member States to the institutions of the Community; (b) from the executive and the legislature to the judiciary, and (c) from higher national courts to lower national courts.’

More than a decade ago, Stone Sweet and Brunell (SSB) (1998a, p. 73) stated that the wide variance in the use of the preliminary reference procedure between different member states is ‘the most important puzzle confronting scholars working in this area.’ This variance can easily be grasped by looking at Table 1, which indicates the frequency with which the courts from different member states call on the ECJ to issue preliminary rulings. To ensure comparability, the numbers in the table have been normalised for population size.
Table 1

Preliminary references normalised for population size in the year 2009

Country

Preliminary references

Population size in mill. inhabitants

Preliminary references per mill. inhabitants

Belgium

35

10.7

3.3

Malta

1

0.4

2.4

Austria

15

8.3

1.8

Latvia

4

2.3

1.8

Estonia

2

1.3

1.5

Netherlands

24

16.4

1.5

Cyprus

1

0.9

1.2

Bulgaria

8

7.6

1.0

Hungary

10

10.0

1.0

Slovenia

2

2.0

1.0

Greece

11

11.2

1.0

Lithuania

3

3.4

0.9

Germany

59

82.1

0.7

Denmark

3

5.5

0.5

Sweden

5

9.2

0.5

Italy

29

59.8

0.5

Czech Republic

5

10.4

0.5

United Kingdom

28

61.4

0.5

France

28

62.3

0.4

Finland

2

5.3

0.4

Portugal

3

10.6

0.3

Poland

10

38.1

0.3

Spain

11

45.6

0.2

Slovak Republic

1

5.4

0.2

Romania

1

21.5

0.0

Ireland

0

4.4

0.0

Luxembourg

0

0.5

0.0

Average EU 27

11.1

18.4

0.9

Average CEE

4.6

10.2

0.8

Average EU 15

17.0

26.3

1.0

Source: Curia, IMF

Why is it that judges in Belgium call on the ECJ almost 10 times as often as judges in Portugal? Why have judges in Luxembourg recently completely stopped calling on the ECJ? SSB (1998a, b) suggest some answers. Since the publication of their work, other scholars have added their efforts to solving this puzzle by proposing and testing additional conjectures.

Most contributions trying to solve this puzzle have focused on institutional explanations. In this paper, many of the well-known hypotheses are tested again, some of them more precisely than before. Furthermore, we introduce a novel argument: referring a case to the ECJ presupposes that judges are familiar with that option. We hence conjecture that higher levels of familiarity with EU law lead to more demands for preliminary references. As proxy for the familiarity of judges with EU law we resort to the number of graduates from the College of Europe in Bruges, which concentrates on teaching EU law. Unlike previous earlier findings, this variable is highly significant and very robust. Beyond this new variable, we find that the economic structure is highly correlated with preliminary references and that the number of years a country has been democratic without interruption is also positively correlated with the number of preliminary references demanded from the Luxemburg court.

The article is structured as follows: Sect. 2 briefly describes the preliminary references procedure. Section 3 summarises extant literature. In the theoretical Sect. 4, a number of conjectures on the likely determinants that cause nation-state judges to refer a case to the ECJ are developed. Section 5 describes our estimation approach as well as the data. In Sect. 6, our findings are described and interpreted. Section 7 concludes.

2 The preliminary references procedure: a primer

National courts may, and in some cases must, ask the ECJ for a preliminary ruling involving interpretation and validity of Community Acts. The preliminary rulings procedure is described in Art. 267 TFEU (formerly Art. 234 TEC, and, prior to that, Art. 177 EC Treaty). National courts can turn to the ECJ for a preliminary ruling on interpretation of EU law. More precisely, Art. 267 grants all courts that are not courts of last resort the option of turning to the ECJ, whereas courts of last resort are required to ask the ECJ if EU legislation is decisive for the judgment of the court. This means that lower courts have de jure discretion, whereas last-instance courts have de facto discretion. They have de facto discretion because as the top courts, it is them to evaluate the potential relevance of EU law for a specific case. Should they decide not to initiate a preliminary reference procedure, it will not be initiated since nobody has the power to correct their decisions.2

No national court has the authority to rule on the incompatibility of secondary EU legislation with primary EU legislation. The ECJ thus has a monopoly on interpreting European law. However, the ECJ will not rule on a concrete case, but only interpret primary as well as secondary legislation in a general fashion. In court, a private litigant who believes that application of EU legislation would further her interests can suggest that the court make use of the preliminary rulings procedure but cannot force the court to do so. National courts are thus important gate-keepers in the use of the procedure. If the national court does ask for a preliminary ruling, it is bound by that ruling. The effect of preliminary rulings seems to be the setting of precedent for a potentially large number of similar cases, making the procedure of huge importance in effecting implicit constitutional change.

Note that the ECJ can provide preliminary rulings even in cases where the national court has phrased the question in an inappropriate or inadmissible way and the ECJ can rule on issues even if the same have not been specifically mentioned by the court asking for a preliminary reference. Also of note is that the court has established a number of strict guidelines that force national courts to request rulings, although it also must be kept in mind that the ECJ has repeatedly turned down requests for preliminary rulings.

3 Brief survey of the literature

In the introduction to this article, we mentioned that judges in Belgium call on the ECJ approximately 10 times as often as do judges in Portugal. This section briefly surveys the conjectures offered to explain the wide variance. SSB set the stage for articles empirically assessing this question with two very similar articles (1998a, b). They argue that transnational activity is facilitated by a common legal framework. Private-law subjects engaged in important transnational activity would demand such a common legal framework. They conclude that judges in countries host to firms with above average transnational activity should also draw on the preliminary reference procedure more frequently than others.

Many details of these early studies can be—and are—criticised. SSB do not, for example, normalise their trade data for the size of the country. They find a very high correlation between EU trade and the average number of preliminary references and interpret this high correlation as causality running from trade to the number of preliminary references. Pitarakis and Tridimas (2003) take issue with both the data and their interpretation. They find that preliminary references cause EU trade, but not vice versa. Further, judges do not enjoy agency in the SSB articles: high trade volumes might, indeed, lead to the demand for unified law but, ex ante, it is by no means certain that judges in different member states will satisfy that demand in a uniform way by calling on the ECJ for preliminary rulings. Tridimas and Tridimas (2004) offer a public choice approach to explain the preliminary reference procedure. They stress that the preliminary reference procedure empowers lower courts with a de facto right of judicial review which they are likely to use if that is conducive to see their own preferences being implemented.

The original SSB articles contain two conjectures that are not even tested anymore as, according to SSB, ‘it is obvious that [they] have no systematic effect that is measurable by our data’ (1998a, p. 73). These are (1) the way in which national legal systems incorporate international law into the domestic system: that is, are the two systems of law interpreted as one single system (‘monism’) or as two distinct ones (‘dualism’)? and (2) the effect of whether judges have the power of judicial review. Both hypotheses are taken up below.3

Carrubba and Murrah’s work (2005) is an important step forward in identifying the determinants of nation-state judges using the preliminary reference procedure. CM use a negative binomial panel model with fixed effects and draw on the EU 15 from 1970 to 1998. In addition to the one conjecture tested by SSB, they test the two not tested by SSB, and additionally explore whether public support of European integration is conducive to the use of preliminary references (it is), and whether political awareness of the public is conducive (it also is). Concerning the two conjectures not tested by SBB, CM find only limited empirical evidence in their favour. The CM article is more comprehensive than those by SSB, but we intend to make even further improvements by formulating some conjectures more precisely, proposing several more appropriate proxies, and adding a few new conjectures.

Vink et al. (2009) is to date the most recent and comprehensive attempt to identify reasons for the large variation in calling on the preliminary reference procedure. It is based on the EU 15 and yearly observations between 1995 and 2006 and relies on two different estimation techniques: the first is a simple pooled OLS regression model; the second draws on Boolean analysis of necessary and sufficient conditions.4 They improve on SSB by measuring intra-EU trade as a ratio to GDP. This ensures that trade is not simply a measure of size but of transnational activity. They further introduce two new variables: population size and litigation rates. In their OLS estimates, intra-EU trade (as a ratio to GDP) is not statistically significant in explaining variation in the use of preliminary references, casting doubt on the early SSB results.

4 Determinants of preliminary rulings

The literature briefly surveyed in Sect. 3 contains a number of potentially relevant determinants of requests for preliminary rulings. In this section, we add some new ones. Special emphasis will be on country characteristics that explain the use of the procedure. We propose to distinguish between three broad categories: (1) the economic dimension, (2) the structure of the judiciary, and (3) socioeconomic factors.

4.1 Economic dimension

First, we cannot exclude that requesting preliminary references is simply a function of being a large economy. It is a matter of statistical probability that the larger the economy, the more frequently disputes arise. From a theoretical perspective, population size and total GDP are both convincing predictors. Since the two variables are highly correlated, however, the model would suffer severe multicollinearity if both were included. In general, total GDP is a somewhat better predictor for the number of preliminary references, which is why we use it in the following specifications. To approximate the degree of economic connectedness, we use the share of intra-EU trade over GDP. Normalizing for GDP ensures that it is not only the size of the country as such but the degree of its economic integration into the EU that matters.5

The size of the economy and the degree of economic integration into the EU might not be the only dimensions indicating a high relevance of EU legislation. Individuals and firms operating in areas disadvantaged by national rules are more likely to have their rights enforced on the European level. Figure 1 shows that at present the single most important legal domain for preliminary references is tax law, followed by environmental and consumer issues. These subject areas were involved in more than one-fourth of all proceedings in 2009. To test for the effect of tax legislation, we rely on statutory corporate tax rates as well as the insurance and financial services activity in the economy. The latter variable is normalised over commercial service exports, as disputes arising out of transnational activities are most likely to matter for preliminary rulings.6
Fig. 1

Preliminary references by legal domain in the year 2009. Source: EuGH (2009)

Furthermore, different economic sectors could rely on EU legislation to different degrees. SSB’s analysis (1998a, b) shows that until the 1990s it was agriculture, which attracted more than one-fifth of the preliminary references. We therefore use the structure of the economy (agriculture and industry as share of total GDP) as additional explanatory variables. If being a predominantly agricultural, industrial, or finance-driven economy creates a high demand for preliminary references, the variables outlined above should indicate that.

4.2 Judicial structure

The conjecture that judges from monist legal orders request preliminary references more frequently than those from dualist systems was first made by Alter (1996) and Mattli and Slaughter (1998). Monist legal orders are those in which both domestic and international law are perceived as forming a single legal order. Dualist orders are those in which the two kinds of legislation are assumed to coexist independently of each other.7

In monist orders, so the argument goes, judges are more accustomed to relying on international law in arriving at their decisions than are judges in dualist systems. Hence, they also should be more likely to demand a preliminary ruling from the ECJ. However, according to CM (2005), exactly the opposite could be the case: because judges in monist orders have more experience in interpreting international law, and thus perceive themselves as experts in this area, they are less likely to draw on the expertise of the judges in Luxemburg.

Textbooks on international law put great emphasis on the monist/dualist distinction. In practice, however, sorting countries into one of the two traditions is difficult. Both CM and Vink et al. (2009) use a dummy variable for monism. The bivariate correlation between the two is r = 0.73, with Austria and Belgium coded differently. Voigt (2010) contains a number of variables proxying for monism. Here, we use a more refined measure that codes as 1 countries in which international law has supremacy over domestic constitutional law in case of a conflict between the two, as 0.5 those countries in which international law has supremacy over ordinary domestic law, and as 0 those countries in which domestic constitutional law trumps international law.8

The conjecture that the presence of judicial review (JR) could have an impact on the propensity of judges to ask for preliminary references was also first introduced by Alter (1996) and Mattli and Slaughter (1998). The conjecture was made more specific by CM, who hypothesised that judges in countries with concrete JR would display a higher propensity to ask for preliminary references than would judges in countries without JR. The latter, however, are expected to be more likely to ask for preliminary references than judges in countries with only abstract JR.

Concrete JR is the power to declare legislation unconstitutional based on a concrete case; abstract JR is the power to declare legislation unconstitutional even in the absence of a concrete case challenging the law. Unfortunately, CM attribute the difference between concrete and abstract JR to the temporal dimension, claiming that concrete JR only comes into play after a law has been implemented, whereas abstract JR is available before the law’s implementation (2005, p. 404). Their conjecture is based on the rationale that courts accustomed to declaring laws unconstitutional only before their implementation (‘abstract’ in their words) are not only unfamiliar with ex post JR but that such ex post review would also be contrary to their legal tradition.

Given the highly unusual characterization of concrete versus abstract JR offered by CM, we propose to test two different dimensions of JR. The first is the temporal dimension, and we distinguish four cases, namely, the constitutionality of legislation can be reviewed (1) before promulgation, (2) after promulgation, (3) both before and after, and (4) neither before nor after. Frequently, two systems of judicial review are distinguished: (1) the American and the (2) Austrian or Continental (see, e.g., Harutyanyan and Mavcic 1999). This distinction involves whether the system provides for a special court with competence to review the constitutionality of legislation. In the United States, every court has the power to review the constitutionality of legislation. In Austria, a special constitutional court with the exclusive power of JR was founded in 1920 subsequent to the new Austrian constitution of the same year. After World War II, constitutional courts became widespread in Europe, hence the name. It is often conjectured that the preliminary reference procedure grants lower court judges the power to bypass the domestic top-level judges and hence change the balance of power within the judiciary (Tridimas and Tridimas 2004). Under the American system of JR, lower courts have relatively more power than under the Austrian or European system, implying that judges under the latter two systems have greater incentive to request preliminary references than those working under an American-style system. We constructed a dummy variable to capture the different systems. All courts following the Austrian model were coded 1, all others 0.9

A further difference in the architecture of judicial systems is the construction of their top. At one end of the spectrum, there are the countries with a single top court—such as the US with its Supreme Court. At the other end of the spectrum, Germany has six federal courts that are at the top of a court hierarchy. Following Art. 267 TFEU, top courts have less discretion than other courts in calling for a preliminary reference. We hypothesise, hence, that a higher number of high courts causes a higher number of preliminary references.10

The wide ranging effects of legal origins have played a prominent role in economics over the last decade (for a summary, see La Porta et al. 2008). In the context of this article, it seems reasonable to assume that the differences between civil and common law might lead to differences in the behaviour of judges: the function of a judge in common-law countries is more than to be the ‘bouche de la loi’ as is the case in civil-law countries. In common-law countries, judges are expected to help develop the law. We thus conjecture that, on average, judges in common-law countries have a more developed self-consciousness and are loath to ask others for an opinion. The hypothesis is that judges in common-law countries will request preliminary references less often than judges in civil-law countries. The problem with testing this hypothesis, of course, is that only three EU member states (the United Kingdom, Ireland, and Malta) are common-law countries.

Asking the ECJ for a preliminary ruling on a specific case presupposes that the judge knows that such a request is possible, if not exactly how to go about making it. Thus, one possible determinant of the frequency with which preliminary rulings are requested could be familiarity with European law. Ideally, we would use data for all relevant legal actors who have received training in EU law, but, at least to our knowledge, no such database exists. As a proxy for the familiarity of judges with the preliminary reference procedure, we use the number of students who have graduated from the College of Europe in Bruges, which offers a variety of degrees in European studies. We expect to see a positive correlation between the number of nationals who have a law degree from Bruges and the number of preliminary references requested.11

Vink et al. (2009) suspect that a substantial judicial backlog reduces the propensity for asking a preliminary reference. The argument is that judges value the judiciary’s reputation as a branch of government that produces ‘results.’ If a court is already experiencing a substantial backlog, asking for preliminary references will just make things worse and judges will, therefore, refrain from such requests. This could be particularly true if judges have an incentive to make quick decisions. Nevertheless, there are a number of counter arguments to ‘judicial backlog’ theory. First, the judiciary’s reputation is a collective good produced by all judges and an individual judge has limited incentive to contribute to the production of that public good. In other words, the behaviour of an individual judge is unlikely to be influenced by such concerns. Further, the number of cases potentially appropriate for the preliminary reference procedure is miniscule compared to the entire caseload and thus the effect of asking for one preliminary reference on the overall backlog will also be miniscule. Further, sending a case to Luxemburg could be an attractive way of buying time: that is, if the case is in Luxemburg, it cannot possibly be the local judge’s fault that it has not yet been decided.

Judges who have many cases to adjudicate upon have more possibilities to refer cases to the ECJ than those with a low caseload. In that sense, a high number of cases is a necessary precondition to using the preliminary reference procedure. Following Vink et al. (2009), the corresponding hypothesis is tested here with the number of first-instance civil and administrative incoming cases per 100,000 inhabitants in 2008.

4.3 Socioeconomic factors

Brussels can be considered the capital of Europe. Member states geographically very distant from Brussels might be very ‘distant’ from its legal institutions too, and hence draw on the preliminary references procedure less frequently than judges located in countries geographically closer to Brussels. We take this possibility into account by controlling for the distance between Brussels and the respective nation’s capital.

It is argued that national sentiment toward European integration could act as an additional determinant of requests for preliminary references (CM 2005). The judiciary’s legitimacy, unlike other branches of government, does not rely on either the purse or the sword and at the end of the day, national judiciaries depend for their effectiveness on the support of the citizens. Judges will thus be highly cognizant that their actions can either increase or decrease their legitimacy. It follows that if public opinion is very critical of European integration, judges will be less likely to call for preliminary references from the ECJ. On the other hand, by asking for preliminary references, judges could shift part of their responsibility to Europe (see Fiorina 1982 for a general argument in this vein). CM (2005) draw on Euro barometer surveys to determine public opinion. They use the number of citizens who are supportive of European integration minus the number who are rather critical as their proxy for the legitimacy of European institutions.12

SSB (1998a, p. 75) bemoan the fact that they cannot control for the number of internationally active interest groups, believing that such groups could have an interest in a stable legal framework at the European level and thus could exert some pressure on courts to request preliminary references. A variable measuring the number of such groups is now available (Paxton 2002) and we use it here.

Societies differ in the degree to which their members accept hierarchies (Putnam 1993; LLSV 1997; Hofstede 1997), and religious affiliation seems to be one important determinant of this variation. Purportedly, Catholics (and the Orthodox) are more accepting of hierarchies than are Protestants. In the context of preliminary references, this would mean that judges from more egalitarian countries are less likely to call on Luxemburg than are judges from predominantly Catholic countries. We test this conjecture by looking at the percentage of Catholics and Protestants in a society.

Finally, we take into consideration whether the country is unitary or federal. Compared to unitary countries, federal countries are characterised by a more autonomous lower level of government. The state-level court structure is also often more autonomous in federal countries. Assuming that judges at the state level can increase their influence vis-à-vis judges at the federal level and are interested in doing so, we would expect to see more requests for preliminary references originating from federal countries.

This is a fairly large number of conjectures. Many a time, both high and low demand for preliminary references appear possible given certain preconditions. The empirical tests thus carry the burden to identify the correct sign as well as statistical relevance of the respective coefficients.

5 Estimation approach and data

The seminal article by SSB (1998a, b) relies on a standard OLS regression model to explain the number of preliminary references per year and member state. While the small sample properties of this estimator are well known, it is per se not a suitable estimator for dealing with the count data properties of the dependent variable. It is perhaps for this reason that CM (2005) apply a fixed-effects negative binomial panel estimator (FENB). Besides the theoretical underpinnings for doing so, the model fits their data quite well. However, the fit of the model is only one—and arguably a subordinated—criterion in deciding whether the FENB is an appropriate panel estimator.

The FENB applied in CM (2005) dates back to Hausman et al. (1984) and is not without its critics. Allison and Waterman (2002) as well as Guimarães (2008) show that this panel estimator in fact may not condition out the fixed effects. As a solution, Allison and Waterman investigate a variety of alternative estimators. Guimarães (2008) develops a score test indicating whether the fixed effects are in fact conditioned out in the FENB. Instead of adopting one of these fixes, Vink et al. (2009) prefer a dual research strategy using both quantitative and qualitative methods. Their quantitative analysis relies, again, on the standard OLS regression model. We do not think that ignoring the count data property of preliminary rulings is an appropriate solution and that a count data model should be estimated, especially in light of recent developments in the econometric literature.

In a first step, we specify different models for each of the theoretical concepts outlined above. This is done because some potential explanatory variables are highly correlated and thus should not be included in a single specification (see the correlation matrix in Table 2). Furthermore, estimating independent models helps identify relevant drivers of preliminary rulings with regard to different theoretical concepts. We specify pooled and fixed-effects panel data models. The coefficients of the pooled models are very likely to suffer from various types of endogeneity (e.g., omitted variable bias) and therefore will deviate (sometimes considerably) from the coefficients of the panel data models. As the FENB should by definition consider time-invariant country characteristics, some of the country-specific variables that are significant in the pooled models will lose their relevance in the panel specifications. However, due to the fact that the FENB estimator is a ‘pseudo’ panel estimator, the model permits the simultaneous identification of explicit time invariant country effects like the legal origin of a country. We do not specify a random-effects model, as EU member states cannot be considered a random sample. In a second step, we consider the most promising variables from the panel regressions as well as factors found to be relevant in the previous literature and run a horse-race specification.
Table 2

Correlation matrix

 

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(1) GDP

1

           

(2) Population

0.8576

1

          

(3) Intra-EU trade

−0.6095

−0.7300

1

         

(4) Corporate tax rate

0.1883

0.3944

0.0753

1

        

(5) Insurance and financial services

−0.2507

−0.3281

0.5324

0.2084

1

       

(6) Agriculture

0.2389

0.6308

−0.6460

0.2060

−0.4213

1

      

(7) Industry

−0.4739

−0.0863

−0.0173

0.2841

−0.0991

0.4764

1

     

(8) Monism

−0.2675

−0.2877

0.7735

0.4202

0.5962

−0.3142

0.0582

1

    

(9) Judical review

−0.3876

−0.2533

0.3310

0.2018

0.1631

−0.0074

0.5999

0.5198

1

   

(10) Chalstag

0.0221

−0.1505

−0.1801

−0.3574

−0.2370

−0.2090

−0.3209

−0.4371

−0.5303

1

  

(11) Legal origin

−0.3800

−0.5200

−0.0673

−0.6159

−0.2111

−0.3451

−0.0653

−0.6405

−0.3392

0.5661

1

 

(12) Law students

0.5663

0.3951

0.0140

0.3985

0.1703

−0.2290

−0.3676

0.2596

−0.1798

0.0745

−0.3992

1

(13) Incoming cases

−0.1096

−0.0315

0.5552

0.5538

0.5119

−0.1271

0.2241

0.8516

0.6205

−0.8061

−0.7147

0.2370

(14) High courts

0.2448

0.1597

0.3071

0.5935

0.3607

−0.3606

−0.1440

0.5197

0.0546

0.1157

−0.4145

0.7456

(15) Students

0.6315

0.5898

−0.4485

−0.0779

−0.2825

0.4218

−0.3256

−0.1145

−0.1597

−0.1951

−0.4456

0.2737

(16) Support for European integration

0.2339

0.4710

0.0200

0.6864

0.1893

0.3850

0.3603

0.4706

0.3894

−0.6526

−0.7982

0.3053

(17) Political discussion

0.2587

0.0373

−0.3003

−0.0855

−0.1449

−0.3674

−0.3068

−0.4924

−0.3604

0.5822

0.5326

0.2974

(18) IGO

0.3948

0.2855

−0.2033

−0.0358

−0.0164

0.0201

−0.5566

−0.2954

−0.8161

0.1209

0.0994

0.1737

(19) INGO

0.4613

0.1003

−0.0976

−0.4183

−0.0219

−0.4220

−0.8043

−0.1562

−0.2321

0.1122

0.0910

0.1865

(20) Distance to Brussels

0.0664

0.2310

−0.6774

−0.2877

−0.5698

0.4734

0.4134

−0.7187

0.1989

−0.0445

0.3492

−0.4569

(21) Democratic age

0.4976

0.6540

−0.8978

−0.1446

−0.6548

0.7324

0.1044

−0.6847

−0.1618

0.2250

−0.0083

−0.1450

(22) Catholic

0.0787

0.2053

0.1835

0.4302

0.1597

0.2523

0.2467

0.6871

0.6366

−0.3435

−0.8384

0.2043

(23) Protestantism

−0.2649

−0.3842

−0.1102

−0.5582

−0.1725

−0.2680

−0.1472

−0.6675

−0.4893

0.3357

0.9209

−0.3843

(24) Federalism

−0.2909

−0.3806

0.0623

−0.2479

−0.1405

−0.2474

0.1192

−0.0803

0.2041

0.7217

0.3767

−0.0608

 

(13)

(14)

(15)

(16)

(17)

(18)

(19)

(20)

(21)

(22)

(23)

(24)

(13) Incoming cases

1

           

(14) High courts

0.4064

1

          

(15) Students

−0.0291

−0.2337

1

         

(16) Support for European integration

0.6930

0.3318

0.2386

1

        

(17) Political discussion

−0.5086

0.2455

−0.2063

−0.4681

1

       

(18) IGO

−0.2673

−0.0816

0.2559

−0.1936

0.1480

1

      

(19) INGO

−0.1657

−0.0632

0.3895

−0.3334

0.2781

0.2992

1

     

(20) Distance to Brussels

−0.4227

−0.6209

0.1311

−0.1321

0.1619

−0.2588

−0.0001

1

    

(21) Democratic age

−0.5691

−0.3965

0.5147

−0.0157

0.1076

0.0057

0.0759

0.7150

1

   

(22) Catholic

0.5973

0.3532

0.2757

0.6066

−0.5093

−0.5045

−0.1599

−0.2236

0.0481

1

  

(23) Protestantism

−0.6135

−0.5063

−0.3226

−0.6866

0.4311

0.3560

0.1313

0.3112

−0.0549

−0.9656

1

 

(24) Federalism

−0.4242

0.1782

−0.3557

−0.4356

0.3780

−0.5267

−0.0600

0.1110

0.1276

0.1232

−0.0119

1

To counter the econometric concerns discussed by Allison and Waterman (2002) as well as by Guimarães (2008), we estimate the FENB and the Poisson fixed-effects estimator as a robustness check.13 More importantly, however, we implement the score test developed by Guimarães (2008) to check whether the time-invariant country fixed effects have been successfully removed. In the following regressions, we report incidence rate ratios as they can easily be interpreted as a multiplicative effect or semi-elasticity. Therefore, all estimates below 1 should be interpreted as a negative relationship, while estimates >1 reveal a positive effect.

We consider the EU 27 member states over the period from 1982 to 2008. Because for the new member states some variables have become available only recently (e.g., the Eurobarometer survey), our panel dataset is unbalanced. However, we know that the reason behind the different data availability is the fall of the Iron Curtain and the consequent EU membership of some central and eastern Europe (CEE) states. To account for this fact, we include a dummy variable for the 12 new member states, which captures the effect of late EU entry and the unbalanced panel.

The empirical analysis in the next section relies on multiple data sources; some established measures previously used in the literature, some new ones not applied in the present context, and some hand-collected variables never used before. “Appendix” provides an extensive overview of the coding and sources of the variables.

6 Results

Figure 2 reports the annual number of preliminary references that member states have brought to the ECJ. The data reveal a tremendous increase in the use of the procedure over time. In the year 2009, almost 300 preliminary rulings were requested from the ECJ, which is twice as many as in the mid-1980s.
Fig. 2

Preliminary references per year 1961–2009. Source: EuGH (2009)

This leaves us with the question of why some member states have drawn upon the procedure more often than others. To identify the most relevant factors, we start our investigation by estimating different models for each of the three theoretical concepts outlined above. In a second step, we extract the most promising variables that were identified in these models and run a horserace specification among them. Finally, we check for the robustness of our findings by applying two different estimation techniques (FENB and Poisson).

We begin with the economic dimension (Table 3) and restrict ourselves to interpreting the more trustworthy fixed effects estimates. Consistent with the descriptive statistics, we find a very robust and highly significant effect of corporate taxation and the share of agriculture in the economy.
Table 3

Economic dimension

 

Negative binomial

Poisson

Pooled

Fixed effects

Pooled

Fixed effects

(1)

(2)

(3)

(4)

Total GDP/103

1.000***

(0.000)

1.000

(0.202)

1.000***

(0.000)

1.000

(0.114)

New EU member (CEE)

0.340***

(0.000)

1.688

(0.719)

0.217**

(0.000)

 

Intra EU trade/GDP × 102

1.002

(0.214)

1.003

(0.162)

1.002***

(0.030)

1.000

(0.770)

Corporate tax

1.045***

(0.000)

1.014**

(0.018)

1.030***

(0.000)

1.011***

(0.001)

Fin. & Insur./exports × 102

0.969***

(0.000)

1.005

(0.434)

0.978***

(0.000)

1.002

(0.681)

Agriculture/GDP × 102

0.766***

(0.000)

0.771***

(0.000)

0.827***

(0.000)

0.842***

(0.000)

Industry/GDP × 102

0.985

(0.156)

1.005

(0.712)

1.000

(0.986)

0.982**

(0.046)

Observations

331

331

331

331

Standard errors

Robust

Robust

Log likelihood

−1,089.61

−890.78

−1,629.64

−1,123.96

Wald Chi square

480.36***

73.35***

818.52***

140.75***

This table reports the results from negative binomial and Poisson regressions. The dependent variable is the number of preliminary references. ***, **, * indicate statistical significance at the 1, 5, and 10 % level, respectively. We report incidence rate ratios as they can easily be interpreted as a multiplicative effect or semi-elasticity. All estimates below 1 should be interpreted as a negative relationship, while estimates >1 reveal a positive effect. “Appendix” provides detail on sources and variables definitions

In a second step, we explore the impact of judicial structure on the number of preliminary references issued by a particular country (Table 4). As many of the institutional factors are time invariant, we do not estimate a panel data model in this step. We find recent EU membership to have a robust and statistically significant effect on preliminary rulings. The dummy variable indicating an ‘Austrian’ system of judicial review is highly significant in all specifications. Moreover, the number of law students at Bruges, which proxies for the familiarity of judges with the preliminary reference procedure, performs very well in all specifications. While it may be the case that it is a general interest in Europe that drives students to study at Bruges, we find lagged variables (up to 7 years) to be statistically significant, which somewhat counters this conjecture.
Table 4

Judicial structure

 

Negative binomial

Poisson

Negative binomial

Poisson

(1)

(2)

(3)

(4)

Total GDP/103

1.000

(0.217)

1.000

(0.283)

1.000**

(0.012)

1.000

(0.458)

New EU member (CEE)

0.108***

(0.000)

0.034**

(0.015)

0.141***

(0.000)

0.075***

(0.000)

Monism

0.538

(0.372)

0.228

(0.193)

0.814

(0.711)

0.351**

(0.115)

Judical review

2.296***

(0.000)

1.979***

(0.000)

1.838***

(0.000)

1.802***

(0.000)

Chalstag

1.366

(0.147)

2.450

(0.179)

1.189

(0.383)

1.737

(0.111)

Legal origin

0.848

(0.256)

0.785*

(0.099)

0.832

(0.203)

0.792*

(0.092)

Law students

1.187***

(0.000)

1.144***

(0.000)

  

Law students (5 year lag)

  

1.197***

(0.000)

1.133***

(0.000)

Incoming cases/102

0.997

(0.859)

1.030

(0.439)

0.993

(0.630)

1.016

(0.450)

Number of high courts

0.872

(0.568)

0.707

(0.429)

0.873

(0.530)

0.938

(0.822)

Observations

224

224

219

219

Standard errors

Robust

Robust

Robust

Robust

Log pseudolikelihood

−736.80

−1,172.71

−720.14

−1,152.30

Wald Chi square

320.18***

374.97***

329.80***

343.50***

This table reports the results from negative binomial and Poisson regressions. The dependent variable is the number of preliminary references. ***, **, * indicate statistical significance at the 1, 5, and 10 % level, respectively. We report incidence rate ratios as they can easily be interpreted as a multiplicative effect or semi-elasticity. All estimates below 1 should be interpreted as a negative relationship, while estimates >1 reveal a positive effect. “Appendix” provides detail on sources and variables definitions

In a third step, we investigate the socioeconomic dimension. Table 5 provides further evidence that a European legal education is particularly beneficial to the use of Art. 267 TFEU. Evidently, it is not the general number of students studying in a European environment that drives preliminary references to the ECJ, since we no longer find significant results for the lagged variables. For the regressions (2) and (6), the students variable is still significant but with a much smaller coefficient, which indicates that the general number of students at Bruges can be considered a proxy variable measuring support for European integration. The Eurobarometer variable on people’s sentiments toward European integration has a positive effect and is significant in most of the specifications as well. Furthermore, we confirm CM’s (2005) result that political discussion in society plays a significant role. The age of democracy variable is positive and statistically significant throughout almost all specifications, implying that courts in countries with a longer history of uninterrupted democracy are more likely to ask for preliminary references. In the panel regressions, we also find a very strong and positive effect for Protestantism, which is highly significant in all specifications.
Table 5

Socioeconomic factors

 

Negative binomial

Negative binomial

Poisson

Pooled

Fixed effects

Pooled

Fixed effects

Pooled

Fixed effects

(1)

(2)

(3)

(4)

(5)

(6)

Total GDP/103

1.000

(0.865)

1.001***

(0.001)

1.000

(0.635)

1.001

(0.001)

1.000

(0.986)

1.000

(0.267)

Students College of Europe

1.002

(0.861)

1.003**

(0.611)

  

1.003

(0.726)

1.009**

(0.013)

Students College of Europe (5 year lag)

  

0.995

(0.629)

1.003

(0.611)

  

Support for integration

1.013***

(0.000)

1.003

(0.203)

1.012***

(0.000)

1.003

(0.203)

1.012***

(0.000)

1.004**

(0.014)

Political discussion

4.747***

(0.001)

3.196**

(0.021)

4.614***

(0.001)

3.196**

(0.021)

9.156***

(0.000)

5.194***

(0.000)

Int. Gov. Organizations

1.051***

(0.000)

1.005

(0.552)

1.051***

(0.000)

1.005

(0.552)

1.042***

(0.000)

1.009*

(0.094)

Int. Non-Gov. Organizations/102

1.076***

(0.000)

1.016

(0.057)

1.080***

(0.000)

1.016*

(0.057)

1.066***

(0.000)

1.021***

(0.000)

Distance to Brussels/102

0.997

(0.857)

1.136***

(0.017)

0.996

(0.725)

1.136**

(0.017)

1.021

(0.218)

 

Democratic age

0.995**

(0.041)

0.962***

(0.000)

0.996*

(0.100)

0.962***

(0.000)

0.993***

(0.001)

 

Catholic

0.999

(0.729)

1.014*

(0.096)

0.999

(0.772)

1.014*

(0.096)

1.003

(0.221)

 

Protestant

0.981***

(0.000)

1.047***

(0.003)

0.980***

(0.000)

1.047***

(0.003)

0.983***

(0.000)

 

Federalism

3.882***

(0.000)

0.590

(0.334)

3.666***

(0.000)

0.590

(0.334)

3.257***

(0.000)

 

Standard errors

Robust

Robust

Robust

Observations

254

254

254

254

254

254

Log pseudolikelihood

−781.85

−645.12

−781.70

−647.63

−1,084.90

−829.36

Wald Chi square

538.79***

108.28***

556.09***

102.86***

586.08***

133.36***

This table reports the results from negative binomial and Poisson regressions. The dependent variable is the number of preliminary references. ***, **, * indicate statistical significance at the 1, 5, and 10 % level, respectively. We report incidence rate ratios as they can easily be interpreted as a multiplicative effect or semi-elasticity. All estimates below 1 should be interpreted as a negative relationship, while estimates >1 reveal a positive effect. “Appendix” provides detail on sources and variables definitions

In a next step, we now include the most promising variables derived from our independent models in a horse-race specification. We consider these estimates as the most meaningful results, as the variables were found significant before and have undergone the most intense testing. As in the previous models, we start with two structural indicators. We find that ‘size matters,’ with total GDP having a positive and significant impact in all but the Poisson fixed-effects specification. Being a new member state has a negative effect in the pooled regressions, but is not statistically significant in the FENB model, which is due to the fact that the fixed effects already capture this country-specific characteristic.

Next, we consider the two variables found to be significant in the economic panel specifications (Table 3). Corporate taxation is significant only in the pooled models now. However, economic structure continues to be significant in all models, with a larger agricultural share reducing the number of requests for preliminary references sent to the ECJ. We consider this as strong evidence for the structure of the economy having a significant impact on the frequency with which preliminary references are requested.

With regard to judicial structure, we have not yet specified any panel models because most of these variables are time invariant. Monism was found a relevant predictor in the previous literature and is thus considered here as well. The monism variable is now significant in the pooled regressions but remains insignificant in the panel models. The law student proxy is the only variable with a time dimension, which remains robust as well as highly significant throughout all specifications. In the panel estimates, one more law student at Bruges increases the number of preliminary references by 2.7 %.

In unreported regressions, we tested for the two measures of judicial review. The dummy variable indicating an ‘Austrian’ system of judicial review (which was highly significant in Table 4) is insignificant in the horse-race specification. By contrast, the chalstag variable is statistically significant in Eqs. (2)–(4). The variable is highly correlated with the new EU member state dummy, however. To avoid severe multicollinearity, we restrict the horse-race model to the more parsimonious specification reported in Table 6.
Table 6

Horserace

 

Negative binomial

Poisson

Pooled

Fixed effects

Pooled

Fixed effects

(1)

(2)

(3)

(4)

Total GDP/103

1.329***

(0.001)

1.149**

(0.036)

1.196**

(0.044)

1.040

(0.361)

New EU member (CEE)

0.237***

(0.000)

2.705

(0.424)

0.187***

(0.000)

 

Corporate tax

1.031***

(0.000)

1.002

(0.740)

1.023***

(0.000)

1.001

(0.778)

Agriculture/GDP * 102

0.782***

(0.000)

0.879***

(0.001)

0.801***

(0.000)

0.866***

(0.000)

Monism

1.906***

(0.002)

1.851

(0.585)

1.378*

(0.073)

 

Law students

1.077***

(0.000)

1.027***

(0.009)

1.067***

(0.000)

1.027***

(0.000)

Support for integration

0.997

(0.354)

1.003

(0.193)

1.000

(0.900)

1.002

(0.177)

Political discussion

1.508

(0.481)

1.788

(0.109)

1.940*

(0.100)

3.370***

(0.000)

Democratic age

1.006***

(0.000)

0.986***

(0.007)

1.005***

(0.000)

 

Protestant

1.006

(0.120)

1.038***

(0.010)

1.001

(0.640)

 

Observations

289

289

289

289

Standard errors

Robust

Robust

Log likelihood

−966.53

−786.84

671.51

180.54

Wald Chi square

451.01***

115.05***

−1,338.56***

−994.46***

This table reports the results from negative binomial and Poisson regressions. The dependent variable is the number of preliminary references. ***, **, * indicate statistical significance at the 1, 5, and 10 % level, respectively. We report incidence rate ratios as they can easily be interpreted as a multiplicative effect or semi-elasticity. All estimates below 1 should be interpreted as a negative relationship, while estimates >1 reveal a positive effect. “Appendix” provides detail on sources and variables definitions

CM (2005) and, to a lesser extent, Vink et al. (2009) found support for European integration and the frequency of political discussion to be relevant socioeconomic variables. We therefore consider them in our horse-race model. Neither variable is significant in the specifications at hand, except for political discussion in the Poisson fixed-effects model. For the two time-invariant variables that were consistently significant in our previous socioeconomic specifications, we find the age of the constitution to have a significant impact on Art. 267 TFEU activity. Protestantism is significant in the FENB panel regression model as well. Therefore, our conjecture that judges in countries comprised of citizens with a more hierarchically structured mindset would demand more preliminary references than judges in countries where the citizens has a more horizontal outlook on the world is refuted by the data.

Finally, we implement the score test developed by Guimarães (2008) to check whether the FENB model successfully conditions for the country fixed effects. The test indicates that there is a specific functional relation between the fixed effects and the individual overdispersion parameter. Since we cannot reject this null hypothesis, the test indicates that the country fixed effects are indeed cancelled out.

7 Conclusions and outlook

Legal scholars are in apparent agreement that the ECJ’s preliminary reference procedure is one of the most important—if not the most important—factors in making that court the powerful entity it is today. Understanding why there is such variance in its use is of paramount interest because the preliminary reference procedure has been and no doubt will continue to be an important instrument of European integration.

We contribute to this understanding by adding a number of new, and now empirically tested, conjectures and by applying a more adequate econometric approach (testing whether country fixed effects have indeed been removed in the FENB model). As it turns out, previous results are not robust if rigorously tested. In stark contrast to CM (2005), monism appears to be no reliable predictor for preliminary references. On the other hand, the economic structure of a country has a robust and significant effect on the number of preliminary references. Furthermore, by using a novel measure for legal education regarding European issues, we find that an international legal training has an important effect on Art. 267 TFEU activities.

Finally, we have also shown that the drivers of preliminary references are not a given and may change over time. While in the early decades of the ECJ the share of agriculture in the economy had a positive effect on preliminary rulings, taxation matters in absolute terms most today.

Footnotes
1

The constitutional change was ‘implicit’ in the sense that it occurred although the European Treaties—the equivalent of the European Constitution—were not explicitly modified. The concept of implicit constitutional choice is more explicitly described in Voigt 1999.

 
2

It might be noteworthy that the German Federal Constitutional Court invoked its first preliminary reference procedure ever only in 2014. In the case, the Court had to decide about the constitutionality of the newly established European Stabilization Mechanism.

 
3

These conjectures were first introduced by Alter (1996) and Mattli and Slaughter (1998).

 
4

It is unfortunate that Vink et al. (2009) rely on a different time period than CM (2005) as differences in their results could be due to the different period chosen.

 
5

On the other hand, it is well known from trade theory that the trade to GDP ratio tends to be higher in smaller states as large countries can be more self-sufficient. If this was true, requests for preliminary references might be brought forward more frequently by small, rather than large, states.

 
6

A well-known case is Manninen (2004), in which the ECJ held that any tax imputation system that only imputes corporate taxes on dividends from locally resident companies violates the EU Treaty.

 
7

Monist orders need a way to deal with potential conflicts between domestic and international law. Hence, there are actually two kinds of monism: one in which international law enjoys precedence over national law and one in which domestic law enjoys precedence over international law. After the ECJ’s Van Gend en Loos decision, however, this distinction is superfluous within the EU. Further, EU law is different from international law since it determines by which procedure it is to be implemented into national legislation. This does not, however, exclude the possibility that different traditions of how to implement international law into national legislation can affect the propensity of nation state judges to refer a case to the ECJ.

 
8

The partial correlation between this more refined measure and the Vink et al. (2009) variable is r = 0.65. The partial correlation with the CM (2005) monism variable is r = 0.39.

 
9

Following Harutyanyan and Mavcic (1999), these EU member states are coded 0, i.e. as not adhering to the Austrian model of judicial review: Denmark, Estonia, France, Greece, Portugal and Sweden.

 
10

A referee of this journal pointed out that although top level courts are legally obliged to ask the ECJ for preliminary references under specific circumstances, they still enjoy considerable discretion in their decision. Their incentive to call on the ECJ should be considerably lower than those of lower courts, since they are unlikely to gain any influence by invoking the preliminary reference procedure. The expected sign of the coefficient would, hence, be somewhat ambiguous.

 
11

Alternatively, we could have counted the number of law firms with first-rate competence in European law, but that would have been a much more subjective measure.

 
12

We use the average of the spring and autumn wave of the Eurobarometer (1982–2008) question QA6: ‘Generally speaking, do you think that your country’s membership of the European Union is …? A good thing/A bad thing.’

 
13

The Poisson estimator relies on the more restrictive assumption that the variance must equal the mean. Only if this condition is met, the model might do a better job of canceling out the country-fixed effects.

 

Acknowledgments

The authors thank Andreas Engert, Christoph Engel, Paulo Guimarães, Mariusz Goleckifor, Gerhard Wagner, Daniel Zimmer, two anonymous referees as well as the editors of this journal for helpful comments and suggestions. We would like to thank the European Association of Law & Economics (Hamburg 2011), the German Association of Law & Economics (Bonn 2011) and the American Law and Economic Association (Stanford 2012) participants.

Copyright information

© Springer Science+Business Media New York 2014

Authors and Affiliations

  1. 1.University of MunichMunichGermany
  2. 2.Institute of Law & Economics and CESifoUniversity of HamburgHamburgGermany

Personalised recommendations