On Domestic Determinants and Empirical Relevance of Government Preference for Implementing European Union Rules

This study probes the domestic determinants and empirical relevance of government preference for European Union (EU) implementation. The main contention is that governments’ preference for implementation is defined in the context of broader contestation over European integration. Political parties that mediate this contestation represent competing social preferences for implementation and reflect these preferences in the implementation practices of national governments that they constitute. Using a unique dataset of individual infringement actions, the paper then explores the impact of government preference on the resolution of member states’ disputes with the European Commission over implementing EU rules. The key finding is that government preference as shaped in domestic political conflicts and party politics significantly determines EU implementation and, as a result, the effectiveness of EU rules.


Introduction
What accounts for member state governments' preference or (un)willingness to implement European Union (EU) rules?And, more importantly, does government preference matter in EU implementation?The existing literature has a mixed record on those questions.Theoretical studies offer only a partial view of the determinants of government preference for implementation.Scholars focus on the administrative and institutional determinants of government preference, and largely neglect the domestic political processes and mechanisms that shape government preference (Falkner, Hartlapp, Leiber & Treid, 2004;Thomson, Torenvlied & Arregul, 2007;Heritier, Kerwer, Knill, Lehmkuhl, Teutsch & Douillet 2001).Empirical studies offer inconsistent results for the relevance of preference.While some studies find insignificant results for government preference (Mbaye, 2001;Falkner, Treid, Hartlapp, Lieber, 2005), others furnish significant, though qualified, support (Thompson Torenvlied & Arregul, 2007;Toshkov, 2007Toshkov, , 2008)).The current paper has both theoretical and empirical contributions to EU implementation studies.
Its theoretical contribution lies in flashing out the role of domestic political conflicts and mechanisms in determining government preference for implementation.It does so by systematically bringing the insights of two streams of research in EU studies into analysis of EU implementation.The first stream examines political contestation on European integration (Marks & Steenbergen, 2004).The insights offered in this stream help us identify the contours of domestic conflicts among social actors with their competing preferences over EU implementation which, further, defines the content and direction of government preference for implementation.The second stream examines the role of political parties in European integration (Hix & Lord, 1997: Mair, 2006).Research in this stream examines how political parties serve as a mechanism that articulates competing social preferences over EU implementation and connects them to the implementation patterns of national governments.
This study is not the first to highlight the relevance of domestic political conflicts in EU implementation.Prior research, like Borzel's (2003) pull-and-push model, Steunenberg's (2007) model of transposition and Falkner, Treid, Hartlapp, Lieber's (2005) model of the world of domestic politics, has drawn attention to the conflicting preferences of domestic actors in influencing member states' implementation practices.However, analysts often fall short of specifying exactly how domestic actors form their preferences for implementation and what mechanisms link domestic conflicts over EU implementation to governments' implementation practices.Also, the current research is not the first to point to the role of political parties in EU implementation.Studies have argued that parties shape the implementation practices of governments that they form (Treid, 2003;Jenkins, 2007;Falkner et al, 2005;Toshkov, 2007Toshkov, , 2008)).However, these studies either introduce political parties in an ad hoc manner or include party variables without elaborating domestic conflicts over implementation that parties represent.
The empirical contribution of the paper lies in using a unique dataset of individual infringement actions initiated by the European Commission against the then member states from 1995 to 2004.This dataset covering more than 11,000 infringement actions is one of the largest datasets available on the infringement proceedings.This dataset is employed to test the relevance of government preference for the implementation patterns of member states.In particular, the extent to which government preference affects the resolution of member states' disputes with the European Commission over implementing EU rules is examined.

Theory and Hypothesis
Previous research presents two kinds of approaches to account for government preference for EU implementation: the European-level and Domestic-level approaches.The European-level approaches underscore the process of EU decision-making, the characteristics of EU rules, and the distribution of benefits resulting from the implementation of those decisions.The opposition-through-the-backdoor thesis suggests that the variation in governments' willingness to implement EU rules is a function of whether they expressed opposition to decision outcomes at the decision-making stage (Falkner, Hartlapp, Leiber & Treid, 2004).Although studies have found little empirical evidence for the thesis (Mbaye, 2001;Falkner, Hartlapp, Leiber & Treid, 2005), the central idea behind the thesis has not been lost and been reformulated in the notion of 'Incentives to deviate' by Robert, Torenvlied and Arregul (2007).According to Robert, Torenvlied and Arregul (2007), government preference for implementation is determined by the existence of incentives to deviate which vary in strength depending on the difference between actual decision outcomes and the outcomes preferred by governments.Additionally, Thompson, Torenvlied and Arregul (2007) claim that preference for implementation is also shaped, in part, by the magnitude of discretion granted to governments in the implementation process.A higher discretion leads to a weaker preference for implementation.Furthermore, Perkins and Neumayer (2007) attribute the variation in government preference to the distribution of benefits from EU implementation.As a member state's gains from implementation increase, its government becomes more willing to implement EU rules.
One obvious limitation of the European-level approaches is that they almost completely neglect forces and actors at the domestic level where implementation actually happens.The domestic-level approaches attempt to remedy this deficiency.One such approach, the goodness-of-fitness thesis, proposes that member state governments become more or less willing to implement EU rules depending on the degree of fitness or compatibility between national institutional and administrative practices, and the requirements of EU rules (Knill and Lenschow 1998;Heritier, Kerwer, Knill, Lehmkuhl, Teutsch & Douillet 2001).Although the empirical record for the goodness-of-the-fitness thesis is disappointing (Haverland, 2000;Falkner, Hartlapp, Leiber & Treid, 2005;Mastenbroek, 2005), the shift to the domestic level marks an advance over the European-level approaches.However, a major problem with this line of research is the insufficiency of attention to political processes and mechanisms.This has prompted recent reviews of the literature to call for more attention to the domestic politics of EU implementation (Mastenbroek, 2005).
The current research seeks to shed more light into the domestic politics of EU implementation.There are two key premises.The first premise is that government preference for implementation is defined in the context of a broader domestic political debate over European integration.That is, the politics of implementation is an intrinsic part of broader contention over European integration.Hence, the aspects of government preference for implementation are shaped by the dimensionality of domestic contention on European integration.Studies identify at least two dimensions in contestation over European integration: the process and substance dimensions.The process dimension concerns the process of integration and its decision-making mechanisms.Contestation on this dimension is over how to allocate sovereign power at different levels and essentially territorial in nature.Political actors on this dimension have different preferences as to what extent EU decision-making is appropriate for addressing policy issues.While the process dimension is shaped by disagreements over the appropriateness of EU rule-making processes, the substance dimension focuses on the specific rules generated through these processes and the substantive outcomes of these rules.Contestation on this dimension is essentially functional in nature.Political actors on this dimension have disagreements over the intended policy outcomes of EU rules.Scholars have proposed four different models that variously conceptualize the specific characters and orthogonality of these dimensions: the International Relations model, the Regulatory model, the Hix-Lord model and the Hooghe-Marks model (Marks & Steenbergen, 2002).
The first two models assume that contestation over EU-related issues is solely focused on either the process dimension or the substance dimension.The international relations model emphasizes the process.The model posits that contestation over EU-related issues boils down to the disagreement along the sovereignty-integration dimension (Hoffman, 1966;Moravcsik, 1998).The regulatory model, on the other hand, puts emphasis on the substance dimension.This model claims that the disagreement over the degree of economic regulation as the substantive outcome of EU rules is the principal dimension, along which individuals contest EU-related issues (Garrett, 1992).
While the international relations and the regulatory models assume the lack of the orthogonality of the process and substance dimensions, the Hix-Lord and the Hoogle-Marks models conceive different levels of the orthogonality between these dimensions.These models highlight the difficulty of collapsing these two dimensions into a single dimension.The Hix-Lord model suggests that while the substance dimension is shaped individuals stances on the left-right ideological continuum, the process dimension is based on the disagreement on the sovereignty-integration dimension.The Hooghe-Marks model also conceptualizes the two dimensions in similar terms, though this model highlights the correlation between those dimensions (Hooghe & Marks, 1999).
It is reasonable to claim that domestic conflict over implementing EU rules is a mirror reflection of the broader contestation over European integration in that the former, like the latter, is structured along the process and substance dimensions.In the process dimension, domestic actors' stances on EU implementation are shaped by their approaches to European decision-making processes.Those who favor the allocation of public authority to EU institutions and, thus, their country's EU membership are likely to support implementing EU rules; whereas those who prefer the allocation of public authority to the domestic level are likely to oppose to implementing EU rules.In the substance dimension, domestic actors form their preferences for implementation, depending on how EU rules affect their interests.Those actors whose interests EU rules serve are likely to be more supportive of implementing these rules than those actors whose interests are adversely affected by the intended policy outcomes of these rules.Structured along the process and substance dimensions, domestic conflict over EU implementation essentially consists in the competition of distinct social preferences for implementation.
The second premise that this paper makes is that political parties serve as a mechanism which crystallizes distinct social preferences for implementation into government preference by mediating their competition.Prior research has identified the myriad ways, in which political parties influence European integration (See Mair, 2006 for a succinct review of the literature).Scholars have extensively documented that the operations of EU institutions are pervaded by "a mix of national and transnational party politics" (Hix & Lord, 1997: 7).Transnational party groups play a crucial role in organizing EU legislative politics (Hix, 2008;Gaffney, 1996); national political parties constitute the main source of recruitments for officials in EU institutions (Hix & Lord, 1997).Political parties and their interactions are also shown to have significantly shaped major policy developments in the EU (Featherstone, 1988;Dyson & Featherstone, 1999).
The more relevant from the vantage point of the current paper is the finding that political parties play a key role in representing diverse social interests and values of relevance to European integration.Scholars show that political parties serve as the main vehicle that issues arising from European integration are systematically incorporated into policy debates at the domestic level (Marks, Wilson and Ray 2002).Political parties are shown to mediate citizens' orientations toward European integration and their electoral behaviors in European elections (Ray, 2003;Van der Eijk & Franklin, 2004).Analysts suggest that parties integrate EU-related issues into their programmatic orientations (Hix, 1999b).Hence, it is reasonable to surmise that competing social preferences in conflict over implementing EU rules are articulated and filtered through party politics.Political parties represent those competing preferences in their programmatic orientations, which, in turn, feed into the implementation practices of national governments that they constitute.
Given the preceding discussion, it is possible to drive two hypotheses.The first is the process hypothesis.It is straightforward: governments composed of parties with a more favorable stance toward EU rule-making are expected to put more efforts to implement EU rules than governments with a less favorable position.While it might sound almost cyclical to expect a pro-European government to show more commitment to EU implementation, there is ample evidence that while some highly pro-European member states, like Italy, Greece and Portugal, can be a regular infringer of EU rules, some relatively more skeptical member states, like Denmark and Sweden, can perform well in putting EU rules into effect (Bӧrzel, 2001).
The second is the substance hypothesis.It predicts that governments composed of parties with a more favorable stance toward the intended policy outcome of EU rules are likely to put more efforts to implement these rules than governments composed of parties with a less favorable stance.As to the substance hypothesis, one needs to identify a fundamental policy goal underlying EU rules.While it is rather difficult to pinpoint one single overarching goal for EU policies, studies of major EU policy initiatives (Sandholtz, 1993;Dyson and Featherstone, 1999) and general debates over EU policies (Scarpf, 1999) suggest that probably the most general policy goal underlying EU rules is to establish a market economy of a European scale (Note 1).Given the market-oriented nature of the integration project, the substance hypothesis can be reiterated in more specific terms as follows: governments composed of parties, which favor the principles of market economy, are likely to put more efforts to implement EU rules.

Variables and Data
The process and substance hypotheses are tested in examining the resolution of member states' disputes with the European Commission over implementing EU rules.The dependent variable is the severity of implementation problems that member states have, signaled by the stage, at which the infringement proceeding terminates.
EU implementation studies have frequently used infringement data for illustrative and explanatory purposes (Tallberg, 2002;Mbaye, 2001;Borzel, 2001Borzel, , 2006;;Sverdrup, 2004).Despite their common use, these data have been subject to criticism (Borzel, 2003;Falkner Hartlapp, Leiber & Treid, 2005;see Hartlapp & Falkner, 2009 for a detailed review).There are at least two major limitations in infringement data.First, infringement data are only an indirect measure of EU implementation in that they tap member states' implementation through the reaction of the European commission and citizens.What aggravates this limitation is the possibility of biases in the Commission and citizens' approach to infringement instances.Second, infringement data are only a partial measure of EU implementation in that they do not cover all instances of actual infringements and, also, focus only on the behaviors of national governments at the exclusion of sub-national private and public actors.
As to the first limitation, it is possible to control some likely sources of biases in the Commission and citizens' approaches in regression analysis (see the control variables below).As to the second limitation, it is almost impossible to capture implementation in its all aspects.In that, infringement data are similar to other measures of implementation, like data on transposition and data on the decisions of the European Court of Justice (Note 2).
Despite the limitations associated with infringement data, the analysis here is based on a conviction that infringement data contain invaluable information about member state governments' implementation-related behaviors.Along with controlling for possible sources of biases, one thing that needs to be done is to disaggregate infringement data, often presented as aggregate count data of total number of infringement actions.Recent studies have noticed the advantages of focusing on individual infringement actions, which increases the number of observations and offers a lot more information about the features and progression of infringement proceedings (Jenkins, 2007).Following these studies, the current analysis employs a unique dataset of individual infringement actions.
The dataset covers 11,812 infringement actions initiated against member states by the Commission from January 1995 to May 2004.I compiled the data from the Annual Reports on Monitoring the Application of Community Law, which have been published since 1984 (COM (84) 181).Based on the Annual Reports, each infringement action is coded in terms of stages of infringement proceedings (letter of recommendation, reasoned opinion and reference to the Court), forms of implementation problems (e.g.non-notification, non-communication, incorrect application of directives, and infringement of regulation, decision and treaties), reasons for the initiation of infringement proceedings, policy areas of infringement and types of EU legislation (directives, decisions, regulations and Treaties).
The coding scheme used in measuring the severity of implementation problems is based on the following rationale.In the infringement proceeding, what starts off as a potential instance of infringement at the letter of formal notice stage turns into a substantiated judgment about the violation of EU rules at the reference to the Court stage.In the process, the Commission keeps a series of informal and formal communications with member state governments about suspected infringements and ways to rectify them.From the early stage of the letter of formal notice to the late stage of the reference to the Court stage, governments have various opportunities to either establish that the suspected instance of infringement does not have an actual basis or take actions to address it.One can infer that cases appearing at later stages in the process involve more substantial and severe implementation problems.So, the different stages of the infringement procedure are coded accordingly: The appearance on the letter of formal notice stage is considered indicative of low severity; the appearance at the reasoned opinion stage indicative of medium severity; and the appearance at the references to the Court stage indicative of high severity.This coding scheme is consistent with prior research (Jenkins, 2007).
Predicting the highest level reached by the infringement proceeding aggravates the potential problem of overrepresentation of severe cases.In other words, there is likely to be a problem of overrepresentation in the data set of hard or severe cases reaching to the later stages of the infringement procedure, compared to routine cases that are settled at the early stages of the infringement procedure.Although this problem is present in aggregate data as well, it becomes more acute in disaggregate data.the dataset is checked for the overrepresentation problem.This procedure identified 1398 cases, which appeared at multiple stages of the procedure (out of 11812 cases).Before the analysis is run, the early appearances of these cases are removed so that all cases in the data set appear only once.
The explanatory variables are the preferences of national governments concerning the process of EU rule-making and substantive outcome of EU rules.A common practice in measuring government preference is to look at the preferences of parties that constitute the government (Jenkins, 2007;Toshkov, 2007;Tosckov 2008).Party preference data are extracted from the Comparative Manifestos Project (CMP).The CMP studies utilized party manifestos, platforms and government declarations to chart the partisan preferences of political parties across 19 democracies including the EU countries included in the analysis (Budge, Klingemann, Volkens, Bara and Tanenbaum, 2001).Research has used these data to study a variety of questions, like government expenditure and coalition formation (Klingemann, Hofferbert and Budge, 1994).
In measuring preference on the process of EU rule-making, the Pro-Europeanness scores of political parties are used.The CMP created scores for the percentage of favorable and unfavorable statements about European integration by each party.The measure is the ratio of positive mentions of EC/EU to total number of negative and positive mentions of EC/EU.Scholars have found that the Manifesto measure of pro-Europeanness generally correlates with the other measures of pro-Europeanness, like expert surveys (Ray, 1999;Marks, Hooghe, Steenbergen and Bakker, 2004).
In measuring government preferences on the substantive outcome of EU rules, the scores of preferences for various substantial policy outcomes are utilized.In particular, specific measures like supports for Market Economy, Free Enterprise and Market Regulation are used, depending on policy areas under consideration.Government preference is calculated through weighting the influence of each party involved in the government by including the percentage of cabinet positions occupied by this party (Note 3).Also, a number of control variables are included.Weighted Vote in the Council of Ministers taps the influence of member states in the EU rule-making process.To control for the impact of the economic dependency of member states on the EU (Perkins and Neumayer, 2007), the Intra-EU Trade variable measured with intra-EU trade data as a percentage of the total trade of a member state is used (Note 4).Two variables are used to capture the impact of member states' varying material capacities (Falkner, Treib, Hartlapp, and Leider, 2005;Mbaye, 2001).The first variable is GDP per capita, which captures the overall availability of material resources for state and societal actors.The second variable is Government Revenue, measured as a percentage of GDP, which shows how much of the material resources are effectively at the disposal of governments (Note 5).Membership Age is included to control for the role of learning (Checkel, 2001).
A set of domestic institutional variables is also included as controls.Judicial Review controls for the veto power of the judicial branch over the implementation process.It is a dummy variable: "1" if there is a judicial review, "0" otherwise.Unicameralism controls for whether the legislative branch is unicameral or bicameral.It is a dummy variable: "1" if the legislative branch is unicameral, "0" otherwise.Decentralization, an index developed by Lane and Ericsson (1999) with higher score indicating more decentralization captures the extent to which national governments share political power with territorial and functional sub-national units.In order to tap the impact of the mode of interest representation on EU implementation (Sverdrup, 2004), Corporatism, index created by Lijphart and Crepaz (1991), is included.Additionally, Coalition is included to capture the coalition status of governments: "1" if a government is a coalition, "0" otherwise.
Additionally, an attempt is made to control for some of the possible biases built into the infringement data.First of all, the Commission might treat member states differently.The Southern European countries (Greece, Italy, Spain and Portugal) do not have a good reputation in implementing EU rules (Bӧrzel, 2003).The Commission might keep a closer eye on the implementation practices of these countries.A dummy variable is created with "1" if a member state is a Southern European country, "0" otherwise.The Commission may also favor the member states making larger Budget Contributions.To control this bias, member states' budget contributions as a percentage of EU budget are included (Note 6).Also, it has been argued that there is a growing trend in the number of infringements, reflecting the increasing body of EU rules.Time Counter is used to capture any trend over time.
According to the critics of infringement data, another source of possible biases in the data is the citizens of member states on whose initiatives the Commission often depends to detect potential infringement cases (Bӧrzel, 2001).They claim that the infringement actions that member states get reflect more of the sheer number of population, general life satisfaction of individuals, or distrust in national government or general orientations toward European integration.To eliminate these threats to the validity of results, a series of control variables are used, like Population, Life Satisfaction, Distrust in National Government and Support for Speedy Integration (Note 7).
Another possible challenge to the infringement data is that although most infringement instances are captured in the Commission's infringement actions, there might be other instances that are not covered by the data.Scholars have noticed that along with the Commission, there is another venue, through which infringements get revealed and resolved.That venue is national courts resorting to the preliminary ruling procedure, where the courts request the ECJ to interpret and clarify EU rules and regulations (Stone Sweet and Brunell, 1998).To control for possible infringement cases processed through this venue, the number of Preliminary Rulings that the courts of a member state request from the ECJ in a year is employed (Note 8).

Analysis and Results
The empirical analysis ascertains the severity of implementation problems of the then member states as displayed in infringement actions initiated against them from January 1995 to May 2004.Since the dependent variable is polychotomous and ordinal in nature that runs from "low severity" to "high severity", it requires an estimation technique appropriate for ordered data.
There are two techniques typically used with ordered data: ordered logit (proportional odds model) and generalized ordered logit.These techniques differ in their application of the restrictive proportionality or parallel regression assumption, which requires that the effect of an independent variable on the log-odds of being in the J category of the dependent variable is invariant to the cutpoints at which the dependent variable is dichotomized.While the ordered logit universally applies this assumption to all variables without testing its appropriateness, the generalized logit indiscriminately relaxes it for all variables even in cases where it does not have to.A method known as the partial proportional odds technique, which relaxes the proportional odd assumption for only those variables that the likelihood test for the assumption does not hold, is employed.The Wald test for the proportional odds assumption is used for each independent variable with p<.01 (Note 9).Table 1 displays the results as log-odds that a case falls in a higher category of the dependent variable (Note 10).The results in the column C1 give the log-odds of a country being above the first cutpoint (receiving a reasoned opinion and being referred to the Court) versus below the first cutpoint (receiving a letter of formal notice).The results in the column C2 give the log-odds of a country being above the second cutpoint (being referred to the Court) versus below the second cutpoint (receiving a letter of formal notice and receiving a reasoned opinion).For interpretation purposes, positive (negative) coefficients indicate that higher values on the explanatory variable make it more (less) likely that the country will be in a higher category of the dependent variable.
[Table 1 about Here] The government preference variables are highly significant and negative as expected.The Wald test for both government preference variables showed that the proportional odds assumption held for them, meaning that the effects of these variables on the log-odds of being in different categories of the dependent variable are invariant to the different cutpoints (p<.38 for Pro-Europeanness and P<.80 for Market Economy).That is why the coefficients for these variables are the same across the cutpoints, which makes it easy to interpret the results.The higher the Pro-Europeannes of the government of a country is, the less likelihood that the country be in higher categories of the dependent variable.In other words, a higher score on Pro-Europeanness makes it less likely for a member state to receive a reasoned opinion or be referred to the Court as compared to just receiving a letter of formal notice.Also, it makes it less likely for a member state to be referred to the Court as compared to just receiving a letter of formal notice or a reasoned opinion.
Likewise, the higher the support for market economy of the government of a country is, the lower the likelihood that it receives a reasoned opinion, and is referred to the Court as compared to just receiving a letter of formal notice is.Also, a higher support level for market economy makes it less likely to be referred to the court as compared to just receiving a letter of formal notice and a reasoned opinion.For the most Euroskeptical governments in the sample, the odds of reaching to higher stages of the infringement procedure are 3 time that of the most pro-European governments.For governments with the least support for market economy, the odds of reaching to higher stages of the infringement procedure are 20 times that of governments with the most support for market economy.
For a better illustration of the substantial significance of the results for Pro-Europeanness, I estimated changes in predicted probabilities of being referred to the court that were triggered by a change in the partisan preferences of governments.For example, when the first Schussel government with a pro-europeanness score of 0.38 replaced the Klima government with a pro-europeanness score of 1.00 in 2000, Austria's likelihood of being referred to the Court dropped by approximately 13 %.Likewise, when the first Blair government with a pro-europeanness score of 0.90 replaced the Major government with a pro-europeanness score of 0.425, the UK's likelihood of being referred to the Court dropped by about 10%.
A similar impact is observable for Market Economy in the sample.When the first Rasmussen government with a support for market economy score of 13.58 replaced the fourth Rausmussen government with a support for market economy score of 0.37 in 2001, this change led to a drop in Denmark's estimated likelihood of being referred to the Court by about 13 %.When the fifth Gonzalez with a score of 1.18 was replaced by the first Aznar government with a score of 5.90, this change led to a decrease in Spain's estimated likelihood of being referred to the Court by about 6 %.When the first Schroder government with a score of 3.02 replaced the fifth Kohl government with its score of 7.46, this change led to an increase in Germany's estimated likelihood of being referred to the Court by about 4 %.
Regarding the control variables, there are both confirmatory and puzzling findings.First of all, the critics of the infringement data find support for their assertion that caution needs to be exercised in using these data.The result for Southern suggests that the Commission may be treating the southern European countries differently.Among the southern European countries, the odds of being at higher stages of the infringement procedure are about 4 times that of non-southern members.Also, the negative and significant coefficient for Life Satisfaction suggests that as the overall life satisfaction of citizens in member states increases, they become more complacent about possible infringements.Moreover, the coefficient for Preliminary Ruling is negative and significant, which suggests that some of potential infringement cases are solved even before they come to the attention of the Commission.Given the statistical significance of some of the bias variables, controlling for them should boost confidence in the analysis.
The coefficient for Weighted Vote is negative and highly significant.Among the least influential member states, the odds of reaching to higher stages of the infringement procedure are 6 times that of the most influential member states.The result for Intra-EU Trade is negative as expected, but not significant.It appears that although dependence on the EU can put pressure on states to implement EU rules, this pressure may not be strong enough to produce a consistently good implementation record.The findings for the socialization variable and the material capacity variables come as a surprise.The coefficient for Membership Age is positive, but not significant.It appears that implementation does not necessarily improve with the increasing length of membership, which further hints that it may not be driven by the learning process.The coefficients for GDP Per Capita and Government Revenue are positive and highly significant.This finding goes against the expectation that states with a higher material capacity are likely to have a less severe implementation problem.
The results for the domestic institutional and political variables are mixed.The coefficient for Judicial Review is positive and significant at 1 percent level.Judicial review increases the log-odds of a country moving to higher stages of the infringement process.Among member states with a judicial review, the odd of reaching to higher stages of the infringement procedure is 4 time that of member states without a judicial review.Also, the coefficient for the Unicameralism is negative and significant at 1 percent level.The log-odds of reaching to higher stage of the infringement procedure are higher for member states with bicameralism than for member states with unicameralism.
The results for Decentralization, Coalition and Corporatism are either not significant or significant in an unexpected direction.Considered along with the results for Judicial Review and Unicameralism, one interpretation of these findings is that since implementation has a conspicuously legal nature, the characteristics of the judicial and legislative systems of a member state outweigh other institutional features.
The robustness of the findings for the government preference variables are further tested in the subsamples of the data.A pertinent distinction across EU policy areas is the one between de-regulatory and re-regulatory policies (Scharpf, 1999).These policies are differentiated by their substantive rationales as well as their relations to national regulatory policies.The de-regulatory policies of the EU are aimed at reorganizing the EU regulatory environment to reduce regulatory burdens on market forces so that they can operate more freely with fewer distortions and discriminations (Scharpf, 1999) (Note 11).While the rationale of de-regulatory policies is to enhance the allocative and productive efficiencies of market forces, re-regulatory policies are driven to address the regional, functional and ecological welfare effects of the operation of the European market.The underlying purpose of EU re-regulatory policies is to ease adjustment pressures for regional and functional groups, and physical environment (Note 12).
Based on the general discussions of individual EU policies (Hix, 1999a;Dinan, 2005), different policy areas are coded as de-regulatory or re-regulatory and focused on those policy categories separately.De-regulatory policies included in the data set are competition, enterprise, energy, transportation and telecommunication and information society policies.The government preference for the substantive policy outcome in this policy area is measured with support for the principle of Free Enterprise.Re-regulatory policies in the data set are health and consumer protection, environment, industrial, employment, consumer policies.The government preference for the substantive outcome in this policy area is measured with support for Market Regulation (Note 13).
Table 2 presents the determinants of implementation of de-regulatory and re-regulatory policies.That the coefficient for Pro-Europeanness is significant across the two models and in the same direction as hypothesized confirms the process hypothesis.The substance hypothesis is also confirmed in the two models.In Model 1, governments with a stronger preference in favor of Market Regulation are less likely to have severe implementation problems with re-regulatory policies than governments with weaker biases.In Model 2, governments with stronger partisan biases in Free Enterprise tend to have a better implementation record in de-regulatory policies.

Conclusion
In a bid to shed more light into the domestic source of government preference for implementation and the domestic politics of EU implementation in general, this study draws on two growing literatures in EU studies: the literatures on political contestation of European integration and political parties in European integration.Based on the insights of those literatures, it is elaborated that government preference for implementation is defined in the context of broader domestic contention over European integration and mediated by political parties.The empirical analysis that employs a unique dataset of infringement actions confirms the expectation that government preference for implementation substantially affects the resolution of member states' disputes with the Commission over implementing EU rules.As such, this study provides a further testimony to the relevance of preference in EU implementation.Hence, it joins the chorus of those who have put a renewed emphasis on preference in EU implementation studies (Thompson et al, 2007;Toshko, 2007Toshko, , 2008) ) and those who highlight the role of preference in European integration in general (Moravcsik, 1998).
It is possible to expand the current research in a number of different directions.First, in view of the generic limitations of infringement data (Hartlapp & Falkner, 2009), future research with alternative specifications and data is welcome.Second, although descriptive studies of the infringement procedure do not identify reciprocity in the relationship between government preference and the Commission's approach to member states (Borzel, 2003;Hartlapp & Falkner, 2009), future research needs to explicate the possibility of this reciprocity.In other words, scholarly attention needs to be paid to the extent to which the Commission can be more lenient toward some member states because of their policy preferences.Third, the infringement data used in the current analysis can be employed to address a number of interesting puzzles about EU implementation and infringement in particular.Some of these puzzles include the variation across different types of infringement procedures (e.g.non-notification, non-communication and non-application) or different forms of legislation (e.g.directives, regulations, decision and treaties) or different stages of implementation (e.g.transposition and application) and the influence of specific characteristics of EU decision-making on the likelihood of rule implementation.

Notes
Note 1.Although this is the common conclusion of most studies, one might argue that some EU policies, mainly cohesion, social and environmental policies, are geared toward market-correcting rather than market-building.While this challenge certainly holds an element of truth, research shows that even those market-correcting policies are driven as much by a concern with market efficiency as by a concern with the principles of justice and social equality (Wise & Gibb, 1993;Majone, 1996).Note 2. Probably, the only exception to this is the type of data collected through detailed case analyses.For example, see the 'Compliance with Europe' data of Falkner et al, 2005 Note 3. My focus is on the party composition of national governments at the time when an infringement action is initiated.Criticism might be raised that this might be problematic given a possible time inconsistency between the infringement by national government of an EU rule and the time of the infringement action by the Commission.While being cognizant of this issue, I assume that the fact that a government at the time of the infringement action does not take steps to correct the state of infringement for which they might not be fully responsible still indicative of its commitment to implementing the rule in question.Agresti, 2007, for a discussion of categorical data analysis Note 10.I estimated the models with 'robust', 'autofit', 'cluster (country)' options in Stata 10.Robust standard errors account for heteroscdasticity.With the cluster option, the models recognize the interdependence of observations for a specific country.Alternatively, I run the analysis with country dummies, the results were not different.Given the number of independent variables, I tested for multicollinearity.I did not find any correlation higher than .65 with the exception of the Population variable, which was, unsurprisingly, highly correlated with the Weighted Vote and Budget Contribution variables.I dropped the Population variable.
Note 11.These policies do not necessarily abolish regulations, but primarily streamline diverse and discriminatory national regulatory practices into a simplified framework.They either do away with discriminatory national practices through the mechanism of harmonization or neutralize their discriminatory and distorting effects through the mechanism of mutual recognition (Scharpf, 1999).Note 12.These policies are re-regulatory in the sense that instead of removing or neutralizing diverse national practices, they often complement national regulatory policies by addressing the welfare issues that have roots in European integration.Note 13.Data for both support for Free Enterprise and support for Market Regulation come from the CMP studies (Budge, Klingemann, Volkens, Bara and Tanenbaum, 2001).
coefficients for the same variables indicate that the Wald test failed to reject that the effects of those variables are invariant to the cutpoints.Those variables are marked with italic.The coefficients in the C1 indicate the log-odds of being above the first cutpoint (receiving a reasoned opinion and being referred to the Court of Justice) versus below the first cutpoint (receiving a letter of formal notice); the coefficients in the C2 indicate the log-odds of being above the second cutpoint (being referred to the Court of Justice) versus being below the second cutpoint (receiving a letter of formal notice and a reasoned opinion).Significant * p <.1, ** <.05, *** p< .01.
numbers of observations for Model 1 and Model 2 are 3777 and 4904 respectively.The coefficients in the C1 panel indicate the log-odds of being above the first cutpoint (receiving a reasoned opinion and being referred to the Court of Justice) versus below the first cutpoint (receiving a letter of formal notice); the coefficients in the C2 panel indicate the log-odds of being above the second cutpoint (being referred to the Court of Justice) versus being below the second cutpoint (receiving a letter of formal notice and a reasoned opinion) Note 4. Data for the Intra-EU Trade variable come from the databases of Eurostat: Retrieved from http://epp.eurostat.ec.europa.euNote 5. Data for the GDP per Capita and Government Revenue variables are extracted from Eurostat datasets collected and maintained by the European Commission: Retrieved from http://epp.eurostat.ec.europa.eu6.Data for this variable come from Eurostat datasets: Retrieved from http://epp.eurostat.ec.europa.euNote 7. Data for the Population variable are extracted from Eurostat datasets, Retrieved from "http://epp.eurostat.ec.europa.eu".Data for the public attitude variables come from the standard Eurobarometer surveys: Retrieved from http://ec.europa.eu/public_opinion/standard_en.htmNote 8. Data for the Preliminary Ruling variable are based of the reports of the European Court of Justice: Retrieved from http://curia.europa.eu/en/instit/presentationfr/index.htmNote 9. See

Table 1 .
Partial Proportional Odds Model: Severity of Implementation Problems

Table 2 .
Partial Proportional Odds Model: Severity of Implementation Problems Across Policy Areas All other variables in the Table 1 are included in the models, but their coefficients are not reported here.Significant * p <. 1, ** p < .05,*** p < .01.The