More inclusive European governance through impact assessments?

In this article, we present an overall picture of the inclusiveness of IA, based on data on over 800 IAs carried out by all DGs of the European Commission from 2003 to 2013. According to of ﬁ cial guidelines, IAs can deliver a variety of goals, and we posit that each goal can be linked to a different inclusion strategy. Speci ﬁ cally, we consider that the goal of coordination requires the inclusion of Commission actors, the goal of collecting neutral expertise requires the involvement of working and expert groups, while pursuing input-legitimacy requires a large number of stakeholders as well as online consultation open to the general public to be part of the procedure. Our ﬁ ndings reveal that DGs tend to prioritise coordination over collection of expertise and input legitimacy and that experience in carrying out IAs favours a more participatory approach, meaning that the more DGs make use of IA the more they will tend to include stakeholders and to launch public consultation. On the whole, the analysis highlights the importance of learning to fully develop the potential of the ambitious EU IA regime. stakeholders and expert groups in DG ’ s IA networks.


Introduction
The increased use of impact assessments (IAs) has been one of the main innovations in European governance over the last decade. Yet, within the debate on 'new (modes of) governance', IAs have hardly been discussed. Rather, it has been a topic for regulation scholars. Its relative absence from the governance debate can be explained by the fact that IAs constitute an innovation on the Community Method, or an example of 'new old governance' (Trubek and Scott, 2002) rather than an entirely new mode of governance.
In this article, we focus on one particular argument often made in relation to European impact assessments, and which is equally a key topic in the new governance debate, namely that new tools of governance, and in this case IAs, would contribute to more participatory governance. Such arguments are explicitly or implicitly made by public institutions when introducing IAs (European Commission, 2009). Since the 2001 White Paper on Governance, the notion that the inclusion of civil society actors in EU policy processes can effectively contribute to enhance the democratic credentials of the Union has been a cornerstone of the debate on governance arrangements, leading to the creation of a consultation system open to stakeholders and citizens and to the renewal of already established policy tools such as IA.
The academic debate also considers inclusiveness among the different (potential) objectives of IAs, although the argument is rarely developed in terms of democratic theory (Bäcklund, 2009;Radaelli and Meuwese, 2009;Torriti and Lofstedt, 2012). In terms of stakeholder participation, research is mainly based on case study and shows consistently that 'consultation is often limited to the "usual suspects" who have participated before or who have large resources' (Turnpenny et al, 2009, p. 643).
In this article, we go beyond case study to provide an overall overview of the inclusiveness of the EU IA system from 2003 to 2013. Our focus goes beyond the question of including civil society actors but looks more broadly to the entire diversity of actors that are included. More specifically our assessment of 'inclusiveness' is an empirical analysis of how many actors and which types of actors are involved in the IAs. We analyse new empirical data on how the different Directorates General (DGs) of the European Commission organise participation within the IAs they adopt. We do not start from a normative conception of 'inclusiveness', for example on whether the process is inclusive in relation to weaker groups in society, or overall representative of societal demands. Our main research question finds its origin in the regulatory literature on IAs. Among the many potential objectives of IAs, we identify three key objectives that relate more directly to the participation patterns within IAs. We analyse how the practices of Commission's DGs differ in relation to these three objectives. Put differently, analysing the patterns of participation, we conclude on the main usage of IAs. In addition, we clarify the patterns of participation in IAs by relying on both insights from the EU lobbying literature and from the literature on policy learning. More particularly, we test whether there is a relationship between the density of interest groups in a particular policy field and the DG's practice of organising participation in IAs in that area. Finally, we test whether processes of learning in Commission's DGs have an impact on participation patterns in IAs.
In the next section of this article, we place the issue of participation in IAs in the context of the multiple objectives of IAs in the EU, and set out our research questions and framework while identifying different types of inclusiveness. In the following section, we explain our methodology on the gathering of the data. We then proceed by analysing first the amount of actors involved in IAs, and subsequently the types of actors involved. We conclude that, overall, DGs tend to include a limited amount of actors in IAs, paying particular attention to the objective of coordination rather than to the objectives of ensuring neutral expertise and increasing input-legitimacy. Moreover, there is no huge variation in practices between DGs, and such variation is rather due to learning than to the density of interest groups in the policy field.

Participation and the Multiple Objectives of European IAs
The European Commission introduced a system of integrated IAs in 2003, in order to assess ex ante the economic, social and environmental impact of all legislative and major policy proposals. Several scholars have analysed the multiple objectives of European IAs (Rowe, 2006;Bäcklund, 2009;Dunlop et al, 2012). While Rowe provides a comprehensive list of (seven) potential objectives of European IAs, 1 Dunlop et al analyse the actual implementation practice of IAs identifying four main usages. In this article, we start from the explicit objectives formulated by the European Commission in its 2009 guidelines on IAs. We identify five key categories of objectives for IAs, which we will link in a second step to different patterns of organising participation in IAs and types of inclusiveness. • improves the quality of policy proposals by providing transparency on the benefits and costs of different policy alternatives and helping to keep EU intervention as simple and effective as possible Political control • helps to ensure that the principles of subsidiarity and proportionality are respected, and to explain why the action being proposed is necessary and appropriate.
a We borrow the concept from Rowe (2006, p. 463). Table 1 shows the objectives of IAs as literally stated in the 2009 EU guidelines (column 2), which we have regrouped in five key categories of objectives (column 1).
All the five objectives may have consequences for how one organises participation in IAs. If the objective is, for instance, political control, one may favour information and access points that allow the EP or the general public to have control over administrative governance by the Commission. Or, if the main objective is the reduction of the regulatory burden, participation may be organised in a way that costbenefit analysis is prioritised, for example by relying particularly on think-tanks or consultants providing such analysis. However, in this article we limit our attention to the three objectives that are more explicitly linked to types of participation in the 2009 guidelines, namely, ensuring evidence, coordination and input-legitimacy.
The 2009 Commission guidelines on IAs clearly state the actors to be involved in an IA. The lead service of a DG responsible for a policy proposal is also the one to take care of preparing the IA. In carrying out an IA, 'it will need the input of stakeholders and in many cases to involve internal and external expertise' (European Commission, 2009, p. 6).
Internal expertise is said to be ensured through the creation of an Impact Assessment Steering Group (IASG) to include the DGs whose policies are likely to be affected by or contribute to the objectives of the proposed policy, and the relevant policy coordination unit of the Secretariat-General. The IA guidelines also mention the European IA Board (IAB) as a potential actor, as it provides guidance on procedural and quality issues of the IA, although it does not provide expertise on the topic of the IA itself. 2 Regarding external expertise, the guidelines state that IAs 'can draw on work produced by consultants or external expertise' (European Commission, 2009, p. 18). It is said that 'expert groups and, in particular, scientific committees set up by the Commission and EU Agencies are a prime source of scientific advice'. The guidelines though also acknowledge that other external experts can be included. It is suggested that the Web application SINAPSE e-Network15 (Scientific Inform Action for Policy Support in Europe), set up by the Commission in 2005, facilitates DGs to identify, consult and involve external experts. The guidelines though stress that while part of the assessment can be outsourced to external experts, the DG remains responsible for the content and quality of the IA.
The 2009 guidelines also pay extensive attention to the 'consultation with interested parties', which is said to be 'an essential tool for producing high quality and credible policy proposals. Consultation helps to ensure that policies are effective and efficient, and it increases the legitimacy of EU action from the point of view of stakeholders and citizens' (European Commission, 2009, p. 19). This is the only place in the guidelines where legitimacy is mentioned. Unlike the search for internal expertise in other DGs and external expertise in expert committees, agencies and consultants, the inclusion of stakeholders is said to contribute to both effective/ efficient governance and to more legitimacy of EU action.
Hence, the 2009 guidelines make a clear distinction between three types of actors, namely internal Commission actors, external experts and stakeholders. All of them are said to contribute to the objective of gathering evidence for policy-making. 3 However, each of these types of actors contributes to evidence in a different way, and some of these actors contribute also to other objectives of IAs in a way the other categories do not.
First, a DG is expected to involve other Commission actors because they provide evidence through the valuable knowledge, internal studies and analyses they possess, but also because such involvement ensures coordination. Second, while it is acknowledged that stakeholders or 'interested parties' can provide evidence, and can contribute to, for example, 'finding new ideas (brainstorming), collecting factual data, and validating a hypothesis' (p. 19), their involvement comes also with the warning that 'it is important to distinguish evidence from opinions' (p. 20). Stakeholders are expected to provide partisan information, and it is therefore said that DGs should ensure 'peer-reviewing, benchmarking with other studies and sensitivity analysis' in order to 'significantly enhance the quality of data' and ensure 'the robustness of the results' (p. 20). At the same time, DGs should ensure to 'engage all affected stakeholders' and 'consult all relevant target groups'. Such wide involvement seems to have a representative dimension; it is only through wide participation of all stakeholders that the feasibility and legitimacy of policy proposals is ensured. Only wide stakeholder involvement ensures input-legitimacy. Third, at the same time there seems to be an assumption that the category of external experts, namely expert committees, agencies and individual experts, ensures the gathering of 'neutral expertise', for which neither a cautious nor representative approach is required.
On this basis one can make the following assertions: A1: The more IAs aim at coordination, the more the focus will be on the inclusion of Commission actors. We can call this internal inclusiveness.
A2: The more IAs aim at the gathering of 'neutral expertise', the more the focus will be on the inclusion of external experts, that is, expert committees, agencies and individual experts. We call this expert inclusiveness.
A3: The more IAs aim at input-legitimacy, the more the focus will be on the inclusion of a wide array of stakeholders. We call this stakeholder inclusiveness.
Consequently, from the patterns of participation we identify below (more or less involvement of Commission actors, experts and stakeholders) one can draw conclusions on the main usages of IAs in terms of ensuring three key objectives: coordination, neutral expertise and input-legitimacy.
Our focus is on the types of inclusiveness, namely whether IAs aim rather at internal, expert or stakeholder inclusiveness. As set out above, we do not make normative claims on whether an overall pattern of participation is more or less 'inclusive' depending on whether more internal actors, experts or stakeholders are included. Hence, we do not make claims on whether, for example, a network with a high number of experts and low level of stakeholders would be more or less inclusive than another network that is overall smaller but includes more stakeholders. Nevertheless, the overall number of actors involved, that is the extension of the network, allows drawing some conclusions on the usage of IA, more particularly we assert the following: A4: To ensure input-legitimacy, a very high number of actors need to be involved, considerably more so than when the objective is either coordination or gathering neutral expertise.
To explain the differences between DGs in their patterns of organising participation in IAs, we formulate three hypotheses. The first hypothesis relates to external factors to the DG: Hp1: The higher the density of civil society actors in a policy field, the more the IA networks of the DGs are likely to be extensive.
The second and the third hypotheses refer to internal factors and take into account that DGs go through a learning process to realise the three objectives of IA that is facilitated by experience and a dedicated staff. Both factors are indicators of the level of institutionalisation of IA procedures in the policy practices of DGs and, following a neo-institutionalist logic, are expected to influence the functioning of IAs. We therefore posit: Hp2: The higher the number of IAs exercised by a DG, the more likely it will build up the experience to involve a greater amount of actors in its IAs.
Hp3: The existence of a unit within a DG with the specific responsibility for IAs encourages a process of learning, leading to wider IA networks.
The next section will set out how we have gathered our data, before testing these data in the light of the assertions and hypotheses set out above.

Methodology and Data Collection
To get a picture of which actors have been included in European IAs, and thus the type of inclusiveness, we have gathered data on the actors explicitly mentioned in the IAs. This proved possible as IA reports are structured according to a standard format, and more specifically they include an initial section with detailed information on procedural issues. We consulted these reports and collected available information on procedures followed in all 830 IA exercises adopted between 2003 and 2013, 4 that is all IAs adopted since the first general guidelines lead to the systematic publication of IA reports on the European Commission website set up for this purpose (http://ec. europa.eu/governance/impact/index_en.htm). Specifically, we recorded the presence or absence of four types of actors. First, we took note of the composition of the IASG in charge with the IA. The IASG is composed of a variable number of DGs, ideally all directorates with competences on the issue at hand should be included. 5 Second, we recorded the presence of institutions and services that have been consulted in the process, distinguishing between EU agencies and expert groups, 6 constituting together the category of actors ensuring 'external expertise' according to the IA guidelines. 7 As a third step, we took note of all private and public interest groups and organisations whose names are explicitly mentioned in the report for their contribution to the assessment. This group includes the selection of stakeholders who have been contacted by the leading DG and the IASG and whose advice has been formally taken into account in the process. To assess the propensity to include stakeholders we also recorded whether the procedure included an online consultation open to the general public. These are online questionnaires carried out in accordance to guidelines on 'General Principles and Minimum Standards for consultation of interested parties' (European Commission, 2002) to foster civil society and citizen participation into policy formulation. 8 In doing so, for each of the 830 IA exercises we have a map of the network composed of institutional and social actors -DGs, agencies, expert groups, stakeholders, the general public (by way of online consultation)who had a role in the assessment.
Relying on the data explicitly stated in the IA report has its limitations, namely the IA reports mention these actors as single sources of information. Yet, often these actors rely on a wider network of actors. For instance, a report of a European Agency addressed to the Commission in the context of an IA will most likely be based on the information provided by the wider decentralised network of such an agency, composed of mainly national administrations, and sometimes also of other stakeholders. Equally, expert groups consulted during an IA are composed of different actors, some of which may themselves link back to a wider network of actors. This can also be said of the stakeholders involved in the IA, which may be European interest organisations with decentralised membership organisations. The information provided in the IAs does not allow drawing a complete network of actors that directly or indirectly may have influenced the IA. However, we believe that the information provided on the number and type of actors, which, according to the Commission, were directly involved in the IA, allows drawing some initial conclusions about the inclusiveness of the European IA procedure.
In addition to the data on actors provided by the IAs, we have collected two other sets of data, in order to explain variation in participation patterns in IAs across DGs. In order to assess whether there is a relation between the pattern of participation in IAs and the density of interest groups in the policy field, we collected data from the Transparency Register of the European Commission (http://ec.europa.eu/transparencyregister/info/homePage.do.). The Transparency Register has been widely used by scholars working on lobbying, as it usefully requires interest groups to provide basic organisational information and to indicate the policy fields they are active in (Broscheid and Coen, 2007;Coen and Katsaitis, 2013). Although registration in the Transparency Register is not mandatory, registration numbers are high enough to draw conclusions on the density of interest group activity within a policy field, ranging from 700 groups who work on Fisheries to over 3000 that fall within the areas of competence of DG Sanco.
Finally, to take into account aspects of learning in DGs, we also collected data on the existence of an IA unit within each DG, which was done on the basis of information provided on the Commission Directory website (http://ec.europa.eu/ staffdir/plsql/gsys_page.display_index?pLang=EN.). For each DG we checked for the presence of a division or subdivision explicitly dealing with the delivery of impact assessments, such as the 'Impact Assessment and Evaluation' unit within DG EMPL or the 'Economic Analysis and Impact Assessment' unit within DG MOVE. Where individual policy officers are in charge with IA (such as a legal officer within DG HOME's Unit on 'Programming and legal questions') or where no mention of IA could be identified, we recorded the absence of a specific IA unit. 9 Owing to strict word limits, in the following sections we will present and explain the findings, providing only basic information on statistics behind them in footnotes. Full information on statistical analysis is available upon request from the authors.

Number of Actors Involved in IAs: The Inclusiveness of IA Networks
In this paragraph, we start our overview of patterns of participation in IA by exploring the number of actors who are involved in each IA procedure. The extent of the IA network is important as it sheds light on the inclusiveness of practices that have been established by DGs in daily activities.
According to our data, the average number of actors who work together to deliver an IA report is 10.3. This data is interesting insofar as it provides a practical sense of the working relations that are established for IAs: a relatively small group of officials, experts and stakeholders work together to gather the relevant evidence, discuss options and deliver a report on likely impacts of policy proposals. These data suggest that the primary objective of IAs is the search for sound evidence and/or coordination, but definitely not ensuring input-legitimacy as the latter would require the involvement of a much wider representative array of all stakeholders. There are however differences among DGs that can be tested according to our hypotheses Hp1, Hp2 and Hp3.
First, we can test whether differences in the extension of sector-specific IA network can be explained by some fundamental characteristic of the policy field. In particular, previous research has shown that policy sectors are clearly differentiated in terms of the density of civil society activity and that this has significant consequences for DG's choices in terms of inclusion and exclusion of stakeholders in governance arrangements (Broscheid and Coen, 2007;Greenwood, 2007;Gornitzka and Swerdrup, 2008;Coen and Katsaitis, 2013). It might be argued that the higher the number of actors who are active in a policy field, the stronger the pressures on DGs for the adoption of an inclusive approach to policy processes. In the literature on participatory forms of governance, there is some empirical support for this statement. For instance, Broscheid and Coen (2007) demonstrate that the higher the number of interest groups in a policy sector, the higher the propensity of DGs to include experts and stakeholders in ad hoc working groups and expert fora. The authors suggest that this phenomena is a response on the part of the Commission to group overload. We have therefore tested whether the same relation can be observed in the case of IA procedure. Our data shows thattaking into account differences in the number of IAs by DGthe number of private, public interest groups and think-tanks in a sector does not affect the average number of actors involved in each IA exercise by DGs. This suggests that a large IA working group depends mainly on the general level of IA activity of a DGmeasured in terms of the number of IAs performedand does not reflect a 'crowded' policy arena in which the DG operates. In other words, our findings reveal a degree of insulation of IA procedures from external pressures.
Second, we did find a relevant correlation between DG practices and processes of learning within the Commission (hypotheses Hp2 and Hp3). To explore this aspect, we need to take into account the number of IAs performed by each DG, which appears to be highly variable. It ranges from a minimum of four assessments carried out by DG ECOFIN to a maximum of 100, for which DG MARKT was responsible in the period 2003-2013. In other words, IA can represent a distant procedure to be performed once every 2-3 years or conversely a well-established routine for DGs that run several IA processes every year. The frequency of use of IAs by a DG depends on many factors, many of which are not due to choices made in the DG. Above all, it depends on the type of policy action that DGs are most likely to intervene in, as IAs are required particularly for regulatory (and above all) legislative action, and the regularity of such action. Hence, one can, for instance, expect for DG COMP to adopt much less IAs than DG MARKT, as the latter sees much more regulatory initiatives than the former.
We posit that differences in the frequency of use of IAs are relevant for patterns of participation because innovations in forms of governance, such as IAs or online consultations, require a high level of experience to be fully implemented. For instance, previous research on the quality of EU IAs pointed out that the sophistication in terms of the analysis performed and their learning potential are all positively affected by a high level of knowledge of the procedure itself, which arguably is acquired through experience (Hertin et al, 2009). We therefore hypothesise that the frequency in the use of IA affects the inclusiveness of the procedure, by helping DGs to set up a network of actors to be involved. In other words, we expect that DGs who perform a high number of IA will be more inclusive in each IA that they are responsible for compared to DG less active with this specific form of governance. Our data partially confirm this hypothesis, as shown in Figure 1.
Up to a point, the more a DG makes a frequent use of IA, the more it will expand the group of actors that are involved in the appraisal. For example, DG Relex that has been required to deliver only one IA every 2 years limited its network to an average of five actors, whereas DG RTD that has been engaged with two IAs every year in the period 2003-2013 expanded its average network to 14 members. The positive correlation between frequency of IA and inclusiveness however is valid only to a certain extent. DGs that are very frequently involved in IA tend to stabilise around a relatively small group. More specifically, DGs that perform five or more IAs every year tend to form a network composed of around 10 actors. In other words, it is possible to observe a threshold effect: above a certain number of IAs to be delivered each year, the inclination to include more and more actors disappears. Put differently, these data suggest that learning within DGs leads to an increased inclusion of actors but that it also leads to the conclusion that having reached a threshold of 10 actors, the further extension of the IA network is considered counterproductive or at least not providing added value.
Third, this conclusion is further supported by the finding that the formation of an IA unit within a DG makes a difference in terms of the extension of the IA network, as suggested by our hypothesis Hp3: DGs where a dedicated staff is in charge with IA procedures are more likely to extend their contacts until a threshold of around 10 actors is reached. To sum up, internal processes of capacity and learning within the Commission rather than external pressure by density of the interest group field thus seem to explain variation between DGs in the extension of their IA network with stakeholders.

Types of Actors Included
The overall picture that emerges from the description above suggests that there are relevant differences among DGs in terms of the use of IAs as a tool and in their efforts to include the participation of institutional actors, experts and stakeholders.
Such differences in the procedure to perform IA can be linked to differences in the interpretations of the purposes of IA as a tool to support decision making. An interesting analysis to explore this issue can be made by looking at differences in the propensity of DGs to network within the Commission, or with experts and/or stakeholders.
As we have set out above, the 2009 IA guidelines make a clear distinction between three type of actors in IA, namely Commission actors providing 'internal expertise'; expert committees, agencies and individual experts providing 'external expertise', and finally interested parties or stakeholders contributing to both more effective and legitimate policy-making. As argued above, while all actors contribute in some way to evidence-based policy-making, only Commission actors contribute to the objective of coordination; only the 'external experts' are supposed to contribute to 'neutral expertise', while only the involvement of a wide array of stakeholders can ensure input-legitimacy.
On this basis, one can infer the dominant interpretation of IAs given by each DG on the basis of the observable choices made in terms of types of actors involved. More specifically, we consider that IA is given (a) a coordination goal if a DG prioritises the involvement of other DGs; (b) a goal of 'neutral' expertise collection if the DG prioritises links with expert and advisory groups, and EU agencies; and (c) the goal of input-legitimacy if the DG prioritises contact with a wide range of stakeholders and the general public with a view to map preferences on an issue.

A strong coordinative function
To calculate the propensity of each DG, we calculated the proportion of DGs, agencies, expert groups and stakeholders that have been included in each IA by each DG and evaluate the relative incidence of each category of actor.
As Figure 2 shows, a typical situation is that IAs are carried out by a group of around 10 actors (as explained above), half of them are representatives of other DGs. To them a variablebut lowernumber of stakeholders and a limited number of experts from agencies or working groups are added.
It is clear from this graph that when DGs make use of IAs to gather evidence they are not primarily relying on assumed neutral expertise provided by external experts (being advisory committees, agencies and individual experts); other Commission actors and stakeholders are at least as important. In number of actors consulted, they are both more numerous than external experts. As we noted at the start of this article, we do not measure de facto influence, but the data allow in any case concluding that DGs, in their search for evidence, rely on all type of actors and do not privilege external experts. This applies to all DGs.
However, we do notice some relevant differences in participation patterns according to DGs. In what follows we focus on what appears the most relevant distinction, namely between coordinative and input-legitimacy function of IA, and therefore the contrast between intra-institutional and stakeholder involvement. Figure 3 shows in more detail the relation between the number of DGs and the number of stakeholders who have been directly involved. More specifically, each bar in the graph reports the number of DGs included for any stakeholder. For example, the data for DG MARKT indicates that in its network there are 2.2 DGs for each stakeholder. This means that the higher the number, the stronger the priority given to networking within the Commission compared to networking with interest representatives. According to this series of data, only DG EMPL decisively contacts more stakeholders than other DGs (in a proportion to two interest group for each institutional actor), while a small subset of three DGs has a balanced approach, inviting a similar number of social and institutional representatives. At the other end of the spectrum, DG REGIO, ENER and DG TRADE appear to use IAs mainly for coordinative purposes, inviting five other DGs for each stakeholder.
The majority of DGs conform to the general model briefly outlined above and tend to prioritise the coordinative function within the Commission. From this point of view, there are no significant differences among DGs operating in policy fields with a  high number of interest groups (Hp1), or among DGs who set up a dedicated IA unit (Hp3). This suggests that the interpretation of IA as a coordinative tool is well established within the Commission and that the input-legitimacy function remains rather marginal. Our data on the type of actors involved suggest that while the objective of sound evidence is obviously central to IAs (whether through internal or external expertise or stakeholder participation), one should clearly not underestimate coordination as a core objective of IAs. In some analysis of IAs, such as Dunlop et al (2012), coordination is only given secondary importance. However, our data show that involving other DGs is a key priority of IAs and takes a high proportion of actors involved (particularly taking into account that the Commission has only a limited number of DGs compared to the potentially high number of stakeholders and experts).

Relatively weak input-legitimacy function
In terms of the objective of providing evidence for policy-making, our data suggest that there is not a clear preference for 'neutral expertise' provided through expert committees and agencies. Overall, IAs tend to include rather a higher amount of stakeholders. However, one cannot deduce from this that ensuring input-legitimacy through participation is the key objective of IAs. The 2009 Guidelines stress the importance of wide and balanced consultation as a way to ensure the feasibility and legitimacy of policy measures. We therefore use two criteria to assess whether current participation patterns are driven by this objective. First, if input-legitimacy were the key objective, one would expect a frequent use of online consultations, which provide the widest participatory basis. Second, if input-legitimacy were the key objective, one would expect key stakeholders with high representative value to be involved regularly. Both hypotheses are contradicted by our findings. We have contrasted the propensity to search for information relevant to IA from the variety of experts included in the expert and advisory group system with the propensity to reach out to the general public by launching an online consultation process. As clearly stated in EU guidelines on consultations, the two strategies are not mutually exclusive and the policy formulation stage can include both focused and open forms of consultation. This sort of wide-ranging consultation effort would seem appropriate for IA, as by definition this specific form of appraisal is applied to a selection of highly salient issues that are priorities in the annual working plan of the Commission. Still, the combination of focused and open forms of consultation does not appear the norm: while the proportion of IAs that does not include any contact with one or more expert groups is negligible, only 25 per cent of IA procedures include an online survey. In general terms, all DGs but CLIMA and DEV seem to prefer focused over open forms of consultation when delivering an IA. This finding suggests that technical, focused, specialised expertise coming from a limited number of experts and institutional actors is prioritised over the inputlegitimacy objective of IA.
An important qualification is needed. The frequency in the use of IA is strongly correlated with the inclusion in the appraisal procedure of online consultations (Correlation r = .67). This means that the more a DG is in charge with IAs, the more it will consult the general public. The presence of a dedicated IA unit within the DG is similarly significant. These findings confirm that a learning process and the institutionalisation of procedure are needed to fully adopt the EU guidelines on IA that, as noted above, present a multiplicity of goals and objectives, the participation of the general public being the more recent and innovative.
Moreover, we note that most stakeholders have been included only once in IA. This suggests that they are selected on the basis of the specific contribution that they can provide in terms of expertise and preference mapping on the issue at hand, rather than on the basis of their generic importance or general 'representative nature' in the field. This also suggests that IA is a rather fluid procedure, and that established working relationships between DGs and representatives of interest groups have not been replicated in the context of IA.
There are however a number of groups who result to be more frequently involved in IA. Examples are Copa-Cogeca (a leading farmers' organisation), UEAPME that represents small and medium-sized enterprises, Business Europe, CECED that represents domestic equipment manufacturers, Eureau (water operators), WWF, Amnesty International, ECRE (a network of NGOs assisting refugees), ICES (a network of scientists working on ocean sustainability), and BEUC, representing consumers. These are frequent players in the context of IAs having taking part in three or more exercises. Further, it is possible to identify a small number of groups who have been involved by more than one DG. This group includes again BEUC (Bureau Européen des Unions des Consommateurs), UEAPME (European Association of Craft & Small and Medium-sized Enterprises), European Automobile Manufacturers' Association, Business Europe, the European Trade Union Confederation and the WWF. Their presence however is far from being systematic and in our view they cannot be considered regular players in the IA procedure. Taken these findings together, it seems fair to suggest that the participatory objective in IA has been interpreted in strictly instrumental terms, as a way to collect relevant information and knowledge on the issues to be assessed. At the same time, it can be noted that all repeat players are interest groups rather than think-tanks or consultants. This suggests again that DGs go wherever they believe they find some expertise, rather than prioritising some sort of assumingly more neutral expertise at the detriment of actors who represent more clearly a particular interest.

Conclusion
In this article, we discussed the contribution of IAs to the participatory features of EU governance. We highlighted that IA official guidelines are ambitious insofar as they require DGs to simultaneously deliver a multiplicity of objectives when carrying out IAs, policy coordination, collection of robust evidence and input-legitimacy being the most relevant ones.
We posit that such objectivesthough not incompatiblerequire DGs to adopt a complex strategy to deliver each IA. In terms of inclusiveness, the search for policy coordination leads DGs to build intra-institutional relationships; the gathering of neutral expertise will call for the inclusion of expert groups and EU agencies; the input-legitimacy goal is best pursued by the involvement of a significant number of representative stakeholders and the use of online consultations to collect contribution from the general public.
Our empirical analysis on the number and the type of actors involved in all IA exercises carried out from 2003 to 2013 reveals that DGs tend to prioritise one IA objective over the others. With the only exception of DG Employment, all DGs are more likely to involve other DGs with competences on the issue at hand rather than external experts and/or stakeholders. According to our data, the policy coordination objective is the dominant interpretation given to IA within the European Commission. Internal inclusiveness rather than expert or stakeholder inclusiveness is the priority.
Given this shared understanding of IA, there are however some relevant differences among DGs, particularly in their contacts with stakeholders. Our findings suggest that DGs who are more frequently involved in IAs and those who have established an IA unit set up a larger network of stakeholders and make more systematic use of online consultations. This finding suggests that DGs need time and experience to learn how to apply IA guidelines properly and that the presence of dedicated staff increases the potential for enhancing the participatory character of IA. While we can confirm that the multiple objectives of IA are more likely to be delivered when the conditions within the Commission are favourable, we have found that the existence of a dense web of interest groups in the policy sector does not significantly affect the number and the type of actors involved in IA.
2 For this reason, we have not counted the IAB as actor in our analysis of inclusiveness as its role is one of observer and quality control on the process rather than participant on the topic at hand. 3 All three categories are mentioned in the section 'Data sources, collection and analysis ' (p. 18). 4 A small number of IA reports deviate from the standard format for reporting and do not include a procedural section. These cases are missing in our database. 5 In the last 10 years the configuration of the Commission changed. New DGs have been created (such as Dg Clima, Energy, Move in 2010), other discontinued (such as DG Tren that has been divided in 2010).
To take into consideration such modifications in our models, data are weighted to account for the number of years that each DG has been active in the period 2003-2013. 6 These come under different names, namely expert group, advisory group or working group. 7 While the guidelines also mention individual experts in this category, their presence in the IA was so low that they were statistically not relevant to be included. 8 Consultations open to the general public, carried out in the context of IAs, Green and White Papers and other major policy initiatives are accessible through the 'Your Voice in Europe' portal. See http://ec .europa.eu/yourvoice/. 9 DGs that do not mention IA might have units dealing with various forms of monitoring and evaluation.
For instance, DG CLIMA includes a unit on 'Strategy and Economic Assessment', DG COMP a unit on 'Strategy, Delivery and Evaluation' where relevant skills and competences for IA can be found. As we are interested in assessing the extent to which IA has been institutionalised within the Commission, we considered these units too generic to be recorded as IA units.