Policy-centred sampling in interest group research: Lessons from the INTEREURO project

Proper sampling is the foundation for all scientific enquiry aimed at making generalizable claims about a wider set of cases. Indeed, inferential statistical analysis presupposes representative samples and units of analysis that can be considered as independent observations. Establishing a sample of issues on which lobbying may take place, which is at the same time representative of an overall population of issues and of the varying levels of conflict and political mobilization, however, is a major challenge for interest group research. Drawing on existing research practices, we discuss a series of different approaches that may be used to establish a sample of policy issues. The focus then is on the policy-centred stratified sampling procedure used in the INTEREURO project. Although our approach has important advantages, we extensively discuss several challenges we faced as well as the procedures we developed in order to deal with these.

surrounds specific policy issues is not taken into consideration (Beyers et al, 2008). In the past, the inherent complexity of interest group politics made many group researchers rely primarily on case studies. Although these studies made major contributions to the field, by definition case study findings usually remained restricted to particular types of groups, specific issues or policy areas under a particular set of conditions. Moreover, while case studies have contributed substantially to the explorative research on group activities and theory development, they are often set up in a way that inhibits direct comparisons and cumulative insights necessary to draw more general conclusions.
During the past few decades, the field moved beyond this non-cumulative state of affairs as a diverse and rich set of qualitative and quantitative studies has been added to the literature. Yet, some earlier quantitative studies adopted a survey methodology starting from a sample of interest groups and usually no or only a small number of policy issues were part of the research design (for an overview, see Baumgartner and Leech, 1998). Interest group surveys have their merits in terms of the large set of groups that can be studied, they are limited in their ability to capture the policy context in which interests groups operate. While case studies may run the risk of offering a somewhat idiosyncratic and unrepresentative picture of lobbying, large-scale quantitative research designs may result in a de-contextualized understanding of lobbying.
Instead of actor-centred sampling, that is, sampling from a universe of organized interests, or selecting only a small number of policy cases, our project hence relies on what we label as policy-centred sampling. Essentially, this means that sampling does not start from actors, but from a set of public policies. Policy-centred sampling is typical for research designs that aim to blend the benefits of case study research in gaining context-sensitive insights into groups' lobbying activities with the advantages of a large-N approach in generating generalizable knowledge. In order to combine qualitative and quantitative evidence, we sampled a moderate number of legislative proposals and then drew on various sourcessuch as structured personal interviews with group representatives and decision makers (Beyers et al, 2014b) as well as qualitative and quantitative content analyses of newspaper articles and official documentsto obtain information on policy substance, interest groups' diverse mobilization efforts and their effect on policy outcomes. Relying on qualitative and quantitative methods to analyse our multifaceted empirical material not only facilitates the substantive understanding of interest group politics in European policymaking but should also increase the validity and reliability of our descriptive and causal inferences.
This article presents an overview of the rationale behind policy-centred sampling, the way we implemented this in the INTEREURO project, the representativeness of our final sample and some of the pitfalls and challenges in drawing policy-centred samples. 1 The main problems we face are the undefined nature of the population and the fuzzy boundaries between individual cases in the universe from which we sample. Many structural features of this universe are unknown and often policy issues are not entirely independent observations. Yet, by carefully designing the different steps of the sampling process and remaining reflective during the overall research project, these challenges can be addressed adequately. In general, we believe that there are many reasons why policy-centred sampling is a highly useful approach in studying interest group strategies, framing and influence.

Approaches to Policy-Centred Sampling
The main reason for policy-centred sampling is the theoretical notion that variation in substantive policies shapes the nature of politics surrounding specific issues (Lowi, 1964(Lowi, , 1972Wilson, 1989). Who gets involved, how, and in which arena, how issues are resolved and how much influence stakeholders have, is largely determined by specific features of policy issues. A key variable in this regard is how the costs and benefits are distributed across segments of society. Thereby it is often hypothesized that the concentration of costs and benefits generates competitive battles among interests groups, whereas diffuse costs (or benefits) make interest group politics less competitive and/or more prone to capturing. Another crucial factor is that, under conditions of uncertainty, human beings are generally more sensitive to the threats and the risks of losing benefits (Kahneman and Tversky, 1984), which means that the salience attached to specific issues, the level of conflict and the mobilization of a highly attentive public may considerably affect interest group strategies and influence Dür and Mateo (2014). Instead of objective costs and benefits, much lobbying is driven by bandwagoning, herding, mimicry and how entrepreneurial stakeholders frame the nature of policy issues (Granovetter, 1978;Banerjee, 1992;Baumgartner and Leech, 2003;Halpin, 2011). The skewed pattern that is regularly observed in lobbying communities could be partially explained by these social mechanisms. Finally, the decisionmaking rules, the power distribution between legislative institutions as well as the venues in which issues need to be resolved might also affect interest group mobilization. For instance, it makes a considerable difference whether issues are processed by judicial venues involving mainly legal reasoning or whether issues are subject to parliamentary debate involving extensive media coverage.
All this speaks in favour of policy-centred sampling. Yet, the awareness of a theoretical notion does not solve the problem of how to sample policy issues. A major problem is that in most circumstances we lack a proper map of the universe from which to sample. And if official records are available, they mostly do not map whether and how the different issues are related to each other. Often it is unclear where a particular issue stops and the next issue starts, to what extent issues are subsets of a more encompassing issue, or how issues are tied in a larger legislative package. Another problem is that the entities we are interested in are, regarding some of their key features, characterized by highly non-linear distributions (King et al, 1994, p. 125), which makes simple randomized sampling problematic.
Policy-centred sampling thus cannot presume a population in a usual sense. In view of this problem, scholars have used two types of policy-centred sampling: a bottom-up perspective starting from a set of sampled actors and a top-down perspective starting with a set of publicly available sources. Bottom-up policycentred sampling, as exemplified in the work by Baumgartner et al (2009, pp. 261-276; see also Mahoney, 2008), starts from a random sample drawn from a population of organized interests that is active within a particular venue. In the next step, the sampled organizations are interviewed focusing on the most recent issue in which the organization spent time and resources. All the issues identified in the interviews then constitute the sample of issues and are closely scrutinized on the basis of publicly available sources as well as further interviewing with other stakeholders.
The major advantage of this procedure is that it leads to a wide and diverse sample of issues on which at least some lobbying took place, with issues on which a lot of lobbying took place more likely to be included in the sample (Baumgartner et al, 2009, pp. 267-271). The approach heavily depends on the availability of a reliable record of the interest group universe. However, many political systems, including the European Union (EU), lack such a record. Another potential problem might be situated during the interviewing stage (see the contribution by Beyers et al, 2014b). When talking to stakeholders, in particular, during the first round of interviews, the interviewer does not possess prior evidence or information on the policy issues, which means that the interviewer might be somewhat constrained in checking factual details or might not always grasp the intricacies of a particular issue.
In contrast, top-down policy-centred sampling starts researching the universe of policy cases that were on the policy agenda during a particular period of time. For this approach, no register of lobbyists is needed; at minimum, one only needs a systematic overview of a list of laws, regulations or other policy initiatives initiated by policymakers. A challenge is the skewed nature of policy attention that is typical for most political agendas. This means that an un-weighted random sample will include a large number of low salient issues, where usually little or no lobbying takes place, and only a small number of highly contested issues. Yet, if one aims to analyse how lobbying influences policymaking, one needs cases with at least some lobbying. At the same time, in order to establish general conclusions we need sufficient diversity in terms of controversy as well as control cases where no or little lobbying takes place.
An interesting example of top-down policy-centred sampling is provided by Thomson's (2011, pp. 27-32) work on legislative politics in the EU (see also . Thomson started with a list of legislative proposals and used 'the political importance' of a proposal for the selection of cases with a sufficient level of political controversy. His selected cases needed at minimum four hits in two media sources, namely, Agence Europe and the European Voice. Note that two media mentions is still a rather low threshold that makes that highly visible cases will be selected in addition to a large number of cases that get much less attention. After sampling these issues, the research continues with expert interviews and extensive document archiving. One major challenge for this procedure is that one needs to define the weighting factorsmostly level of controversy, political salience, public attentionin advance. For this, scholars usually rely on media attentiveness, but we need to acknowledge that salience remains a concept that is notoriously difficult to measure (Warntjen, 2012).

The INTEREURO Sampling Procedure
The basis for the INTEREURO sample is the list of all proposals for generally binding EU law, that is, directives (N = 144) and regulations (N = 459), adopted by the European Commission between 1 January 2008 and 31 December 2010. 2 We opted for this period to maximize the chance of finding interview partners in the Commission as well as among organized interests and to be reasonably certain that at the time when we did the interview a decision would have been taken on the proposal (so no proposals that were too recent could be included; see also Beyers et al, 2014b). We dropped all proposals for codifications, as these do not change the substance of existing legislation. After also eliminating a few proposals concerning the remuneration of and/ or pensions for Commission officials or the internal management of the EU institutions, the list contained 111 proposals for directives and 427 proposals for regulations.
We decided to sample directives and regulations separately, as a simple random sample would have made us select very few directives (about four regulations for each directive). Having a large enough number of directives in the sample was important as we wanted to see whether there are any differences in terms of lobbying between regulations and directives, and because for directives we can also study lobbying during the national transposition process. Moreover, we stratified our sample according to public saliency, which increased the likelihood that highly salient cases would be selected. We did this by checking the media coverage of all proposals in five sources, namely, Agence Europe, European Voice, the Financial Times, the Frankfurter Allgemeine Zeitung and Le Monde. Using non English-language media sources was important to avoid a bias in our sample towards proposals of specific interest to some countries. About two-thirds of the proposals for directives and 60 per cent of the proposals for regulations were mentioned in at least one media source. We found most media hits in Agence Europe, and fewest hits in Le Monde (see Figure 1). In all sources but European Voice, proposals for directives got more media attention.
The resulting sample includes 48 proposals for directives and 38 proposals for regulations that were mentioned in at least two of these media sources. We opted for this rather low threshold of two newspaper articles in order to allow for large variation in salience across proposals in our sample. Moreover, to add additional variation in terms of salience, we added a randomly selected set of 10 proposals for directives and 9 proposals for regulations that did not meet the media coverage criterion. Finally, we added all proposals for directives and regulations that had not made it into the sample following that strategy, but for which public consultations had been held and consultation documents are available. We did this for pragmatic reasons, as we wanted to benefit from the additional data that is available for consultation cases.
This process resulted in a sample of 125 proposals, which breaks down to 64 proposals for directives and 61 proposals for regulations. The interviewing projects in the influence and the strategies modules take this set of 125 cases as a starting point for their data collection. Two parts of the INTEREURO project use different samples (see Beyers et al, 2014a). The multi-level governance (MLG) module concentrates on lobbying and controversy at the member state level, and for this purpose we identified 20 proposals for directives with the largest media coverage, to maximize the chances that the cases to be analysed created some controversy at the national level. The US-based framing project (see XX in this volume) considers all proposals for directives (N = 27), regulations (N = 21) and all the Green and White Papers (N = 20) on which public consultations have been held and for which position papers submitted by organized interests were available. 3 Of the 68 cases in the US sample, the 48 directives and regulations overlap with the interviewing sample of 125 legislative proposals. Table 1 summarizes the INTEREURO samples.

The INTEREURO Sample
In this section, we describe how representative our sample is for the population of EU legislative decision-making processes. The vast majority of the cases in the sample deal with policies regulating economic exchanges between member states in the EU's internal market (see Figure 2). Our sample covers regulatory policy areas (Internal Market and Services, Health and Consumer Protection, Internal Market and Services Policy-centred sampling in interest group research and so on) as well as the few policy areas in which the EU is directly involved in redistribution (Agriculture and Rural Development, Regional Policy). 4 As the INTEREURO sample covers all policy areas in which the EU legislates, we will be able to draw general conclusions on interest group mobilization, groups' strategies and interest group influence in internal market policymaking. Yet, although our sample is broadly representative of the policy areas that constitute the internal market, trade policies, which are the cornerstone of the EU's activities with third countries and other regional and international organizations, are strongly underrepresented. Our sample will therefore not allow us to draw broader conclusions on interest groups politics in EU external affairs.   As directives need to be implemented through national laws in member states to become effective, national interest organizations might spend less effort and resources in lobbying directives at the EU-level but focus their attention on the domestic implementation stage instead. If that is the case, our sample might underestimate the extent to which national interest organizations are lobbying at the EU-level. At the same time, national organizations might want to secure their stakes in the early stages of EU-level decision making, thus leading to no differences in the extent to which national and European organizations lobby the formulation of directives and regulations at the EU-level. Given the sufficient number of regulations and directives, respectively, we should be able to empirically assess potential biases. Finally, we expect the institutional rules by which legislative decisions are taken to have an impact on interest groups' lobbying efforts and strategies. Two decisionmaking procedures account for the vast majority of policies adopted at the EU-level (Hix and Høyland, 2011). In the consultation procedure, the Commission formulates a proposal, which is then adopted by member state governments in the Council. The European Parliament can only submit nonbinding opinions. In the co-decision procedure, on the other hand, the European Parliament is a co-equal legislator; proposals formulated by the Commission only become effective if adopted by member state governments in the Council and the members of the European Parliament. Interest group mobilization and decision-making dynamics might vary considerably between both procedures due to the differential access different types of groups might have to the EU's legislative institutions and their need to invest organizational and political resources carefully when trying to influence EU policies (Beyers and Kerremans, 2004;Dür and Mateo, 2012). The INTEREURO sample under-represents the share of proposals decided in the consultation procedure during the period 2008-2010 (see Figure 4). Since 2009, however, the number of consultation procedures has sharply decreased so that our sample broadly reflects Policy-centred sampling in interest group research the institutional rules by which policies are taken in the recent phase of European integration. In our description of the sampling procedure, we highlighted that our strategy aimed at selecting cases which vary regarding the public attention they received. This selection strategy was meant to avoid a sample that is overwhelmingly made up of proposals that experienced little political conflict. Taking the number of days it took the Council and the European Parliament to adopt the European Commission's legislative proposal as a proxy for the level of conflict, we find that conflicts during the decision-making phase vary considerably for proposals in our sample. It took governments and the European Parliament on average 464 days to adopt a proposal (median: 405; minimum: 2, maximum: 1666), with a standard deviation of 345. Between 2001 and 2007, the median duration for the adoption of EU legislative proposals varies between 303 and 430 days (Häge, 2011, pp. 472-473), indicating that the INTEREURO sample is broadly representative of the level of conflict that characterized legislative politics in the EU during the last decade. 5 The subsamples of the framing module (mean: 672 days; standard deviation: 364; median: 611) and of the MLG module (mean: 724 days; standard deviation: 401; median: 580) are somewhat skewed towards more conflictive policy proposals.

Concrete Challenges
Once the sample was selected, detailed fieldwork was carried out on each of the 125 legislative proposals. This revealed a surprising level of heterogeneity, which poses a number of challenges for empirical and theoretical inference. First, it turned out that not all proposals are equivalent entities and that the differences identified contribute to the skewed distribution of lobbying activity that is directed towards the Commission. Second, it emerged that a significant number of proposals are in some way directly connected to each other, which questions the extent to which one particular legislative proposal can be taken to be a fully independent unit of analysis. In respect to the effect of proposal equivalence on lobbying activity, we find that 23 proposals within our sample are of a kind that attracts little or no spontaneous lobbying activity (18 per cent). Moreover, this subset excludes a further 15 (12 per cent) proposals which although of a type that may attract spontaneous lobbying, were insufficiently salient to prompt such action. Our fieldwork, hence, indicated a substantial number of cases that are intrinsically off-limits to lobbyists. To clarify this, the following categorization has been made: cases where access to decision making is highly restricted (seven cases); those where decisions are taken outside the normal legislative arena (eight); those that are administrative (six); as well as seemingly ad hoc types that defy even the broadest categorization (two).
Taking the first category, we find that access to the decision-making process is formally constrained and highly regulated in cases of both anti-dumping and social policy. Anti-dumping allegations, although initiated by an interest group claiming economic injury, if taken up, immediately become quasi-legal investigations. In social policymaking, the role of institutionally embedded corporate interests is formally defined in the Treaties. Here the lack of access for other interests was summed up by a Commission official who stated, 'we are not able to accommodate other interests, which can be a problem, but even we ourselves are not always privy to the evolving nature of negotiations'. 6 The second category includes instances in which decisions are taken outside of the EU's inter-institutional setting. These include proposals for restrictive measures (sanctions in respect of Guinea and President Lukashenko), which turn out to be verbatim translations by the Commission of a Council common position; as well as inter-governmental agreements including the Financial Stabilisation Package (2010). The third category is essentially administrative housekeeping, but unlike the codifications that were removed as part of the sampling process, the characteristics of these cases were not immediately obvious. For example, the sample includes an alignment proposal, the purpose of which is simply to distinguish between delegated and implemented acts. Finally, to illustrate the small set of proposal that although problematic to categorize nevertheless attract no lobbying, our sample includes a €40 million pilot fisheries policy that was described by the Commission official responsible as 'peanuts' and although it 'looks like a legislative doc' is 'actually a facilitation doc to stimulate dialogue. Nobody feels threatened', and hence nobody would lobby. 7 The extent to which proposals are independent from each other, and indeed how independence is defined is potentially very complex. A strict application of the principle of independence might find the population wanting, as it is likely that the occurrence of some proposals will affect the probability of a second making it onto the agenda. However, it would be naive to think that policymaking is conducted in a vacuum, and as such it would be surprising to observe a set of policy proposals that were not in any way affected by the outcome of others. In the context of understanding lobbying behaviour, we adopted a research strategy that takes cases where the relationship between proposals is explicit and obvious as one unit of observation and analysis.
We have 20 proposals (16 per cent of the sample) that are strongly connected to one or more other proposals within this subset, but the nature of the relationships varies. To give a sense of this, we have two instances of so called 'quick-fix' proposals that were initiated in response to the financial crisis, and were subsequently amended. We have several instances of consecutive amendments to an existing proposal, for example, changes to the financial management of the Regional Development, Social and Cohesion Funds. There are two proposals to set up separate financial regulatory authorities (banking and pensions) in which the decision-making process, and hence lobbying, was delegated to one of the ultimately near identical proposals (banking). In this instance, we also find evidence of a third such proposal (securities) outside our sample that similarly delegates authority to the banking proposal. Similarly, we observed a more or less symbiotic relationship between the revised directive on 'waste electrical and electronic equipment' and the parallel directive restricting 'the use of certain hazardous substances in electrical and electronic equipment'. In the dense policy field of Common Agricultural Policy (CAP) reform linkages abound, with a further four proposals in some way dependent on progress within this subset. Further demonstrating the interconnectivity of proposals, we observe three proposals in which specific issues were prevalent at the time of their initiation, but were later subsumed in other proposals, within and outside our sample.

Conclusion
Our sample is, to the best of our knowledge, a highly encompassing and, in terms of variation on theoretically important dimensions, a very balanced set of cases allowing us to study diverse aspects of interest group mobilization, interest group strategies and interest group influence. Most quantitative studies of interest groups in the EU sample groups rather than issues (Beyers, 2004;Eising, 2004;Dür and Mateo, 2012) and are thus limited in drawing conclusions about the contextualized nature of lobbying and the influence of societal interests. The approach we adopted comes with several key advantages. First, starting with legislative acts makes sense in the EU context, as it is difficult to identify the population of all national-and European-level groups that may be active in EU politics. Although there are some data sets of groups active at the EUlevel (Wonka et al, 2010), data sets on the population of national actors interested in EU politics are difficult to assemble. Second, even if we had been able to establish a sampling frame that is representative of the target population of interest groups, getting information on the issues that these groups are active on would have been challenging. A pilot project that aimed at coding issues mentioned on the groups' Webpages showed that many Webpages contain little information on the specific issues the groups are active on. Rather, groups mention quite vague topics on their Webpages, such as reducing unfair tax competition or combating climate change. We could have called a sample of groups and directly asked them to identify a recent issue on which they were active on, but this would have been a time consuming undertaking. Moreover, one advantage of this approach was that our interview projects could be enriched by considerable prior information on the specific cases. Third, compared with non-random policy-centred approaches such as the one used in the Decision making in the European Union project, where researchers analysed primarily highly conflictive proposals , we are quite confident that we do not overestimate conflict and mobilization. In fact, our sample includes 19 proposals for which we could not find evidence of public visibility. This allows us to generalize our findings to a broader population of issues, which makes our approach particularly distinct from qualitative studies that often focus on highly salient issues.
Nevertheless, our approach is no panacea to all research design problems. As with all other approaches, it comes with important trade-offs. First, we only consider proposals that made it onto the political agenda and thus cannot say much about the agenda-setting stage (Bachrach and Baratz, 1962). Although we cannot exclude the possibility that the population of groups active in the agenda-setting stage is different from the population of groups that we identify, we do not have any indication for that, either. Second, our approach turned out to be quite resource-intensive. In fact, sometimes the border between the data-collection stage and the sampling stage became blurred, as only interviews with Commission experts made us aware that certain cases actually were not decided independently of each other and thus were no independent cases in our sample. Finally, looking at proposals put forward during the relatively short time period of 3 years may mean that short-term fluctuations in the Commission's agenda have a large impact on the composition of the sample. In our sample, for example, we find several legislative acts related to the financial crisis of 2007-2008, which clearly is not typical of the EU's long-term legislative agenda. Having sampled proposals from two different colleges (Barroso I and II) and over an extended period of time, however, we are still confident that our sample is pretty representative of the Commission's agenda. In addition, given our knowledge about the individual cases, we will be able to take these idiosyncrasies into account in the analysis of our data. Overall, therefore, we are confident that our sampling strategy worked well and that it allows us to make inferences to a broader population of proposals on the EU's legislative agenda.
In sum, policy-centred sampling is feasible, but demanding in terms of resources and time as it requires intensive collaboration among like-minded scholars. Importantly, it also implies that when inferences are made to the wider population, consideration can be given to the full extent of heterogeneity within the sample.
When selecting individual proposals for analysis, an assessment of independence needs to be made, and the implications considered. As a consequence, interviews with policy elites are an important prerequisite and crucial complement to the sampling procedure itself. Similarly, the differential effect of proposal type is an important factor, alongside issue salience, in accounting for the skewed distribution of lobbying activity that is directed towards the European institutions.