Shadows as leaders? The amendment success of shadow rapporteurs in the European Parliament

In light of secluded decision-making and early agreements, a binding mandate for the European Parliament’s negotiation team is essential to prevent agency loss in trilogue negotiations. In this article, I investigate the influence of the often-overlooked shadow rapporteurs on this mandate. Shadow rapporteurs are their party group’s representatives and act as checks on the rapporteur. Drawing on novel insights from network analysis, I expect shadow rapporteurs and their stance on EU integration to affect the success of amendments they are sponsoring. I draw on a novel dataset of 1524 committee amendments and employ three-level multinomial logistic regression to test these expectations. I find shadow rapporteurs to be influential policy leaders who successfully shape the committee report and, therefore, mitigate the risk of agency loss in potential trilogues. Shadow rapporteurs can successfully check the rapporteur and thereby influence the content of EU legislation.


Introduction
In the European Union (EU), early agreements have become the most common form of concluding the legislative process. Therefore, crucial decisions are taken in interinstitutional negotiations, so-called trilogues, rather than in the plenary of the European Parliament (EP). This practice has motivated scholars to criticise the lacking democratic accountability and strong information asymmetries between members of the European Parliament (MEPs) (e.g. Yordanova, 2013: 107). One way to alleviate these information asymmetries consists of drafting committee reports. As of December 2012, they are a necessary precondition to commence trilogues (EP, 2014: 23) and provide the mandate for the EP's negotiation team (EP, 2015: Rule 73.2). Thus, understanding who is shaping committee reports and how is vital for the EP to avoid agency loss in trilogue negotiations. In line with the power shift resulting from early agreements, scholars have argued that research should move its focus from the plenary to the committee (Burns, 2013). Despite amendments being the most valuable instrument for individual MEPs to influence legislation (Hurka, 2013: 274), only a few scholars have analysed amendments at the committee level. Häge and Ringe (2019) identify rapporteurs and shadow rapporteurs (shadows) as the policy leaders in the EP. Furthermore, Ripoll Servent and Panning (2019) identify shadows meetings, comprising the rapporteur and the shadows, as the new locus of intra-institutional conflict. While rapporteurs have received considerable scholarly attention throughout the past few years, the impact of shadows on the legislative process is still unclear (Finke and Han, 2014: 149). They have a dual function as party group representatives in the legislative process and essential checks on the rapporteur (Häge and Ringe, 2020: 710). However, as members of the EP negotiation team, they are also powerful relais actors who occasionally bypass the rapporteur to strike a deal with the Council of the EU (Council) (Judge and Earnshaw, 2011). In a first attempt to systematically measure the influence of shadows on the legislative output, I seek to answer if shadows amend the committee report more successfully than ordinary MEPs. Moreover, I examine if some shadows are more successful than others.
Drawing on the perceived-preference cohesion theory (Ringe, 2009), I hypothesise that shadows are more successful with their amendments than ordinary legislators. However, MEPs from ideologically extreme party groups are primarily interested in signalling opposition towards moderate policies to their constituencies rather than constructively shaping policies (Finke, 2015: 753). Thus, I hypothesise that Eurosceptic shadows are less successful with their amendments than other shadows. The empirical focus is on the 19 ordinary legislative procedure files adopted by the Committee on Employment and Social Affairs (henceforth: employment committee) between December 2012 and July 2019. I introduce a novel dataset linking data on MEPs and 3844 amendments retrieved from the EP web page with data from the Chapel Hill Expert Survey (CHES) (Bakker et al., 2020), the ParlGov database (Döring and Manow, 2020), a software package (Gorka, 2018) and information provided by Hurka and Haag (2020). The three-level multinomial logistic regressions show that shadows successfully shape committee reports and thus the mandate for potential trilogues. Besides acting as party group representatives, they can effectively control the activities of the rapporteur and, by extension, the content of EU legislation.
influence is not contingent on holding an office, although he does not clarify who the other members of the influential sub-group are (Marshall, 2010). Jensen and Winzen (2012) identify shadows, beside the rapporteur, as the most central, active, and popular actors in the negotiation process and within their party group. Similarly, Häge and Ringe (2019) argue that rapporteurs and shadows are the policy leaders on a given proposal. Shadows occasionally bypass the rapporteur to strike a deal with the Council (Judge and Earnshaw, 2011). While current research provides insights into the opportunity structures of shadows and anecdotal evidence of their impact on EU legislation, the extent to which shadows shape the policy output is still unclear. This research gap calls for a systematic investigation of shadows' amendment success to examine if they can transfer their centrality in the negotiation process into actual influence.
According to Ringe's (2009) theory of perceived preference coherence, party group lines are not constant but change from proposal to proposal. The party group's shadow (or the rapporteur) is the key player that shapes the party group line. Other party group members from the responsible committee engage with the shadow (or the rapporteur) to draw up the initial party group line only if they are interested or the file is very salient (Ringe, 2009: 59). At this stage, the key players already consider the preferences of other party groups' shadows and the rapporteur (Ringe, 2009: 59). After a party group's key players have agreed on a position, they will discuss it in the party group's committee working group (Ringe, 2009: 59-60). Subsequently, the party working group discusses the matter allowing party group members from other committees to participate in the deliberation (Ringe, 2009: 60). If the working groups establish a party group line, uninvolved party group members will vote according to it because they perceive that their preferences are coherent with the party group's experts involved in the deliberation (Ringe, 2009: 62). Shadows are their party group's representative (Häge and Ringe, 2020: 710) and the most central actor in the negotiation process (Jensen and Winzen, 2012: 137). Since acquiring the same level of expertise as the shadow or rapporteur of one's party group is costly (Costello and Thomson, 2010: 222-223), the position of the party group's key players is rarely contested (Ringe, 2009: 64). According to Häge and Ringe (2020), MEPs shadow the reports they are interested in, which makes the appointment of shadows mostly a bottom-up, self-selection procedure. Party group coordinators are only involved if several MEPs indicate their interest (Häge and Ringe, 2020: 721). Thus, shadows are self-selected experts who are highly interested in the respective policy proposal. Hence, they are 'more likely to invest the time required to make well-informed policy choices' (Jensen and Winzen, 2012: 124).
The recent literature sees rapporteurs as influential but constrained actors (e.g. Costello and Thomson, 2010;Judge and Earnshaw, 2011;Thierse, 2019). With the draft report, the rapporteur sets the agenda for further deliberations on the EP's position regarding a policy (Marshall, 2010: 558). However, the open amendment rule is said to have at least the same impact on the outcome as the rapporteur's draft report (Marshall, 2010: 572). Rapporteurs are the dominant shapers of the committee report, but they rather compromise on certain parts than risk endangering a majority for their report (Hurka, 2013: 292). Since no single party group has a majority in the EP, rapporteurs must build coalitions. These coalitions are primarily motivated by policy-seeking, meaning that parties will favour the coalition producing the policy outcome most in line with their ideological and policy preferences (Finke, 2012: 489). Therefore, the rapporteur needs to anticipate and incorporate the positions of other party groups to secure a majority for the report (Marshall, 2010: 560). However, rapporteurs conventionally draft their report isolated from other MEPs (Marshall, 2010: 558). To successfully anticipate the preferences of other relevant actors, the rapporteur relies on expert information, which the shadows of the other party groups can facilitate (Finke, 2012: 488). Thus, shadows are crucial actors who can help the rapporteur to anticipate the committee's backing by acting as a relais between them (Delreux and Laloux, 2018: 313). Together, rapporteurs and shadows form the link between the different party groups (Jensen and Winzen, 2012: 134) and try to find consensus within the committee (Farrell andHéritier, 2004: 1196).
According to Ripoll Servent and Panning (2019), the inter-party conflict has shifted from the committee to a more informal setting, the so-called shadows meeting. These meetings between the rapporteur and the shadows are crucial to coordinate positions and negotiate compromises to find a majority for the report (Ripoll Servent and Panning, 2019). After the rapporteur presents the amendments introduced by the draft report, the shadows discuss them and possible further amendments (Ripoll Servent and Panning, 2019: 307). Even though the shadows negotiate on behalf of their party group, it is not sure that compromises agreed upon with the rapporteur, or other shadows, are ultimately backed by their party group (Neuhold and Settembri, 2009: 142). Amendments tabled without the shadow's support may occur if national delegations, or single MEPs, defect from the party group line (Ringe, 2009: 60), the shadow's amendments deviate too much from the party group's ideals or no party group line can be established. In principle, I expect amendments sponsored by a shadow to be more successful since shadows possess a profound knowledge of the policy and the preferences of other party groups. Their amendments should accordingly be less likely to incorporate the preferences of just a single MEP, party group, or national delegation and be more likely to contain a position capable of gaining a majority. Amendments solely sponsored by ordinary MEPs should conversely be less successful since ordinary MEPs do not possess the same level of policy expertise and knowledge of other actors' preferences. These amendments should therefore be less likely to convince a majority of legislators.

H1:
If an amendment is (co-)sponsored by a shadow, then it is more likely to be adopted.
Thus far, I assumed that MEPs are motivated by policy-seeking, meaning their chief motivation is to align the policy output with their preferences (Finke, 2012: 489). This assumption entails that amendments only fail due to incomplete information (Finke, 2012: 507). According to Tsebelis et al. (2001), there are two reasons why legislators table amendments. First, legislators think their amendment will be adopted. Second, legislators want to make a statement regardless of the adoption of their amendment (Tsebelis et al., 2001: 575-576). Hence, policy output is not the primary concern of all MEPs (Finke, 2015). Some MEPs are motivated by the signal an amendment can send to their constituencies or other stakeholders, such as their national party or interest groups (Finke, 2015: 753). Especially party groups with extreme ideologies send signals to their constituencies by using radical amendments combined with roll call votes to distance themselves from the mainstream party groups (Finke, 2015: 751). Curry (2019) assumes that legislators who perceive another legislator as ideologically extreme also perceive the legislator's amendments as ideologically extreme. Therefore, legislatures should be less likely to adopt the amendments of more ideologically extreme legislators (Curry, 2019: 219). In theory, the committee is thus less likely to adopt amendments of ideologically extreme MEPs because their amendments are preference outliers or are perceived as such.
The most salient ideological dimensions in the EU are the left-right and the EU integration dimension (e.g. Hix et al., 2007). Previous studies show that the left-right dimension accounts for most of the plenary roll call votes (Hix et al., 2007). However, the distance of MEPs on the left-right dimension to the committee median has no significant effect on their amendment success (Hurka, 2013: 285). Initial findings suggest that the EU integration dimension has surpassed the left-right dimension as the primary indicator for plenary roll call votes with the eighth legislative term (Hix et al., 2019). Brack (2018) also shows that the approval rate of plenary amendments is the lowest for MEPs from Eurosceptic party groups. She argues that the Europe of Nations and Freedom Group (ENF) and non-attached MEPs primarily send signals to their constituencies and are not interested in influencing policies (Brack, 2018: 401). MEPs' positions on EU integration are also relevant to understanding their participation in policy-making networks (Häge and Ringe, 2019). Sometimes, the shadows of mainstream party groups negotiate in an informal setting because they cannot exclude far-right Eurosceptics from official shadows meetings (Häge and Ringe, 2019: 224). However, the Eurosceptic MEPs mostly choose to exclude themselves from decision-making processes (Häge and Ringe, 2019: 223-224). Even if Eurosceptic shadows participate, they are more likely to engage in signalling. Thus, their amendments are not policy-oriented or are perceived as such, which causes them to be less successful.

H2:
The more Eurosceptic a shadow who (co-)sponsors an amendment is, the less likely the amendment will be adopted.

Research design
To test the two hypotheses, I focus on the last part of the seventh and the entire eighth legislative term of the EP (December 2012 to July 2019). Before December 2012, committee reports were often obsolete when trilogues led to an early agreement with the Council (Yordanova, 2013: 105). Since then, committee reports have been a necessary precondition for negotiations with the Council (EP, 2014: 23). Thus, I can only rule out that trilogues shaped the content of a committee report after this date. I focus on the 19 concluded legislative proposals adopted by the employment committee and subject to the ordinary legislative procedure. 1 The employment committee was the sole committee that published the voting results on all amendments from all legislative proposals within the chosen time frame. I assume that it is a deliberate decision to publish these voting results. Hence, focusing on the employment committee enables to rule out the influence of unobserved factors that may determine a committee's decision to publish the amendments' voting results.
The employment committee is among the EP's more powerful (Yordanova, 2013: 39) and prestigious committees (Whitaker, 2019: 173). It comprised between 54 and 61 members between 2012 and 2019. It is mainly interest-driven and produces mostly distributive policy output (Yordanova, 2013: 39). The distributive output also explains why many MEPs want to join the employment committee. They perceive the committee as important to their voters (Whitaker, 2019: 173). Especially MEPs with trade union ties are likely to be part of the employment committee (Yordanova, 2013: 52). The employment committee's voting outcomes are usually less consensual and tighter than in other committees (Neuhold and Settembri, 2009: 139). Its reports, however, are significantly more successful in plenary than those from other committees, except those from the Committee on Industry, Research and Energy (Yordanova, 2013: 104). In terms of party group composition, the committee essentially represents the plenary. MEPs from the Greens/European Free Alliance (Greens) and the Europe of Freedom and Direct Democracy group (EFDD) are slightly underrepresented at the end of the seventh term, while the European Conservatives and Reformists (ECR) are slightly overrepresented. A grand coalition between the European People's Party (EPP) and the Progressive Alliance of Socialists and Democrats (S&D) constitutes a majority throughout the entire timeframe. Neither a left-wing coalition (S&D, Greens, and the European United Left/Nordic Green Left (GUE/NGL)) nor a centre-right coalition (EPP, ECR, and the Alliance of Liberals and Democrats for Europe (ALDE)) has a majority. The left-wing coalition could gain a majority by including the ALDE group during the seventh and the beginning of the eighth term. Thus, a (realistic) coalition supported by only one of the two large party groups always needs to include the ALDE group.
As the aim is to investigate the committee's position before opening negotiations with the Council, the empirical focus is on the first reading. Therefore, I exclude amendments from reports at second reading or conciliation. Since non-attached members do not form a party group and cannot appoint a shadow, I exclude their amendments from the analysis. Furthermore, I exclude draft report, compromise, withdrawn, inadmissible and not voted upon amendments. 2 For the analysis, I rely on data provided by the CHES dataset (Bakker et al., 2020), the ParlGov database (Döring and Manow, 2020), the R package 'europarl' (Gorka, 2018), Hurka and Haag (2020) and data I collected. I use an amendment sponsor's national party positioning on the EU integration dimension and the left-right dimension from the 2010 and 2014 expert surveys contained in the CHES trend file (Bakker et al., 2020). The 'EU_position' variable indicates the party leadership's position on European integration from one, strongly opposed, to seven, strongly in favour (CHES, 2020: 19). The 'LRGEN' variable indicates the party's ideological position from zero, extreme left, to ten, extreme right (CHES, 2020: 22). The 2010 survey lacks estimates for parties from Luxembourg, Cyprus, and Malta. I impute the missing estimates with data from 2014.
If national party estimates are missing due to party splits, I impute them with the estimate of the pre-existing party (see Baller, 2017: 475).
I use the ParlGov database stable version from 2020 (Döring and Manow, 2020) to establish if an amendment sponsor's national party was part of the government during a committee vote. I rely on the 'cabinet party' variable, indicating if a party was part of the national cabinet. It excludes so-called caretakers, which are cabinets with a limited legislative mandate lasting less than three months (ParlGov, 2020). The database includes all parties that acquired at least 1% or two seats in the last election (ParlGov, 2020). If an MEP's national party is not part of the ParlGov database, I consulted the web page of the respective party and imputed the variable accordingly. I use Hurka and Haag's (2020: 94-95) variable word entropy to indicate the linguistic complexity of Commission proposals. I retrieved all other data from the EP website and the EP's legislative observatory, 3 where it is freely accessible. The R package 'europarl' (Gorka, 2018) already includes some basic information on MEPs, which enabled the web-scraping of more data on individual MEPs from the EP webpage. To retrieve the relevant amendments, (draft) reports and minutes of the employment committee meetings, I web-scraped the ID numbers of the files from the legislative observatory and the EP webpage. The ID numbers were then used to create URL addresses and automate the downloading process using R.
Following Hurka (2013), the dependent variable comprises three nominal categories: full success, partial success and failure of an amendment. The category full success indicates that the employment committee voted in favour of an amendment. The committee secretariat compiles voting lists that determine the order in which committee members vote on amendments to speed up the voting procedure (Hurka, 2013: 279). The committee often votes on the rapporteur's compromise amendments first. If adopted, all subsequent amendments tabled to the same matter fall, meaning that the committee does not vote on these amendments as the secretariat deemed their content included or in conflict with the already adopted amendment. Thus, if an amendment falls, the wording or intent of the amendment can still be part of an adopted (compromise) amendment (Hurka, 2013: 280). Therefore, an amendment can be partially successful, even though it was not adopted. I evaluated 670 randomly drawn, fallen amendments checking for partial success by carefully comparing each amendment with the final report. If the wording or intent of the final report depicted at least one meaningful change, addition or deletion introduced by an amendment, I coded it as partially successful. The category of failed amendments comprises rejected and fallen amendments that are not partially successful.
The independent variable of interest for the first hypothesis is binary and indicates if an amendment was (co-)sponsored by a shadow. It takes the value of one if at least one of the sponsors is a shadow and the value of zero if none of the sponsors is a shadow for the legislative proposal. However, the variable cannot indicate if a shadow drafted an amendment. The EP rules do not specify that the author of an amendment must be the first person listed in the documents (Baller, 2017: 474). Shadows are likely to be the ones drafting an amendment as they are their party groups' responsible actors. However, there are also groups of co-sponsors that do not include a shadow. Several MEPs may also author an amendment together in an equally joint effort. Hence, the independent variable indicates if the group of co-sponsors includes a shadow.
The independent variable of interest for the second hypothesis is a numeric variable indicating the group of sponsors' average ideological position on the EU integration dimension. I use the average of all sponsors, as it is impossible to differentiate the authorship of an amendment. I rescaled the initial EU integration variable from the CHES dataset to facilitate a more intuitive comparison of effect sizes with the left-right variable (see Obholzer et al., 2019: 249). The variable now ranges from one, strongly opposing, to ten, strongly favouring EU integration. The control variables account for traits of the proposal, the amendment and the sponsors. 4 The hypotheses aim to explain two different relationships. First, comparing shadows with other MEPs and second, comparing shadows with other shadows. Thus, I conduct separate regression analyses. In both instances, I employ three-level multinomial logistic regression, as the potential efficiency loss of a nominal model is preferable to a potential bias (Hurka, 2013: 283). One or more MEPs from the same or different party groups may jointly table amendments to a specific proposal. To account for the possibility of sponsors being from different party groups, the party group variable clusters the amendment into each unique coalition of party groups. Sponsors are from the same party group for most amendments. The first level is the amendment, the second level is the party group or a coalition of party groups, and the third level is the proposal. 5 Analysis Figure 1 shows that most rapporteurs are from the EPP or the S&D. The EPP, the S&D, and the Greens always appoint a shadow if the rapporteur is not from their party group. The only party groups without a rapporteurship are the party groups representing the extreme ends on the left and right, the GUE/NGL and the ENF. Like the party groups of the centre, the GUE/NGL appointed a shadow in most cases. The EFDD and the ENF appointed shadows for less than half the proposals studied. However, the ENF only formed in the eighth term and nominated a shadow for eight of the nine proposals tabled during its existence.
There are 1524 amendments included in the analysis. Altogether 395 amendments were fully and 385 partially successful, while 744 amendments failed. Furthermore, 63.8% of the amendments were co-sponsored by a shadow. Figure 2 illustrates that the two largest groups also tabled the most amendments, whereas the extreme right ENF tabled the fewest ones. Shadows are responsible for most amendments across all party groups except the EPP and the S&D. This exception is not surprising as these groups appoint the rapporteur in most cases and therefore do not designate a shadow. Large party groups also have more MEPs available to amend a proposal. Small party groups do not have the same capacity. Thus, they are more reliant on their shadow to propose amendments. Especially in the Green and the GUE/NGL group, the ratio of amendments sponsored by the shadow is very high. The employment committee neither fully nor partially adopted an amendment sponsored by the ENF group, irrespective of the sponsor being a shadow or an ordinary MEP. This corroborates previous findings of a cordon  sanitaire against strong Eurosceptic party groups, which also encompasses the systematic rejection of amendments (Brack, 2018: 401;Ripoll Servent and Panning, 2021: 87). 6 In line with the first hypothesis, I find that amendments sponsored by a shadow are more successful than amendments solely sponsored by ordinary MEPs. Table 1 displays the results of two three-level multinomial logistic regressions. The first model comprises all amendments, whereas the second focuses on amendments with a single sponsor. All statements regarding significance refer to 95% confidence intervals. The coefficient of the independent variable is positive and significant when comparing fully successful with failed amendments. Thus, the employment committee is more likely to adopt amendments sponsored by shadows fully. The coefficient of the independent variable is also positive and significant when comparing partially successful with failed amendments. If amendments fall due to another (compromise) amendment, the ones sponsored by shadows are more likely to be partially successful. Hence, their wording or intent is more likely to be included in the committee report than the wording or intent of amendments sponsored by ordinary MEPs. The independent variable remains positive and significant for comparing fully and partially successful amendments with failed ones when focussing on single-sponsored amendments. Thus, the results corroborate the first hypothesis.
I calculate average predicted probabilities at the observed values (henceforth: predicted probabilities) to analyse effect sizes. Unlike predicted probabilities at the mean or other specified values, they allow inferences about the average effect in the study sample rather than just a single case (Hanmer and Ozan Kalkan, 2013). Figure 3 displays predicted probabilities of full amendment success with 95% confidence intervals. The predicted probability of an amendment sponsored by an ordinary MEP being fully successful is 23.3% and rises to 28.8% if a shadow sponsors it. The probability of an amendment being fully successful thus increases by 5.5 percentage points if at least one of the sponsors is a shadow. To assess whether this effect occurs in all party groups, I disaggregated the predicted probabilities by party group. Amendments sponsored by ALDE members show the highest predicted probability of full success, followed by MEPs from the Greens, the S&D and the EPP. This is true for both ordinary members and shadows. While the shadows of the Greens are slightly more successful than the ones from the S&D, it is the other way around for ordinary members. The predicted probability of full success for an amendment sponsored by an ALDE member is 33.1%. The predicted probability increases to 39.2% if the ALDE shadow sponsors the amendment. The predicted probabilities of full amendment success for ordinary members of the ECR, the EFDD and the GUE/NGL range between 10.5% and 16.7%. For shadows, they range from 16.1% to 23.0%. Amendments sponsored by ENF members have the lowest predicted probability of full success. It runs from 2.1% for amendments sponsored by ordinary members to 3.9% for amendments sponsored by the ENF shadow. With 7.6 percentage points, the EFDD group shows the highest increase in predicted probabilities of full success between amendments tabled without a shadow's support and those sponsored by a shadow. The ENF group has the smallest increase in predicted probabilities with 1.8 percentage points. All remaining party groups show an increase in predicted probabilities between 4.6 and 7.2 percentage points.
Regarding partial amendment success, the predicted probability of an amendment sponsored by an ordinary MEP is 20.4% and rises to 29.6% if a shadow sponsors it. The probability of an amendment being partially successful thus increases by 9.2 percentage points if at least one of the sponsors is a shadow. Hence, the difference in the predicted probabilities is higher for partial success than for full success. The Online appendix displays the predicted probabilities of partial amendment success disaggregated by party group. Amendments sponsored by S&D members show the highest predicted probability of partial success, followed by those sponsored by EPP and ALDE group   Notes: * * * p < 0.001; * * p < 0.01; * p < 0.05; standard errors in parentheses; reference level type of law: regulation.
members. This is true for both ordinary members and shadows. The predicted probability for partial success of an amendment sponsored by an S&D member is 25.3%. The predicted probability increases to 35.4% if the S&D shadow sponsors the amendment. The remaining party groups have predicted probabilities ranging from 11.9% to 22.3% for ordinary members and 20.5% to 31.3% for shadows. Amendments sponsored by ENF members also have the lowest predicted probability of partial success. It ranges from 1.9% for amendments sponsored by ordinary members to 4.0% for amendments sponsored by the ENF shadow. With 10.1 percentage points, the S&D group shows the highest increase in the predicted probabilities of partial success when comparing amendments sponsored by ordinary MEPs and those sponsored by shadows. With 2.1 percentage points, the ENF has the smallest increase in predicted probabilities. The remaining party groups show an increase in predicted probabilities between 7.6 and 9.9 percentage points. The three-level multinomial regression shows that shadows have a higher amendment success than ordinary MEPs at the committee stage. This result applies to their amendments' full and partial success and thus corroborates the first hypothesis. The analysis of predicted probabilities shows that the effect exists in all party groups, albeit differing in size. The descriptive statistics show that no amendment sponsored by the ENF was fully or partially adopted. Thus, there is no difference between the amendment success of the ENF shadow and ordinary party group members. Furthermore, the predicted probabilities show that the three party groups of the super grand coalition, EPP, S&D and ALDE, are also the most successful. This result complements the previous finding that most winning coalitions of plenary roll call votes on employment and social policies comprise these party groups (Hix and Høyland, 2013: 179). The Greens are very successful as well. Comparing the predicted probabilities between ordinary members and shadows, the EFDD shows the highest increase for full success, while the S&D shows the highest increase for partial success. The fact that the EFDD only appointed members of the Italian Movimento 5 Stelle as shadows can partially explain why shadows are considerably more successful than other EFDD members. While centrist party groups usually perceive MEPs from the EFDD as Eurosceptic position-takers, they are willing to cooperate with MEPs from the Movimento 5 Stelle if majorities are tight (Ripoll Servent and Panning, 2021). Thus, their amendments are more likely to be accepted than the ones of their fellow party group members. Overall, the additional value of shadows is observed across all party groups and is even more pronounced for partial than full success. The chances of full success rise if a shadow sponsors the amendment. Even if the committee does not adopt an amendment, those sponsored by shadows are more likely to have their wording or intent, at least partially, represented in the final report. Therefore, I can confirm the first hypothesis except for the case of the ENF.
The analysis also shows that less Eurosceptic MEPs are more successful when sponsoring amendments. However, this effect does not exist for single-sponsored amendments. The distance to the committee median on the left-right dimension has a significant negative impact, showing that groups of ideologically more extreme sponsors are less successful. For single-sponsored amendments, the effect only persists for full success. The sponsors' party group size better explains the partial success of single-sponsored amendments.
In line with the second hypothesis, I find that the amendments of more Eurosceptic shadows are less likely to be adopted than those of other shadows. Table 2 displays the results of the three-level multinomial logistic regression. While the first model comprises all amendments (co-)sponsored by a shadow, the second model focuses on singlesponsored amendments by a shadow. The coefficient for the EU integration variable is positive and significant when comparing fully successful with failed amendments. This result holds for both models. Thus, the employment committee is less likely to adopt the amendments of Eurosceptic shadows. The coefficient for the EU integration variable is also positive and significant for comparing partially successful amendments to failed ones in the first model. Thus, if amendments fall due to other (compromise) amendments, the amendments of shadows favouring further EU integration are more likely to be partially successful. The effect is still positive yet insignificant in the model focussing on single-sponsored shadow amendments. Therefore, I can confirm the second hypothesis only for full success. Party group size positively and significantly affects partial success in both models. It also positively and significantly affects full success in the second model. Hence, the number of potential votes a shadow can facilitate is another influential factor when comparing the amendment success of shadows. The distance to the committee median on the left-right dimension has no significant effect. Thus, the position on EU integration matters more than the left-right axis when comparing shadows with each other. The positive and significant effect of competing rapporteur amendments Table 2. Three-level multinomial logistic regression comparing shadows' positions on EU integration.
All amendments (co-) sponsored by shadows Single-sponsored amendments by shadows
on partial amendment success indicates that the wording or intent of the amendment is already partially included in the text of the rapporteur or in another (compromise) amendment that replaced the text of the rapporteur. Thus, rapporteurs and shadows seem interested in establishing a broad coalition through compromises to enhance the EP's bargaining success in negotiations with the Council. In addition, I calculated predicted probabilities at the observed values to interpret the effect sizes of shadows' positions towards EU integration. The predicted probabilities are only valid for the range of observed values on the EU integration dimension between 1.5 and 9.5. Figure 4 displays the predicted probabilities for full amendment success. Within the observed range, the predicted probabilities for full amendment success increase by 31.9 percentage points from 14.2% to 46.1%. Thus, shadows with a more integrationist standpoint have a higher chance that the committee will adopt their amendments.

Conclusion
The analysis shows that amendments (co-)sponsored by shadows are more likely to succeed. This effect exists for full and partial success across most party groups. Thus, I can confirm the first hypothesis except for the case of the ENF. Moreover, the analysis shows that shadows with a pro-integrationist stance are more likely to have their amendments fully adopted, partially corroborating the second hypothesis. Shadows sponsor most amendments and are responsible for the majority of successful ones across all party groups except the EPP. Thus, I conclude that shadows are crucial actors for their party group and in shaping the committee report.
Throughout this article, I analysed the amendment success of shadows to learn more about their influence. As success can also be caused by luck, it does not necessarily equate to influence (Dür et al., 2015: 954) but it is a good proxy (Marshall, 2010: 567). Salient proposals are likely shadowed more extensively (Rasmussen andReh, 2013: 1019). Nevertheless, I consider all proposals equally salient as most parties usually appoint a shadow. All amendments are considered equally important due to the absence of a credible estimate of their relative importance (see Häge and Kaeding, 2007: 350). I do not classify amendments by type since even a small change to a recital can be politically very controversial (see Mahr and Ringe, 2016: 38). Future research may consider how salient a policy is to party groups and their shadows to shed more light on eventual logrolling dynamics at this stage of the policy-making process. Hurka and Kaplaner (2020) argue that the conventional ideology dimensions might be too broad to explain policy conflict at the committee stage. They propose to focus on more policy-specific preference measures (Hurka and Kaplaner, 2020: 7). Further studies could examine the influence of special interests, such as trade unions (see Yordanova, 2013: 53), on the amendment success of shadows to account for the interest-driven nature of the employment committee. Due to the focus on the employment committee, the generalisability of the results to other committees is limited. The employment committee is the only powerful, mainly interest-driven committee with a largely distributive policy output (Yordanova, 2013: 39). As coalition patterns change across policy areas (Hix and Høyland, 2013: 179), future research should investigate if and how the amendment success of shadows is different in committees with other characteristics.
With this article, I shed more light on the impact of shadows on the committee report and the legislative mandate for trilogue negotiations. In line with Häge and Ringe (2019), I find shadows to be successful policy leaders who actively shape the committee report. Amendments sponsored by shadows are more likely to be incorporated into the committee report, although this effect differs in scope between party groups. Regarding their dual function, one can say that shadows fulfil both functions as a party group representative in the legislative process and an essential check on the rapporteur. Thus, they actively shape the mandate for trilogue negotiations and mitigate the risk of potential agency loss. Furthermore, the immense success of the most centrist and pro-EU integrationist party groups points to what has been described as cartel politics, meaning that pro-EU integrationist party groups and especially the groups of the super grand coalition (EPP, S&D and ALDE) are dominating the decision-making process (Rose and Borz, 2013). I show that this dominance of the super grand coalition and other pro-EU integrationist party groups is also present in the formation of the committee report. In terms of output legitimacy, this adds to previous concerns that the aggregation of the pro-EU integrationist party groups can lead to misrepresentation (e.g. Rose and Borz, 2013). The cause of such oversized majorities is probably the ambition to strengthen the EP's bargaining position in upcoming negotiations with the Council (Costello, 2022: 120). I show, once again, that Eurosceptic party groups often exclude themselves from the policy-making process by tabling fewer amendments and appointing fewer shadows. The employment committee systematically rejects amendments if Eurosceptic MEPs table them. Also, MEPs representing the extreme left and right are less successful than their more centrist colleagues. However, the left-right dimension loses its significance for amendments sponsored by shadows (cf. Hix et al., 2019). Instead, the variation in shadows' success across party groups is explained best by Euroscepticism and party group size. The most successful shadows are from the large centrist groups, followed by shadows from smaller groups. The least successful are the shadows of small and Eurosceptic groups. Hence, the results point again towards a cartel (Rose and Borz, 2013) that dominates the formation of the committee report. This cartel consists of the centrist super grand coalition and other non-Eurosceptic party groups. Especially in light of the increased use of early agreements and secluded decision-making in trilogues, this finding is essential, as scholars found that a united front towards the Council can enhance the EP's influence on the legislative output (e.g. Delreux and Laloux, 2018;Finke, 2012). In conclusion, shadows are crucial players who use their expertise and centrality in the policy-making process to shape the committee report and thus the legislative mandate for potential trilogue negotiations. As long as shadows act as loyal brokers of their party group, the EP's mechanisms to avoid agency loss in trilogues are well on track. 5. I display the model specification and report the results of single-level multinomial models with clustered standard errors and two-level multinomial models in the Online appendix. 6. I illustrate descriptive statistics of the independent and control variables in the Online appendix.