The Trio Presidency and the Efficiency of Council Decision‐Making: An Empirical Study

Does the trio presidency system enhance efficiency in EU legislative decision&#8208;making? Since 2007, fixed groups of three Member States have been required to set a joint 18&#8208;month agenda, with each Member State taking 6&#8208;month turns at the presidency. To date, however, there has been no empirical evidence that either confirms or refutes the notion that this system is efficient. In a study of the duration of Council decision&#8208;making on 1,927 legislative proposals for the 2000&#8211;12 period, we obtain empirical support for the hypothesis that a common agenda leads to a significant decrease in the amount of time needed to reach a first agreement on regulations, directives and decisions. In addition, we show that the requirement to pre&#8208;negotiate the agenda helps to moderate the effect of political conflict on the speed of decision&#8208;making, thereby offering support to the recent decision to proceed with the trio presidency system until at least 2030.


Introduction
The European Union (EU) is well known for being slow and inefficient in making decisions. On average, currently, the time required to get a first-reading agreement in the Council, which is still the EU's principle legislative institution, is about 15 months. This is a far from speedy performance. Häge (2011) for instance, shows that the time to reach agreements in the Council has almost tripled between 1976 and 2006, from 145 to 430 days. König (2007) reaches a similar conclusion for the 1984-98 period. Although there are many causes for this increased and lengthy process, one significant reason is the lack of continuity in decision-making, frequently attributed to a system of rotating presidencies. Specifically, Member States holding the rotating chair are often criticized for overly pursuing national priorities during their 6-month stints in office, leading to the infamous problem of presidency discontinuities that ultimately results in inefficient decision-making.
Improving the EU's ability to legislate in a timely manner was one of the key issues on the agenda of the Lisbon Treaty, concluded in 2007 and implemented in 2009. The Treaty increased the use of qualified majority voting in the Council at the expense of the burdensome requirement of unanimity. It also changed the voting system from a triple majority to an easier-to-reach double majority. 1 Finally, with respect to the discontinuity problem, it altered the rotating presidency system. 2 Under the new presidency format, fixed groups of three heterogeneous Member States are formally required to set up a joint 18-month agenda for the Council, with each Member State still taking 6-month turns at the presidency. The new system, informally referred to as the trio presidency system, took effect in January 2007 and was specifically introduced to counter presidency discontinuities in priorities and to improve decision-making efficiency altogether. 3 The main question we address in this article is simple and straightforward. Did the introduction of the trio enhance efficiency in Council decision-making? Surprisingly, to the best of our knowledge there is no contribution in the literature that provides empirical evidence either in favour of or against the efficiency of the trio presidency. An empirical study on whether the trio presidency has increased efficiency in Council decision-making is thus worthwhile. In this study, we focus on the duration of decision-making, which is perceived to be one of the two main indicators of efficiency and has been extensively studied by political economists. Fewer discontinuities in decision-making arguably increases the speed of reaching agreements. 4 In our study of the duration of Council decision-making on 1,927 legislative proposals for the 2000-12 period we obtain empirical support for our conclusion. First, we isolated the effect of the trio presidency by performing an interrupted time series (ITS) analysis on a monthly basis with the introduction of the trio as the treatment. We found that the time to reach agreement had decreased significantly since the formal requirement was introduced to coordinate a longer-term agenda. In addition, we show that this effect cannot be explained by a number of other compelling changes at the same point in time, such as the increasingly used early agreements. Second, we performed an event history analysis at the proposal level and found that the trio presidency moderates the effect of political conflict. Political conflict is often measured as the distance between ideal policies among Member States, and has been found to be an important determinant of the duration of the legislative process. Our findings show that ex ante agenda-setting negotiations among a heterogeneous mini-coalition neutralizes the effect of political conflict.
This article contributes to the literature on EU Treaty reforms and their effect on the legislative process and decision-making. It provides a first empirical test of the 18-month joint agenda format on legislative duration. In contrast to some doubts raised in the literature on the performance and merits of a trio presidency system, the study shows that it did reach its primary objective of improving efficiency in legislative decision-making. Additionally, this article contributes to the study of preferences, institutional rules and, most importantly, the interplay between them. Using data on Member States' preferences in combination with two distinct agenda-setting structures, we found that political conflict can be moderated by institutional arrangements. Thus our findings provide support for the EU's recent decision to proceed with this system until at least 2030.
The article is organized as follows. In the next section we briefly discuss the literature on decision-making duration and introduce the hypotheses the trio presidency gives rise to. Section two describes the data and research design, while section three presents the empirical results. The final section concludes and presents an avenue for future research.

Literature and Hypotheses
The literature to date frequently highlights the role of the institutional and procedural characteristics that determine the duration of the decision-making process. Schulz and König (2000), Golub (1999Golub ( , 2007, and Golub and Steunenberg (2007), for example, found that the speed of decision-making in the EU slowed down if the European Parliament is formally involved in the policy-making process. Similarly, the 2004 enlargement was shown to increase the time necessary to reach agreements (Hertz and Leuffen, 2011). In contrast, qualified majority voting, as suggested by König (2007), reduced the duration of EU decision-making.
Unlike the legislative procedure, enlargement or voting rule, the Council's agendasetting structure is an institutional feature that has not received much attention in the literature on the duration of decision-making, arguably because it has been relatively unaffected by subsequent Treaty changes and institutional reforms over the past 30 years. However, the literature pays considerable attention to the question whether rotating presidents have agenda-setting powers, and, if so, how they translate into (policy) outcomes. We briefly discuss this literature below and subsequently theorize how this system has direct consequences for the duration of legislative decision-making.
Scholars have argued that the rotating Council presidency is the main agenda-setter and serves as the motor for advancing EU policy-making. 5 Steunenberg and Dimitrova (2003), for example, assume the presidency takes the lead in EU-decision-making and presents a proposal to the Council members and the European Parliament. Napel and Widgrén (2006) argue that the presidency can choose the policy it prefers from among those between the two pivotal Member States in the Council. Similarly, König and Proksch (2006) treat the Council president as the agenda-setter once the Commission has introduced a proposal and makes the final offer to the Member States. These claims are supported by a substantial number of studies presenting evidence that rotating presidents tend to have significant agenda-setting powers (Häge, 2017;Schalk et al., 2007;Tallberg, 2003Tallberg, , 2010Thomson, 2008;Warntjen, 2008Warntjen, , 2013a and are able to obtain policy outcomes closer to their ideal preferences (Van Gruisen et al., 2017).
However, the fact that presidents are powerful players in a rotating system, which in turn distributes institutional capability equally among the 28 Member States, also has significant drawbacks. For one, Member States are in office for only 6-month terms, and many proposals often need significantly more time before they are adopted. 6 As 5 This argument is at the expense of the agenda-setting role of the EU Commission. For instance, Kreppel and Oztas (2017) show that the Commission is merely a technical agenda-setter, with no power to affect policy outcomes. König and Proksch (2006) argue that the Member States in the Council can exclude the Commission and offer compromises and trade votes. Crombez (2000) argues that the Commission does not play a legislative role at all under co-decision post-Amsterdam. 6 See Figure A1 for a kernel density plot of Council duration . Warntjen (2013b) rightly argues, a Member State may be close to striking a compromise but may eventually fail in this effort due its limited term in office. Moreover, once another Member State takes over the presidency with a different set of priorities the progress that has been made in the past is then ultimately lost. As a result, powerful chairmen can create discontinuities in EU legislative politics and thus render the decision-making process inefficient. This stands in the way of the primary objective of the trio presidency reform, which was to increase continuity in decision-making by introducing a formal requirement to set a common 18-month agenda between a group of three heterogeneous Member States.
The trio presidency has attracted significant attention among scholars since its introduction in 2007. Studies have been conducted in the past on the workings of the trio mechanism in depth (Mazzucelli, 2008;Udovic and Svetlicic, 2012), the effect of a formalized 18-month programme on continuity in the decision-making process (Raik, 2016;Warntjen, 2013b;), the opportunities the trio offers to show collective leadership (Batory and Puetter, 2013;Jensen and Nedergaard, 2014;Vieira and Lange, 2012) and the voting behaviour of trio members (Van Gruisen et al., 2017).
Our analysis complements this body of research by empirically studying the trio effect on efficiency, as measured by legislative duration. We believe a longer-term common agenda should negate short-term interests. Hence, if the trio leads to more continuity, indeed, one can expect agreements to be reached faster. We formulate the first hypothesis as follows: Hypothesis 1: The introduction of the trio presidency decreases the time to reach agreements in the Council In addition to institutional characteristics, scholars have studied the effect of Member State preferences on decision-making speed. Legislative scholars often translate these preferences into a concept referred to as political conflict. This concept captures the level of disagreement among Member States, showing that more political conflict (disagreement) leads to longer decision-making time. The literature provides considerable empirical evidence to validate this claim. For instance, Golub (1999Golub ( , 2007 and Golub and Steunenberg (2007) have looked at the presence of extreme governments as measures of preference heterogeneity. The authors include a dummy variable for the Margaret Thatcher government in their regression analyses to account for policy heterogeneity. König (2007), on the other hand, uses the distance between the extreme governments in the Council as a proxy, while Klüver and Sagarzazu (2013) measure conflict as the extreme distance between the Commission, the Council and the European Parliament. 7 Despite variations in these studies in measurement of political conflict, they all obtained similar results; namely, that political conflict decreases decision-making speed. The underlying mechanism is fairly simple: more preference heterogeneity leads to more disagreement and hence it takes longer to reach agreements.
Apart from increasing efficiency in the Council's work, another objective of the trio was to prevent conflict in a heterogeneous Council by putting trios together that differ 7 Drüner et al. (2018) focused on the size of the win-set instead and obtained similar findings. The smaller the win-set, the longer the duration of decision-making. in size, length of EU membership and geographical location. In this system, a minicoalition of heterogeneous trio members discuss and negotiate, and put compromise proposals on the Council's agenda. This is in line with Raik's (2016) argument that disagreements that would otherwise have arisen later in the process are already resolved at the preparation stage. Likewise, Vieira and Lange (2012) refer to the so-called intra-trio effect; namely, rounds of tough negotiations between the trio members to find compromises on sensitive issues.
Along this line, we argue that a system of trio presidencies moderates the effect of political conflict in the Council because compromises have already been made by a mini-coalition representing a heterogeneous Council as such. We formulate the second hypothesis as follows: Hypothesis 2: A heterogeneous trio presidency moderates the effect of political conflict on legislative duration

Data and Research Design
To study empirically the effect of the trio presidency format on legislative duration, we constructed a variable that captures decision-making speed. The dependent variable, Council duration, is measured by the number of days from the day a Commission proposal is sent to the Council to the first agreement reached in the Council. For legislation under the co-decision procedure, this decision often corresponds to the date of a political agreement or an agreement on the Council's common position, while the date of the adoption of the act is used under the consultation procedure. Note that this construction measures decision-making speed in the Council within the first reading agreement, and thus anticipates potential duration bias from second readings and conciliation committee meetings when the European Parliament is involved. We used information released by the EU Commission in PreLex from 2000 to 2012, as extracted by Häge (2011).
In order to test the two hypotheses we operated on two levels of analyses: an aggregated monthly level of Council duration and the individual proposal level, which is the original format of the data. For a similar approach, see for example, Bølstad and Cross (2016).

Aggregate level
We analyzed Hypothesis 2 by transforming the data on the duration of individual proposals to a monthly average, classified by start date, and considered them as a time series. 8 This format allowed us to apply an ITS design, similar to a regression discontinuity approach in time (for example, see Hausman and Rapson, 2017;Imbens and Lemieux, 2008). An ITS design is appropriate when, at a specific point in time, an intervention occurs, such as the enactment of a policy, treatment with a medicine or, in this case, a change in institutional rules as part of a change in a Treaty. We considered the change in agenda-setting structure as a treatment with an expected effect of increasing decision-making speed. Note that the formal requirement to set a joint agenda entered into force in January 2007. We assumed that all proposals that were transferred from Commission to Council after this date had been put on the joint agenda and have thus been treated. 9 An ITS design can lead to relatively strong causal effects if the model is appropriately specified and if certain requirements are met. The main argument for causality lies in the design of the ITS model. In particular, it is designed to control for all potential (unobserved) confounders that do not change discontinuously at the treatment threshold.
A first requirement of the ITS is that no other change occurs at the same time of the treatment. Usually a treaty change encompasses various institutional arrangements that simultaneously enter into force, making it impossible to disentangle the individual effects of the change, such as a new agenda format with an ITS approach. Fortunately, this problem does not apply to the introduction of the joint agenda requirement in the Council. Although it is part of the Lisbon Treaty, it went into effect almost 3 years prior to December 2009, and in isolation of any other institutional change, except for the accession of Bulgaria and Romania. 10 Nonetheless, adding two Member States would most likely increase duration. Any effect we may find would thus merely suggest a lower bound on the treatment effect.
A second requirement is that the treatment is exogenous. We thus needed to assume that the decision to introduce the joint agenda system was independent of legislative duration and that there was no strategic (anticipating) behaviour near the threshold of January 2007. Given that the trio system was one of the few institutional modifications that was undisputed at the Constitutional Treaty in 2005 (Raik, 2016), we can safely assume the treatment was exogenous with respect to decision-making speed.
The dependent variable is the average number of days to reach an agreement in the Council. The data are right-skewed and a Breslow test suggests that there is overdispersion; hence we employed negative binomial regression. 11 In addition, we controlled for seasonal characteristics by including a set of monthly dummies, and we controlled for the number of legislative proposals introduced in a given month. The aggregated model to test Hypothesis 2, thus, is as follows: 9 Indeed, there are proposals (217 out of 1,927) that were introduced before and negotiated and decided upon after January 2007. Potentially, they will have had the treatment to a certain degree. We considered these events were not treated, thus taking a conservative stance, given that those proposals are likely to benefit from this joint cooperation and may require less time to be agreed upon. 10 The attentive reader will notice that there is also a change in the country that holds the presidency at the time of treatment. Finland passed the presidency to Germany. Characteristics of the Germany presidency, such as its size and previous presidency experience, would make a compelling case for a decrease in duration. However, we would need to see an increase in duration after this period, which we did not observe. The mere fact that Germany held the presidency in the first semester of 2007 would thus not be sufficient. Nonetheless, for the purpose of robustness checks, we controlled for presidency size, experience and conflict in the proposal-level analysis. For an overview of a number of institutional changes and their introduction, see Table A1. 11 A Hausman test indicates the same. An alternative would be to log-transform monthly average duration and perform ordinary least square regression. Due to the limited number of observations and because a negative binomial regression is more efficient, we used that in the main analysis. The ordinary least square regression shows a similar direction and magnitude of the coefficients, but it is not statistically significant. The results are shown in Table A2. where Y t is the average number of legislative duration in days in a particular month classified by start date, β is a vector of coefficients for the variables of interest, T is the time trend variable before the intervention (centred at the time of the treatment for interpretation purposes), X t is the treatment dummy (1 for proposals introduced since January 2007 and 0 before), TX t is the interaction of time and treatment to allow the linear trend to change after the treatment, and z is a vector of coefficients for a vector of control variables c. In the Appendix, and merely as a robustness check, we repeated the exercise and allowed for a nonlinear trend in the data. To that effect, we performed nonparametric locally estimated scatterplot smoothing regressions, presented in Figure A2. Apart from inferring relatively strong causal effects, an ITS approach also allows for transparent and clear graphical illustration. The general strategy behind the ITS is that it uses the observed trajectory of Y t before the intervention X t to estimate the counterfactual after the intervention (Morgan and Winship, 2015). This approach is particularly interesting with respect to the introduction of the joint agenda due to the absence of a comparable control group that did not receive the treatment. Under this approach the estimated treatment effect δ t is then: where Y O t is the observed outcome after the intervention and Y C t is the counterfactual outcome that would have emerged in the absence of the intervention of the joint agenda.
Finally, an ITS approach requires a time window to be specified. A time window is a time range before and after the introduction of the treatment. We discuss this issue in the Results section.

Individual proposal level
The second independent variable of interest is the effect of political conflict, measured by the level of preference heterogeneity between the pivotal Member States and its impact on duration. In particular, we were interested in studying whether the new agenda-setting structure, in which an already heterogeneous mini-coalition compromises over the agenda, moderates the effect of political conflict on legislative duration. For this reason we proceeded to the proposal level and performed survival analyses in order to estimate the potential moderating effect of the joint agenda on preference heterogeneity (Hypothesis 2). In addition, the analysis allowed us to control for other potential confounders at the proposal level, such as the characteristics of the rotating presidency, which also changed in January 2007.
At the proposal level, the time period of the analysis goes from 2000 to 2012. Earlier years would not contribute insights specific to the key focus of this article, the introduction of joint agendas, while duration data for years beyond 2012 are increasingly censored. That is, an increasing number of legislative acts that were introduced after 2012 are still waiting proposals. The data thus cover 13 years in which legislation was introduced and adoptedseven before a joint agenda and six thereafter. This provides us with a total of 1,927 regulations, directives and decisions in the analysis.
Given the nature of the data, we used Cox regression analysis, which is well suited for survival data. Survival analysis is a statistical tool specifically designed to study duration until an event occurs, in this case the first decision reached by the Council. See also Golub (2007), Klüver and Sagarzazu (2013), and Drüner et al. (2018).
The independent variable of interest, joint agenda, is a dummy variable that has the value (1) for legislation introduced under the system of a joint agenda since 2007 and (0) otherwise. We expected this variable to have a positive effect on the hazard rates.
We measured political conflict as the size of the interval between the two pivotal Members States under the respective voting rules, qualified majority voting and unanimity. The larger the size of this interval, the higher the disagreement between the two pivots. 12 We calculated this variable in line with Crombez and Hix (2015). We use the ParlGov dataset and calculate the positions of Council actors on the left-right dimension based on the weighted seat share of the parties in government. For acts decided on unanimously, we calculated the interval as the difference between the most left and most right Council Member State. The interval under qualified majority is the difference between the two pivotal Council members under the applicable majority thresholds. 13 In the analysis we used the squared distance for every 6-month period. Figure 1 presents the intervals. Our main empirical model for the proposal levels analysis is as follows. 14 where h(t) is the hazard function, h 0 is the baseline hazard, agenda is a dummy for the joint agenda, conflict L_R represents the heterogeneity measure on the left-right dimension, as illustrated in Figure 1, and z is a vector of coefficients for a vector of control 12 While the distance between the extreme Member States is by no means an invalid proxy for the level of disagreement, it does not identify the level of conflict between the pivotal players to strike a deal. In other words, the level of conflict in the Council may increase, but it may leave the level of conflict between the two pivotal Member States unaffected (see also Schulz and König, 2000). 13 During the  14 Although the left-right dimension is arguably the major dimension of conflict in the Council, scholars have found that there are other dimensions, such as pro-anti EU integration or budget contributions (Thomson et al., 2004;Zimmer et al., 2005) In the Appendix (Table A3) we repeated the exercise with heterogeneity on pro-anti EU integration and budget contribution dimensions. We did not find any effect for these dimensions, however. The trio presidency and efficiency variables x. We discuss the control variables below, starting with a number of institutional characteristics.

Co-decision
During the time period the two procedures used were consultation and co-decision. The latter may slow down the decision-making process because of the involvement of the European Parliament (Golub, 1999(Golub, , 2007Golub and Steunenberg, 2007;König, 2007).
In the analysis, co-decision is coded as (1) if a proposal was introduced under co-decision and (0) if it was introduced under consultation.

Qualified majority voting
Previous studies have shown that qualified majority voting has sped up decision-making compared with unanimity (Golub, 1999(Golub, , 2007König, 2007;Schulz and König, 2000). Given that the EU increasingly uses qualified majority voting we had to rule out the possibility that a potential decline in duration, as from 2007, can be attributed to the use of qualified majority voting instead of changes in the setup of agenda-setting. Qualified majority receives value (1) if the proposal was voted on by qualified majority voting and (0) for unanimity. Decision-Making in the EU Dataset (Häge, 2011) does not provide information on the voting rule for an overwhelming majority of observations during the time period under study. We therefore matched the proposals obtained from PreLex to proposals presented in the Monthly Summary of Council acts, which do consistently list the voting rule.

National parliaments
Apart from changing the agenda-setting mechanism in the Council, the Lisbon Treaty also increased the formal involvement of national parliaments in the decision-making process by introducing the early warning system and strengthened the role of parliaments. We expected this to increase duration. We assigned value (1) for proposals introduced since January 2010 and (0) for proposals introduced before that date.

Enlargement
We created a dummy variable enlargement, coded as (1) since May 2004 and (0) before that date, when the number of Council actors increased from 15 to 25, making it potentially more difficult to reach a decision (Golub, 2007;König, 2008). Second, we controlled for a number of presidency characteristics.

Presidency experience
More experienced Member States may be more efficient in chairing the Council, thus decreasing decision-making time. The variable is the count of the number of times Member States have held the presidency previously.

Presidency size
Likewise, large Member States may have more resources available and may push for proposals to be adopted faster. The variable is a dummy that takes on the value (1) if one of the big five (Germany, France, the UK, Italy or Spain) held the presidency and (0) otherwise.

Presidency heterogeneity
We controlled for the distance between successive presidents on the left-right dimension. If the country holding the presidency in the subsequent 6 months had significantly different preferences, this may have led to an increase in duration. We calculated the squared distance between two subsequent presidencies on the left-right dimension to capture this effect. Third, we controlled for legislative activity, crisis periods and a proxy for proposal salience.

Legislative activity
This variable is simply the number of pieces of legislation introduced in a given year. More legislation introduced may speed up decision-making. However, the opposite argument is plausible as well. We had no particular expectation, but it is important to control for legislative activity.

Economic crisis
In a crisis governments may be more pressured to be efficient and hence speed up decision-making, although this mechanism may not be present for areas other than economic governance, if at all. Nonetheless, some readers may have wanted us to control for such events. 15 We coded it with value (1) if legislation was introduced in 2008 or 2009 and (0) otherwise.

No priority proposals
We had to rule out the possibility that certain proposals may drive the result of faster legislation, an assumption that cannot be theoretically attributed to the joint agenda argument. These are arguably proposals introduced at the end of a trio presidency, which are probably not priorities of the trio members. This variable is (1) for proposals introduced during the final 3 months of a trio's tenure and (0) otherwise. Finally, we controlled for policy area, instrument and monthly fixed effects. 16 Some policy areas may be more sensitive and thus require more time. Likewise, some months are characterized by less legislative activity (summer and Christmas) and may affect duration. Finally, directives and regulations are proven to be often more contested (Golub, 2007;Schulz and König, 2000) and can thus be expected to increase duration. Tables A4 and A5 present descriptive statistics for these variables and their correlation matrix, respectively.

Aggregated level
We started by analysing the aggregated data on monthly average duration with the ITS approach and estimate the effect of the joint agenda structure. As mentioned previously, an ITS approach requires the researcher to specify a time window. In general, choosing a time window involves finding an optimal balance between precision and bias. A larger 15 Our results are robust to model specifications without this variable. 16 We captured policy area based on (1) subject area discussion Council, (2) subject area adoption Commission and (3) subject area adoption Council. Because the subject area may vary, we decided to use the subject area if two of three were identical. If all areas varied or information was missing, we took information based on the rank outlined above.
window yields more precise estimates as more observations are available to estimate the regression. On the other hand, the linear specification is less likely to be accurate when a larger window is used, which can bias the estimate of the treatment effect. We therefore performed the analysis with two distinct windows: a wide window of 30 months on each side of the treatment cut-off and a smaller window of 18 months on each side. 17 We then estimated two local linear regressions on each side of the treatment's cut-off point. Table 1 shows the estimates of the negative binomial regression, whereas Figure 2 presents the plots of the ITS for both sets of window intervals. There is a clear discontinuity at the time of the introduction of the joint agenda. Proposals introduced since January 2007, thus receiving the treatment, were agreed upon significantly faster than proposals that did not receive the treatment. Figure 2 shows the pre-intervention trend of the average monthly duration (solid line), and the counterfactual estimated prediction when no agenda coordination would be required (dotted line). All but one data point lies below this counterfactual line, suggesting a decrease in the duration to reach a first formal decision among the Member States following a change in agenda-setting structure.
These results suggest that the formal requirement to set the agenda under a predetermined mini-coalition of three Member States did indeed lead to faster decisionmaking. Rather than creating discontinuities in priorities among subsequent presidencies, (heterogeneous) trio members coordinated the agenda together and mutually decided which issue to address during their 18-month stints in office, resulting in less need to delay proposals that were unfavourable and lending support to Hypothesis 2.

Proposal Level
Next, we proceeded to the analysis at the proposal level, allowing us to take into account proposal-specific characteristics and to include additional years in the analysis. The semiparametric Cox survival analysis we present below nicely complements the analyses presented above.
A critical assumption of the Cox model is that the hazards of the covariates are proportional (Box-Steffensmeier & Zorn, 2001). We referred to such a test and found 17 Windows larger than 30 months are not desired due to the enlargement in May 2004 pretreatment, while the Lisbon Treaty took effect in December 2009, a little over 30 months post-treatment. Notes: * P < 0.10, ** P < 0.05, *** P < 0.01. Models are estimated by a negative binomial regression with robust std. errors in parentheses. Window time is 30 months prior and 30 months after treatment. FE, fixed effects. that a number of variables violated this assumption. We therefore repeated the exercise and took into account the time-dependent effect. 18 For clarity of interpretation, we report the effects as hazard ratios. Values greater than one indicate an increase in the hazard of adoption of legislation, while values below one indicate a decrease in the hazard, that is, a larger delay to reach the first formal decision. Table 2 presents the results of the Cox regression with the dependent variable being the number of days to reach a first decision in the Council. Model 1 presents the result of the Cox survival model without interactions and focuses on the effect of the joint agenda. The joint agenda variable is highly significant and has a hazard rate of 1.51, suggesting that at any given point in time, proposals that are pre-negotiated under a heterogeneous mini-coalition are 51% more likely to be agreed upon than proposals that are individually set by the Council presidency. 19 This finding is in line with the results from the ITS analysis.
In order to rule out an alternative compelling explanation, that of the increasingly used early agreements under the co-decision procedure, we repeated the exercise with a subset of proposals, that is, those falling under the consultation procedure. Under this procedure, no early agreements are possible. Table A6 in the Appendix presents the results of the regressions. Our findings prove robust to this exercise. 18 The results show violations for four control covariates: co-decision, enlargement, large state presidency and legislative activity. Accordingly, we interacted these controls with the natural logarithm of Time, as is the standard approach in the literature. An alternative would have been to stratify using those variables. 19 Figure A3 presents a Kaplan-Meier survival function for the two agenda-setting systems. Most of the control variables are in line with the reported findings in the literature. Codecision and enlargement have increased legislative duration. On the other hand, the use of qualified majority decreases decision-making time. Furthermore, the increasing role for national parliaments in the legislative process, as introduced by the Lisbon Treaty, slows down decision-making, although it is not statistically significant in all models.
We now proceed to the analysis of Hypothesis 2. Institutional arguments frequently imply that the relationship between political inputs (in this case preference heterogeneity) and outcomes (legislative duration) varies depending on the institutional context. A change in institutional context, such as a change in agenda-setting structure, may alter the effect of political conflict. In particular, we expect that setting a joint agenda by an already heterogeneous trio moderates the effect of preference heterogeneity in the Council on decision-making speed.
Model 2 in Table 2 shows how the effect of political conflict is moderated by the joint agenda. Remember that political conflict is mean-centred. The coefficient of 0.97 represents the effect of conflict before the introduction of the joint agenda. More conflict leads to an increase in duration, in line with previous studies (Drüner et al., 2018;Klüver and Sagarzazu, 2013;König, 2007). However, and interestingly, the interaction has the opposite effect (1.03), indicating a clear moderating effect of the trio system on political Notes: Estimations of semi-parametric Cox proportional hazard models with time-dependent effects (T). The dependent variable is the time to reach a first agreement in the Council in days. Coefficients are reported as hazard ratios. * P < 0.10; ** P < 0.05; *** P < 0.01. AIC, Akaike information criterion; BIC, Bayesian information criterion; FE, fixed effects; MS, Member State; QMV, qualified majority voting; T, time.
conflict. The effect is visualized in Figure 3. The mere fact that an already heterogeneous mini-coalition of Member States has previously negotiated and built on a compromise counters conflict between the other Member States in the Council. Finally, in Model 3 we include a number of additional control variables. The results of the joint agenda variable, as well as the moderating effect of political conflict are robust to the inclusion of these additional controls.

Conclusions
The main objective for altering the institutional setup of the rotating presidency was to increase the efficiency and effectiveness of Council decision-making. With this institutional reform in mind, policy-makers hoped to streamline heterogeneous preferences between successive presidents, to counter the prevalent discontinuities in their priorities and to speed up legislative decision-making altogether. Whether the trio presidency did indeed foster efficiency in the EU's main legislative arena has hitherto remain a puzzle in the much existing literature specifically on the trio presidency, and the duration of EU policy-making more broadly.
To fill this gap we studied the trio's impact on Council efficiency in a consistent and empirical manner. To do so, we exploited a unique opportunity: the timing of the introduction of the trio. As we argued earlier, Germany proposed to move ahead with the trio presidency system, an idea that was unanimously agreed upon by all the Member States at the negotiations of the Constitutional Convention, rather than to hold off until the Treaty of Lisbon entered into force in December 2009. As a result, trio presidencies have been operating in the EU ever since January 2007, without any other institutional innovations that would have made it increasingly difficult to identity the trio's treatment effect.
In a study of Council decision-making duration on 1,927 legislative proposals, initiated between 2000 and 2012, we obtained several interesting findings. First, the trio presidency format, accompanied by the requirement to work out a joint and longer term 18-month policy agenda, gives rise to faster decision-making. Proposals that require pre-negotiations by trio members need less time to be agreed upon. This is arguably the result of improved continuity in the Council's work. Raik (2016) and Warntjen (2013b), for example, studied the effect of the joint trio programmes on continuity and argued that the latter can improve efficiency. Our results are consistent with this reasoning. Does this mean that trio members have more consistent and coherent priorities than before? Not necessarily. It merely suggests that diverging priorities between successive presidents are erased to a higher extent, which suffices to improve efficiency itself. The introduction of the trio can thus be considered a successful institutional reform that addresses a major drawback of the 6-monthly rotating presidency; that is, the discontinuities in priorities. On the other hand, the system retains its merits, such as equality and legitimacy in decisionmaking. Each Member State continues to receive the opportunity to lead the legislative business of the Council for 6 months, without suffering from inefficiencies and strategic delay, which is to the benefit of all Member States, the European Commission and the European Parliament. It is therefore not surprising to see that the Council recently agreed to proceed with the trio presidency format until at least 2030.
Second, we find that the trio structure plays a crucial role in overcoming diversity in Member States' preferences. The latter, also referred to more formally as political conflict, has been found to be an important determinant of the length of negotiations. Our findings show that prior bargaining among a mini-coalition of three Member States that already differ in size, length of membership and geographical location neutralizes this effect. In this way, the trio presidency complements the role of another important actor, the General Council Secretariat, in stimulating smooth decisionmaking operations. The latter is often regarded as highly effective in spurring presidencies to be productiveand hence efficientby providing the necessary information and resources to small and unexperienced Member States. Although the Council Secretariat does indeed facilitate the decision-making process in its role as mediator, it can prevent political conflict and stimulate compromises to only a limited extent. To that effect, early discussions and formal pre-negotiations between delegates of the trio members themselves, representing diverse interests, arguably prevents political conflict from hindering smooth and timely decision-making, and thus nicely complements the performance of the Council Secretariat.
Finally, in the advent of the very first trio presidency configuration resuming office for a second term in July 2020, nearly 13 years after its inauguration, it may be worthwhile to perform comparative studies on trios over time in future endeavours.

Supporting Information
Additional supporting information may be found online in the Supporting Information section at the end of the article.   Figure A1: Kernel Density Plot of the Time Needed in Months to Reach a First Agreement in the Council. Figure A2: Set of Non-Parametric Local Fittings. Figure A3 Kaplan-Meier Survival Function.