Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/oxford-university-press/how-formal-parliamentary-negotiation-affects-policy-making-evidence-dOBNG3gvEQ
- Publisher
- Oxford University Press
- Copyright
- © The Author(s) 2018. Published by Oxford University Press on behalf of the Hansard Society; all rights reserved. For permissions, please e-mail: journals.permissions@oup.com
- ISSN
- 0031-2290
- eISSN
- 1460-2482
- DOI
- 10.1093/pa/gsy021
- Publisher site
- See Article on Publisher Site
Abstract
Abstract Does parliamentary negotiation help facilitate legislation? This article applies propensity score matching to empirically investigate the effect of a unique formal negotiation mechanism in Taiwan’s parliament. The analysis of legislative initiatives between 2012 and 2015 shows that the so-called Party Negotiation Mechanism, while being invented to revive legislation amid stalemates, substantially reduces the chance of successful legislation. By extension, the undercut probability of legislative success under Taiwan’s majoritarian system implies that the conventional majoritarian–consensus framework has overlooked a crucial parameter that directly shapes the patterns of policy-making—the internal organisation of a parliament. ‘Inclusions are not sufficient to democracy: if collectivities lack the capacity to act, inclusions remain powerless’ —Warren and Mansbridge (2013, p. 88) Parliamentary negotiation is an oft-seen practice adopted by modern parliaments to reconcile differences among lawmakers and resolve legislative stalemates. From a normative perspective, parliamentary negotiation enhances the inclusiveness and the equality of rights within the parliament, which helps avoid the dangers of tyranny (Warren and Mansbridge, 2013). Moreover, it has been suggested that the US Congress should institutionalise certain rules of negotiation to resolve legislative stalemates (Binder, 2003). In particular, closed-door and repeated negotiations are recommended as they allow opposing parties to share their views freely and thereby facilitate reaching agreements (Warren and Mansbridge, 2013; Binder and Lee, 2015). On the other hand, a counterargument sounds equally plausible. That is, parliamentary negotiation, especially when it emphasises inclusiveness and power equality, may function as a gate that blocks legislation, rather than a conflict-resolving mechanism. The intuition is analogous to the immobilism problem with power-sharing arrangements (Horowitz, 2014) and the theory of veto players (Tsebelis, 2002). Since the nature of negotiation values the exchanges of views, the give-and-take process can be time consuming. Furthermore, negotiation may lead to a situation where nothing is agreed until everything is agreed. If negotiation is a required process, legislative inaction may occur even in the presence of a majority coalition that can surely enact a bill through a majority rule. As the epigraph suggests, an inclusionary decision-making mechanism might undermine the capacity of a democracy to pass necessary legislation in response to crucial national problems. Despite these theoretical controversies, we know relatively little from the existing literature about which account is empirically grounded. The lack of such studies results partly from limited public information given that most parliamentary negotiations are informal and off-the-record. Moreover, these informal parliamentary negotiations are applied as an intermediate process to cut deals that are subject to the floor’s rejection.1 Given the indirect linkage between informal parliamentary negotiations and legislative outcomes, it is difficult to evaluate the policy-making effect of parliamentary negotiations. Drawing on the practice of a unique formal parliamentary negotiation mechanism in Taiwan’s parliament, this article offers crucial evidence to fill the aforementioned research gap. In 1998, Taiwan’s Legislative Yuan (LY) legalised a Party Negotiation Mechanism (PNM) in the hope of preventing the boycotts and fierce fights that frequently occurred in the second and third LY (1993–1999). The PNM is comparatively peculiar in three respects. First, the size threshold for a party to be a negotiator is low. Any party with three members (or five before 2008) is allowed to share the decision-making right in the negotiation. Secondly, it requires a unanimous consensus to make a decision. In effect, the PNM grants an equal decision-making right to each party in the LY. Thirdly, the PNM more often than not replaces voting as the major decision-making mechanism (Sheng and Huang, 2017). The decisions made through the PNM can be final without any further voting procedure. With these features, the PNM increases the inclusiveness and power equality in the LY, as well as directly determines the outcomes of many legislative initiatives. In the final stage of the legislative process, some legislative initiatives are sent to the PNM, while others go through the plenary process without being processed by the PNM. The difference in the success rates between initiatives that are sent to the PNM (hereafter, the PNM-treated initiatives) and their counterparts provides valuable evidence demonstrating whether a formal mechanism of parliamentary negotiation facilitates or hinders policy-making. An investigation of the PNM’s effect on legislative success also helps complement comparative studies of policy-making. Lijphart (1999) distinguishes between majoritarian and consensus democracies based mostly upon constitutional arrangements. While Lijphart insists that consensus democracies, which emphasise inclusiveness, bargaining and compromise, tend to outperform majoritarian democracies in many aspects of governance, he admits that the policy-making in a majoritarian system can be more decisive and faster than one in a consensus democracy. Pundits have enthusiastically debated over the implications drawn from Lijphart’s typology (Armingeon, 2002; Lijphart, 2004; Selway and Templeman, 2012). However, the constitutional systems may not sufficiently account for the empirical patterns of policy-making. For instance, Taiwan is arguably a majoritarian democracy given its one-party executive branch, mixed-member majoritarian electoral system, unitary government and two-party system with a clear majority party in a unicameral parliament. Counterintuitively, Taiwan’s pattern of policy-making resembles one that might be observed in a consensus democracy. From 1999 to 2015, the average success rate of legislation in Taiwan is under 30%, and the government formed by the majority party has at best enacted only 56% of its bills (Huang and Sheng, 2017). This non-majoritarian pattern of policy-making implies that the conventional majoritarian–consensus framework has overlooked a crucial parameter that directly shapes the patterns of policy-making. Taiwan’s parliament features a unique negotiation mechanism through which a portion of legislative decisions are made by consensus. To clarify whether the PNM hinders legislation is a key to enhancing our understanding of policy-making beyond the conventional framework. Methodologically wise, making a valid causal inference about the PNM’s effect is effortful. Causal inferences regarding the effect of a certain legislative rule, such as the open or closed rules in the US Congress or the Co-decision rule in the European Parliament, can be suspicious, as researchers are not allowed to manipulate the legislative procedure and conduct a randomised experiment.2 Without a randomised treatment assignment, the observational data of legislative initiatives can be highly imbalanced. Namely, initiatives that go through the procedure of interest (treated) and those that do not (untreated) might differ so substantially that any observed treatment effects are simply reflecting the intrinsic discrepancies between the two sets of initiatives. Even worse, such discrepancies might be associated both with the treatment assignment and with the outcome, which confounds the treatment effect. Regression may not assure the balance in the values of the covariates between the treated and untreated. In addition, the lack of common support, which is the lack of equal cases in both groups, may make the other-things-being-equal statement only a mathematical extrapolation (Gelman and Hill, 2007). Similar issues also call into question the existing studies of the PNM because the PNM-treated initiatives are allegedly more controversial than their counterparts (Hawang and Ho, 2007). If a legislation-hindering effect is found, a reasonable doubt is that the PNM-treated initiatives are less likely to succeed not because they are treated by the PNM, but because they are intrinsically more controversial. As many discrepancies between the treated and untreated exist to confound the effect of the PNM, it remains contentious whether the PNM facilitates or hinders legislation. This article is an unprecedented attempt of using matching to study the effect of a formal parliamentary negotiation mechanism. Matching is a class of statistical methods that are applied to reveal the hidden experiment in observational studies and thereby enhance the casual inference (King and Nielsen, 2016). In plain words, the intuition is to make sure the treated and untreated cases differ only in whether they have received the treatment. To be specific, this article adopts propensity score matching that approximates a completely randomised experiment to assure that the assignment of the treatment (i.e. whether an initiative is sent to the PNM) is independent of the legislative outcome and other covariates. When the treatment assignment is ignorable given the covariates, the imbalance between the two groups can be minimised or eliminated, and the differences in the outcomes can be solely attributed to the treatment (Rosenbaum and Rubin, 1983). Although propensity score matching faces some criticisms, it is useful to make an improvement relative to the original data if applied carefully (King and Nielsen, 2016, p. 24). Furthermore, while other matching approaches (e.g. exact matching) might produce high-level balance, propensity score matching is more optimal to find sufficient matches given the numerous confounders in this study. With a propensity score matching analysis of the initiatives from Taiwan’s most recent full parliamentary term (The eighth LY, 2012–2015), this article addresses the issues of causal inferences that have been overlooked in the existing literature. The matching analysis demonstrates that an initiative is much less likely to be enacted if it is referred to the PNM than would otherwise have been. In other words, the PNM substantially lowers the chance of successful legislation in the LY. Post-estimation analyses are offered to ensure the post-matching balance and the robustness of the major finding.3 These results offer valuable empirical evidence inferring an undesirable policy-making consequence if a parliament adopts negotiation as a major decision-making mechanism. In the next section, I discuss the contesting accounts and mixed findings regarding the PNM’s effect from the literature, followed by a section introducing the research method and model specification. The third section presents the main findings with balance-checking and sensitivity analyses. Finally, the conclusion discusses the implications for studies of policy-making and parliamentary rules. 1. Party negotiation mechanism and policy-making in Taiwan Policy-making is challenging to a young democracy. When political liberalisation begins, newly enfranchised political parties and individual legislators may be extraordinarily ambitious to take part in making national policies (Figueiredo and Limongi, 2000; Casar, 2002; Nalepa, 2016). In particular, these parties and politicians may try to frustrate the dominant party by boycotting its legislation. This was what happened in the few years before and after Taiwan’s democratisation in 1996. Successful legislation was almost unlikely because numerous procedural motions and parliamentary inquiries, not to mention the notorious parliamentary brawls (Batto et al., 2016), repeatedly paralysed the legislative process in the LY. From 1993 to 1998, a norm of party negotiation was invented in order to reduce the frequency of boycotts and enhance the harmony in the LY. Speaker Liu Song-fan invited party whips and opposing legislators to negotiate over disputed legislation. Since 1998, this informal norm has been written into Taiwan’s Law Governing the Legislative Yuan's Power, and starting from the fourth LY (1999–2001), the PNM has been put into practice for almost two decades. Since the deals made through this mechanism can be final legislative outcomes without any further procedures, the governing and opposition parties have shared the decision-making power together through negotiations. It has been expected that the number of enacted bills would increase through closed-door exchanges of views and interests (Wang, 2003; Lo, 2006). Echoing this expectation, Hawang and Ho (2007) allege that the PNM enhanced the success rate for controversial legislation in the fifth LY.4 Moreover, a recent survey shows that many legislators and their staffs believe that the PNM is effective to resolve disputes in the LY (Sheng and Huang, 2017). Nevertheless, several studies cast doubt on the merits of the PNM. Among others, pundits question whether the PNM actually improves the chance of successful legislation.5 According to the current statutes, initiatives can be sent to the PNM by a decision of the corresponding committee. Alternatively, any single party or a legislator with ten co-sponsors may also send initiatives into the PNM by filing a motion in the plenary of the second reading.6 In practice, once an initiative is sent to the PNM, it will not move to the next stage until a unanimous consensus is reached (Sheng and Huang, 2017).7 As such, the PNM offers opportunities for parties and legislators to defer or bury unfavourable legislation. Furthermore, small parties gain bargaining chips to exchange the legislation in their own favour (Wang, 2002, 2014). To elaborate, the PNM grants each negotiating party the right to veto given that a unanimous consensus is required to make a decision. In a legislative body, such a veto right may not only increase the chances of stalemates, but allow the negotiating parties to pass their preferred policies that would otherwise have little chance of enactment under a majority rule (Nunnari, 2017). The expectation to gain such benefits from disagreements may in turn increase the chance and duration of stalemates. For example, in 2003, the legislation of the Act for the Establishment and Management of Free Ports was deferred in spite of the support from the two major parties, the Kuomintang (KMT) and the Democratic Progressive Party (DPP). The reason was that a small party, the Non-Partisan Solidarity Union (NPSC), asked for the protection of working rights for Taiwanese aborigines in exchange of its approval (Sheng and Huang, 2015). Incidences like this delayed or even buried many bills in the negotiation process. Recent studies show descriptive statistics against Hawang and Ho’s conclusion and suggest that the PNM initiatives had a much lower success rate than their counterparts in the seventh and eighth LY (Huang and Sheng, 2017; Sheng and Huang, 2017). The existing accounts and findings about the PNM’s effect are mixed. Moreover, most of the previous studies offer only theoretical discussions or descriptive statistics. All these studies have neither taken into account the assignment of the initiatives to the PNM, nor addressed the imbalance between the PNM-treated initiatives and the untreated. Accordingly, there is lack of valid evidence showing whether being treated by the PNM increases or decreases the chance of enactment for individual initiatives. Such evidence is necessary to settle the controversy over whether a formal parliamentary negotiation mechanism facilitates or hinders legislation. 2. Research method This article performs a propensity score matching analysis to assess the causal effect of the PNM on legislative outcomes. This matching approach adopts a conventional logistic or probit regression to estimate the propensity to the treatment, matches the treated and untreated cases on the estimated propensity score and, finally, obtains the average treatment effect on treated (ATT) with the matched cases (Rosenbaum and Rubin, 1983; 1985; Gelman and Hill, 2007, pp. 206–207). The matching process addresses the imbalance in the raw data and enhances the validity of causal inferences. Like all other methods, propensity score matching should be applied carefully. King and Nielsen (2016, p. 16) claim that there is a propensity score paradox–after propensity score matching accomplishes its goal of complete randomisation, to continue randomly pruning cases increases imbalance. According to their simulation, when the number of cases pruned increases, propensity score matching suffers imbalance, inefficiency, model dependence and bias. Although their allegations sound plausible, their simulation is an extreme case in which propensity score matching is applied carelessly and inappropriately. In particular, they claim that researchers often continue the pruning process in order to find the most favourable estimate of the causal effect, leading to serious imbalance and biased estimates (King and Nielsen, 2016, p.19). In practice, however, propensity score matching analyses, including the current investigation, control for necessary covariates to predict the treatment assignment, perform balance-checking tests and report the finding when balance is achieved. Moreover, this study presents a sensitivity analysis to ensure the finding is robust to model misspecifications. Finally, the problem of random pruning can be avoided by using the nearest neighbour algorithm with replacement (Jann, 2017). King and Nielsen (2016, p. 24) concede in the same article that propensity score matching can be an improvement if used carefully and can help the most when the pre-matching data has a high level of imbalance. In fact, propensity score matching is particularly useful in a study of the PNM as numerous covariates might confound the effect of interest. To be clear, propensity score matching is more applicable to address the curse of dimensionality than exact matching, including the Coarsened Exact Matching (CEM) proposed by King and his colleagues. Since exact matching matches cases on the values of each covariate, a large number of covariates may result in a limited number of matched cases (Blackwell et al., 2009, p. 526; Iacus et al., 2012. p. 13). This may in turn constrain the generalisability of the inference. Propensity score matching, by contrast, reduces the dimensionality by matching cases on one scalar. Therefore, while exact matching may produce a high level of balance, propensity score matching is a more optimal option that allows me to draw a valid inference from a reasonable number of matches. 2.1 Data With the official legislative record from the Parliamentary Library of the Legislative Yuan (2015), this article analyses the legislative initiatives from 2012 to 2015. Since the PNM is a procedure embedded in the second reading, only the initiatives arriving at the second-reading plenary can be referred to the PNM. Among all the 5515 initiatives, 48.12% or 2654 are qualified initiatives: 1419 or 53.47% of the qualified initiatives were sent to the PNM, while 1235 initiatives (46.53%) went through the plenary process without being treated by the PNM. Apparently, the PNM has been the major decision-making mechanism for the initiatives at the final stage of the legislative process.8 As shown in Figure 1, 655 of the initiatives arriving at the second reading were not enacted, and 94% of these failed initiatives were the PNM-treated initiatives. In other words, if the initiatives reaching the final stage of the legislative process failed, they mostly failed in the negotiation process. A simple comparison will also tell us that 96.68% of the untreated initiatives were enacted, while only 56.73% of the PNM-treated initiatives succeeded. The descriptive comparison indicates that the PNM-treated initiatives had a lower chance of enactment than the untreated. Figure 1. View largeDownload slide The initiatives at different stages of legislative process Note: The numbers in the parentheses are the percentage that a given category of initiatives takes among the initiatives at the previous stage. Figure 1. View largeDownload slide The initiatives at different stages of legislative process Note: The numbers in the parentheses are the percentage that a given category of initiatives takes among the initiatives at the previous stage. 2.2 Model specification The propensity score is ‘the conditional probability of assignment to a particular treatment given a vector of covariates’ (Rosenbaum and Rubin, 1983, p. 41). I adopt logistic regression to estimate the propensity score. The dependent variable for the propensity score model is a binary variable denoting whether an initiative was sent to the PNM (treatment). In order to maximise the balance after matching, the covariates included should capture as many differences between each initiative as possible. Moreover, it is crucial to control for factors affecting both the likelihood of receiving the treatment and the legislative outcome (Gelman and Hill, 2007, pp. 208; Stuart, 2010, p. 5). Considering these general rules, I include four sets of covariates: levels of controversy, initiators, legislative procedures and generic characteristics of initiatives. The PNM was invented in order to resolve the conflicts in the LY. Initiatives involved with a higher level of controversy are presumably more likely to be sent to the PNM and less likely to be enacted. As such, the effect of the PNM is confounded with the level of controversy. It is not easy to directly gauge the level of controversy for each initiative. Fortunately, several proxy variables reflecting the level of controversy from different angles are available. First, Hot Legislation is a dummy variable indicating whether the law concerned with a given initiative was ranked as the top ten popular legislation in the eighth LY.9 This ranking depends on the number of initiatives concerning a law, which infers how many contesting proposals have been proposed. An initiative competing with many other proposals might accompany an intense controversy. The number of news coverage may also reflect the level of controversy. I then create a dummy variable, High Media Attention, indicating whether the number of news coverage about a given initiative is greater than the average across all initiatives during the eighth term.10 Moreover, I include a dummy variable indicating whether a bill is related to the Major Political Disputes during the eighth LY. These disputes either resulted from the unification-independence cleavage or induced large-scale mass protests against the government. Examples are the bills concerning the Sino-Taiwan relationship, the food sanitation, military judicial system, stock income tax, long-term care services, nuclear power and labours’ rights. These bills might be more controversial and more likely to be sent to the PNM than other bills. In addition to the level of controversy, the initiator(s) of a bill may have different legislative influence. I first control for the affiliations of the initiators. KMT represents bills whose 90% of sponsors and co-sponsors are from the KMT and those proposed in the name of the KMT caucus.11 I apply this coding scheme to the DPP’s bills (DPP). Bills proposed by small parties are combined together into a single dummy variable (Small Parties), and those that do not meet the 90% criterion are coded as cross-party bills (Cross-Party). Additionally, bills from the Judicial Yuan, Control Yuan and Examination Yuan are combined together and labelled as Three Yuans.12 The base category includes the bills initiated by the Executive Yuan (hereafter, the government-initiated bills). Next, in order to control for the initiator’s parliamentary experience, I include Freshman, which denotes whether the leading initiator is in his or her first term as a legislator. Moreover, I control for dummy variables denoting the leading initiator’s positions within the LY, including Party Whip, corresponding Committee Member, corresponding Committee Chair, Procedure Committee Member and Procedure Committee Chair. Moreover, since professional and educated legislators might be more persuasive in promoting their bills (Huang, 2017), I control for variables indicating whether the leading initiator holds a special profession (e.g. lawyers, judges, professors, accountants, architects and other licenced specialists) and whether the leading initiator owns a master’s or a doctoral degree. PR Members is a dummy variable indicating whether the leading initiator was elected through the proportional representation system. Under Taiwan’s mixed-member electoral system, legislators are elected either from the single-member districts or through the party-list proportional representation system. The PR members might engage more in lawmaking activities, increasing the chance of successful legislation. The third set of variables represents the features of the legislative procedures that a given bill had gone through prior to the second reading. For starters, the chair of the corresponding committee has significant influence on a bill’s referral to the PNM. Moreover, the majority party might be advantageous over the minority party in negotiations (Chiou and Cheng, 2014). If this is true, committee chairs from the majority party might be tempted to send initiatives to the PNM.13 Hence, I include two dummy variables, Majority Chair and Minority Chair, which denote the partisanship of the corresponding committee chair for each bill. Secondly, the workload of the corresponding committee might be inversely associated with how much time a committee can spend on each piece of legislation. The lack of committee time might raise the chance of a bill’s referral to the PNM and lower the likelihood of legislative success. According to the number of initiatives processed by a given committee, I categorise each bill’s corresponding committees into High Workload (over 500 initiatives), Mid-level Workload (300–500 initiatives) and Low Workload (fewer than 300 initiatives). The first two categories are included in the model. Thirdly, a bill can be deliberated either in a single committee or in a joint meeting of multiple committees. Deliberations in a joint committee might be more complicated than that in only one committee. I include two dummy variables, Single Committee and Joint Committee, indicating the types of committee deliberation. The initiatives that were sent directly to the second reading are left out as the base category. Next, the executive branch tends to be advantageous over members from the legislative branch (McCubbins and Nobles, 1995). As a free-rider, legislators often propose bills along with the government in order to boost the chance of legislative success. I then include variables indicating whether the initiatives concerning a given law were proposed only by the government (Only Government) or by both the executive and legislative branches (Both Branches). Finally, legislative deliberations can be time consuming. The initiatives proposed in the final session presumably have a lower chance to be enacted than those proposed earlier. I include two dummy variables, First Session and Mid Sessions (i.e. sessions between the first and final session), denoting the session(s) in which a given initiative is proposed, and the initiatives proposed in the final session are treated as the base category. The fourth set of variables captures the generic characteristics of an initiative that might predetermine the fate of legislation. First, the numbers of sponsors and cosponsors reflect the strength of political support that might shepherd a bill through the legislative process. For each initiative, I control for three dummy variables indicating whether a bill has a Large Number of Sponsors (>mean), Moderate Number of Sponsors (between one and the mean), Large Number of Co-sponsors (>mean). Next, minor revisions have better chances to succeed (Sheng and Huang, 2015), while the bills proposing a new law or a grand revision of an existing law might be prone to be negotiated. Accordingly, I control for two dummy variables, New Law and Grand Revision (five or more articles). In a similar vein, I control for the Number of Articles included in each bill. Last, the propensity to the treatment and the chance of enactment might vary across issue areas. To account for this variation, I categorise all initiatives into 11 issue areas, including Political Development, Agriculture, Transportation and Infrastructure, Health and Welfare, Internal Affairs and Ethnicity, Labor, Judicial System, Environment, Education and Culture, Foreign Policy and National Defense, Economy and Public Finance. The base category is Public Administration and Personnel. In addition to the four sets of covariates, I include interaction terms in order to enhance the balance in the matched data. First, I interact the Number of Articles with High Workload Committee and Mid-level Workload Committee. This is to account for the possibility that these variables may interact to increase a bill’s propensity to be treated by the PNM. Secondly, I include the interactions between Media Coverage, Major Political Disputes and Hot Legislation in order to better capture the level of controversy. 3. Propensity score matching analysis A valid causal inference requires that the difference in the outcomes between the treated and untreated cases can be solely attributed to the treatment. Namely, the treated and untreated should only differ in the reception of the treatment. However, when the treatment assignment is not randomised, some cases are more likely than others to enter into the treatment group given the characteristics of the cases. As such, the treated and untreated may differ substantially in many respects. The treatment effect observed in the descriptive statistics might simply reflect the intrinsic discrepancies between the two groups. Table 1 justifies such a suspicion. Before matching, the probability of being sent to the PNM varies across individual initiatives given the covariates included.14 For example, the average propensity score for the initiatives with a high level of media attention is 0.65, and those concerning major political disputes have an average propensity score of 0.75. Hence, the salient and controversial initiatives are more likely than their counterparts to be treated by the PNM. Moreover, there is partisan imbalance between the treated and untreated as the minority parties’ bills are more likely to be sent to the PNM. The majority party (KMT)’s bills and the government-initiated bills have an average propensity score of 0.47 and 0.49, respectively, which are lower than the mean score for the DPP’s bills (0.62) and for small parties’ bills (0.67). Similarly, the propensity score varies with the pre-treatment procedures and generic characteristics of the initiatives. Together, the treatment and control group may consist of numerous distinct and incomparable initiatives. Table 1 Propensity score by covariates Variables Variable = 1 Variable = 0 T-test Level of controversy Hot legislation 0.54 0.53 High media attention 0.65 0.48 * Major political disputes 0.75 0.5 * Initiators KMT 0.47 0.56 * DPP 0.62 0.51 * Small parties 0.65 0.53 * Cross-party 0.54 0.53 Three yuans 0.41 0.54 * Freshman 0.53 0.54 Party whips 0.49 0.55 * Committee members 0.53 0.54 Committee chairs 0.54 0.53 Procedure committee members 0.55 0.53 Procedure committee chairs 0.69 0.53 * Professionals 0.54 0.53 Master 0.56 0.52 * PhD 0.5 0.54 * PR members 0.58 0.52 * Legislative procedures Majority (committee) chair 0.51 0.57 * Minority (committee) chair 0.55 0.53 High workload committee 0.54 0.53 Mid-level workload committee 0.49 0.55 * Single committee 0.52 0.63 * Joint committee 0.61 0.52 * Only government 0.37 0.54 * Both branches 0.59 0.38 * First session 0.67 0.49 * Mid-sessions 0.49 0.64 * Generic characteristics Large number of sponsors 0.54 0.53 Moderate number of sponsors 0.53 0.54 Large number of co-sponsors 0.55 0.53 Moderate number of sponsors 0.53 0.54 New law 0.74 0.5 * Grand revision 0.64 0.52 * Number of articles 0.49 (−) Political development issue 0.86 0.52 * Agriculture issue 0.54 0.53 Transportation and infrastructure issue 0.4 0.55 * Health and welfare issue 0.56 0.53 * Internal affairs and ethnicity issue 0.71 0.52 * Labor 0.45 0.54 * Judicial system issue 0.53 0.54 Environment issue 0.68 0.53 * Education and culture issue 0.59 0.53 * Foreign policy and national defense issue 0.44 0.54 * Economy and public finance issue 0.44 0.55 * Interaction terms High media attention*hot 0.53 0.54 Major political disputes*hot 0.63 0.53 * High media attention*major political disputes 0.72 0.52 * High media attention*hot*major political disputes 0.64 0.53 * Number of articles*high workload committee 0.54 (+) Number of articles*mid-level workload committee 0.27 (−) Variables Variable = 1 Variable = 0 T-test Level of controversy Hot legislation 0.54 0.53 High media attention 0.65 0.48 * Major political disputes 0.75 0.5 * Initiators KMT 0.47 0.56 * DPP 0.62 0.51 * Small parties 0.65 0.53 * Cross-party 0.54 0.53 Three yuans 0.41 0.54 * Freshman 0.53 0.54 Party whips 0.49 0.55 * Committee members 0.53 0.54 Committee chairs 0.54 0.53 Procedure committee members 0.55 0.53 Procedure committee chairs 0.69 0.53 * Professionals 0.54 0.53 Master 0.56 0.52 * PhD 0.5 0.54 * PR members 0.58 0.52 * Legislative procedures Majority (committee) chair 0.51 0.57 * Minority (committee) chair 0.55 0.53 High workload committee 0.54 0.53 Mid-level workload committee 0.49 0.55 * Single committee 0.52 0.63 * Joint committee 0.61 0.52 * Only government 0.37 0.54 * Both branches 0.59 0.38 * First session 0.67 0.49 * Mid-sessions 0.49 0.64 * Generic characteristics Large number of sponsors 0.54 0.53 Moderate number of sponsors 0.53 0.54 Large number of co-sponsors 0.55 0.53 Moderate number of sponsors 0.53 0.54 New law 0.74 0.5 * Grand revision 0.64 0.52 * Number of articles 0.49 (−) Political development issue 0.86 0.52 * Agriculture issue 0.54 0.53 Transportation and infrastructure issue 0.4 0.55 * Health and welfare issue 0.56 0.53 * Internal affairs and ethnicity issue 0.71 0.52 * Labor 0.45 0.54 * Judicial system issue 0.53 0.54 Environment issue 0.68 0.53 * Education and culture issue 0.59 0.53 * Foreign policy and national defense issue 0.44 0.54 * Economy and public finance issue 0.44 0.55 * Interaction terms High media attention*hot 0.53 0.54 Major political disputes*hot 0.63 0.53 * High media attention*major political disputes 0.72 0.52 * High media attention*hot*major political disputes 0.64 0.53 * Number of articles*high workload committee 0.54 (+) Number of articles*mid-level workload committee 0.27 (−) Notes: The T-test shows whether the difference in the propensity score is statistically significant at the 0.05 level. The propensity score for each continuous variable is estimated at the mean of the variable. The plus and minus signs indicate the direction of the association between the variable and the propensity score. View Large Table 1 Propensity score by covariates Variables Variable = 1 Variable = 0 T-test Level of controversy Hot legislation 0.54 0.53 High media attention 0.65 0.48 * Major political disputes 0.75 0.5 * Initiators KMT 0.47 0.56 * DPP 0.62 0.51 * Small parties 0.65 0.53 * Cross-party 0.54 0.53 Three yuans 0.41 0.54 * Freshman 0.53 0.54 Party whips 0.49 0.55 * Committee members 0.53 0.54 Committee chairs 0.54 0.53 Procedure committee members 0.55 0.53 Procedure committee chairs 0.69 0.53 * Professionals 0.54 0.53 Master 0.56 0.52 * PhD 0.5 0.54 * PR members 0.58 0.52 * Legislative procedures Majority (committee) chair 0.51 0.57 * Minority (committee) chair 0.55 0.53 High workload committee 0.54 0.53 Mid-level workload committee 0.49 0.55 * Single committee 0.52 0.63 * Joint committee 0.61 0.52 * Only government 0.37 0.54 * Both branches 0.59 0.38 * First session 0.67 0.49 * Mid-sessions 0.49 0.64 * Generic characteristics Large number of sponsors 0.54 0.53 Moderate number of sponsors 0.53 0.54 Large number of co-sponsors 0.55 0.53 Moderate number of sponsors 0.53 0.54 New law 0.74 0.5 * Grand revision 0.64 0.52 * Number of articles 0.49 (−) Political development issue 0.86 0.52 * Agriculture issue 0.54 0.53 Transportation and infrastructure issue 0.4 0.55 * Health and welfare issue 0.56 0.53 * Internal affairs and ethnicity issue 0.71 0.52 * Labor 0.45 0.54 * Judicial system issue 0.53 0.54 Environment issue 0.68 0.53 * Education and culture issue 0.59 0.53 * Foreign policy and national defense issue 0.44 0.54 * Economy and public finance issue 0.44 0.55 * Interaction terms High media attention*hot 0.53 0.54 Major political disputes*hot 0.63 0.53 * High media attention*major political disputes 0.72 0.52 * High media attention*hot*major political disputes 0.64 0.53 * Number of articles*high workload committee 0.54 (+) Number of articles*mid-level workload committee 0.27 (−) Variables Variable = 1 Variable = 0 T-test Level of controversy Hot legislation 0.54 0.53 High media attention 0.65 0.48 * Major political disputes 0.75 0.5 * Initiators KMT 0.47 0.56 * DPP 0.62 0.51 * Small parties 0.65 0.53 * Cross-party 0.54 0.53 Three yuans 0.41 0.54 * Freshman 0.53 0.54 Party whips 0.49 0.55 * Committee members 0.53 0.54 Committee chairs 0.54 0.53 Procedure committee members 0.55 0.53 Procedure committee chairs 0.69 0.53 * Professionals 0.54 0.53 Master 0.56 0.52 * PhD 0.5 0.54 * PR members 0.58 0.52 * Legislative procedures Majority (committee) chair 0.51 0.57 * Minority (committee) chair 0.55 0.53 High workload committee 0.54 0.53 Mid-level workload committee 0.49 0.55 * Single committee 0.52 0.63 * Joint committee 0.61 0.52 * Only government 0.37 0.54 * Both branches 0.59 0.38 * First session 0.67 0.49 * Mid-sessions 0.49 0.64 * Generic characteristics Large number of sponsors 0.54 0.53 Moderate number of sponsors 0.53 0.54 Large number of co-sponsors 0.55 0.53 Moderate number of sponsors 0.53 0.54 New law 0.74 0.5 * Grand revision 0.64 0.52 * Number of articles 0.49 (−) Political development issue 0.86 0.52 * Agriculture issue 0.54 0.53 Transportation and infrastructure issue 0.4 0.55 * Health and welfare issue 0.56 0.53 * Internal affairs and ethnicity issue 0.71 0.52 * Labor 0.45 0.54 * Judicial system issue 0.53 0.54 Environment issue 0.68 0.53 * Education and culture issue 0.59 0.53 * Foreign policy and national defense issue 0.44 0.54 * Economy and public finance issue 0.44 0.55 * Interaction terms High media attention*hot 0.53 0.54 Major political disputes*hot 0.63 0.53 * High media attention*major political disputes 0.72 0.52 * High media attention*hot*major political disputes 0.64 0.53 * Number of articles*high workload committee 0.54 (+) Number of articles*mid-level workload committee 0.27 (−) Notes: The T-test shows whether the difference in the propensity score is statistically significant at the 0.05 level. The propensity score for each continuous variable is estimated at the mean of the variable. The plus and minus signs indicate the direction of the association between the variable and the propensity score. View Large In fact, the imbalance between the treatment and control group before matching is unignorable. In Figure 2, the propensity scores for the PNM-treated initiatives are concentrated on large values, while most untreated initiatives have a low propensity score. Moreover, the average propensity score of the treated is 0.64, while the average score of the untreated is 0.42, making a difference of 0.22 (p < 0.001). When the two groups differ so substantially from each other, ignoring the imbalance may result in questionable inferences about the effect of the PNM. Figure 2. View largeDownload slide Propensity scores before matching Figure 2. View largeDownload slide Propensity scores before matching The high-level imbalance in the original data justifies the application of propensity score matching. I then adopt the nearest neighbour algorithm with replacement to match cases from the two groups on the estimated propensity scores. This algorithm matches one treated case with one untreated case that has the nearest propensity score.15Figure 3 shows that the balance is much improved after matching. The post-matching distributions for the two groups look almost the same. Meanwhile, the mean propensity score for each group is equal at 0.51, which further indicates the balance between the two groups. Moreover, the matched cases from each group were spread over a wide range of propensity scores, suggesting that the matched initiatives are not only those with a moderate propensity score. By all accounts, matching on the propensity score greatly improves the balance between the two groups, and it ensures a broad range of common support. Figure 3. View largeDownload slide Propensity scores after matching Note: The number of matched cases is 948, and there are 470 treated cases and 478 untreated cases. Figure 3. View largeDownload slide Propensity scores after matching Note: The number of matched cases is 948, and there are 470 treated cases and 478 untreated cases. After matching, the initiatives sent to the PNM are enacted at a rate of 62%, while the matched cases that go through the regular process can be enacted at a rate of 95% (See Table 2). The estimated ATT is −0.33, indicating that ceteris paribus, being treated by the PNM on average reduces the probability of legislative success for an initiative by 33%, compared to a matched initiative that is not treated by the PNM. In plain words, although a PNM initiative has a good chance to be enacted, the same initiative would have a much better prospect of enactment if it was not sent to the PNM. Moreover, the raw comparison and conventional logistic regression analysis indicate a larger negative effect of the PNM as the estimates are contaminated by the intrinsic differences between the treated and untreated. The matching analysis affirms that the negative effect of the PNM is not spurious. By all accounts, while the PNM was invented as a solution to legislative stalemates, this solution has backfired as it has significantly undermined the chance of enactment for the initiatives that would have otherwise been enacted without the PNM. Table 2 Effect of the PNM on legislative success Raw comparison Logistic regression Propensity score matching Number of cases 2654 2654 948 Success rate for treated 0.57 0.57 0.62 Success rate for untreated 0.97 0.93 0.95 Difference in success rate −0.40*** −0.36*** −0.33*** Raw comparison Logistic regression Propensity score matching Number of cases 2654 2654 948 Success rate for treated 0.57 0.57 0.62 Success rate for untreated 0.97 0.93 0.95 Difference in success rate −0.40*** −0.36*** −0.33*** *** p<0.001. View Large Table 2 Effect of the PNM on legislative success Raw comparison Logistic regression Propensity score matching Number of cases 2654 2654 948 Success rate for treated 0.57 0.57 0.62 Success rate for untreated 0.97 0.93 0.95 Difference in success rate −0.40*** −0.36*** −0.33*** Raw comparison Logistic regression Propensity score matching Number of cases 2654 2654 948 Success rate for treated 0.57 0.57 0.62 Success rate for untreated 0.97 0.93 0.95 Difference in success rate −0.40*** −0.36*** −0.33*** *** p<0.001. View Large In order to ensure the balance in the matched data, I follow Rubin’s (2001) criteria to conduct a balance-checking analysis. The first indicator, also known as the Rubin’s Β, measures the absolute standardised mean difference of the propensity scores between the treated and untreated. A larger value of Β indicates worse balance. Rubin suggests that ‘the means must be less than half a standard deviation apart’ (Rubin, 2001, p. 174), and a value of 0.40 is deemed as minor enough (Rubin, 2001, p. 181). The overall value of Β in this study is 0.31, indicating a satisfactory level of balance in the propensity score. As for the balance in each covariate, the standardised mean difference (bias) between the treated and untreated is much smaller than 10% for most covariates. The only exception is Cross-party for which the standardised bias is 12.8%. The second criterion is that the Rubin’s R, which measures the ratio of the variances of the propensity scores in the treated and untreated groups, should be close to 1 and should be bounded between 0.5 and 2. Here, the variance ratio for an individual covariate ranges from 0.97 to 1.20. The overall variance ratio is a perfect 1.00. The third criterion is that the ratio of the residuals’ variances of the covariates must be from 0.5 to 2. The model specified in this article has met this criterion provided that the ratio for each covariate lies within the acceptable range. All together, these statistics indicate a satisfactory level of balance. I also carry out a sensitivity analysis using the Mantel-Haenszel (MH) test statistic (Aakvik, 2001; Becker and Caliendo, 2007).16 The purpose of this test is to examine whether the findings would sustain if there were any unobservable covariates that might alter the assignment of the treatments. A larger value of the test statistic (gamma) indicates a greater effect of unobservable covariates on the treatment assignment. If the result remains significant at a large value of gamma, it is to a large extent insensitive to unobservable covariates. The gamma here is as large as 8.5, meaning that the analysis is greatly insensitive to hidden biases even if they exist. The propensity score matching analysis demonstrates that the PNM lowers the chance of successful legislation in the LY. It assures that this effect does not result from the imbalance in the observational data of legislative initiatives. Moreover, necessary tests have been carried out to ensure the post-matching balance and robustness to misspecification. The analysis here provides valuable and robust evidence inferring an unfavourable policy-making consequence when a parliament adopts a formal mechanism of parliamentary negotiation. 4. Conclusion This article sets out to settle a theoretical controversy regarding whether parliamentary negotiations facilitate or hinder policy-making. It offers a propensity score matching analysis that addresses the issues of causal inferences faced by observational studies of legislative organisation. It contributes by far the most rigorous empirical evidence that the PNM, as a unique formal mechanism of parliamentary negotiation, decreases the chance of successful legislation. These findings further imply a dilemma of parliamentary negotiation: Negotiation is a tool that any parliaments would be tempted to utilise in times of legislative stalemates; however, the more negotiations, the less likely the legislation can make any headway. Furthermore, the evidence from Taiwan highlights the relative importance of within-parliament rules vis-à-vis constitutional arrangements in determining the levels of policy-making majoritarianism. From 2012 to 2015, the KMT occupied 57% of the seats and formed a one-party government. By the end of this parliamentary term, this majority government only enacted around 56% of its legislative attempts. Although the government-initiated bills would die at any stage, those buried in the PNM constitute 35% of all the failed ones. Among the government-initiated bills at the second reading, those treated by the PNM were enacted at a rate of 58%, while 98% of the untreated were enacted. By implication, the majority government would have enjoyed a much greater chance of legislative success without the PNM. Taiwan’s experience strongly suggests that failures to take into account within-parliament rules may result in misleading inferences about the policy-making performance in a certain democracy. While the PNM decreases the chance of legislative success in the LY, many negotiations were successful. To be exact, 57% of the PNM-treated initiatives before matching and 62% in the post-matching data may still be enacted. While it is intriguing to explore the variation among the PNM-treated initiatives, due to the limited space and lack of necessary official record, I have to leave this question open for future research. However, I offer several possible explanations here. First, some negotiations were successful because the negotiated initiatives were less controversial. As the veto player theory suggests, the distance between veto players’ ideal points is inversely associated with the likelihood of a unanimous vote (Tsebelis, 2002). Hence, the negotiation outcomes can be further explained once each party’s stance on each piece of legislation is available. Secondly, negotiations might proceed in various ways. For example, negotiation meetings can be chaired either by the speaker or by the corresponding committee conveners. With partial information, research has suggested that reaching a consensus is more likely in the former meetings than in the latter (Sheng and Huang, 2017); and among the latter meetings, the conveners from the majority party might be more successful in facilitating a deal than those from the minority (Chiou and Cheng, 2014). Unfortunately, the public record about the details of each meeting is not complete for systematic studies. All in all, it warrants future endeavours to collect necessary data in order to better account for the variation among the PNM-treated initiatives. Before concluding, the evidence from Taiwan shows that formal parliamentary negotiation does not serve well as a solution to legislative inaction. Although this article is no way an attempt to promote the majoritarian model of policy-making, it does suggest the necessity of a thorough consideration before any parliaments adopt negotiation as a formal decision-making mechanism. Acknowledgements I would like to thank Dr. Sing-Yuan Sheng in National Chengchi University for her helpful assistance in data collection. I am also grateful for thoughtful advices from Sing-Yuan Sheng, Eric Chang, Nathan F. Batto, Gisela Sin, Kharis Templeman, Yi-ting Wang and Wen-chin Wu. Nevertheless, I take full responsibility for this paper. Finally, I declare no conflict of interest. Footnotes 1 South Korea has a formal negotiation mechanism. Twenty members of the National Assembly can form a floor negotiation group whose floor leaders negotiate over the agenda and procedural matters, while the final decisions over legislation are made through a voting procedure. 2 Although laboratory experiments can be conducted (Fréchette et al., 2003, 2005), experiments with a randomised treatment assignment in real parliamentary settings are unlikely. 3 The main conclusion is corroborated by a supplementary analysis using Coarsened Exact Matching (CEM), which is promoted by recent literature (Blackwell et al., 2009; Iacus et al., 2012; King and Nielsen, 2016). Due to a large number of confounding variables, using CEM leads to a limited number of matched initiatives. To be exact, 95% of initiatives are pruned. Given this limitation, the CEM analysis is presented as a Supplementary Material. 4 A problem with their study is that among the 1926 initiatives that had never been sent to the PNM, 1197 were either in the first reading or in the committees. In other words, they compared initiatives at different stages, and a majority of the initiatives they analysed were too premature to be sent to the PNM and to be enacted. This explains why they find that the initiatives untreated by the PNM had a lower chance of enactment. 5 There have been criticisms against the closed-door process of negotiation (Wang, 2002, 2014) and the harm to the function of the standing committees (Yang, 2002; Yang and Chen, 2004). 6 Please see the Law Governing the Legislative Yuan's Power, Article 68. 7 Although the legal period of negotiation is one month, in practice, the negotiation may simply continue until a consensus is reached. 8 The pattern in the fifth and seventh LY is similar (Hawang and Ho, 2007; Huang and Sheng, 2017; Sheng and Huang, 2017). Thus, the legislative performance here is not a special case. 9 I list the ten laws in Supplementary Table 1. 10 The number of news reports for each initiative is obtained from an archive maintained by the Parliamentary Library (2015). 11 In Taiwan’s LY, any party with three or more members can form a party caucus that is allowed to propose bills without any cosponsors. 12 In Taiwan’s five-power constitutional system, the governing power is shared by five branches. In addition to the executive and legislative branches, the other three branches may send their own bills into the LY. 13 In each standing committee, there are two conveners taking turns chairing the meetings. Each of them is allowed to set their own agenda. 14 Given that the regression coefficients are not of the central concern in this article, the result of the logistic regression is not presented. However, it is available in Supplementary Table 2. 15 Matching with replacement allows each untreated case to be matched with more than one treated cases. Compared to matching without replacement, this helps increase the number of matched cases, and the matching process is not conditional upon the order of the cases (Stuart, 2010). Additionally, through a try-and-error process, I set the caliper to be 0.00042 to achieve good balance. 16 Another sensitivity test is using the rbounds, which is invented for continuous outcome variables based upon Rosenbaum (2002). References Aakvik A. ( 2001 ) ‘Bounding a Matching Estimator: The Case of a Norwegian Training Program’ , Oxford Bulletin of Economics and Statistics , 63 , 115 – 143 . Google Scholar Crossref Search ADS WorldCat Armingeon K. ( 2002 ) ‘The Effects of Negotiation Democracy: A Comparative Analysis’ , European Journal of Political Research , 41 , 81 – 105 . Google Scholar Crossref Search ADS WorldCat Batto N. , Tsai Y. -C. , Weng T. -W. ( 2016 ) ‘Parliamentary Brawls, Media Coverage, and Re-Election in Taiwan’. Paper presented to the 2016 Annual Meeting of the American Political Science Association, Philadelphia, 2016. Becker S. , Caliendo M. ( 2007 ) ‘Mhbounds - Sensitivity Analysis for Average Treatment Effects’ , Stata Journal , 7 , 71 – 83 . Google Scholar Crossref Search ADS WorldCat Binder S. ( 2003 ) Stalemate: Cause and Consequence of Legislative Gridlock , Washington, DC , Brookings Institution Press . Google Preview WorldCat Binder S. , Lee F. ( 2015 ) ‘Making Deals in Congress’. In Mansbridge J. , Martin C. (eds) Political Negotiation , Washington, DC , Brookings Institution Press , pp. 91 – 120 . Google Preview WorldCat Blackwell M. , Iacus S. , King G. , Porro G. ( 2009 ) ‘Cem: Coarsened Exact Matching in Stata’ , The Stata Journal , 9 , 524 – 546 . Google Scholar Crossref Search ADS WorldCat Casar M. A. ( 2002 ) ‘Executive-Legislative Relations: The Case of Mexico (1946-1997)’. In Morgenstern S. , Nacif B. (eds) Legislative Politics in Latin America , Cambridge , Cambridge University Press , pp. 114 – 144 . Google Preview WorldCat Chiou F. -Y. , Cheng Y. -Y. ( 2014 ) ‘The Party Negotiation Mechanism in the Legislative Yuan: Minority Dominance or Majority Strength?’ , Political Science Review , 14 , 155 – 194 . WorldCat Figueiredo A. C. , Limongi F. ( 2000 ) ‘Presidential Power, Legislative Organization, and Party Behavior in Brazil’ , Comparative Politics , 32 , 151 – 170 . Google Scholar Crossref Search ADS WorldCat Fréchette G. , Kagel J. , Lehrer S. ( 2003 ) ‘Bargaining in Legislatures: An Experimental Investigation of Open versus Closed Amendment Rules’ , American Political Science Review , 97 , 221 – 232 . Google Scholar Crossref Search ADS WorldCat Fréchette G. , Kagel J. , Morelli M. ( 2005 ) ‘Gamson's Law versus Non-Cooperative Bargaining Theory’ , Games and Economic Behavior , 51 , 365 – 390 . Google Scholar Crossref Search ADS WorldCat Gelman A. , Hill J. ( 2007 ) Data Analysis Using Regression and Multilevel/Hierarchical Models , New York , Cambridge University Press . Google Preview WorldCat Hawang S. -D. , Ho S. -T. ( 2007 ) ‘Party Caucuses Negotiation and Legislation Process: Analysis of the Fifth Legislative Body in Taiwan’ , Political Science Review , 34 , 1 – 44 . WorldCat Horowitz D. ( 2014 ) ‘Ethnic Power Sharing: Three Big Problems’ , Journal of Democracy , 25 , 5 – 20 . Google Scholar Crossref Search ADS WorldCat Huang S. -H. ( 2017 ) ‘Who Wants Issue Ownership? Legislators' Initiatives and Pursuit of Issue Ownership in Taiwan’ , Taiwan Democracy Quarterly , 14 , 1 – 51 . WorldCat Huang S.-H. , Sheng S. -Y. ( 2017 ) ‘Decentralized Legislative Organization and Its Consequences for Policy-making in the Ma Ying-jeou Era’. Paper presented to the Annual Conference on Taiwan Democracy, Stanford, 2017. Iacus S. , King G. , Porro G. ( 2012 ) ‘Causal Inference without Balance Checking: Coarsened Exact Matching’ , Political Analysis , 20 , 1 – 24 . Google Scholar Crossref Search ADS WorldCat Jann B. ( 2017 ) ‘Decentralized Legislative Organization and Its Consequences for Policy-making in the Ma Ying-jeou Era’. Paper presented to the 2017 German Stata Users Group Meeting, Berlin, 2017. King G. , Nielsen R. ( 2016 ) ‘Why Propensity Scores Should Not Be Used for Matching’, accessed at http://j.mp/1sexgVw on 10 October 2016. Lo C.-H. ( 2006 ) Meeting Management and Legal Regulations , Taipei, Wunan . Google Preview WorldCat Lijphart A. ( 1999 ) Patters of Democracy , New Haven , Yale University Press . Google Preview WorldCat Lijphart A. ( 2004 ) ‘Constitutional Design for Divided Societies’ , Journal of Democracy , 15 , 96 – 109 . Google Scholar Crossref Search ADS WorldCat McCubbins M. , Noble G. ( 1995 ) ‘The Appearance of Power: Legislators, Bureaucrats, and the Budget Process in the United States and Japan’. In Cowhey P. , McCubbins M. (eds) Structure and Policy in Japan and the United States , Cambridge , Cambridge University Press , pp. 56 – 80 . Google Preview WorldCat Nalepa M. ( 2016 ) ‘Party Institutionalization and Legislative Organization: The Evolution of Agenda Power in the Polish Parliament’ , Comparative Politics , 48 , 353 – 372 . Google Scholar Crossref Search ADS WorldCat Nunnari S. ( 2017 ) ‘Dynamic Legislative Bargaining with Veto Power: Theory and Experiments’, Working paper. Parliamentary Library ( 2015 ) ‘Legislative Yuan Database’, accessed at https://npl.ly.gov.tw/do/www/dbMaintainNew?dbKind=%E7%AB%8B%E6%B3%95%E9%99%A2%E6%99%BA%E5%BA%AB on 13 May 2018. Rosenbaum P. ( 2002 ) Observational Studies , New York , Springer . Google Preview WorldCat Rosenbaum P. , Rubin D. ( 1983 ) ‘The Central Role of the Propensity Score in Observational Studies for Causal Effects’ , Biometrika , 70 , 41 – 55 . Google Scholar Crossref Search ADS WorldCat Rosenbaum P. , Rubin D. ( 1985 ) ‘Constructing a Control Group Using Multivariate Matched Sampling Methods That Incorporate the Propensity Score’ , American Statistician , 39 , 33 – 38 . WorldCat Rubin D. ( 2001 ) ‘Using Propensity Scores to Help Design Observational Studies: Application to the Tobacco Litigation’ , Health Services and Outcomes Research Methodology , 2 , 169 – 188 . Google Scholar Crossref Search ADS WorldCat Selway J. , Templeman K. ( 2012 ) ‘The Myth of Consociationalism? Conflict Reduction in Divided Societies’ , Comparative Political Studies , 45 , 1542 – 1571 . Google Scholar Crossref Search ADS WorldCat Sheng S.-Y. , Huang S.-H. ( 2015 ) ‘Decentralized Legislative Organization and Its Consequences for Policymaking’. Paper presented to the Annual Conference on Taiwan Democracy, Stanford, 2015. Sheng S.-Y. , Huang S.-H. ( 2017 ) ‘Party Negotiation Mechanism: An Analysis through the Lens of Institutionalization’ , Soochow Journal of Political Science , 35 , 37 – 92 . WorldCat Stuart E. ( 2010 ) ‘Matching Methods for Causal Inference: A Review and a Look Forward’ , Statistical Science , 25 , 1 – 21 . Google Scholar Crossref Search ADS PubMed WorldCat Tsebelis G. ( 2002 ) Veto Players: How Political Institutions Work , Princeton : Princeton University Press . Google Preview WorldCat Wang J.-P. ( 2003 ) ‘The Legislative Yuan under a Multiparty System: On the Operation of Party Caucuses in the Legislative Yuan’ , Legislative Yuan Newsletter , 31 , 11 – 21 . WorldCat Wang Y.-L. ( 2002 ) ‘Reinventing an Ideal Environment for Constitutional Practices: Ideas to Reform Electoral System, Legislative Process and Party Negotiation Mechanism’. In Chen Long-Zhi (ed.) Proceeding of a Conference on New Era and New Constitution , Taipei, Angle , pp. 331 – 349 . Google Preview WorldCat Wang Y.-L. ( 2014 ) ‘On Party Negotiation Mechanism in the Legislative Yuan’, End-of-Term Report for the 2013 Research Project of the Center for Public Policy and Law. Warren M. , Mansbridge J. ( 2013 ) ‘Deliberative Negotiation’. In Mansbridge J. , Martin C. (eds) Negotiating Agreement in Politics , Washington, DC , American Political Science Association , pp. 86 – 120 . Google Preview WorldCat Yang W.-Y. ( 2002 ) ‘The Role of the Committee in the Legislative Decision-Making Process: The Third Term Legislative Yuan’ , Issues & Studies , 41 , 83 – 113 . WorldCat Yang W.-Y. , Chen T.-W. ( 2004 ) ‘The Transformation and Evaluation of the Inter-Party Negotiation System after the Congressional Reforms’ , Soochow Journal of Political Science , 19 , 111 – 150 . WorldCat © The Author(s) 2018. Published by Oxford University Press on behalf of the Hansard Society; all rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
Journal
Parliamentary Affairs – Oxford University Press
Published: Jul 1, 2019