Abstract
Do upper houses matter for cabinet formation? If so, does it make a difference how powerful they are? The most sophisticated study to give clear answers to these questions is still that of Druckman, Martin and Thies (2005), which argues that upper houses influence cabinet formation, regardless of their formal powers. We dispute these answers theoretically and empirically. Since adjusting cabinet formation to second chamber composition is not necessary and can be costly, we should expect that if upper houses matter at all, their formal powers should matter too. Replication shows that the empirical results are driven exclusively by two countries with weak upper houses, one of which (Ireland) is miscoded. If this case is recoded there is no statistically or substantially significant effect of upper houses, regardless of their power. We conclude that the two questions cannot yet be answered: we need more research based on broader samples in order to assess whether upper houses matter for cabinet formation and whether their formal powers make a difference.
Introduction
Do upper houses matter for cabinet formation? If so, does it make a difference how powerful they are? Constitutional designers of parliamentary systems would profit from robust scientific answers to these questions. Some might embrace upper houses in the hope of generating more diverse and inclusive cabinets, thus fostering the ‘consensual’ character of democracy (Lijphart, 2012). Others might try to avoid any effect of upper houses on cabinet formation, so that lower house elections could be organized as choices between clearly identifiable cabinet alternatives (Powell, 2000). Whichever normative view one prefers, the practical consequences of parliamentary design depend on empirical facts.
The most sophisticated study to give clear answers to our questions is still that of Druckman, Martin and Thies (2005) (hereafter DMT): the study argues that, yes, upper houses matter for cabinet formation, and no, how powerful they are does not make a difference. DMT use conditional logit models to study cabinet formation in eight democracies. They conclude, first, that cabinet alternatives that control upper house majorities are more likely to form than those that do not, everything else being equal. Second, they argue that this is true regardless of how constitutionally powerful upper houses are. Given the practical importance of these two conclusions, it is useful to ask how theoretically plausible and empirically robust they are.
In this note we question DMT’s findings. First, we argue that the combination of answers proposed is theoretically implausible (section 2). If rational coalition-builders make special efforts to gain upper house majorities, despite the significant costs of doing so, they can also be expected to carefully consider the constitutional powers of these chambers. Second, we argue that DMT’s data does not support their conclusions (section 3). The results are driven exclusively by the two cases with weak upper houses, one of which (Ireland) is coded misleadingly. If this case is recoded, there is no statistically or substantially significant effect of upper houses. We conclude that further research should apply DMT’s pioneering empirical approach to broader samples (section 4).
Theory
DMT rightly criticize many coalition-formation theories for assuming parliaments to be ‘little more than loyal rubber stamps of government initiatives’ (p. 334). While this assumption is reasonable in many contexts, it becomes problematic for minority cabinets and bicameralism. For the latter case, DMT argue that if upper houses can affect the success of government policies, and possibly induce their early termination, then coalition-builders have an incentive to build coalitions that control a majority in the upper house. DMT thus hypothesize that ‘of all the potential governments that might arise in a given formation episode, those with upper-house majorities are more likely to form, everything else being equal’ (p. 535). Moreover, they suggest that this is true even if an upper house has only weak legislative powers (p. 534).
The problem with this reasoning is that it neglects the costs of achieving majorities in upper chambers. These costs come mainly in the forms of a higher number of cabinet parties, greater ideological divisions or oversized cabinet status. If we consider these costs theoretically, we see that they can provide coalition-builders with strong incentives not to control a second chamber majority. As we know from the literature on minority cabinets, cabinets do not necessarily need a fixed legislative majority to secure the passage of their programme. Indeed, avoiding a fixed majority can have significant advantages, such as the freedom to form different coalitions on different issues, which may greatly reduce the costs of coalition-building (Tsebelis, 2002: 111–114). Moreover, if the policy space is essentially unidimensional, it would be sufficient to include the upper house median into the cabinet rather than seeking upper house majority status (Eppner, 2014).
Considering the costs of upper chamber majorities is important, because rational coalition-builders that seek a balance between costs and benefits will be very attentive to the formal powers of upper houses. The more powerful a house, the greater should be the incentive to secure majorities within it (either by fixing this majority or by controlling the median). DMT (p. 541) note research that shows how even formally weak upper houses can influence legislation under certain conditions (e.g. Tsebelis and Money, 1997). This is true as far as it goes, but accepting that weak upper houses have the potential for some influence some of the time is very different from assuming that they have the same average causal effect as strong upper houses. Indeed, most of the empirically successful literature on the legislative effects of institutional ‘veto players’ makes sharp distinctions based on formal powers (Tsebelis, 2002). If formal powers matter for the explanation of the legislative effects of second chambers, however, it would be paradoxical if they did not matter for the explanation of cabinet formation.
We believe that there is no paradox. Most of the qualitative evidence on the effects of upper houses on cabinet formation does focus on powerful veto players – as do DMT themselves with their illustrative example of Japan (pp. 531–532). And, as we show next, there is no robust evidence in DMT’s study for any causal effect of upper houses. The study has not shown that the constitutional powers of upper houses do not matter.
Replication
DMT use the dataset by Martin and Stevenson (2001), but focus only on the 8 of 14 countries with upper chambers (p. 535). This is a small basis for drawing general conclusions, but we stick to the sample for the sake of comparable results.
DMT distinguish between countries with symmetric and asymmetric bicameralism (Lijphart, 2012). Symmetric bicameralism means that the upper house has absolute veto power (on all or most important legislation) and sufficient democratic legitimacy to use it. If the upper house possesses only suspensory veto power or lacks democratic legitimacy, or both, bicameralism is asymmetric. Two bicameral systems in the sample were asymmetric (Ireland [1973–86] and Austria [1949–82]) and six were symmetric (Belgium [1946–85], Denmark [1949–53], Germany [1949–87], Italy [1949–87], the Netherlands [1948–86] and Sweden [1945–70]).
DMT also follow Martin and Stevenson in excluding all situations in which single parties won a majority (p. 544, n. 9). 1 DMT repeat the conditional logit analysis of Martin and Stevenson for the reduced sample and add a dummy variable for ‘Upper-Chamber Majority’. 2 We were able to replicate the results of their conditional logit analysis (DMT’s model 4), 3 given in column (2) of Table 1. 4 The statistically significant coefficient of 1.11 implies an odds ratio of 3.03, which means that majority status in the upper house triples the chances for a potential government to form.
Replication of DMT’s conditional logit models.
t statistics in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01.
To test for the relevance of upper house powers, DMT also interact the majority status variable with an indicator for whether or not the upper chamber has constitutionally granted veto power. They do not present the results, but report that they did not find any evidence for an interaction effect. They conclude that ‘upper chambers matter for government coalition builders, even if they are possessed of few formal powers’ (p. 542).
DMT’s interaction model is shown in column (1) of Table 1 (model 1). Before we comment on the upper house majority variable, note that several variables capture the costs of building inclusive cabinets and that the results are in line with our theoretical reasoning. Increasing the number of cabinet parties and increasing the ideological differences within the cabinet both reduce the likelihood that a proto-coalition actually forms, everything else being equal. The estimated negative effect of oversized coalitions also underlines the costs of inclusiveness, but it is not statistically significant. This evidence is in line with rationalist theories.
In contrast, DMT’s findings on the impact of upper houses seem to contradict these theories. In model 1 the variable ‘upper house majority’ is interacted with a dummy that is 0 for all upper houses with absolute veto power, and 1 for those without. 5 The results imply that symmetric upper houses have no influence on government formation. The overall estimated effect of upper houses estimated by the authors is driven exclusively by the (two) weak upper houses of Austria and Ireland. 6 To show this more clearly, we estimate DMT’s main ‘model 4’ (Table 1, column (2)) only for the six countries with symmetric upper houses (column (3)). We see that the coefficient is almost cut in half and not statistically significant. As argued above, if these results were robust, they would contradict rationalist theories. Rational coalition-builders should be more, rather than less, willing to take upper houses into account, despite the potential costs of doing so, when these houses are powerful.
Closer inspection of the data shows, however, that the results are not robust. Instead, the evidence suggests that upper houses do not matter at all, at least in DMT’s sample. The reason is that the Irish case adds only four government formations to the analysis (1981, two in 1982, 1987), and removing those four is sufficient for the coefficient to become insignificant (column 4). This fact is highly problematic for two reasons. First, the idea that the Irish upper house (Seanad) affects cabinet formation is not corroborated by in-depth case evidence (Coakley and Laver, 1997; Gallagher, 2010; Manning, 2010). 7 The composition of the Irish Senate is not only similar to that of the first chamber, but a newly elected prime minister has the right to nominate 11 out of the 60 upper house members to ensure a government majority in the second chamber. Hence, ‘rather than the Seanad having any impact on the composition of the government, it is currently the government … that has an impact on the composition of the Seanad’ (Coakley and Laver, 1997: 61). Second, DMT’s coding of this case distorts this procedural reality. For each Irish formation opportunity, they code the Senate composition after the new prime minister’s nominations have taken place. It is thus hardly surprising that all actually formed governments held upper chamber majorities, while few alternative proto-coalitions could count on majorities in the Senate.
Column 5 shows DMT’s ‘model 4’ with more adequate coding. We gathered data on upper house composition and assigned each Irish proto-coalition its potential upper house status (should it form a government and be able to nominate members in the newly elected Senate). Simply re-coding the four Irish cabinet formations in this way is sufficient for the coefficient to be substantially reduced and lose statistical significance, even if all eight countries are included in the analysis (compare columns (2) and (5)).
Conclusion
In sum, DMT provide no convincing evidence that upper houses affect government formation. On the basis of their sample, we would have to conclude that neither symmetric nor asymmetric bicameralism affects cabinet formation. It would clearly be premature to draw firm conclusions, however, as their analysis includes only eight bicameral systems. Other studies which cover larger sets of countries and focus exclusively on symmetric bicameralism do report evidence that upper houses affect cabinet formation (Ganghof, 2010; Volden and Carrubba, 2004), yet these studies do not use conditional or mixed logit models. The literature simply lacks robust answers to questions of whether upper houses affect cabinet formation and how their formal powers and democratic legitimacy mediate the relationship.
Future research should follow in DMT’s footsteps and apply conditional or mixed logit models to larger and more diverse samples. Our theoretical expectation is that if upper houses do matter for cabinet formation – and it is not theoretically obvious that they do – their formal powers (and democratic legitimacy) must also matter, because rational coalition-builders care about the costs of inclusiveness. This reasoning also implies the possibility that coalition-builders might content themselves with controlling the upper house median rather than an upper house majority. These different theoretical possibilities can only be arbitrated empirically. Despite DMT’s pioneering effort, much work remains to be done before we might be able to provide constitutional designers with reliable conclusions.
Footnotes
Declaration of Conflicting Interest
The authors declare that there is no conflict of interest.
Funding
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.