Saylor.org's Comparative Politics/Political Alliances and Cleavages

=Legislative Party Discipline and Cohesion in Comparative Perspective= by Sam Depauw and Shane Martin

It is hard to envisage representative government, save in terms of unified political parties. Legislative voting unity is a precondition for responsible party government. Existing scholarship has focused extensively on explaining patterns of unified party voting within legislatures by references to presidential versus parliamentary forms of government (see, for example, Bowler et al. 1999; Carey 2007; Tsebelis 2002).

Institutions associated with parliamentary systems, such as the vote of confidence mechanism, are said to enhance party voting unity (Huber 1996). Explanations of variation in party voting unity across parliamentary regimes have been limited. Our aim, beyond a mere description of the behaviour of legislators in casting floor votes, is to build on the scarce exceptions that attempt to link party unity in the legislature and the varying degree to which electoral and other institutions shape the behaviour of legislators (Carey 2007; Hix 2004; Hix et al. 2005; Sieberer 2006), and progress towards a general, comparative framework that allows us to explain variation in the level of party voting not just between different political systems but also between parties operating in the same political system. The institutions which we focus on are the electoral system, the candidate selection system and the opportunities that party leaders have to promote legislators to higher political office.1

Notwithstanding recent attempts to introduce a comparative approach to understanding party unity the problems with this existing body of knowledge are manifold. Most analysis has tended to employ only system (country) level variables. While the unit of analysis should typically be at the level of the individual legislative party the institutional explanations posited are at a different, higher level. For one thing, this eliminates the possibility of explaining differing levels of voting unity among political parties in the same legislature.

Perhaps even more damaging has been the lack of cross-national data on legislator voting behaviour. Even the Döring project that did so much to uncover and report data on so many aspects of legislative politics in Europe was nevertheless unable to systematically collect data on voting unity (Saalfeld 1995a:557). Even for those legislatures where votes are commonly recorded, the records are not made easily available (Carey 2007).

Another possible explanation for the dearth of cross-national research on the topic is the controversies surrounding the most commonly used indicator of party unity, Rice’s Index of Cohesion. The index of cohesion is computed as the absolute difference between the proportion of party members voting in favour and the proportion of party members voting in opposition, multiplied by 100 to obtain a number ranging from 0 to 100.

It is worth repeating and attempting to deal with some of the controversies before beginning our analysis. First, recorded votes are not a random selection of votes (Carrubba et al. 2006; Hug 2005; Saalfeld 1995a). Recorded votes are typically called for by party leaderships for reasons of disciplining or signalling: to allow their party’s legislators to be monitored or to denounce important differences of opinion in the other parties. Both reasons, however, can be expected to have opposite effects on party voting unity scores. On a related issue, as recorded votes increase in number, they tend to include more minor matters (e.g. resolutions, amendments) and therefore to exhibit more unity: on those minor matters only those legislators most interested in leadership positions will attend and they are more likely to toe the party line (Carrubba et al. 2006). Indeed, Hug (2005) notes that party unity scores are higher for those votes in the Swiss Parliament that are automatically recorded: e.g. final votes or votes on urgency measures.

Second, the index of cohesion tends to overestimate unity in smaller parties. A majority of members voting ‘the wrong way’ (i.e. against the party line) pushes cohesion upward and this is more likely to happen in small parties. Yet, the bias appears to decrease as parliamentary party group size exceeds a minimal number of members and groups are more cohesive – both of which apply to our sample of parties (Desposato 2005). Third, interpreting non-votes and abstentions is by no means straightforward – the option of abstention is not recorded in all legislators for instance. Excluding both non-votes and abstentions is the more conservative option when attempting to measure voting unity, however (Cowley and Norton 1999), and this is the approach we employ here. Finally, Krehbiel points out that the Rice index cannot discriminate between situations of perfect and no party discipline at all. That is, the index does not take into account legislators’ preferences. Under conditions of perfect discipline, legislators vote together even when their preferences diverge, while under conditions of no discipline legislators may still vote together but only when their preferences converge (Krehbiel 1993, 2000).

Country·Period covered·No. of Parties·Mean·Stdev
Australia 1996-98→3→99.07→0.15

Austria 1995-97→5→98.68→1.45

Belgium 1991-95→9→99.06→0.75

Canada 1994-95→4→97.60→2.24

Denmark 1994-95→7→99.93→0.11

Finland 1995-96→7→88.63→2.59

France 1993-97→4→99.33→0.63

Germany 1987-90→3→96.33→1.79

Iceland 1995-96→6→96.93→2.84

Ireland 1992-96→3→100.0→0.00

Israel 1999-00→10→96.88→1.15

Italy (1st Republic) 1987-92→9→97.52→1.60

Italy (2nd Republic) 1996-01→11→96.46→1.44

New Zealand 1993-94→2→93.17→0.65

Norway 1992-93→6→95.90→0.52

Sweden 1994-95→7→96.57→1.51

United Kingdom 1992-97→2→99.25→0.49

In what follows we explain how variation in key political institutions which shape the behaviour of legislators will likely have an impact on the level of observed party voting unity. Using a mix of party-level and system-level data we then empirically test the arguments that the design of political institutions affects party voting unity.

We compile or bring together data on the voting behaviour of legislators in over 90 parties in 16 legislatures. As we can see from table 6.1 party voting unity tends to be lowest in Finland and highest in Ireland and Denmark. Combining our voting unity data with system and party-level data permits a theoretical and empirical analysis of the variation in legislative voting unity between parties that has not been possible to date. We conclude the chapter with a review of our findings and suggestions for future research in the area.

The Electoral System, Personal Vote and Party Voting Unity
While the shape, origin and consequences of different electoral rules are generally well documented, their impact on legislative behaviour, most notably on party unity in legislative votes, is not always well understood. For example, German legislators elected via single member districts choose different legislative committee assignments than legislators elected under the party list (Stratmann and Baur 2002). Cox and McCubbins (2007) argue that the ties that bind candidates’ electoral fates together are responsible for party unity. These ties reflect the party reputation based on the state of the economy, major pieces of legislation and in their argument the reputation of the president. Legislators are ready to comply with party unity when an unfavourable party reputation might seriously damage their own electoral prospects. Such an unfavourable party reputation might result from overspending, as legislators chase pork-barrel benefits for their constituencies, or even from open in-fighting in the legislature. But when candidates cannot hope to benefit from spill-over votes from copartisans, they will focus on cultivating a personal vote. In those circumstances, they are more inclined to point out differences with their party than legislators whose electoral incentives are more aligned with their party.

Depending on the ballot structure, legislators have varying incentives to appeal to voters over party leaders. In more candidate-centred electoral environments, incumbent politicians will actively respond to and build personal relations with individual constituents in their district. In more party-centred electoral systems, incumbents focused on re-elections have greater incentives to cultivate favour with their party leadership in the hopes of securing a prominent position on the party list. Carey and Shugart (1995) offer such a method to rank-order electoral systems according to the value of a personal vote on the basis of the interaction between ballot control, vote pooling, and type of votes on the one hand and district magnitude on the other hand. Where intra-party competition is present, greater district magnitude increases the need for a personal vote as the number of co-partisans on the list increases. Yet, when intra-party competition for votes is absent, the possibility of a personal vote decreases as district magnitude grows.

The presence of such intra-party competition is defined by ballot control, vote pooling, and type of votes. Ballot control refers to the degree of control district-level party leaders have over access to the party label and voters’ ability to upset their proposed list. The pooling of votes indicates whether votes for one candidate also contribute to the number of seats won by other candidates of the same party. The type of votes is determined by the form of the ballot paper that voters are presented with – voters may vote for a party, for multiple candidates, or for a single candidate. As voters may only vote for a single candidate (vote), those votes are not pooled (pool), and those votes do ‘upset’ the party list (ballot), the intra-party competition increases and candidates search for a personal vote – if need be by voting against the party line (Carey and Shugart 1995).

With district magnitude, the intra-party competition increases and candidates are forced to seek out a personal vote – that is, when the ballot structure allows for such competition. On the other hand, with district magnitude, the information demands on voters, too, increase rapidly. Voters can hardly keep up with voting records of multiple incumbents. District magnitude, thus, might have a different impact depending on the type of vote. In closed-list systems, district magnitude increases party unity. In open-list systems, party unity decreases with district magnitude. But in those circumstances, an independent voting record may not be the only, or even the most effective, means to court a personal vote. Shugart et al. (2005) argue that district magnitude increases the number of candidates who have local roots or have served in local elected positions within the district in ‘pure’ open-list systems: social characteristics become more important as candidates hope to attract personal support.

Despite the seminal character of Carey and Shugart’s contribution, research on the relationship between ballot structure and voting unity has yielded only mixed empirical success. Focusing on the European Parliament, Hix (2004) finds a relationship between voting unity within the party group and the electoral system by which the MEP was elected (see also, Hix et al. 2005). Sieberer (2006) argues that incentives to cultivate personal votes should be associated with lower unity in the parliamentary party group. Differentiating between three categories of electoral systems Sieberer (2006) finds that voting unity is marginally stronger in candidatecentred than party-centred electoral environments. However, an intermediate electoral environment creating mixed incentives for personal vote and party vote cultivation is most strongly associated with higher voting unity, questioning the validity of the argument that voting unity is a function of electoral rules and in particular the need to cultivate personal votes. More recently, Carey (2007) reaches a different conclusion, finding evidence that the level of intra-party electoral competition, considered a defining feature of personal-vote electoral systems, helps explain variation in voting unity. Given the theoretical interest in the effect of electoral rules on party unity and the only mixed evidence that such relationships withstand empirical scrutiny, we attempt to measure more accurately the effect of ballot structures on party voting unity.

One reason for these mixed results may be that the interaction effect at the heart of Carey and Shugart’s thinking renders operationalisation more difficult. A second reason regards the uncertainty surrounding single-member district plurality systems.

Carey and Shugart code SMDs among the systems least encouraging the development of intra-party competition and therefore a personal vote, while Wallack et al. (2003) maintain that there is room for a personal vote in those circumstances and code SMDs accordingly. Both appear to be right: the search for a personal vote in SMDs is not inspired by intra-party competition (at least not in any single election), but by the necessity to court the median voter in the district. As long as the opinions of the local median voter sufficiently differ from the national median voter, there might be a reason for MPs to dissent. Finally, the ballot indicator combines a characteristic of the electoral system with one of the party selection process. On the electoral system level, ballot indicates whether votes for candidates can actually ‘upset’ the party list. On the party level, ballot captures whether party leaders can present lists at all. The latter aspect might in fact be better captured by the candidate selection process.

In sum, we suggest that political parties who operate in electoral systems that provide less incentive to cultivate the personal vote will be more likely to have higher levels of unified legislative voting than political parties operating under electoral rules where electors choose between individual candidates rather than political parties. Where a difference exists between the preference of constituents (the median constituent or an electorally significant sub-constituency) and the party leadership we would expect the electoral system to shape the voting decision of the legislator to vote with or against the party.

Candidate-Selection and Party Voting Unity
The process by which candidates for legislative office are selected and/or re-selected remains one of the most overlooked aspects of politics (Gallagher and Marsh 1988; Rahat and Hazan 2001). While, as we discussed above, attention has focused on the nature and impact of electoral systems much less is known about how candidate reselection procedures impact the behaviour of individual legislators. Yet, if re-election is the goal of incumbent legislators then the proximate aim is to get re-selected as a candidate – in effect to secure access to the ballot, or as high as possible a position under list electoral systems. We should note that the critical issue here relates not just to ballot access but the ability to be associated with the party label. An incumbent may easily access a ballot by paying a registration fee and/or collecting signatures; we are primarily interested in how much the party leadership controls access to the party label for prospective candidates. In a general sense, as Strøm (1997) was one of the first to note, what an incumbent must do to be reselected is likely to influence their legislative strategies and role orientation.

Of course, processes of candidate selection are complex undertakings, involving many dimensions and even more actors. Rahat and Hazan (2001) have argued that at least the dimensions of inclusiveness and centralisation should be separated. Inclusiveness of the process refers to the number of actors that are part of the selectorate. Centralisation, on the other hand – and this is the key concern here – regards the degree of control the central party leadership has over the (re-)selection processes visà- vis other actors in the process, most commonly local party executives.

Indeed, much of the impact of the ‘party-centeredness’ of electoral rules may be logically attributed to candidate selection procedures and in particular the risk of being deselected by the national party leadership. Carey (2007), for instance, found party unity to be lower in both presidential and parliamentary systems where legislative candidates compete against co-partisans for personal votes. But he effectively contrasted parties where candidates compete against co-partisans for personal votes with parties where nominations are controlled by party leaders. In fact, Poiré (2002: 21) reported that electoral rules fail to predict party unity in over 60 political parties in the 1950s and 1960s, when candidate selection procedures are included. Hix (2002: 20), on the other hand, concluded that the defection rate of MEPs from their national parties is more affected by candidate-centred rules than decentralised selection procedures. The latter effect is in the predicted direction, but not statistically significant. Sieberer (2006) finds that party voting unity is slightly higher in parties where the leadership has some formal control over candidate selection, and that candidate selection is a better predictor of party voting unity than electoral rules.

Building on this body of research and unclear empirical results we predict a direct causal link between the degree of control party leaders exert over the candidate reselection process and the level of unified party voting. Lundell (2004) developed a five-point ordinal scale to measure this degree of centralisation. Essentially it is a reduced version of Janda’s nine-point scale, collapsed over the inclusiveness dimension (Janda 1980).2 In our empirical analysis, Lundell’s data on candidate selection rules are supplemented with information from Gallagher and Marsh (1988), Gallagher et al. (2005) and Narud et al. (2002) – in particular on countries that have legally regulated candidate selection procedures: Finland, Germany, and Norway.

Detailed information on the inclusiveness of selectorates is generally lacking. Yet, something of its impact can be found in the impact of the membership organisation. Ozbudun (1970) distinguished two strands of the argument. The first emphasises that party unity is greater in mass membership parties than in parties where the membership organisation is not the dominant decision-making centre. The second maintains that a mass membership is sufficient – dominant or not in the party. On the other hand, as the proportion of the party electorate that is also a member of the party increases, party unity is expected to decrease: mass membership is not only a unifying force, it is also likely to be more diverse and thus provide dissenting members cover. Members at the party’s more extremist wings often claim to be loyal to the party’s orthodoxies when they dissent.

Opportunities for Promotion and Party Voting Unity
The motivation of legislators may very well extend beyond the desire to get reselected or re-elected (Strøm 1997). For example, legislators may feel secure in the knowledge that they will be re-selected or re-elected. More probably, it could be argued that once elected, legislators in parliamentary systems are strongly motivated by the desire to gain leadership positions within the party, which they hope would ultimately lead to a ministerial seat (Huber and Shapin 2002:197). In parliamentary systems the executive, by which we mean prime minister, cabinet and junior ministers, typically emerges from and is populated by members of the legislature (Gallagher et al. 2005). This is at odds with presidential government, where separation of powers requires that the head of executive be directly elected and the executive cabinet be composed of non-legislators. The difference in approach to staffing the cabinet in parliamentary and presidential systems probably explains why most theories of legislative behaviour, rooted as they are in Congressional politics, start and end in assuming that legislators are motivated by re-election (the classic example being Mayhew 1974).

To re-emphasize our point, in parliamentary systems legislators care greatly about reselection and re-election but they are also motivated by the desire to gain even higher political office, similar to what Carroll et al. (2006) describe as mega-seats. Such political office is typically at the discretion of the party leader. In effect, the party leadership can use the potential for promotion to the ranks of government as a form of control over individual legislators.3 The tight grip typically held over the legislative agenda by the cabinet under parliamentarism makes individual cabinet ministers the prime initiators of policies – almost to the exclusion of all other legislators (Laver and Shepsle 1996). The autonomy that cabinet ministers are awarded differs remarkably between countries and so may the desirability of the position. Hallerberg (2004: 16) distinguishes between systems of delegation (where the prime minister gives ministers detailed instructions), commitment (where detailed policy agreements restrict ministers’ discretion), and fiefdom (where ministers have relative autonomy over decisions in their jurisdiction).

While the practice of including only serving legislators in the cabinet may differ from country to country, promotion is mostly in the hands of the party leadership. And that provides a powerful incentive for motivated politicians not to dissent from the party leadership in legislative votes. The more opportunities that exist for promotion, the more legislators will be inclined to yield to the party leadership. We argue therefore, that where legislators stand a stronger chance of being promoted to the ranks of government party voting will be more unified. Where the prospects for leadership are more limited, individual legislators are more likely to rebel against the party leadership, resulting in lower levels of unified party voting.

It is worth noting that this argument is not restricted to governing parties, assuming that no one political party continually monopolises executive seats. In most circumstances, legislators from non-governing parties will be acutely aware that their party may be in government at some point in the future and if or when that time arrives the party leadership may look to them. Hence, we expect to see government and non-government legislators responding to the varying prospects for higher political office. Nevertheless, the promise of promotion may play out differently in governing and non-governing parties as that promise is more uncertain as it lies further in the future.

To quantify the opportunities for ministerial promotion we collected data on the number of government posts filled by legislators in each country included in this study.4 Logically, a legislator with 99 colleagues is, ceteris paribus, more likely to have realistic ambitions of obtaining promotion than a legislator operating in a parliament of 200 members. Consequently our measure of ministerial opportunity controls for the size of the legislature and the Member’s party. We present two measures of opportunity for ministerial promotion: the variable Cabinet measures the number of available senior ministerial positions per legislator. The broader Government measures the number of cabinet and sub-cabinet ministerial posts available per legislator.

Having identified how the design of institutions shape the actions and behaviour of legislators, we proceed in the next section to test empirically the claims that electoral systems, candidate-selection rules and promotional prospects impact the level of party voting unity under parliamentarism. First, we will look at bivariate regressions because small sample size limits the degrees of freedom. Second, the effects of electoral systems, candidate-selection rules and promotional prospects will be combined in multivariate regressions.

Empirical analysis
Centralisation of the candidate selection procedures has a strong impact on party unity in our selection, when using Lundell’s five point scale. With every additional point on the scale towards national party control over nomination, party unity increases – that is, when the first and second point on the scale are combined. As the national leadership enters the selection process, a party’s unity scores increase almost three points on the Rice-index. As the national leadership further strengthens its control over the process, beyond merely ratifying local decisions, unity scores further increase. The difference between the first and second point on Lundell’s scale is related to the inclusiveness of the party selectorate rather than to centralisation. While the composition of party selectorates is not an unimportant concern in intra-party politics, its impact on cohesion is sketchy at best and cohesion, itself, is only imperfectly related to discipline, which is in fact what we observe.

[TABLE 6.2 UNAVAILABLE]

Candidate selection procedures affect party unity irrespective of a party’s position in or out of office, the majority’s margin, or the size of parliamentary parties – although the effect of the strongest centralisation category is not significant. Because of space limitations, however, only the bivariate regressions are listed in table 6.2. Parties of all sizes have long solved the issue by developing formal means of discipline. In fact, party unity is strongest in the larger parties. Larger parties are slightly more likely to have developed centralised nomination processes, for one. As a result, the effect of party size disappears after controlling for candidate selection – whereas the effect of the nomination process remains unaffected.

Contrary to what is often expected being part of the government reduces rather than reinforces party unity – even if the impact of office is not significant. That expectation has largely been fuelled by the debate on the impact of presidential and parliamentary institutions – the vote of confidence in particular – on party unity. Jackson (1968), however, pointed out that opposition parties may remain absent when they face considerable dissent with little harm to the party reputation. The government side has no such option. While from a longitudinal perspective, it is plausible, for instance, that political parties develop centralised nomination processes in response to the shock of losing office, cross-sectionally candidate selection processes and being in or out of office are largely unrelated.

The impact of centralisation is reinforced by the party membership organisation. As the proportion of party voters that are also party members increases, party unity suffers. This, in turn, may be an indication of the impact of inclusiveness and diversity of the party membership. Parties with a mass membership are more likely to have developed centralised nomination processes. After controlling for the effect of a large membership organisation, however, party unity continues to increase as nomination processes are more centralised. In particular, the effect of the most centralised condition is strengthened. Thus, the proportion of party members to the party electorate reflects the inclusiveness of the nomination process, which is not captured by the centralisation of the nomination processes. Especially in Finnish parties, a large membership compared to the party electorate plays a crucial role in selecting the parties’ candidates. The members use the cover that this provides vis-àvis the party leadership to dissent more often.

In addition to candidate selection procedures, electoral rules that provide incentives to cultivate a personal vote reduce party unity. As Hallerberg and Marier’s (2004) index of personal vote increases, party unity decreases.5 To be fair, the impact is not strong and largely depends on the precise coding rules for various electoral rules. Singlemember district systems, for instance, have been considered both among the most candidate-centred (Wallack et al. 2003) and the most party-centred electoral rules (Carey and Shugart 1995). In fact, Carey and Shugart’s original rank-order appears more consistent with the practice of party unity than Wallack’s coding. But even the Carey/Shugart rank-order overestimates the incentives that ordered-list proportional systems provide to cultivate a personal vote. In that respect, the Hallerberg coding appears more correct – acknowledging that parties often have established other means to restrict the impact of these personal votes. For one, party votes might be redistributed in the order of the list, thus adding another obstacle for candidates ranked lower. In addition, party leaders ranked at the top of the list often get more than their proportional share of these personal votes, thus further reducing their impact.

As mentioned, the electoral rules that provide incentives to cultivate a personal vote include the ballot structure, the pooling of votes, the number of votes, and district magnitude (Carey and Shugart 1995). None of these rules, however, is able to consistently explain party unity on its own. Nevertheless, as the selection of cases does not include cases where the party leadership does not control access to the ballot, party unity increases as voters cannot ‘disturb’ the list. In addition, party unity decreases as voters cast a single vote below the party level and those votes are pooled across the list. In particular, the latter runs counter to the expected effect of intra-party competition. The effect of vote pooling, however, differs remarkably from one coding rule to the next: to be more precise, from one rule of coding SMDs to the next. The counterintuitive result appears to be largely driven then by unity in the Finnish parties. With district magnitude, party unity decreases – indicating that growing intra-party competition may in fact outweigh the effect that increasing voters’ information demands may have on the propensity to defect from the party line. The difficulties that voters face to keep track of the voting records of tens of incumbents do not seem to mean that a strategic dissenting vote will pass unnoticed.

[TABLE 6.3 UNAVAILABLE]

In fact, it is something of a surprise that personal vote has an impact at all. After all, a personal vote can be based on a number of activities and characteristics, e.g. local office, pork-barrel benefits, celebrity status, that may or may not have an impact on a legislator’s voting record.

Finally, the level of observed party unity in parliamentary systems is related to opportunities for ministerial promotion when combined with ministerial autonomy. The prospect of promotion effectively silences dissent only when the position actually promises an impact on policy. For this purpose, the number of cabinet positions compared to the parliamentary party group size is too crude a measure. The number of neither cabinet, nor junior minister positions in itself affects party unity significantly. Only in combination with government type and government status does the prospect of promotion loom sufficiently large in the minds of members. Party unity increases as the number of Cabinet positions available rises and ministerial autonomy is strengthened from a situation where it is severely curtailed by the prime minister or a detailed policy agreement to a situation of ministerial fiefdom. Furthermore, only a more immediate prospect of promotion has that effect: in opposition parties, future promotion doesn’t cast its shadow forward that much. To capture this, the number of cabinet positions is weighted by 0.5 in opposition parties. Note, however, that party unity is unrelated to government type in itself and that unity is actually stronger in parties currently out of office. Yet, combined with the number of Cabinet positions, government type and government status are positively and significantly related to party unity – even if the impact is not substantively large. An increase by ten percent, for instance, in the proportion of Cabinet positions is expected to raise party unity by .15 in opposition. The increase is expected to rise further to .86 if the party was in office and ministerial autonomy was at its strongest. In fact, the impact of promotion further increases as the weight of the opposition parties were to be lowered from 0.5.

The difference between cabinet and junior government positions tells much the same story. In itself, the relationship with party unity is even in the wrong direction: unity decreases as the number of junior minister positions available increases. Yet, combined with government type and government status the relationship is in the right direction – though not significant. Legislators, thus, appear more motivated by the prospects of attaining a seat at the cabinet level than by the opportunity to serve as a junior minister - despite the fact that holding a junior ministerial post may be a stepping stone to securing a full cabinet seat.

The impact of candidate selection, personal vote, promotion, and membership on party unity is hardly affected, when their effects are combined in multivariate analysis. Voting unity is strongest in parties where candidate selection processes are centralised, in parties where the chances of promotion to an autonomous cabinet position are the greatest, in parties where the party electorate does not extend far beyond the party membership, and in parties operating under electoral rules that do not encourage the cultivation of a personal vote.

[TABLE 6.4 UNAVAILABLE]

To be fair, these effects are vulnerable to the selection of cases – as is not uncommon in small-n studies. It appears that, in particular, party unity is relatively low in Finland and New Zealand. Low party unity in Finland can be traced back to candidate selection rules and the electoral system. Finnish political parties’ primary selection rules are required by law (Sundberg 1997:97-117). In New Zealand, low unity is consistent with neither candidate selection, nor the personal vote. This not easily explained – it could be of interest that the Parliament studied is in fact the last under the First-Past-the-Post rules, before the introduction of Mixed-Member Proportional Representation. However, the electoral reform does not appear to have affected party unity in the following Parliament (Barker and McLeay, 2000:139). On the other hand, party unity scores are relatively high in Denmark and Ireland – especially in light of the open candidate selection rules in the former and STV electoral rules in the latter.

It is surprising that the inclusion of country dummies reduces the impact of the centralisation of candidate selection processes most – a variable that has performed most consistently so far. Yet, incentives to seek out a personal vote continue to encourage Members of Parliaments to defect from the party line – even if that personal vote is most vulnerable to the selection of cases. More importantly, opportunities to be promoted to a Cabinet position that promises a tangible impact on policy consistently serve to hold Members together. As a result, promotion opportunities are as crucial in understanding cross-national differences in party unity as they are in understanding rebels and loyalists in the British Parliament.

Conclusion
Strong parties whose members vote collectively within the legislature have long been understood as a necessary element of parliamentary government. Previous attempts to account for variation in legislative party unity have focused on presidential versus parliamentary forms of government as being the main explanation for cross-national variation.

Our aim in this chapter has been to point to the fact that within parliamentary systems parties display variation in the level of legislative voting unity – something which cannot be accounted for by relying on the classification of presidential versus parliamentary systems. Beyond a mere acknowledgment of this fact, our aim has been to explain this variation in party unity within otherwise similar political systems.

Incentives to cultivate a personal vote encourage Members of Parliament to defect from the party line. Centralized selection rules, where the party leadership has greater control over the future of incumbents, appear to result in higher party voting unity – although this may be influenced by the particular selection of countries. The opportunity for promotion to government, and in particular the opportunity to enter cabinet, is a tempting offer to maintain unity. The evidence suggests that legislators in parliamentary systems are motivated by the desire to be promoted. This result might point to a significant difference between legislators in presidential systems and legislators in parliamentary systems of government and one that needs to be more explored at the theoretical and empirical level.