Formal models of parliamentary government typically assume that ‘any government that actually forms must be in equilibrium in some sense’ (Laver, Reference Laver2003: 34). It follows that when governments terminate before their term of office is expired, it must be because some development has occurred that disrupts this state of balance. Many kinds of unforeseen events might play a role in this, but the one that is most commonly modeled is a change in the opinion poll standings of political parties. Thus, if the polls suggest that a government party has risen in popularity since the last election, it may be tempted to seek a premature dissolution of the legislature. As Lupia and Strøm (Reference Lupia and Strøm1995, Reference Lupia and Strøm2008) have elaborated in their influential model of government termination, it could also use this information in coalitional contexts (those in which no party holds a parliamentary majority) to renegotiate the current coalition deal or to strike a new one that provides it with greater benefits.
How would one determine whether government terminations occur with these goals in mind? Answering this question is by no means straightforward. Consider the case of voluntary dissolutions. It is not simply a matter of determining whether the parties of prime ministers, or perhaps all government parties, generally emerge from early elections with increased legislative or government payoffs. Certainly, increases in these payoffs following an early election would strongly suggest that the election was called for that purpose, but it is also possible that it was called in order to limit losses rather than to capture gains. In addition, there is the time factor to consider. In the Lupia–Strøm model, the calling of early elections entails opportunity costs as legislative seats and cabinet portfolios are being abandoned before they have to be. The magnitude of these costs varies with the length of time available in the current parliamentary mandate; the less the time remaining in the mandate, the lower the costs (since there is less time to enjoy the benefits). Thus, whether there is a net gain from early elections depends not just on what is sacrificed and what is won, but on when the sacrifice is made.
For Diermeier and Stevenson (Reference Diermeier and Stevenson1999), this time factor suggested a different way to address the issue. As the opportunity costs of calling early elections diminish as the mandate nears its end, decisions to dissolve should correspondingly become more likely.Footnote 1 Using a competing-risks event history model, they were able to show that this is precisely what occurs in West European parliamentary systems. A similar result was also obtained by Strøm and Swindle (Reference Strøm and Swindle2002), using a somewhat different sample.
This evidence is certainly suggestive, but it is also quite indirect. It does not indicate whether the effect is one of strategic advantage or mere convenience: elections may be called a few weeks or even a couple of months early simply to reduce the potential for bad weather or to avoid overlap with some other event (such as a major conference or a religious holiday).Footnote 2 A more telling indicator of strategic dissolution would be whether the benefits derived after an election tend to increase the earlier the election is held, as would be expected under the opportunity-costs logic. However, this possibility does not appear to have been investigated.
We face similar gaps in our knowledge with respect to the other possible outcome of an equilibrium-destroying event – the renegotiation of the coalition deal or the replacement of the existing coalition government with a different one. The Lupia–Strøm model, or indeed virtually any other formal model of coalition formation and maintenance, anticipates that a government party will upset an existing coalition arrangement when it stands to gain from that action. If the result is not an election in which the party improves its situation, then it should be the emergence of a re-negotiated or new coalition that does so. We have very little information, however, on whether these anticipated outcomes are in fact realized.
This lack of clarity over whether and in which ways parties gain from defection is evident in the two examples that Lupia and Strøm (Reference Lupia and Strøm1995) use to illustrate the key elements of their model. In the case of the defection of the German Free Democrats from their coalition with the Social Democrats in 1982, it is clear that the defection paid dividends. Policy divisions between two coalition partners were increasing in the early 1980s and the Free Democrats were experiencing a serious decline in the polls; some in the party feared that it would lose parliamentary representation entirely. By defecting, the party not only averted that outcome but obtained a further 16 years in government with the Christian Democrats. However, the withdrawal of the Irish Labour party from its coalition with Fine Gael in 1987 presents a different picture. Because there was no other coalition partner to turn to – the main opposition party, Fianna Fail, had a (credible) ‘no coalition’ policy – the defection provoked an early election. In this election, Labour lost one-quarter of its parliamentary seats. It also lost its share in government, as Fianna Fail emerged strong enough to form a single-party government. Thus, Labour's defection appears to have cost it both seats and power.
These are mere anecdotes, of course, but there is little in the way of systematic evidence to put them into context. In fact, the only extant study of the consequences of bringing down governments suggests a cost that Lupia and Strøm did not countenance: defectors may alienate their former coalition partners. In her analysis of coalition governments in 18 European countries, Tavits (Reference Tavits2008) found that when coalition governments fall apart because of party defections or inter-party conflict, the parties involved in the dispute are significantly less likely to come together in subsequent coalition governments, implying that they are being ‘punished’.
This finding does not necessarily undercut the Lupia–Strøm model, since a party might be shunned by some potential parties but do very nicely from others. But the fact is that we do not know whether or not this tends to be the case. Do parties defect from governing coalitions because they are in a position to join a different governing coalition that rewards them more handsomely? Do they do so because they believe it will enhance their electoral prospects? Or might they do so without the prospect of either type of benefit, and despite the risk of being shunned by some parties in the future, for reasons that are much less immediate and less evident?
The objective of this paper is to address these questions, not by directly assessing party beliefs and intentions – which would be very difficult to do – but by examining what actually happens to parties that dissolve legislatures, defect from coalition governments, or participate in disputes that bring them down in West European democratic systems. The outcomes of interest primarily concern legislative seat shares (if an election has taken place) and the benefits of membership in subsequent cabinets, including the degree of power or influence it conveys and the relative viability of the new cabinets vis-à-vis their predecessors. Where appropriate, the evaluation of outcomes will also take into consideration the time factor. The data that have been collected for these purposes, and the methods used to analyze them, are discussed in the next section. The subsequent two sections present the data analysis. A discussion of the implications of the findings for future theoretical work on coalition government concludes the paper.
Data and methods
The data collected for this investigation cover 14 West European democracies from the end of the Second World War or the start of the current regime (whichever came later) to the present. Given that considerable attention has been devoted to the strategic timing of elections in majoritarian systems (e.g. Smith, Reference Smith2003, Reference Smith2004), the focus here is on coalition governments. Caretaker governments and governments that terminated for technical reasons (e.g. where a government must resign when a new president is chosen) are also excluded.
The unit of analysis is the individual government party, of which various types have been identified for analytic purposes. For ‘dissolution terminations’, that is, government terminations that end with elections, the key distinction is between the parties that brought about the dissolution (‘dissolvers’) and other government parties. Dissolvers, of course, are not always easy to identify. If a defection or dispute presaged the dissolution, the party(ies) involved might be identified as dissolver(s). Otherwise, attention naturally focuses on the parties of prime ministers, who have control or at least disproportionate influence over decisions to dissolve in most systems (Strøm and Swindle, Reference Strøm and Swindle2002: 577). For ‘replacement terminations’ (terminations that result in the formation of another government without new elections), the most relevant distinction is among parties that defected from the government (‘defectors’), parties that engaged in disputes that brought about the government's demise (‘disputers’), and the remaining parties, which, for want of a better term, will be called ‘loyalists’.
Operationally, a defector is defined as a government party that (a) resigned from the government, (b) challenged the government in a way that caused the government itself to collapse, or (c) voted against the government in a confidence motion. Where coalition governments fell because two or more of their member-parties simply could not agree, the parties in question are categorized as disputers. Two Belgian cases illustrate the distinction. On January 19, 1974, the Belgian government tendered its resignation to the King after one of its members, the Socialist party, resigned from the government the previous day (Keesing's Contemporary Archives 1974: 26533). The Socialists were coded as defectors in this instance. In contrast, the government that preceded it had resigned because of ‘dissension between and within the coalition parties concerning various issues’, as Keesing's (1974: 25816) put it. Here, all three government parties were coded as disputers.Footnote 3
As these examples suggest, the principal data sources for information on governments, including their composition and reasons for termination, are Keesing's Contemporary Archives (1945–87) and its successor, Keesing's Record of World Events (1987–2010). From the late 1980s onwards, the annual data sections of the European Journal of Political Research provided an additional source of information. The identification of defectors and disputers, which was by far the most difficult data-collection challenge, was also aided by the Comparative Parliamentary Data Archive (CPDA).Footnote 4 As the greatest danger here is missing instances of this type of behavior, the following decision rule was used: in cases where Keesing's and the CPDA disagree, a clear identification of a defector or a disputer by either source was accepted. In practice, the sources agreed in the vast majority of cases.
Payoffs to parties are typically conceptualized in formal models by some combination of legislative seats and cabinet portfolios.Footnote 5 To capture this understanding of what is at stake, four types of payoffs were measured for each party. The first is the ‘legislative seat share’, that is, the percentage of legislative seats held by the party. Naturally, changes in this type of payoff are only relevant if there has been an intervening election. The second is ‘portfolio share’, the party's percentage of the total number of cabinet portfolios in the government. A more nuanced measure of portfolio allocations is provided by the third type of payoff, the ‘salience-weighted portfolio share’. This measure, taken from Druckman and Warwick's (Reference Druckman and Warwick2005) Portfolio Allocation Data Set, uses expert-based estimates of the saliencies of government portfolios in each system as weights in calculating the portfolio allocation to each government party. The final measure is the ‘left–right range’ of the government. This is generated by calculating the distance between the right-most and left-most party in the cabinet, using estimates of party left–right positions from the latest release of the Comparative Manifestos Project (CMP) data (Klingemann et al., Reference Klingemann, Volkens, Bara, Budge and Macdonald2006).Footnote 6 Given the dominance of the left–right dimension in the countries under examination, cabinets spanning larger left–right ranges can be expected to have more difficulty reaching key policy decisions and to be less durable; membership in them should therefore be inherently less valuable to parties.
The central strategy in considering how government parties fare after a government termination is to examine how their payoffs change. For the most part, this means considering their payoffs in the replacement or post-election government as well as any changes that may have occurred to legislative seat shares as a result of an intervening election. Although the kinds of payoffs that follow dissolution and replacement terminations overlap considerably, the contexts are quite different. For voluntary dissolutions, the key question is whether post-election payoffs increase with the opportunity costs of giving up office before it is legally required; for replacement terminations, it is whether defectors or disputers do better in replacement governments regardless of the time remaining in the legislative term.Footnote 7 Because the questions we ask are different for the two types of termination, a separate data set was created for each.
The two data sets have certain features in common, to be sure. Both are time series, albeit interrupted ones, and both combine data from a range of countries. In other words, the data take the form of cross-section time series (CSTS). It has become standard in the political science literature to analyze CSTS data along the lines suggested by Beck and Katz (Reference Beck and Katz1995): ordinary least squares regression with a lagged dependent variable as a regressor to handle first-order serial correlation and panel-corrected standard errors to deal with contemporaneous correlation of errors across panels (countries) and panel heteroskedasticity. It is also advisable (see Wilson and Butler, Reference Wilson and Butler2007) to include a set of country dummy variables to deal with cross-national differences in levels of the dependent variable.Footnote 8 The regression analyses conducted in this investigation follow this template, with two exceptions: (1) the lagged dependent variable will be omitted because the dependent variable is the change in a given payoff, not the payoff itself, and (2) Prais–Winsten regressions are specified to capture any remaining first-order serial correlation in the error terms.
Strategic dissolutions?
The analysis begins with the consequences of dissolutions. A substantial body of research has shown that parties participating in government generally experience declines in votes in the next election. The most recent contribution to this literature is provided by Narud and Valen (Reference Narud and Valen2008: 379), who report that governments in Western European countries lose an average of 2.59% of their votes in the next elections. Our focus is on legislative seat shares rather than vote shares, but the results are consistent. As the first column of Table 1 (upper panel) shows, the tendency is for both the parties of prime ministers and other government parties to suffer losses in legislative seat shares in the ensuing elections. For the former, legislative seat shares fall by an average of 1.47%; given that the mean seat share of prime ministerial parties is 36.4%, this represents a decline of about 4.1% in their legislative strength. The declines experienced by other government parties are of the same order (a mean of 1.72%).
Mean payoffs to governments after legislative dissolutions
The losses extend to other domains as well. The remaining columns of the upper panel show that, although most prime ministerial parties participate in the post-election governments (column 2), the fact that a substantial minority does not tends to depress their mean portfolio shares, both unweighted (column 3) and weighted (column 4). Other government parties have a lower rate of returning to power after elections and also experience substantial declines in portfolio shares.Footnote 9
Are these losses simply the result of too many governments holding on to power until the last available moment? The lower panel of the table suggests that this is not the case: both prime ministerial parties and other government parties did worse as a result of the elections, even when the government voluntarily dissolved parliament. Indeed, it appears to make little difference whether the election was called prematurely or not; they suffered about the same, regardless.
The finding that government parties lose in absolute terms across elections even when called voluntarily does not prove that early dissolutions are not strategic; governments may dissolve early not just to capture potential gains but also to curtail potential losses. This was the conclusion that Smith (Reference Smith2004) reached for single-party governments, and Lupia and Strøm (Reference Lupia and Strøm1995: 650–651) couch their description of the Irish Labour party's defection in these terms as well. Regardless of whether the legislature is dissolved to capture gains or curtail losses, however, one would expect the opportunity-costs logic to apply: because earlier dissolutions entail larger sacrifices, they should be associated with better payoffs.
That the amount of time remaining until the next legally mandated election makes a difference is suggested by Figure 1, which presents a histogram of ‘time left’, the number of months that remain until elections have to be held (by law), for all governments that were followed by early dissolutions.Footnote 10 The histogram shows that dissolution terminations become markedly more frequent as the maximum available time runs out, consistent with the diminishing opportunity-costs hypothesis. As noted earlier, however, this type of evidence is very indirect. A much better sign would be evidence that earlier dissolutions are provoked by the anticipation of better payoffs after the elections, whether it be in terms of legislative seats or cabinet portfolios.
Distribution of early dissolutions by ‘time left’ in the mandate.
Examining this hypothesis requires that we move to a regression format. The dependent variables in the regressions are the changes in the various payoffs that occur as a result of the election. The independent variable of interest is ‘time left’, although, as noted earlier, a set of country dummies is also included.Footnote 11 To place attention squarely on situations where the expected relationship should be most apparent, the analysis will be confined to parties in governments that terminate voluntarily (and are followed by elections).
The first outcome to be examined is changes in legislative seat shares. Model 1 in the upper panel of Table 2 presents the results of the regression of this variable across all government parties. The significant negative intercept indicates that voluntary dissolutions display a general tendency to produce a decline in the seat shares of government parties, consistent with the figures reported in Table 1 and earlier research. What is less expected is the estimated time effect, which is significant but negative. This means that earlier dissolutions tend to lead to significantly worse results in terms of legislative seat shares, rather than better ones, for government parties as a whole. Although prime ministers may be mostly or solely responsible for decisions to dissolve, the lower panel of Model 1 shows that earlier dissolutions confer no significant advantage to their own parties. Thus, for prime ministerial parties as well as for government parties as a whole, the payoffs in legislative seat shares from early dissolutions fail to conform to the opportunity-costs hypothesis.
Post-election payoffs and time left in mandate (voluntary dissolutions only)
Note: Models are estimated with ordinary least squares and panel-corrected standard errors (in brackets).
*Significant at P = 0.05; **significant at P = 0.01; ***significant at P = 0.001.
Legislative seat shares would probably not be the prime concern for parties involved in strategic dissolutions; the more substantial issue is portfolio shares. The relationship of time with changes in (unweighted) portfolio shares is taken up in Model 2. In this model, the binary variable ‘next government’ has been added to control for whether the party entered the post-election government.Footnote 12 The time-left effect is now positive but highly insignificant (P = 0.73). It is also highly insignificant when the focus is solely on prime ministerial parties (bottom panel). Model 3 shows that this pattern persists when portfolio salience is taken into account in calculating portfolio shares. It would appear that changes in portfolio payoffs are unrelated to the amount of time remaining when the voluntary dissolution occurred.
There is thus no support in the present data for the idea that coalition governments seek early dissolutions to the extent that the opportunities they sacrifice when they end their terms in office prematurely are offset by payoffs in post-election legislatures. Although governments must sacrifice some benefits if they terminate before their term has expired, and although this sacrifice must be greater the more premature the termination is, the post-election benefits that follow voluntary terminations do not appear to vary in proportion to that sacrifice.
These conclusions hold only to the extent that strategic decisions to terminate and seek parliamentary dissolutions are made either by heads of government or coalition governments as a whole. But strategic terminations may have been provoked in another way. It may be that individual parties defect from governing coalitions, or engage in disputes that bring coalitions down, with the intention of forcing early elections. This suggests that we should also consider defectors and dissenters as instigators of strategic early dissolutions.
In fact, there are very few cases where party defections or disputes had this outcome. Of the 140 instances of government terminations followed by early dissolutions in the data, just 21 can be linked to one or more defectors. A further 14 involved disputes within the coalition without any clear defector. With so few instances of defections or disputes leading to early elections, it seems unlikely that the overall pattern of a rising hazard for dissolutions can be attributed to strategic behavior by these parties. The evidence (not shown) bears this out. Defectors and disputers experience losses in absolute terms in all three types of payoff considered here and the magnitudes of the losses are not statistically distinguishable from those experienced by loyalist parties. More tellingly, none of the payoffs increases for defectors and disputers according to the length of time remaining in the legislative term.
‘Replacement’ terminations
Parliaments often dissolve before they have to by law, but the evidence presented in the preceding section does not indicate that any net advantage in legislative seat shares or cabinet portfolio shares accrues to the parties that might be (or be held) responsible – be they prime ministerial parties, other government parties in general, or parties whose disputes or defections ended the government. But what of replacement terminations – those in which a government falls and is replaced by another without an intervening election?
In the Lupia–Strøm model, replacement terminations occur when a member of the government is in a position to extract a better deal from a different government (it could also demand more from the current one, but that cannot be observed here). Let us turn, therefore, to the question of how defectors and disputers fare in replacement governments. Obviously, benefiting in terms of portfolios and the influence and perks they entail is contingent upon becoming a member of the new government. In this respect, the differences between defectors, disputers, and loyalists are striking. As the first column (top panel) of Table 3 shows, 80.1% of loyalist parties manage to appear in replacement governments, but for disputers the figure drops to 63.6% and for defectors it falls much further to just 36.2%. These differences are highly significant (P < 0.001).
Mean payoffs to various types of government members following a replacement termination
aParticipation in a subsequent government refers to participation in the next government or, failing that, participation in a later government in the same legislative term or, failing that, participation in the first government in the next legislative term.
If disputers and especially defectors are noticeably less likely to get into the next government, one would expect their portfolio shares to decrease accordingly. The second column of Table 3 shows the mean changes in this type of payoff for the three types of government party. Although defectors, in particular, do appear to lose more on average, the differences between the three means are insignificant (P = 0.32). The same conclusion holds for changes in weighted portfolio shares (P = 0.78).
As one would expect, the principal source of decreases in portfolios is the failure to be included in replacement governments. But we have seen that the return rates of loyalists, disputers, and defectors differ strikingly. How can it be, then, that their declines in portfolio shares differ so little?
For defectors, the answer is that they fare better than loyalist parties if they enter replacement governments and suffer smaller losses if they do not (because they tend to be smaller parties and hence to have smaller portfolio shares to begin with). These two effects can be exposed by moving to a regression format. In the regressions reported in Table 4, defectors and disputers are identified by the binary variables ‘defector status’ and ‘disputer status’, leaving loyalists as the baseline category. Model 1 shows that, with participation in the next government controlled, defectors do significantly better than loyalists in terms of portfolio shares.Footnote 13 Model 2, which repeats the analysis with salience-weighted portfolio shares as the dependent variable, reveals that this advantage persists when portfolio salience is taken into account. These effects do not, however, carry over to disputers. Although disputers also appear to do better if they get into replacement governments and suffer smaller losses if they do not, these differences fail to achieve statistical significance at the 0.05 level in either model.
Payoffs in next governments and subsequent governments (replacement terminations only)
Note: Models are estimated with ordinary least squares and panel-corrected standard errors (in brackets).
†Significant at P = 0.10; *significant at P = 0.05; **significant at P = 0.01; ***significant at P = 0.001.
Defectors, then, tend to gain more than loyalist parties if they enter replacement governments (and lose less if they do not), but they also stand a much lower chance of entering. The two effects largely offset each other, with the result that (on average) defection appears to convey no net advantage over remaining loyal. Disputers, too, appear to gain little from dissension. But perhaps the advantage has to do less with shares in the government than the intrinsic value of the government itself. As Lupia and Strøm (Reference Lupia and Strøm1995: 652–653) note, ‘a coalition containing parties with similar policy agendas is likely to generate greater utility for its members (all else constant) than would a coalition containing parties with conflicting policy agendas’. One can also add that a coalition with greater survival potential can be expected to generate greater utility for its members because the expected benefits are likely to flow for a longer period of time.Footnote 14 Could it be that defectors leave governments because the shares they expect to receive in their replacements are about the same but the anticipated benefits are greater and/or likely to last longer?
As discussed earlier (footnote 6), the overall policy compatibility of a government can be assessed reasonably well by the range of party positions it contains, as measured on the CMP left–right scale. Left–right range should also be a good indicator of the expected longevity of governments. Although numerous possible influences on government survival have been proposed, there is considerable consensus that two key factors are majority status and ideological compactness.Footnote 15 Generally speaking, governments whose member-parties have fairly compatible policy positions and which together command a parliamentary majority tend to survive much longer than other governments.
As the final column of Table 3 (top panel) reveals, the replacement governments that defectors and disputers enter do seem to be more ideologically compact than the ones they left: on average, the left–right range declines 7.97 units on the CMP scale for defectors and 5.79 units for disputers (vs. a decline of just 1.59 units for loyalists). These differences are statistically significant at the 0.07 level. The problem, however, is that these figures take no account of the government's majority status. Other things being equal, one would expect minority governments to be more compact than majority ones but not longer-lived; their minority status more than offsets any apparent advantages in compactness. In a sense, their actual degree of compactness is under-estimated, since it does not take account of the external support upon which they may depend. As the impact of this problem tends to diminish as the government approaches majority status, changes in the closeness of governments to majority status should also be considered.
Model 1 of Table 5 shows the results of a regression model that incorporates these factors. The dependent variable is ‘change in cabinet left–right range’ (i.e. the difference between the left–right range of the replacement government and that of its predecessor). In addition to the defector status and disputer status variables, the independent variables include changes in the binary variable, ‘majority status’, and in ‘closeness to majority status’, where closeness refers to the percentage by which the government's legislative basis falls short of a majority. Note that only government parties that entered the replacement government are considered, since they are the beneficiaries of any net increase in policy compactness or survivability.
Ideological compactness in next governments and subsequent governments (replacement terminations only)
Note: Models are estimated with ordinary least squares and panel-corrected standard errors (in brackets). Only parties in both the present and the next or subsequent government are included.
†Significant at P = 0.10; *significant at P = 0.05; **significant at P = 0.01; ***significant at P = 0.001.
The estimated slopes in Model 1 make it very clear that, to the extent that policy compatibility and government survival can be captured by left–right range (adjusted for government size), prospects do not differ across party types. There is, in other words, no evidence that defectors or disputers, when they do enter replacement governments, have an advantage over loyalists in terms of the policy compatibility or the durability of the governments they enter. As their odds of entering replacement governments are considerably lower than those of loyalists, their expected payoff in terms of compatibility and duration must be judged to be worse.
Subsequent governments
The analysis to this point has focused on possible gains for a defector or disputer in the first government formed after its current government falls. But parties may not have their eyes only on the immediate replacement government. In many systems, three or more governments may be formed between elections, which means that a defector that does not manage to get into the replacement government may still enter a government later in the legislature. Even if that does not happen, it may anticipate that the next election, when it does occur, will bring gains.
Let us therefore broaden our understanding of benefits by looking beyond the immediate replacement government. The focus will remain on the next episode of government participation for each government party, but if a party does not participate in the next government formed, we will consider whether or not it participated in a later government in the same legislature, and if that is not the case, whether or not it participated in the first government formed after the next elections. (This does not include cases where the termination of the government led to elections, which were dealt with under ‘dissolution terminations’.) Participation in any one of these will be referred to as participation in a subsequent government.Footnote 16
The consequences of substituting subsequent participation (so defined) for participation in the next government are shown in the lower panel of Table 3. As one would expect, loosening the definition in this way increases the odds of participating across the board: 87.5% of loyalists, 83.6% of disputers, and 53.4% of defectors participated in (at least) one of these opportunities. Despite some narrowing in the return rates of the three party types, however, the basic tenor of the findings remains the same. Because participation rates increase, the declines in portfolio shares, unweighted (column 2) and weighted (column 3), are smaller, but it remains the case that they differ only modestly and insignificantly across the three types of party.
Turning now to the regression analysis, Model 3 of Table 4 shows that, controlling for participation in a subsequent government, neither defectors nor disputers fare significantly better than loyalists in terms of unweighted portfolio shares. The corresponding estimates in Model 4, where weighted portfolio shares are used, fall just short of significance at the 0.05 level. Although the patterns are less clear than for replacement governments alone, the overall story remains essentially the same: defection or dissension does not bring rewards in subsequent governments.
The same conclusion holds when attention is turned to the policy ranges of subsequent governments. Although the subsequent governments that defectors and disputers enter appear to be more compact (final column, lower panel, of Table 3), Model 2 of Table 5 shows that, as was the case with replacement governments alone, these parties obtain no advantage over loyalists once changes in government size (i.e. majority status and closeness to it) are taken into account. As defectors and disputers are less likely to enter a subsequent government, this, too, must be counted as an area of overall disadvantage for them.
All in all, extending the consideration of participation opportunities beyond the immediate replacement government provides no evidence to temper the conclusion that defectors and disputers are generally no better off, and in some ways worse off, as a result of their government-terminating actions. This conclusion presumes, moreover, that future rewards are on a par with present rewards. Because of the fewness of cases and the lack of information on how heavily to discount time, no attempt was made to discount payoffs in subsequent governments by the length of time parties had to wait to realize them. If benefits are generally worth more today than next month or next year, however, one must reckon that defectors and disputers are actually worse off overall than Tables 3 and 5 indicate.
Later elections
The section on dissolution terminations examined what happens to the various types of party when government terminations are followed by elections. Although the governments under scrutiny in this section do not fall into that category, one can still ask how the various party types fared when the next elections came to be held. It is at least plausible that parties are prone to dispute with their coalition partners or defect from coalitions to the extent that they fear the electoral consequences of remaining in government. If so, then one can speculate that defectors and disputers will be found to do better electorally than parties that made no overt attempts to dissociate themselves from their governments.
As we have seen, ‘doing better’ need not mean increasing seat shares in the next legislature. Like the parties whose governments ended with elections, the governing parties under investigation here experience a decline in seat shares in the next election: on average, it amounts to 2.26%. Is there any difference between defectors, disputers, and loyalists in this regard?
Disputers, with a mean decline of 1.76%, do seem to do somewhat better than loyalists whose mean decline is 2.65%, and the gap is considerably stronger for defectors: their mean decline is just 0.66%. These differences are not, however, statistically significant (P = 0.17). This is partly a function of the fact that we are dealing with a sample is that substantially smaller in size, owing to the need to avoid duplicated cases (i.e. parties that participated in more than one government in a given legislature).Footnote 17 Thus, while there is some indication that defectors may retain seats better than loyalists in later elections, the evidence is inconclusive at this point.
Discussion and conclusions
This paper began with the mission of determining what happens to parties in coalition governments that defect, enter into government-ending disputes, or provoke legislative dissolutions. Despite the plethora of information on coalition governments in Western Europe since 1945, there are a host of measurement challenges that stand in the way of conclusive answers to these questions. If a government ended in a way that appears to be voluntary and its resignation was followed by an election, is it safe to conclude that it sought the dissolution? If so, were all member-parties agreed on this goal, or just some of them, for example, the party of the prime minister? If a government fell due to internal disagreements, have the available sources correctly identified the defector(s) or disputer(s) that brought it about? And so on.
No claim can be made that these and the other measurement issues have been solved in the data sets created for this investigation. Indeed, it would be better to regard the data analyzed here as first cuts at inherently very challenging measurement tasks. Nonetheless, what has emerged from their analysis is a remarkably consistent picture. Government coalitions or their leading parties do not seem to profit from early dissolutions, nor reap better rewards the earlier the dissolution takes place; defectors and disputers do not seem to profit either, regardless of whether their actions lead to dissolutions or to replacement governments. In the latter scenario, any gains in terms of portfolio shares if membership in the next government is achieved are offset by the risk of its not being achieved; with respect to the chances of moving to a cabinet with better internal policy compatibility and survival prospects, the picture is even more dismal. Broadening our compass to include later governments formed in the same legislature or the first governments formed in the next legislature does not change the tenor of the findings in any significant way. Simply put, dissolvers, disputers, and defectors do not appear to profit from their actions.
This conclusion depends, of course, on what we mean by ‘profit’. It may be that party leaders see things differently from what has been assumed, especially in formal models. This is not to say that they are not office-seekers and policy-seekers – they undoubtedly are – but it does suggest that they may pursue these goals in other ways.
Which other ways? If a party that takes one of these government-ending actions does not benefit in terms of future legislative seats or cabinet participation, it could be because its focus is not so crisply defined. Perhaps its leaders have come to believe that the policy compromises the party has accepted in the current government are alienating its supporters. Just as supporters, activists, and interest groups may be offended if a party changes its policy positions too abruptly (see Tavits, Reference Tavits2007 for a discussion of the substantial literature on this issue), so may the same groups become disillusioned if the party abides by coalition agreements that are perceived to undercut those positions (Warwick, Reference Warwick2006). As Schofield and Sened (Reference Schofield and Sened2006: 16–17) note, offending activists can be doubly damaging, since it may cost a party not only their votes but also the effort and money they contribute to winning over other voters and activists. As disillusionment among these groups becomes apparent, party leaders may become increasingly willing to enter a period in opposition in order to rebuild the party's reputation – even if there is no clear idea of whether, when, or how well it will pay off. Indeed, this is essentially how Lupia and Strøm (Reference Lupia and Strøm1995: 651) describe the Irish Labour party's defection from its coalition with Fine Gael in 1987.Footnote 18
Another possibility is that a decline in popularity may cause leaders to conclude that continued association with the government will damage their reputation for competence. In other words, the instigating factor may be less an impression of principles betrayed than a sense that the policies agreed to are not particularly effective and/or that association with them will nurture an impression of incompetence in managing the affairs of government. Thus, a party might leave the government or become involved in government-ending disputes without any concrete notion of when it might return or on what terms, simply to prevent further damage to its identity, the viability of its policy stance, or its perceived capacity to perform effectively in government.
The evidence presented in this paper does not lend direct support to either of these hypotheses; it simply shows that defectors and disputers do not disrupt current coalition deals because they have better deals lined up, and that dissolvers do not appear to follow an opportunity-costs logic. As it is unlikely that the behavior of these parties is as self-defeating as it seems, it is reasonable to suppose that something along the lines of what has been suggested above must be going on. Reaching a more definitive conclusion, however, will require further, more close-grained research on the motivations and strategies of parliamentary parties, with a particular focus on dissolvers, disputers, and defectors.
