Jump to ContentJump to Main Navigation
Elections, Parties, DemocracyConferring the Median Mandate$

Michael D. McDonald and Ian Budge

Print publication date: 2005

Print ISBN-13: 9780199286720

Published to Oxford Scholarship Online: February 2006

DOI: 10.1093/0199286728.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2019. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use.  Subscriber: null; date: 17 September 2019

Fluctuating Political Forces

Fluctuating Political Forces

Chapter:
(p.181) 11 Fluctuating Political Forces
Source:
Elections, Parties, Democracy
Author(s):

Michael D. McDonald (Contributor Webpage)

Ian Budge (Contributor Webpage)

Publisher:
Oxford University Press
DOI:10.1093/0199286728.003.0011

Abstract and Keywords

Political variables (median voter and median parliamentary, government and ministry positions) change more radically and more quickly than policy outputs. This accounts for the lack of relationships previous studies have failed to find as we did, too, in chapters 8 and 9, between political changes and policy. However, political forces oscillate round stable equilibrium positions, which show a strong relationship to the policy regimes’ equilibria previously uncovered. In the long term, the responsiveness of the median parliamentary position to the median voter position is high, as is government position to MPP, and ministries to government position.

Keywords:   political fluctuations, median voter, MPP, governments, ministries, short term fluctuation, long term equilibria, responsiveness

Having examined the dependent variable—policy outputs, the end results of the democratic process—in Chapter 10, we turn to the hypothesized political forces at work on them. In terms both of theoretical perspectives and previous findings these are median voters', median parliamentary party's, and government's preferences. We want to see, as with policy, what their dynamics are and how important these are compared with cross‐national differences. Is there a ‘political regime’ we can match with the ‘policy regimes’ identified in Chapter 10?

We need to do more than replicate the policy analysis of speed of change and cross‐country variation. We also want to examine relationships between the political forces themselves, especially how far median voter preferences affect those of the parliamentary median and how these in turn affect the government position. This is particularly relevant in light of the median mandate argument (see Table 2.4) that there is a sequence at work, one in which the median voter chooses the median party and the median party influences the government.

We already examined relationships between these political forces in Chapter 7. However, that analysis averaged the dynamics rather than accounting for them. Here we need to look at relationships in terms of how they carry over time and, in particular, how responsive the median party preference is to that of the median voter and how responsive the government is to the median party. If one does not shift in response to the other in the way indicated by a median mandate, an essential link in our argument will be missing.

The chapter proceeds from the ‘political rate of change'—which after the evidence reported in Chapters 6 and 9 we must clearly expect to be greater than that of actual policy—to the analysis of election by election responsiveness within countries. On the way we consider the extent to which elections coordinate the forces at work and the way in which both parties and voters shape election outcomes.

11.1 POLITICAL DYNAMICS

We now come at the policy situation from the other side; that is, we consider what we know and can estimate about political dynamics. We have already (p.182) seen in Chapter 6 that electoral dynamics move quite rapidly. In particular, the evidence showed that electorates whose party‐vote percentages deviate from their average (a form of equilibrium value) can be expected to return very quickly to their typical levels.1 We also saw in Chapter 7 that our twenty‐one nations have different sets of ideological equilibria.

In light of the previous analyses we can take these median voter, parliamentary median, and government mean values as pointing to the approximate political equilibrium in the countries concerned and ask what the political dynamics of these three actors look like. To do this we once again estimate the speed of adjustment, now with respect to political ideological change, with our autoregressive term included in an equation with a set of national dummy variables. We find the following:

Median Voter

(N = 245 elections)

M V i e = α + 0.276 M V i e - 1 + λ i e ( N i e ) + ɛ i e ( 0.067 ) R 2 = 0.395 and S e = 10.5

Parliamentary Median

(N = 245 parliaments)

P e a r l M e d i a n = α + 0.294 P a r l M e d i a n ie - 1 + λ i e ( N i e ) + ɛ i e ( 0.064 ) R 2 = 0.316 and S e = 14.2

Government

(N = 411 governments—excluding caretaker, nonpartisan, and transition governments)

G o v e r n m e n t i g = α + 0.481 G o v e r n m e n t i g - 1 + λ i t ( N i t ) + ɛ i e ( 0.044 ) R 2 = 0.394 and S e = 14.0

where

  • MVie and MVie ‐1 are the Left‐Right positions of median voters for nation i in the current (e) and preceding (e −1) elections.

  • (p.183)
  • Parl Medianie and Parl Medianie ‐1 are the Left‐Right positions of parliamentary medians for nation I at the time of the parliament sitting after the current (e) and preceding (e − 1) elections.

  • Governmentig and Governmentig ‐ 1 are the aggregate Left‐Right positions of government parties weighted by the parliamentary seats held by them, in nation i at the time of the current (g) and preceding (g − 1) governments.

  • Σλi,e or g(N i,e or g) is a matrix of nation dummy variables and respective coefficients for nation i at a given election or government.

  • ɛie or g is (assumed to be) a well behaved error term.

Clearly the speed of adjustment towards a country's electoral, parliamentary, or governmental equilibrium is much faster than the speed of adjustment of policy. Indeed, with coefficients between 0.25 and 0.5 on the autoregressive term (i.e. the Y at t‐1 variables), our estimates tell us that political movements are between three and five times more rapid than policy movements.

11.2 TAKING STOCK OF POLICY AND POLITICAL DYNAMICS

It is now time to bring our bits of scattered evidence into direct association with one another so as to form a unified picture of what might be going on. In Chapter 7 we saw that the average ideological positions of various countries on these measures are more or less left‐leaning versus centre‐leaning versus right‐leaning and that representational distortions around each country's national ideology tend to cancel out through time. In Chapters 8 and 9 we saw that the connections between policy and the political ideological positions of voters, parliaments, governments, and ministries at each individual time‐point are weak and fleeting. In Chapter 10 we have seen that policy is slow moving and we have just shown that politics are highly dynamic. The overall picture of politics that emerges is one where political influences oscillate around country‐specific ideological equilibria. These in turn mark a rather steady central tendency within a single country despite dramatic temporary upsets. However, these steady ideological inclinations differ quite considerably across countries. They thus have the potential to mark out a divergent track for policy in each country.

Contrary to hypotheses that point to governments and ministers as the operative policymakers (e.g. Laver and Shepsle 1996) we suspect that governments are relatively less important than parliaments and individual ministers are even less important than governments. More than that, we suspect that no single parliament can be assigned responsibility for the policy it makes. A parliament necessarily has to work within the policy regime it inherits. One half of our contrary thesis is this: if a party or coalition takes control of government's policymaking apparatus without the people being with them, and the people never do come to see things in the way the party or (p.184) coalition would like, it is difficult to imagine that government having a lasting policy impact in the sense of changing the policy regime. The other half of our thesis is this: if a party or coalition takes control of the apparatus of policymaking, and the people come along initially, but popular support evaporates, it is difficult to imagine that government having a lasting policy impact.

11.3 ELECTIONS AS COORDINATING INSTRUMENTS

At this juncture we have varying but interlocking pieces of evidence that give rise to the interpretation above. Interpretations are not final answers, however; they are hypotheses we arrive at after viewing the evidence. We therefore need to ask what independent evidence would make this hypothesis untenable. At the centre of our median mandate thesis is the proposition that elections have the power to animate the policymaking process. If this proposition fails, then so does the whole thesis.

Chapter 7 found that ideological positions held by median voters over a series of post‐war elections in twenty democracies match the parliamentary and government positions well enough for us to conclude there is not much representational bias, except in France and Britain. This is a necessary piece of evidence for thinking there is coordination of median and government positions by elections. But it is hardly sufficient. At the very least, for elections to coordinate the politics and policies of a country, we have to show direct responsiveness of parliamentary and government positions to electoral ones. We know from Chapter 7 that the ideological positions of parliaments are usually not much biased with respect to the ideological positions of their electorates. But, then, two random series can have the same mean; two trends running in different directions can have the same mean; two series zigzagging in antithetical response to one another can have the same mean. If similar means arise for reasons other than parliamentary medians being directly responsive to election medians, we do not have evidence of elections having the potential to coordinate politics and policy.

11.4 ELECTORAL RESPONSIVENESS: THEORETICAL POSSIBILITIES

We begin with a disclaimer. It is unnecessary to ask in some general sense whether electoral responsiveness exists. It undoubtedly does. A vast and important literature constructed around the concepts of representational responsiveness, with a concern for bias as well, has repeatedly shown that party seat percentages respond to shifting vote percentages (Tufte 1973; see also Edgeworth 1898; Kendall and Stuart 1950; March 1957–8; Taagepera and Shugart 1989; King and Gelman 1991). No one doubts that particular (p.185) generalization. However, those investigations have not been much concerned with another generalization—about whether there is policy responsiveness. Their concern has been the precise magnitude and absolute reliability of responsiveness of seat shares to party vote shares under different rules (Taagepera and Shugart 1989), under the same rules in different places (March 1957–8; Tufte 1973; Taagepera and Shugart 1989), or under the same rules in the same place at different times (Tufte 1973; King and Gelman 1991).

The issue of responsiveness looks much different when infused with a concern for policy. It changes the question from who wins how much of the valued currency, namely seats, to how accurately the policy position of an electorate empowers a similar policy position in parliament and government. The distinction is readily apparent when one contrasts the consequences of different electoral systems, most especially SMD versus PR systems.

From a seat‐vote perspective, SMD systems are said to be overly responsive compared to the near definitional requirement of proportional responsiveness under PR. When policy is allowed into the consideration, this simple distinction needs to be qualified. Varying seat percentages in response to shifting vote percentages is essentially a rule‐based, mechanical operation. Whether such shifts have consequences for policy preferences and, if so, to what extent, depends on whether the vote shifts are relevant to changing the parliamentary median and whether the change in the parliamentary median is large or small. Furthermore, as we have said, the answer to these questions depends on more than the behaviour of voters. The changing policy positions of the parties can shift the parliamentary median even when the voters stand still, and changing party positions can make the parliamentary median stand still even when the party choice of voters changes.

A few examples help us explore these possibilities and think through the possible consequences. A three‐party system operating under PR rules (as in Germany) might record an electoral shift to the left. But as long as the middle party, fixed in its policy position, remains the party with which the parliamentary median affiliates there is no effect on policy preferences at the decision‐making level. A two‐party system (as in the USA) might record an electoral shift from 52 to 60 per cent of the vote. But as long as the electoral rules properly identify the majority party, one with a fixed policy position in both elections, there is no recorded policy consequence. In both cases, seat shifts in response to vote shifts are not relevant to changing the parliamentary median and thus are unlikely to have any direct policy consequences.

As regards policy responsiveness when the parliamentary median does change as a consequence of vote‐percentage shifts, the magnitude of the response is likely to depend as much or more on the distribution of party positions as on the redistribution of party vote percentages. Seat‐percentage changes in response to a percentage point shift in votes are two to three times as large under Canada's SMD rules as under Sweden's PR rules.

(p.186) Nevertheless, when the vote change produces a change in the parliamentary median in both countries, the effect on policy preferences is about a six unit movement along the Left‐Right dimension in Canada—where the Liberals and Progressive Conservatives stand about six units apart, on average—versus a twenty‐five unit Left‐Right movement for Sweden—where Social Democrats and Liberals stand about twenty‐five units apart, on average. The proximity of parties to one another, especially of the two parties normally adjacent to the median voter, matters a great deal for policy responsiveness.

Consider also that voters alone are not the only electoral force that can move the policy position of the parliamentary median. Parties hold that power too. In the three‐party case (here Austria might be the operative example), if the middle party abandons its middle position by moving to the right of the one‐time right‐wing party, this will put the heretofore party on the right in the position of the parliamentary median even if the party movement had no effect on the vote distribution.

These are hypothetical possibilities, of course. It is very hard to imagine that marked programmatic shifts by parties have no effect on their vote (Adams 2001). In the case of Austria, the Freedom Party's move rightwards undoubtedly lost it some voters while it gained others. The electors who stuck with it expressed, through their revealed preferences (votes), a willingness to record a change in their own opinions.

What these examples illustrate is that election outcomes generally reflect both the party alternatives on offer and the electoral reactions to them. There is no way of entirely distinguishing the two.2 Over time, indeed, electors might support new parties or movements that provide different alternatives for them. In any one election, however, the only way they can reveal their preference is by selecting one of the parties competing, or by abstaining. These are points we take up again in methodological comments on the results from our analysis of responsiveness.

11.5 ELECTORAL RESPONSIVENESS: COMPARATIVE EVIDENCE

A failure to find shifts in parliamentary and government preferences in response to electoral ones would be enough to dismiss our median mandate (p.187) thesis, even if their discovery is not enough to substantiate it finally. So the first thing we need to know is whether the electoral process can be viewed as moving the centre of the policymaking process. Our answer starts with the most general form of the relationship—across all elections in our data set—and moves down through levels of aggregations through system types to individual countries.

Figure 11.1 shows the relationship for all 265 elections. The graph is distinctly linear with a slope very near to one‐to‐one direct responsiveness and with a reasonably good fit. The associated linear equation is

P e a r l M e d i a n i e = 1.47 + 1.01 M V i e + ɛ i e ( 0.70 ) ( 0.05 ) R 2 = 0.598 S e = 10.5 N = 266

where Parl Medianie is the Left‐Right position of the parliamentary median in nation i following election e, and MVie is the position of the Left‐Right position of the median voter in election e, and ɛie is assumed to be a well behaved error term. Policy responsiveness is estimated with respect to the slope. There is a remarkably consistent degree of responsiveness. The solid line represents hypothetical one‐to‐one policy responsiveness; the dotted line is the estimated relationship. We find each shift of a median voter's position by one unit along the Left‐Right dimension associated with a one‐unit shift of the parliamentary median position. Representational bias is estimated by the intercept. Given its statistically significant value of +1.47, there is about a one and a half unit rightward bias in the parliamentary median relative to the position of the median voter. This is not much; it amounts to the difference between a Christian party or progressive Liberal party being at the ideological centre of parliament following an election. The difference is, however, statistically significant and thus indicates a reliably consistent bias in election outcomes.

We saw earlier, in Chapter 7, that SMD systems tend to have a rightward‐tilting representational bias. We have added a dummy variable for SMD systems to the above equation to check on how that bias might affect the estimated responsiveness and how this analysis records the bias by system. The statistical results show

P a r l M e d i a n i e = 0.02 + 0. ( 0.98 M V i e + 4.35 S M D i e + ɛ i e ( 0.84 ) ( 0.05 ) ( 1.39 ) R 2 = 0.612 S e = 10.3 N = 266

where SMDie is a dummy variable scored 0 for elections held under PR rules and scored 1 for elections held under SMD rules, and where other variables (p.188)

Fluctuating Political Forces

FIGURE 11.1. Responsiveness of median parliamentary Left‐Right policy position to median voter Left‐Right position over 266 elections in twenty‐one democracies, 1950–95

are as previously defined. Controlling for system type does virtually nothing to the estimated magnitude of responsiveness—it is still nearly one‐to‐one—but the re‐estimation tells us that representational bias is entirely within the ambit of SMD elections. There is essentially no bias under PR rules. The intercept, which records PR system bias, is a negligible 0.02, but the bias under SMD rules amounts to a rightward tilt of outcomes over four units to the right of a median voter's position.

This second equation, as did the first, imposes the assumption on the analysis that the responsiveness in PR systems and SMD systems has the same magnitude. From what we know about translating vote percentages to seat percentages, that might seem a dubious assumption. We therefore separate the analyses by system.

PR Systems

P e a r l M e d i a n ie = - 0.14 + 0.96 M V i e + ɛ i e ( 0.57 ) ( 0.04 ) R 2 = 0.779 s e = 6.9 N = 185

(p.189) SMD Systems

P e a r l M e d i a n ie = 4.63 + 1.08 M V i e + ɛ i e 1.80 ) ( 0.17 ) R 2 = 0.341 s e = 15.7 N = 81

where all variables are as previously defined. The responsiveness in both systems is similar; in statistical terms they can be considered to be the same.3 And, as we have already seen, representational bias exists only under the SMD system. Another point of distinction is the more reliable responsiveness under PR rules than under SMD rules. Mispredictions for SMD compared to PR systems are much larger; standard errors of estimate show PR = 6.9 and SMD = 15.7. A consequence is that direct reliability estimates for responsiveness (i.e. standard errors of the slopes) are about four times better under PR than under SMD systems (0.04 versus 0.17).4

The responsiveness equations just presented run the risk of drawing on essentially cross‐national differences to gain leverage for the responsiveness estimates. That is not wholly legitimate from all theoretical perspectives, because responsiveness is most often a theoretically important system‐specific phenomenon. Table 11.1 breaks the responsiveness analyses down by country. The results are nothing less than stunning. In twenty of twenty‐one countries under investigation, responsiveness exists to a statistically significant extent. France is the lone exception. That the existence of responsiveness is so widespread is an important electoral fact. Electoral rules are supposed to ensure that responsiveness exists. What we learn here is that they fulfill their purpose not just in terms of votes and seats but of policy as well. What makes the results especially impressive is that in every country policy responsiveness can be said to be one‐to‐one. There is no country for which we can reject the hypothesis that the policy responsiveness slope equals 1.0. That means we are justified in saying for all countries that a one unit shift in the Left‐Right position of the median voter produces something like a one unit shift in the Left‐Right position of the parliamentary median. France in particular, but also Germany, Iceland, Canada, the UK, and the USA, permit the one‐to‐one inference because their estimated responsiveness is (p.190)

TABLE 11.1. Responsiveness of the median parliamentary party's Left‐Right position to the median voter Left‐Right position, by country

Intercept

(sa)

Slope

(sb)

r2

se

Austria

0.57

(2.56)

0.89**

(0.17)

0.706

8.8

Belgium

0.57

(0.82)

0.88**

(0.09)

0.875

2.8

Denmark

−3.07

(1.51)

0.83**

(0.15)

0.658

5.9

Finland

−0.56

(1.70)

0.92**

(0.09)

0.903

4.5

Germany

0.52

(2.21)

1.31**

(0.21)

0.798

7.6

Iceland

−0.35

(3.44)

0.65*

(0.28)

0.326

11.1

Ireland

−1.12

(3.96)

0.92**

(0.26)

0.510

14.3

Italy

0.67

(0.89)

0.86**

(0.10)

0.889

2.6

Luxembourg

1.37

(2.30)

0.89**

(0.13)

0.849

3.6

Netherlands

1.30

(1.18)

0.95**

(0.08)

0.926

3.6

Norway

−4.85

(6.45)

0.86**

(0.26)

0.556

5.7

Portugal

2.91

(1.91)

1.15**

(0.19)

0.862

4.0

Spain

2.49

(3.36)

1.18**

(0.24)

0.854

4.3

Sweden

−3.75

(2.54)

0.95**

(0.11)

0.851

6.8

Switzerland

−0.76

(1.07)

1.12**

(0.12)

0.902

3.2

Australia

3.40

(5.33)

1.09*

(0.45)

0.264

20.0

Canada

0.46

(2.38)

0.73*

(0.37)

0.250

8.0

France

7.52

(6.46)

0.53

(0.74)

0.061

16.9

New Zealand

4.97

(5.21)

1.08*

(0.45)

0.313

13.7

UK

12.30

(6.14)

1.57**

(0.39)

0.593

16.5

US

−2.26

(6.47)

2.81*

(1.44)

0.297

17.5

not especially reliable; elsewhere it is because the slope itself is in the interval between 0.8 and 1.2.

How can it be, given the ways in which we usually think about electoral responsiveness, that electoral policy responsiveness is one‐to‐one? Does not a shift from Conservative to Labour (or vice versa) in Britain or from the SDA to FP (or vice versa) in Sweden create large differences in policy positions? There are two points to be made in answer to these questions. First, some shifts in median voter positions create no difference in the parliamentary median. Those non‐responsive elections compensate the highly responsive elections. Over a series of elections, average responsiveness is somewhere between 0 and some large value, the magnitude of which depends on the difference between parties adjacent to a median voter in the Left‐Right space. Second, being close to the median voter is usually what it takes for a party to be the one with which the median parliamentarian affiliates. Were parties to stand at fixed policy positions, so that only movements by voters towards one or another party policy position were the source of redirecting the parliamentary median, there would likely be larger degrees of responsiveness in Britain and Sweden. That is to say, importantly, that shifting voter positions are not the only force that redirects the parliamentary centre. Party positions move too. Policy position movements by parties contribute (p.191) their own policy dynamics to elections, and those dynamics contribute to the policy responsiveness we have been observing.

This explicitly reminds us that sometimes parties lead and voters follow, while at other times voters lead and parties follow. Taking notice of this two‐way street makes it best not to consider the pervasiveness of the one‐to‐one policy responsiveness an unconditional fact. Instead, one‐to‐one responsiveness exists under two conditions, (a) a country's voters know which Left‐Right positions the parties are proposing at the time of an election; (b) parties mean what they say about policy at the time of each election. Of course these are major assumptions, stressed classically (e.g. in Downs 1957: 96–114), on which mandate theory also has to base itself (see Tables 2.1 and 2.4), and there is no reason to doubt them. Doing so would undermine the foundation of democracy itself, certainly as embodied in the countries we are examining here, not just the idea of the mandate.

11.6 RESPONSIVENESS: METHODOLOGICAL CONCERNS

Our responsiveness analysis raises in acute form methodological concerns already discussed elsewhere in the book, particularly in Chapters 3 and 7. We return to them here because they very much affect what we report and infer about responsiveness. Two sets of questions come up, both concerned with interdependency between electoral and parliamentary positions. To what extent can we regard them as truly independent when parties at one and the same time fix the policy alternatives voted on in elections and the policy positions taken up in parliament? Is not any correspondence purely tautological or even manipulated, certainly not a spontaneous reflection of voter preferences?

The other, related, set of questions are more narrowly technical, but relevant nonetheless. If party positioning along the Left‐Right continuum enters centrally into defining what electors are voting for, through the estimation of the median voter position, and is then used to define the median party in parliament, is responsiveness not ensured definitionally by this measurement overlap? If so, what are our findings worth for actually testing the existence of a median mandate in the democracies under study?

These questions can be answered in terms of substantive theories of modern democracy, which put political parties at the centre of the representative process. Parties are in fact often likened rhetorically to a transmission belt between electors and governments. Mandate theories (see Tables 2.1 and 2.4) explicitly see them as defining policy alternatives for electors to choose between and then effecting them at the level of government and parliament.

Parties' dual electoral and governance role is not merely an artefact of our measurement procedures but exist as a central feature of theories of democracy, clearly echoed in contemporary practice. Parties are supposed to define (p.192) the policy alternatives for electors and then to advance these in government or parliament. If they did not, decision‐makers would act outside the frame of reference used by voters, and, therefore, popular preferences would not have a ‘necessary connection’ with policy outputs. It is precisely the dual party role in defining the electoral alternatives and then effecting them that ensures democracy works effectively to transmit popular preferences to decision‐makers.

Seen from this point of view the parties' part in defining the electoral alternatives and ordering them so as to produce median positions, both in electorates and parliaments, appears not as tautological and manipulative but as essential in democratic terms. Without party definition and ordering, democracy would not have a common frame of reference for debate, nor would there be any clear‐cut indication as to what an election outcome implies for decision‐making. From this perspective, the parties' ability to change the meaning of election outcomes by moving their own position appears not just acceptable but essential in substantive terms.

Clearly democracies work on this assumption. Party votes constitute the recorded preferences on which the whole process is based. We cannot try to second guess this normative principle of democracy (Downs 1957: 18) by using some other measure of preference. We have research techniques to use the currency—votes and published party programmes—that democracies themselves use, and we analyse these within the framework of democratic theory to test and evaluate that theory.

These substantive democratic considerations also underpin our measurements. These measure median voter preference on the basis of the party Left‐Right ordering and their share of votes, and relate it to party Left‐Right ordering and their share of legislative seats. To the extent this is tautological it is a tautology embedded in the structure of elections and legislatures, both of which are ordered and organized by parties. The only way electors can express their preferences is by choosing between parties. The only way legislative voting can be ordered and coordinated is through parties. The overlap between the two is substantive and entailed by the very set‐up of party democracy.

What this implies is that responsiveness is not so much a causal function as a translation function, in much the same way that the vote‐seat relationship is not so much causal as translational. Voters record their preferences for parties so that we know the distribution of voter preferences. The voter preference distribution is translated into a distribution of preferences in parliament according to rules. Entry and exit of parties affect the preference distributions we observe on both sides of the translation. Nevertheless, given the parties on offer, it is important to descriptive, empirical, and normative democratic theory to investigate whether shifts in the parliamentary distribution respond to shifts in the voter distribution. That importance is attenuated (p.193) not in the least by knowledge that the parties on offer affect the shape of both distributions.

In the Appendix to this chapter we modify the effects of current party position on estimates of the median voter by varying our measure to take in average party positions over three elections rather than just one. We still relate this to the parliamentary median as reflected in the results from the one election in question. The overall results of this ‘distancing analysis’ of the two electoral and parliamentary measures are much the same as reported in Table 11.1, indicating that legislature responsiveness to median voter preference is not just a measurement artefact based on using the same party ordering in both measures.

From the evidence mustered in this methodological review, as well as from our substantive analyses, we can draw two important conclusions. First, elections do a remarkably accurate job in creating responsiveness between the Left‐Right medians of voter and parliamentary distributions. Where voters are well informed about party positions, the centres of the two distributions move in tandem over a series of elections and track each other in one‐to‐one movements. The tandem movements are more reliable in countries using PR electoral systems, but the same (but less reliable) pattern exists in and among SMD systems. Second, and equally remarkable, party movements along the Left‐Right dimension have a large hand in creating the accurate responsiveness. Were party positions fixed, so that shifts in the vote distributions were the only source of election change, responsiveness would exist in many countries but it would be less reliable and less accurate. This finding makes clear that voters have a responsibility to themselves to know what the parties are offering beyond general, long‐term differences. Electorates that do not accept that responsibility will get in the way of the coordinating potential of elections. The centre of the vote distribution will mean one thing but the parties, who know what they themselves have proposed, will take it to mean something different. This is not an onerous burden on voters; accurate responsiveness will exist in most countries even if voters have an imprecise sense of party positions (see Appendix to this chapter). All‐in‐all, the principal message is that elections are good, accurate coordinating instruments for marking the Left‐Right preferences of electorates and re‐presenting their Left‐Right positions in parliaments over the course of several elections, even though they are not especially reliable for any single election.

11.7 COORDINATION FROM PARLIAMENTS TO GOVERNMENTS

Thus far we have taken for granted that parties mean what they say. Can we really expect that what parliaments produce will reflect the ideological (p.194) disposition we infer they have, given the distribution of party seats? We learned from the analysis of government declarations that parliaments, perhaps looking over their shoulders at the position of the median voter, expressly declare policy intentions in line with the centre of their ideological distribution. However, that relationship was not especially tight; the correlation was only in the neighbourhood of 0.5. One has to wonder whether the slippage during the step from parliament to government arises as a consequence of the nature of the government installed or the policy objectives those governments say they will pursue? To investigate this we can look directly at ideological responsiveness in the step from the parliamentary median to the weighted Left‐Right position of parties in government. As we did for our analysis of the voter to parliament step previously, we start at the most general level, looking across nations and time simultaneously, and proceed to disaggregate the analysis by system type first and then by nation.

We know from Chapter 7 that Left‐Right positions of parliaments and governments match each other with a high degree of accuracy. We also know from other analyses (Laver and Budge 1992; Müller and Strøm 2000) that the party of the parliamentary median is in government under multiparty PR systems about 80 per cent of the time. And, we know from simple logic that the party of the parliamentary median is almost always the single‐government party in dominant two‐party systems. The results of the following equation tell us further that all this is in large part a matter of government policy intentions reflecting parliamentary median positions on a nearly one‐to‐one basis. The translation in the step from parliamentary median to the weighted position of government, as given by the associated linear equation, is:

G o v t i g = 0.40 + 0.93 P a r l M e d i a n i g + ɛ i e ( 0.48 ) ( 0.03 ) R 2 = 0.687 s e = 9.9 N = 456

where Govtig is the weight mean Left‐Right position of parties in government for nation i in government g, ParlMedianig is the Left‐Right position of the parliamentary median in nation i in government g, and ɛie is assumed to be a well behaved error term.

The modestly surprising aspect of the results is the slight but statistically significant tendency for the government positions to tend towards the centre, the zero point on the Left‐Right dimension. There is essentially no bias in this translation, as can be seen in the small and statistically insignificant magnitude of the intercept, but the slope is lower than 1.0. This we suspect is due to governments created through legislative negotiation tending slightly more towards the centre than the parliamentary median itself. Separate analyses of PR and SMD systems show a one‐unit movement in the parliamentary median produces slightly less than a 0.9‐unit movement in the position of (p.195) government under PR, whereas the foregone conclusion of normally single‐majority parliaments under SMD is a one‐to‐one relationship.

PR Systems

G o v t i g = 0.13 + 0.88 P a r l M e d i a n ig + e i e ( 0.67 ) ( 0.04 ) R 2 = 0.555 s e = 11.3 N = 337

SMD Systems

G o v t ig = 0.32 + 1.01 P a r l M e d i a n ig + ɛ ig ( 0.34 ) ( 0.02 ) R 2 = 0.963 s e = 3.7 N = 119

where all variables are as previously defined.

The nation‐by‐nation results of government to parliament responsiveness in Table 11.2 reveal where and therefore how PR systems as a group have a slight tendency to install a more centrist government relative to the position of the parliamentary median. The parliament to government translation in Denmark is so unreliably responsive as to be indistinguishable from zero. It is anyone's bet as to what the ideological disposition of a Danish government will be, regardless of knowing the ideological disposition of a Danish parliament. This may be due to the frequent minority governments there having to carry out legislative wishes regardless of their own ideological predilections. Whatever, the essentially zero slope for Denmark helps to drag the responsiveness of PR systems as a group below 1.0.

That is not all, however. Austrian, Irish, Dutch and Swiss governments are less responsive, more centrist, than the parliaments that produce them. On reflection, these tendencies come from widely recognised facts about these countries, which may reflect their tendencies to ‘consensus democracy’ (Lijphart 1984). Austria and Switzerland often install grand coalition governments. Given that a weighted mean of positive and negative scores is more likely than a weighted median to sum to a value of zero, that is, the median is based on the position of a single party that is involved in the calculation of the positive and negative scores that make up the weighted mean—Austrian and Swiss governments tend towards a centrist position relative to their parliamentary median.

Similarly, Irish governments throughout much of the post‐war period stood as binary alternatives of Fianna Fail in government alone or Fine Gail and Labour in a coalition. That coalition often involved Fine Gail splitting the weighted difference between itself on the right, where it marked the centre of parliament, and Labour to the left, which created a government closer to the centre than the parliamentary median.

(p.196)

TABLE 11.2. Responsiveness of the government's Left‐Right positions to the median parliamentary party's Left‐Right position, by countrya

Intercept

(sa)

Slope

(sb)

r2

se

Austria

−3.33

(3.15)

0.59**

(0.20)

0.346

13.3

Belgium

−2.01

(1.52)

0.80**

(0.20)

0.399

7.1

Denmark

−1.68

(5.30)

0.29**

(0.46)

0.016

22.0

Finland

3.24

(4.30)

0.94**

(0.23)

0.357

17.1

Germany

1.31

(2.40)

1.00**

(0.15)

0.711

10.9

Iceland

1.66

(2.74)

0.83**

(0.19)

0.540

10.6

Ireland

2.30

(1.83)

0.78**

(0.10)

0.774

7.8

Italy

0.86

(1.15)

1.15**

(0.07)

0.866

3.1

Luxembourg

−2.03

(3.16)

1.09**

(0.22)

0.674

7.0

Netherlands

−2.56

(2.87)

0.65**

(0.21)

0.442

9.4

Norway

2.96

(9.30)

0.97**

(0.35)

0.291

11.7

Portugal

0.61

(0.66)

1.02**

(0.07)

0.965

1.9

Spain

0.00

(∼∼)

1.00**

(∼∼)

1.00

0.0

Sweden

2.99

(4.30)

1.19**

(0.17)

0.716

11.7

Switzerland

1.24

(0.99)

0.81**

(0.09)

0.634

6.0

Australia

0.00

(∼∼)

1.00**

(∼∼)

1.00

0.0

Canada

1.37

(1.38)

0.92**

(0.14)

0.736

5.6

France

1.63

(0.88)

0.98**

(0.05)

0.942

4.3

New Zealand

0.00

(∼∼)

1.00**

(∼∼)

1.00

0.0

UK

−1.14

(1.46)

1.04**

(0.06)

0.950

6.0

US

0.00

(∼∼)

1.00**

(∼∼)

1.00

0.0

The Dutch case is more complicated. Until Kok's 1994 coalition government involving PvdA, D 66, and VVD, Dutch Christians were the core around which all governments formed. During the 1950s through to the late 1960s, when the Christians were slightly to the right of centre, they had a slightly greater tendency to coalesce with the left‐of‐centre PvdA than the right‐of‐centre VVD. Since 1967, after which time the Christians themselves moved slightly left of centre, they had a slightly greater tendency to coalesce with the right‐of‐centre VVD than with the left‐of‐centre PvdA.

A fact not to be missed in this discussion of the centrist tendencies of governments in five countries, is the important flip side. The ten remaining PR systems show something close to a one‐to‐one responsiveness in the ideological positions of governments to the ideological centres of parliaments.

SMD systems make this translation one‐to‐one almost by definition. Their statistical results are only needed at the bottom of the table as numerical reminders that on a few occasions after elections in Canada, France, and the UK there was no single‐majority party in parliament and thus a once‐in‐a‐while slip took place between the parliamentary median and the government positions.

(p.197) Finally, the unreliability in moving from the ideological centre of parliaments to governments under PR systems deserves mention. In all but four PR countries (Iceland, Italy, Portugal, and Spain), the errors from our predictions of government Left‐Right positions from parliamentary Left‐Right positions—using the standard errors of estimate as the measure of mispredictions—are about twice as large as the errors in predicting parliamentary positions from median voter positions (see Table 11.1). Even with substantial difficulty predicting any particular government's Left‐Right position, an important fact stands out. The general tendency is for government positions to hover around a one‐to‐one relationship with the centres of their parliaments. Politics swirl in seemingly chaotic ways one day to the next, one year to the next, one government to the next, one parliament to the next, and one election to the next. Nevertheless, in the long run, the swirling politics centre themselves around the middle political ground in each country, marked by the positions of the countries' median voters.

APPENDIX

Distancing Median Voter Estimates from the Measure of Median Party in Parliament

We have argued that using party Left‐Right orderings as an element in the measures both of Median Voter Positions MVLR and Median Parliamentary Party MPPLR reflects the actual practices of democracy rather than a tautology in our measures. Nevertheless it is always useful methodologically to validate measures by varying their procedures and seeing if one gets similar results.

We ‘distance’ the measures here first by using the moving average of party positions over the last, current and next elections to estimate Left‐Right orderings for the median voter calculation. Substantively this implies that voters may be a bit hazy about current party policy but have a good general idea about where they have come from and where they are moving to, over a nine to twelve year period. In contrast the MPPLR position is still estimated in terms of where it is after the current election. This realistically assumes greater knowledge of the actual party position on the part of its leadership, and approximations (but not unrealistic ones) by electors. Methodologically it implies that there is less overlap between the two measures so the results of the responsiveness analysis can be taken more at face value.1 Of course, the great (p.198) interest in the analysis is whether it corroborates the substantial responsiveness already observed.

Statistical results are reported in Table 11.A.1. Responsiveness is evident in sixteen out of twenty‐one countries. Along with France, for which there is no evidence of responsiveness even were voters to be aware of the precise party positions at each election (see Table 11.1), responsiveness is absent in Germany, Iceland, Portugal, and New Zealand. That is the negative news. Elsewhere, which is to say in over three‐quarters of our twenty‐one countries, responsiveness not only exists but is often statistically indistinguishable from one‐to‐one. In a large majority of PR systems and 50 per cent of SMD systems, even if voters have only an imprecise sense of what party positions are, elections will create accurate responsiveness between the position of median voters and parliamentary medians.

The coincidence in general results between Tables 11.1 and 11.A.1 increases confidence in the validity of our median voter measure for all our analyses.

A further investigation of validity was offered by an anonymous reviewer of our book MS for OUP, who allowed us to make public the following comments and analysis which we report here verbatim:

TABLE 11.A.1. Alternative estimation of responsiveness of the median parliamentary party's Left‐Right position to the median voter Left‐Right position, by country: median voter position measured with a three‐election moving average of party positions

Intercept

(sa)

Slope

(sb)

r2

se

Austria

−0.08

(3.37)

0.89**

(0.27)

0.448

11.6

Belgium

0.90

(1.86)

0.81**

(0.23)

0.489

5.7

Denmark

−1.74

(1.64)

1.03**

(0.18)

0.654

5.9

Finland

11.26

(5.42)

1.74**

(0.26)

0.675

8.2

Germany

0.43

(5.00)

0.76

(0.94)

0.062

16.4

Iceland

−3.68

(3.83)

0.21

(0.93)

0.005

13.5

Ireland

−4.04

(4.32)

1.21**

(0.35)

0.494

14.6

Italy

−1.32

(1.25)

0.76**

(0.15)

0.742

4.0

Luxembourg

1.37

(4.06)

0.97**

(0.26)

0.628

5.6

Netherlands

7.50

(4.52)

1.56**

(0.43)

0.540

9.0

Norway

−0.45

(5.93)

1.04**

(0.24)

0.682

4.8

Portugal

3.83

(6.23)

1.04

(0.62)

0.309

8.9

Spain

9.73

(3.73)

1.60**

(0.26)

0.904

3.5

Sweden

0.03

(7.24)

1.02**

(0.33)

0.420

13.4

Switzerland

−1.86

(1.85)

1.07**

(0.19)

0.754

5.0

Australia

3.46

(5.88)

1.08*

(0.58)

0.179

21.1

Canada

1.19

(2.45)

0.90*

(0.42)

0.282

7.9

France

8.04

(7.75)

0.46

(0.81)

0.38

17.1

New Zealand

0.88

(7.95)

0.53

(0.70)

0.043

16.2

UK

16.86

(6.61)

2.15**

(0.51)

0.622

15.9

US

3.46

(5.88)

1.08*

(0.58)

0.179

21.1

(p.199)

In considering the likely critical reception of this book, the most radical aspect of the authors' methodology involves their reliance on election results to infer the location of the median voter. Briefly, the authors use the distribution of votes across parties, combined with the parties' Left‐Right positions as coded by the Comparative Manifesto Project (CMP), to locate the median voter in each election (this is an adaptation of a method originally developed by Kim and Fording, 1998). Now, I believe that this is a reasonable procedure, and indeed I think this is the only feasible procedure given that the obvious alternative approach—namely, to employ survey data to estimate the voter medians—is unworkable, since reliable, cross‐nationally comparable survey data on voter ideologies is not available over the range of countries and time periods the authors analyse.

The authors point out that even if such survey data were available, its use in cross‐national analyses is problematic because respondents' Left‐Right self‐placements are not necessarily comparable across different countries (i.e. the fact that the median respondent self‐placement in Norway is similar to the median respondent self‐placement in Britain does not prove that the distributions of Norwegian and British citizens' ideologies are actually similar). This is a good point, and to this I would add that even if this problem could somehow be overcome—which it cannot—an additional problem with relying on survey data to measure citizens' ideologies is that strong evidence exists that citizens' Left‐Right self‐placements are subject to assimilation effects, that is, that citizens tend to place themselves unduly close to parties they like for non‐policy‐related reasons. Huber, for instance, reports that this is the case with Eurobarometer respondents' ideological self‐placements in Ireland, Germany and Belgium (see Huber, ‘Values and Partisanship in Left‐Right Orientation: Measuring Ideology’, European Journal of Political Research 17 (1989: 599–621). These effects render the use of public opinion survey data problematic, if the goal is to understand how shifts in voters' ideologies affect their voting behaviour.

The above observations notwithstanding, I expect that some readers will nevertheless find the Budge‐McDonald methodology for locating voter medians to be problematic. The authors have anticipated this, and they present several reasons why their approach is reasonable (for instance they demonstrate that election results are not substantially influenced by economic conditions—thereby eliminating one possible outside influence that might bias their ideological estimates—and they have also just demonstrated that their ideological estimates are similar if they use party positions over three elections rather than just using the current election). While these arguments are reassuring, a natural question that arises is: ‘Would the authors' substantive conclusions have been different, had they been able to rely on cross‐national survey data to estimate the ideological medians’?

Since the authors do not explicitly report their estimates of the median voter positions I cannot answer this question directly. However I do have access to the Kim–Fording median voter estimates—which are presumably quite similar to the Budge–McDonald estimates, since the latter were constructed using a variation of the Kim–Fording coding procedure—and I also have survey‐based measures of voters' ideological positions, as reported in Eurobarometer surveys. I therefore decided to compare the Kim–Fording estimates of the median voter positions against mean voter positions as computed from Eurobarometer respondents' Left‐Right self‐placements. (p.200) I based my comparisons on the Eurobarometer data from eight countries—Britain, France, The Netherlands, Greece, Spain, Denmark, Luxembourg and Italy—since Huber's (1989) analyses of the Eurobarometer surveys suggest that the ideological self‐placements from the remaining countries included in the Eurobarometer surveys are unreliable. This data are presented in Table 11.A.2, which reports the Kim–Fording estimates of the voters' median Left‐Right positions, compared with voters' mean positions as computed from the Eurobarometer data. (Ideally these tables should report median Eurobarometer respondents' self‐placements, not means, so (p.201)

TABLE 11.A.2. Voters' median Left‐Right positions as coded by Kim and Fording using the CMP data, versus voters' mean Left‐Right positions as computed from Eurobarometer data

Country

Election year

Left‐Right median Kim‐Fording coding

Left‐Right mean Eurobarometer

Denmark

77

−6.38

5.33

Denmark

79

1.37

5.59

Denmark

81

−0.71

5.63

Denmark

84

−10.22

5.67

Denmark

87

−23.17

5.56

Denmark

88

−1.52

5.87

Denmark

90

minus;6.69

5.57

Denmark

94

−2.67

5.62

Denmark

98

−3.20

5.54

Netherlands

77

−10.87

5.72

Netherlands

81

−19.06

5.31

Netherlands

82

−11.39

5.44

Netherlands

86

−3.34

5.3

Netherlands

89

−10.74

5.28

Netherlands

94

1.34

5.3

Netherlands

98

−10.95

5.01

Luxembourg

79

−15.03

5.63

Luxembourg

84

−16.32

5.65

Luxembourg

89

−2.36

5.61

Luxembourg

94

−17.10

5.46

France

78

9.22

5.09

France

81

−11.34

4.78

France

86

5.40

5.22

France

88

−0.52

5

France

93

−5.15

4.92

France

97

2.10

4.6

Italy

76

−1.00

4.13

Italy

79

−8.17

4.3

Italy

83

−4.50

4.63

Italy

87

4.91

4.73

Italy

92

9.00

4.85

Spain

86

−7.01

4.8

Spain

89

−21.40

4.55

Spain

93

−22.32

4.55

Spain

96

−18.10

4.71

Greece

81

16.39

6.12

Greece

85

−8.56

5.04

Greece

89

5.30

5.38

Greece

89

14.48

5.38

Greece

90

2.41

5.66

Greece

93

−7.92

5.59

Greece

96

−12.37

5.61

Great Britain

79

−3.60

5.89

Great Britain

83

2.38

5.87

Great Britain

87

3.28

5.92

Great Britain

92

−5.52

5.41

Great Britain

97

11.55

5.02

that they would be more directly comparable to the Kim–Fording median voter estimates. However I happened to have the mean data at hand and so relied on that; my assumption is that the median Eurobarometer self‐placements are virtually identical to the means).

The time period covered is from 1976, the first year for the Eurobarometer surveys to 1998, the most recent year for which the CMP data is available. Note that the Kim–Fording codings of the voter medians are on a scale running from −100 to +100, while the Eurobarometer scale runs from 1–10.

Using these data, I computed the correlation between these two measures of voters' positions. The correlation is only 0.14, which is not statistically significant. To me this strongly suggests that use of the Eurobarometer surveys might have supported quite different conclusions than the ones the authors report, in cross‐national analyses based on data from a single time period. (This assumes, of course, that the Budge–McDonald codings of voter medians are similar to the Kim–Fording codings.) However I then computed the correlation between the changes in voters' Left‐Right positions between elections, as computed from the Eurobarometer data and from the Kim–Fording method. To clarify, the variable I constructed was the difference between the median voter's Left‐Right position in the current election and the median voter's position at the previous election, using both the Budge–McDonald measure and the Eurobarometer data. So for Denmark 1977–79, for instance, the change in the Budge‐McDonald coding of the median voter's position is given as [1.37−(−6.38)]=7.75, while use of the Eurobarometer data gives a value of (5.59–5.33)=−0.26.)

The dynamic correlation was 0.55, which is statistically significant at the 0.01 level. What this result states is that in situations where the Kim–Fording method registers that the median Left‐Right voter position in a country has shifted, there is a strong tendency for the Eurobarometer surveys to register an ideological shift in the same direction. To me this suggests that the authors' central conclusion, that over time (p.202) Western democracies tend to faithfully translate the preferences of the median voter into government policy, would likely have been unchanged, had the authors employed survey data to measure shifts in voters' ideologies, rather than relying on their own method for estimating these shifts.

We thank this conscientious reviewer for his comment and analysis. We are naturally pleased that it strengthens confidence in our main findings, even though we do feel for the reasons that we have stated that the survey‐based measure of median (mean) position is less valid than our own estimate.

Notes:

** p < 0.01; one‐tail test for slopes and two‐tail test for intercept.

* p < 0.05;

a Caretaker, nonpartisan, and transition governments excluded.

** p < 0.01; one‐tail test for slopes and two‐tail test for intercept.

** p < 0.01; one‐tail test for slopes and two‐tail test for intercept.

*p < 0.05;

(1) One could of course take the mean‐party position over 20–30 years or the whole post‐war period. This seems unrealistic however in terms of the current electorate's information horizons. In the 1990s they are unlikely to take much account of what politics were like in the 1950s or 1960s. Even if party positions do not change greatly (see Chapter 5) this information will be transmitted through the current decade's positions.

(1) This is necessarily so since Y‐mean = a + b X‐mean. Given a small bias, a is near zero. Given a near one‐to‐one correspondence, b is close to one. In that case, the equation can be written as Y‐mean = X‐mean.

(2) However, in the Appendix to this Chapter, we do try to separate them out, partially, by ‘fixing’ party positions. Certainly in the estimation of median vote preference one might attempt to go beyond the aggregate vote to ascertain preferences. Within the context of an individual country one might use surveys to get at electors' ‘real’ preferences. However, this would take us into a general debate as to whether spontaneous survey responses are any more ‘real’ than a considered vote for a party programme. There is the additional question of which survey question is more ‘real’—general Left‐Right positioning or, for example, attitudes to immigrants. As pointed out in Chapter 3, survey responses are not entirely comparable between countries, rendering the survey option inoperative for a comparative investigation like ours.

(3) This set of equations could be estimated with an algebraic equivalent single equation, which has a less obvious interpretation than the set presented but which does provide a more straightforward view of whether the difference in slopes (0.96 versus 1.08) is statistically significant. As stated, it is not. The difference, 0.12, has a standard error of 0.13, thus t < 1 (p = 0.181, one tail test).

(4) This corroborates what we reported in Chapter 7. Viewed in a regression format, so as to take account of responsiveness, the standard error of estimate is the regression analogue to distortion, which is greater under SMD rules than under PR rules. The intercept is the regression analogue to bias, which is tilted to the right among SMD systems compared to PR systems.