¹ McDonald, Michael D. and Budge, Ian, Elections, Parties, Democracy: Conferring the Median Mandate (Oxford: Oxford University Press, 2005)CrossRefGoogle Scholar.

² According to McDonald and Budge (Elections, Parties, Democracy, pp. 20–1), an interpretation in terms of government mandates is implicit in such well-known models as the ‘Westminster model’ and the ‘responsible party model’. The key conditions are that parties have distinct policy positions, that voters recognize these distinctions and vote accordingly, and that one choice (party or coalition) receives a majority of votes, forms the government and carries out its electoral commitments.

³ McDonald and Budge, Elections, Parties, and Democracy, pp. 22–4.

⁴ McDonald and Budge, Elections, Parties, and Democracy, p. 108.

⁵ McDonald and Budge's evidence for one-dimensional policy spaces is presented in Elections, Parties, and Democracy, pp. 81–7. The median voter theorem, as it is known, is attributable to Black, Duncan, ‘On the Rationale of Group Decision-making’, Journal of Political Economy, 56 (1948), 23–34CrossRefGoogle Scholar.

⁶ Downs, Anthony, An Economic Theory of Democracy (New York: Harper & Row, 1957)Google Scholar.

⁷ McDonald, and Budge, , Elections, Parties, and Democracy, pp. 61–77Google Scholar.

⁸ Very recently, McDonald and Budge (‘From Popular Majorities to Policy Representation: How Democracy Represents the Median Citizen by Responding to Median and Modal Voters’, read at the Mini-Conference on Citizen Preferences, Political Institutions, and Democratic Performance, at Florida State University, 2009) have attempted to defend the median mandate interpretation by suggesting that governments, for reasons beyond their control, may end up implementing a policy stance that is much more centrist than they had promised. This revised understanding of the median mandate is discussed in the conclusion.

⁹ McDonald, and Budge, , Elections, Parties, and Democracy, pp. 184–97Google Scholar. Strictly speaking, a one-to-one relationship would imply that the positions in question are aligned, not just that they move together; i.e. in regression terms, that the intercepts in the two relationships are not significantly different from zero. The focus of McDonald and Budge's discussion of responsiveness is on the slope, however, and that focus will be preserved here.

¹⁰ As McDonald, and Budge, , Elections, Parties, and Democracy, pp. 186–7Google Scholar, put it: ‘A failure to find shifts in parliamentary and government preferences in response to electoral ones would be enough to dismiss our median mandate thesis’.

¹¹ Kim, HeeMin and Fording, Richard, ‘Voter Ideology in Western Democracies, 1946–1989’, European Journal of Political Research, 33 (1998), 73–97CrossRefGoogle Scholar; Kim, HeeMin and Fording, Richard, ‘Government Partisanship in Western Democracies, 1945–1998’, European Journal of Political Research, 41 (2002), 187–206CrossRefGoogle Scholar.

¹² More systematic evidence on this point will be provided later.

¹³ McDonald, and Budge, , Elections, Parties, Democracy, pp. 114–15Google Scholar.

¹⁴ Hermann Schmitt and Evi Scholz, The Mannheim Eurobarometer Trend File, 1970–2002 [Computer file], prepared and distributed by the Zentralarchiv fur Empirische Sozialforschung (Mannheim, Germany: Mannheimer Zentrum fur Europaische Sozialforschung and Zentrum fur Umfragen, Methoden und Analysen, 2005).

¹⁵ Hermann Schmitt and Matthew Loveless, The European Election Study 2004. [Computer file], prepared and distributed by the Mannheimer Zentrum fur Europaische Sozialforschung (Mannheim, Germany: Mannheimer Zentrum fur Europaische Sozialforschung, 2004).

¹⁶ McDonald, and Budge, , Elections, Parties, Democracy, p. 114Google Scholar.

¹⁷ Matt Golder and Jacek Stramski, ‘Ideological Congruence and Two Visions of Democracy’ (paper read at Midwest Political Science Association, Chicago, 2007).

¹⁸ Huber, John and Powell, Bingham, ‘Congruence between Citizens and Policy-Makers in Two Visions of Liberal Democracy’, World Politics, 46 (1994), 291–326CrossRefGoogle Scholar.

¹⁹ This assumption is probably more damaging when the Kim–Fording medians are calculated. Since intervals are created around parties rather than around the eleven points of the left–right scale, there will normally be fewer and hence larger intervals.

²⁰ Browne, Eric and Franklin, Mark, ‘Aspects of Coalition Payoffs in European Parliamentary Democracies’, American Political Science Review, 67 (1973), 453–469CrossRefGoogle Scholar; Warwick, Paul V. and Druckman, James N., ‘The Portfolio Allocation Paradox: An Investigation into the Nature of a Very Strong but Puzzling Relationship’, European Journal of Political Research, 45 (2006), 635–665CrossRefGoogle Scholar.

²¹ McDonald, and Budge, (Elections, Parties, Democracy, pp. 205–25Google Scholar) provide some evidence that the position of the median voter and especially the median party are related to government spending, but reducing a government's policy output to a single point on a left–right scale in order to test for a one-to-one correspondence with median voter or citizen opinion is altogether another matter.

²² Budge, Ian and Laver, Michael, ‘Coalition Theory, Government Policy and Party Policy’, in Michael Laver and Ian Budge, eds, Party Policy and Government Coalitions (New York: St. Martin's, 1992), 1–64Google Scholar. The procedure produces a left-right position for each party in each election by subtracting the proportion of the party's manifesto devoted to thirteen topics identified as left-wing from the proportion devoted to thirteen topics identified as right-wing.

²³ Note that converting both variables to standard deviation units will not allow this type of hypothesis to be assessed. If government positions vary more broadly than citizen medians because changes in the latter are amplified in the former (i.e. the effect is greater than one-to-one), standardization would tend to neutralize this amplifying process, thereby encouraging the emergence of a ‘false positive’ for the median mandate hypothesis. In fact, as Achen (‘Measuring Representation: The Perils of the Correlation Coefficient’, American Journal of Political Science, 21 (1977), 805–15, p. 807) demonstrated, the use of standardized measures compromises any assessment of the relationship between public and party or leader opinion by making it conditional on sample variances, which can easily change from one sample to the next.

²⁴ The vote intention question, unlike the ‘last vote’ question, has been asked consistently since 1973.

²⁵ The accuracy of the Kim–Fording estimates of voter medians, which is the main inferential leap in McDonald and Budge's methodology, could be tested using survey-based median estimates, but McDonald and Budge (Elections, Voters, Democracy, pp. 114–15) have doubts about the cross-national comparability of the latter, as noted earlier. They have no such doubts about the accuracy of expert assessments of party positions, however, which is the standard that will be used here.

²⁶ Benoit, Kenneth and Laver, Michael, Party Policy in Modern Democracies (New York: Routledge, 2006)Google Scholar.

²⁷ In a comparison of several methods for measuring party positions, Gabel, Matthew J. and Huber, John D., ‘Putting Parties in Their Place: Inferring Party Left–Right Ideological Positions from Party Manifestos Data, American Journal of Political Science, 44 (2000), 94–103CrossRefGoogle Scholar, p. 98, also report a close correspondence between this method of placing parties and mean expert placements.

²⁸ Klingemann, Hans-Dieter, Andrea Volkens, Judith Bara, Ian Budge and Michael McDonald, Mapping Policy Preferences II: Estimates for Parties, Electors and Governments in Eastern Europe, the European Union and the OECD, 1990–2003 (Oxford: Oxford University Press, 2006)Google Scholar.

²⁹ Benoit, and Laver, , Party Policy in Modern Democracies, p. 274Google Scholar.

³⁰ The regression of the Benoit–Laver positions on the survey-based positions yields a slope of b = 1.217 (SE = 0.075), which is significantly above unity. This indicates that the survey-based positions are a little more compressed than the expert positions. If so, the likely effect will be to produce government position estimates that are biased towards the centre. Given that survey-based estimates of the citizen median tend to be close to the centre as well, the net effect ought to be to advantage the median mandate hypothesis. Evidence against the hypothesis, therefore, will be especially convincing.

³¹ This also happens when one relies on CMP (manifesto) data, as in McDonald and Budge (Elections, Parties, Democracy). The CMP data yield estimates of party positions at elections only, so that an implicit assumption of their analyses is that the election-time positions of parties can be used to estimate both dependent and independent variables for the entire period of the ensuing legislatures. The case for estimating scores between data collection points is more evident here, however, because the data collection is not aligned with either elections or government formations.

³² Note that it is not the government's position that is interpolated, but rather the positions of its member parties. Government positions can change abruptly when one government replaces the other (i.e. as their party composition changes), but party positions are likely to be much less erratic.

³³ The countries, with numbers of cases with valid data in brackets, are: Austria (4), Belgium (15), Denmark (16), Finland (5), France (16), Ireland (12), Italy (26), Greece (10), Luxembourg (8), Netherlands (9), Norway (2), Portugal (7), Spain (6), Sweden (3), United Kingdom (10), and (West) Germany (12). McDonald and Budge's (2005) analyses, based on a different data source, covered an additional five countries (Iceland, New Zealand, Canada, Switzerland and Australia) and extended across the entire post-war period. Comparisons conducted on that dataset indicate that their responsiveness relationship does not change appreciably if the analysis is confined to the countries and periods covered here, indicating that any differences in findings that emerge here are highly unlikely to be due to differing samples.

³⁴ This reasoning also justifies estimating citizen median values for government formation times by means of interpolation. If median citizen opinion shifts suddenly when a new government takes office, smoothly interpolating values between EB/EES surveys would not make sense.

³⁵ Stimson, James, ‘Regression in Space and Time: A Statistical Essay’, American Journal of Political Science, 29 (1985), 914–947CrossRefGoogle Scholar, at p. 945.

³⁶ Beck, Nathaniel and Katz, Jonathan, ‘What To Do (and Not to Do) with Time-Series-Cross-Section Data in Comparative Politics’, American Political Science Review, 89 (1995), 634–647CrossRefGoogle Scholar; Beck, Nathaniel and Katz, Jonathan N., ‘Nuisance vs. Substance: Specifying and Estimating Time-Series-Cross-Section Models’, Political Analysis, 6 (1996), 1–36CrossRefGoogle Scholar.

³⁷ The method used here for calculating the degree of autocorrelation is to regress the residuals on their lags and the other independent variables in the model ( Beck, Nathaniel, ‘Time-Series Cross-Section Models’, in J. Box-Steffensmeier, H. Brady and D. Collier, eds, The Oxford Handbook of Political Methodology (Oxford: Oxford University Press, 2008), pp. 475–483Google Scholar, at p. 478). The estimated autocorrelation coefficient is ρ = 0.44, p < 0.001.

³⁸ McDonald, and Budge, , Elections, Parties, Democracy, pp. 114–15Google Scholar.

³⁹ Rabe-Hesketh, Sophia and Anders Skrondal, Multilevel and Longitudinal Modeling Using Stata, 2nd edn (College Station, Tex.: Stata Press, 2008), pp. 114–115Google Scholar.

⁴⁰ Wilson, Sven and Butler, Daniel, ‘A Lot More To Do: The Sensitivity of Time-Series Cross-Section Analyses to Simple Alternative Specifications’, Political Analysis, 15 (2007), 101–123CrossRefGoogle Scholar. This step is not cost-free, however, since it induces a correlation between the lagged dependent variable and the error term, violating a core assumption of OLS regression. Wilson and Butler's survey of the available evidence finds, however, that the bias in estimated slopes (apart from that of the lagged dependent variable) is ‘quite small for the types of data sets typically used in political science’ (p. 108).

⁴¹ What is lost is the ability to determine whether the one-to-one relationship holds with respect to intercepts (i.e. whether intercepts are indistinguishable from zero). As noted earlier, however, McDonald, and Budge, , Elections, Parties, and Democracy, pp. 117, 186–7Google Scholar, make it clear that the slope is the key to assessing responsiveness (and that responsiveness is critical to the median mandate hypothesis).

⁴² Beck, ‘Time-Series Cross-Section Models’, p. 478.

⁴³ The variables are lagged to the start of the preceding government. This lag is great enough to ensure that even interpolated values of the lags only contain information from before the government was formed.

⁴⁴ This increase is entirely due to mean-centring the variables; without the lagged dependent variable in the model, the citizen median slope would be slightly higher at 1.54.

⁴⁵ Achen, Christopher, ‘Why Lagged Dependent Variables Can Suppress the Explanatory Power of Other Independent Variables’ (presented at the Annual Meeting of the Political Methodology Section of the American Political Science Association, Los Angeles, 2000), p. 7Google Scholar.

⁴⁶ The long-term effect is given by β ₂/(1−β ₁), where β₂ is the estimated slope for the citizen median and β ₁ is the estimated slope for the lagged dependent variable.

⁴⁷ This calculation takes no account of the external support that minority governments may depend upon; hence, it is possible that it exaggerates the degree of partisanship in coalition, situations. However, if we exclude minority coalitions, the percentage falls only slightly to 78.5 per cent, and therefore it is clear that majority coalitions, at least, are highly partisan.

⁴⁸ Mean non-congruence and mean imbalance are essentially equivalent to what McDonald and Budge (Elections, Parties, Democracy, pp. 122–32) call (overall) distortion and (overall) bias, respectively. The latter terms are not used here because they imply that the norm should be congruence between citizen/voter medians and government positions.

⁴⁹ Spain and Portugal are also interesting in that their patterns are almost perfect opposites of each other.

⁵⁰ Duverger, Maurice, Les Partis politiques, 2nd edn (Paris: Colin, 1954), p. 247Google Scholar.

⁵¹ These include Huber and Powell, ‘Congruence between Citizens and Policy-Makers’; Lijphart, Arend, Patterns of Democracy (New Haven, Conn.: Yale University Press, 1999), p. 288Google Scholar; McDonald and Budge, Elections, Parties, Democracy, pp. 128–9; and Bingham Powell, G. Jr, ‘Election Laws and Representative Governments: Beyond Votes and Seats’, British Journal of Political Science, 36 (2006), 291–315CrossRefGoogle Scholar.

⁵² Blais, André and Marc André Bodet, ‘Does Proportional Representation Foster Closer Congruence Between Citizens and Policy Makers?’ Comparative Political Studies, 39 (2006), 1243–1262CrossRefGoogle Scholar.

⁵³ Golder and Stramski, ‘Ideological Congruence’.

⁵⁴ Powell, G. Bingham Jr, ‘The Ideological Congruence Controversy: The Impact of Alternative Conceptualizations and Data on the Effect of Election Rules’, Comparative Political Studies, 42 (2009), 1475–1497CrossRefGoogle Scholar.

⁵⁵ Lijphart, Arend, Patterns of Democracy, p. 162Google Scholar.

⁵⁶ The Gallagher index ( Gallagher, Michael, ‘Proportionality, Disproportionality, and Electoral Systems’, Electoral Studies, 10 (1991), 33–51CrossRefGoogle Scholar) is √{0.5∑_i(v_i−s_i)²},where v_i and s_i represent party i's vote and legislative seat shares, respectively.

⁵⁷ Estimates may also be calculated using the maximum likelihood method. Rabe-Hesketh and Skrondel (Multilevel and Longitudinal Modelling, p. 82) recommend the empirical Bayes method because it tends to downplay the influence of countries that provide little information, but, in practice, the choice of method makes very little difference to the results to be presented here.

⁵⁸ The correlation between estimated slope and mean non-congruence across all fifteen countries (France excluded) is r = 0.50 (p = 0.024 in a one-tailed test).

⁵⁹ McDonald and Budge, Elections, Parties, and Democracy, pp. 135–6, 143, 226.

⁶⁰ Budge and McDonald, ‘From Popular Majorities to Policy Representation’.

⁶¹ Sartori, Giovanni, Parties and Party Systems (Cambridge: Cambridge University Press, 1976)Google Scholar.