Retrospective ideological representation and its impact on democratic satisfaction^*

* This article is accompanied by an online appendix available as supplemental data. Replication data can be obtained from Harvard Dataverse (https://dataverse.harvard.edu).

ABSTRACT

Congruence between government policies and citizen preferences is a key element that increases the quality of democracy. While scholars have shown that governments generally adopt ideological positions and propose policies close to citizen preferences, they have neglected to consider whether citizens respond to promises or to actual enactments. The paper addresses this gap in two ways. First, we propose a new measure that captures how close a citizen is on average to the policies enacted by the incumbent government, namely retrospective ideological representation – the ideological distance between the position of a respondent and the incumbent government at the end of its term in office. We show that this measure captures retrospective information associated with governments’ actions and in particular whether the government increases or decreases social spending during its mandate period. Second, we show that retrospective ideological representation has a substantial impact on citizens’ democratic satisfaction and greater than prospective ideological representation – an established measure of congruence – which is the ideological distance between the positions of a respondent and the elected government after an election.

Acknowledgment

I wish to thank Elisabeth Gidengil, Stuart Soroka, André Blais, Matt Golder and the journal’s reviewers and editor for their helpful comments.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes on contributor

Benjamin Ferland is Postdoctoral Scholar in the Department of Political Science, Pennsylvania State University, USA and Assistant Professor at the School of Political Studies, University of Ottawa, Canada.

Notes

* This article is accompanied by an online appendix available as supplemental data. Replication data can be obtained from Harvard Dataverse (https://dataverse.harvard.edu).

1 Ferland (2015) argues that instead of an accurate translation of votes into seats, “citizens prefer their party to be advantaged and the other parties to be disadvantaged in the votes-seats translation” (395).

2 At election time, an incumbent government is the one that has been in power until election day. An incumbent government could win the election and form again the government after the election but it could also lose the election and be replaced by a newly elected government.

3 This perfect collinearity between retrospective and prospective representation will be weakened when reelected governments are pooled in the analysis with out-going governments that are replaced at the election making it possible then to assess the respective impact of retrospective and prospective ideological congruence on citizens’ satisfaction with democracy.

4 Note that a citizen initially far away from (close to) an incumbent government will not be more (less) satisfied with democracy only because the incumbent government is replaced by a new government. Whether a citizen will be more or less satisfied as a result of a change in government only depends on her congruence with the newly elected government (this is prospective ideological representation), congruence that may be good or bad notwithstanding her congruence with the previous government.

5 We treat Iceland and Ireland as parliamentary democracies since the presidential role has been mostly symbolic for the period under study.

6 It was possible to test this assumption using the German Longitudinal Election Study (2002–2005–2009). Out of six German parties and over three surveys, we find that no party changed its position significantly between the pre and post-election surveys (based on difference-in-mean tests with a 0.05 level of statistical significance).

7 GDP is computed in thousands of US dollars per capita, constant PPPs, OECD base year. Note that we checked for possible multicollinearity among these control variables and did not find any (VIF scores equal about 3).

8 Note that the results in Table 1 are substantively the same when we use GLS regressions with random effects (see Table C in Appendix).

9 Don’t knows were coded as missing values.

10 Some readers may be worried that there is an endogeneity issue given that we use an individual-specific measure of government positions. Indeed, it is possible that respondents who are a-priori satisfied with democracy will be more likely to locate the government’s position closer to their own position. Note that the results are substantively the same when we use the mean government position as perceived by all respondents instead of an individual-specific measure (see Tables H–K in the Appendix).

11 Let us take an example to clarify our measures of congruence. A coalition forms in 1999 after an election. The coalition is composed of parties A, B and C. The coalition stays in office until the 2003 election. After the 2003 election, the new coalition is composed of parties C, D and E. In 2003, our measure of retrospective ideological congruence is based on the ideological distance between the ABC incumbent government and a respondent’s position while our measure of prospective ideological congruence is based on the ideological distance between the new CDE government and a respondent’s position. Given that the CSES project gathers parties and respondents’ positions in post-election surveys we assume that these positions are essentially the same before and after the 2003 election. Consequently, prospective ideological congruence would have equalled retrospective ideological congruence if the ABC coalition had also formed the government after the 2003 election.

12 The results are substantively the same when using OLS regressions (see Tables U–X in Appendix).

13 See Table P in Appendix for the full results of the ordered logistic regressions.

14 Note that we have also tested for the impact of party identification with other parties in general and found substantively the same results (see Table T in appendix).

15 Note that the marginal effect of retrospective ideological congruence on citizens’ likelihood of being not very satisfied with democracy is of similar magnitude and in the expected direction (negative). The effects, however, are much smaller with respect to the probability of being not at all and fairly satisfied (−0.51 and 0.61% points, respectively) – see Tables E–G in Appendix.

16 Some readers may wonder whether there is a collinearity problem between winner and prospective ideological congruence. Even if winners should be generally closer to governments than losers, the correlation between these two variables remains modest (0.43).

17 Note that we also examined the validity of this CMP-based measure of retrospective ideological congruence as done in Table 1. Based on data from the CMP, we examined whether the position of the newly elected government at the time of the previous election (t − 1) influences social spending during the mandate period (from t − 1 to t). The effect of government position was in the expected direction (negative) but not statistically significant (see Table D in Appendix). While this result casts doubt on the validity of this measure for capturing what governments implemented in terms of policies during their term in office, we should note that empirical findings about the effect of government ideology on public/social spending are tenuous at best (for a review, see Imbeau, Pétry, and Lamari 2001). On the other hand, CMP-measures of government positions may capture other aspects of government policies than social spending. For example, scholars showed that levels of pledges fulfillment in party manifestos are high in advanced democracies (Mansergh and Thomson 2007; Thomson et al. 2014). These findings suggest that CMP-measures of government positions also capture substantive information about governments’ enacted policies although not necessarily the same information than in survey-based measures. This assertion is consistent with Dalton and McAllister’s findings (2015) about the lower correlation found between survey-measures and CMP-measures of party positions (0.64) than between survey-measures and expert-measures (0.89) or candidate-measures (about 0.9).

Additional information

Funding

The author thanks Government of Canada and Social Sciences and Humanities Research Council of Canada (SSHRC) [756-2014-0457] for its financial support.