![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
Abstract
The existing literature on ideological congruence has typically looked at congruence immediately after elections when governments are formed. This article goes beyond that comparative static approach by examining changes in citizen-government ideological congruence between two fixed points in time, namely at the beginning and end of government mandates. Building on a veto player approach and dynamics of party competition under majoritarian and proportional representation (PR) electoral systems, the results indicate, first, that government positions are more stable in between elections, as the number of parties and their ideological distance increase in cabinet. Second, it appears that single-party and homogeneous coalition governments decrease ideological congruence between elections under low levels of polarisation, while they increase congruence under very high levels of polarisation. Third, it was found that governments under majoritarian systems slightly decrease congruence between elections while congruence stays stable on average under PR systems. The different levels of party system polarisation across majoritarian and PR electoral systems mostly explain this difference.
Acknowledgements
I wish to thank Elisabeth Gidengil, Stuart Soroka, André Blais, Matt Golder, Shaun Bowler, Indridi Indridason, Russell Dalton, Bingham Powell, and the journal’s reviewers for their helpful comments.
Notes
1. Even if it is more difficult to change the status quo under fractionalised and polarised cabinets this does not imply that adjustments in governments’ positions are impossible. Instead, the mechanism allowing for changes in government position could be different in PR systems. Indeed, in PR systems cabinet terminations do not necessarily lead to an election. A change in the cabinet composition could occur in between elections if the party composition of the cabinet changes or if there is a change of prime minister. When such events occur, the government position may be changed significantly. This mechanism is not available to single-party majority governments under majoritarian systems since cabinet terminations lead to a new election. My research design (see below) does not, however, allow me to test this assumption directly, and especially whether changes in government positions under PR systems improve congruence or not. For this reason, I do not offer any hypothesis associated with the possible impact of change in cabinet on change in congruence (see Kim and Fording Citation2012 for a study on this topic).
2. Note that parties’ share of cabinet portfolios and, therefore, the distance of the government position from each coalition member is not crucial here. What is important is that once a coalition government is formed I assume that each coalition member is a veto player whose agreement is necessary for changing the government position.
3. I follow the recommendations of Berry et al. (Citation2012) for the theory positing an interaction effect. Berry et al. suggest that the hypotheses should indicate whether the marginal effect of X is positive, negative or null when the modifying variable is at its minimum and maximum value as well as the direction of the interaction (positive or negative).
4. Note that Australia is a single-member district system using the Alternative Vote system. Due to its small district magnitude (1) and the majoritarian features of its electoral system, Australia is coded as a majoritarian system. In addition, I treat the Australian coalition governments formed by the Liberal and National parties as single-party majority governments as proposed by Blais and Bodet (Citation2006: 1256).
5. I include Switzerland because the president does not have the power to replace the government. Moreover, the Swiss Federal Council has significant power similar to an executive cabinet in a parliamentary system. The results are substantively the same when I exclude Switzerland from the analysis.
6. The 0.98 correlation between the position of the median citizen and the median voter is extremely high. I follow Warwick’s (Citation2011) procedure and transform the respondents’ left‒right position into a continuous variable. The main implication is that instead of having median citizens for each country located at position 4, 5 or 6, as is usually the case, I get median citizens for each country that are more reflective of a real continuous distribution. Note that not all national election studies used the 0–10 scale. When another scale was used (1–7, 1–10, etc.) I followed Powell’s (Citation2000: 273) procedure to convert it to a 0–10 scale.
7. I use the median position instead of the mean since the median is less sensitive to extreme values. I also apply Warwick’s (Citation2011) procedure to make the scale continuous. Scholars have argued that using respondents’ perceptions of party positions generally obscures cross-national variations (Best et al. Citation2012). This relates to differential item functioning where respondents in different countries could locate themselves and their parties at the same position on the 0–10 left‒right scale even though they do not share the same position in reality. This is because respondents locate each actor on the left‒right scale based on a shared perception of the political centre of their respective country. Best et al. (Citation2012) developed a method to adjust survey-based positions based on the cross-national variations found in CMP coding of party positions. The main implication of their method is to shift the entire party system of each country to the left or to the right. As a result, the ideological distances between each party and the median citizen are not affected by such adjustments. As underlined by Powell (Citation2006: 296), I nevertheless assume that these ideological distances are comparable and meaningful across countries – e.g. an ideological distance of two points on the left‒right scale between the median citizen and the government represents the same distance in each country.
8. I exclude elections (Denmark 1987 and 1998, and Norway 1997) where it was not possible to compute accurately the position of the government due to a lack of information regarding the party positions of many cabinet members.
9. For a study that examines the congruence of those governments that form between elections see Kim and Fording (Citation2012).
10. Let me take an example to clarify my measure of change in congruence. A coalition government forms after the 2003 election. The coalition is composed of Parties A and B. The coalition stays in office until the 2007 election. Citizen‒government congruence at the beginning of the term in 2003 is based on the post-election survey conducted after the 2003 election. Citizen‒government congruence at the end of the term in 2007 (before the election) is based on the post-election survey conducted after the 2007 election. Citizen‒government congruence would change between the beginning of the mandate in 2003 and the end of the mandate in 2007 whether the position of the median citizen changes or at least one of the parties in government changes position. In addition, note that citizen‒government congruence at the end of the mandate period before the 2007 election would equal citizen‒government congruence at the beginning of the mandate period after the 2007 election if the AB coalition also takes power after the 2007 election with the same share of cabinet portfolios.
11. This is the command xtpcse with the option ‘heteonly’ in STATA. In addition, the standard pcse command assumes that disturbances are contemporaneously correlated across panels. I do not use the standard pcse because there is no reason to assume that a country’s absolute change in government position is influenced by another country’s absolute change in government position.
12. One caveat, however, with the use of wild clustered bootstrap t-statistics is that I could not compute valid standard errors for the coefficient regressions (see Esarey and Menger Citation2017). This prevents me from presenting the marginal effects of the next interaction models. In addition, a random effects model may have been appropriate to account for the nested structure of the data but a Lagrange Multiplier test indicated that the cross-sectional error component is not different than zero.
13. The countries under majoritarian electoral systems are Australia, Canada, France, New Zealand (pre-1996) and the United Kingdom. The countries under PR electoral systems are Austria, Denmark, Finland, Germany, Iceland, Ireland, Netherlands, New Zealand (post-1996), Norway, Spain, Sweden and Switzerland. In total, 15 (26%) government mandates are under majoritarian electoral systems and 43 (74%) are under PR electoral systems. Even if the number of observations is small, I still have more observations than many studies of congruence and responsiveness: Golder and Stramski (Citation2010: 38 observations), Blais and Bodet (Citation2006; 31 observations) and Soroka and Wlezien’s study (2015) of policy responsiveness (28 observations).
14. Note that the results in Table are substantively the same when I include additional control variables such as left-wing governments, the margin of victory between the winning party (parties) and the second party, change in number of ministers, and the effective number of electoral parties (see Table C2 in the online appendix).
15. Note, however, that the coefficients are in the opposite direction. This makes sense because an increase in congruence at the end of the government mandate indicates in this alternative specification that the distance between the median citizen and their government decreases and consequently that representation is enhanced. This therefore yields a negative coefficient while the same result yields a positive coefficient in Table 2. The substantive interpretation is thus the same as in Table 2 even if the coefficients are in the opposite direction.
16. Note that polarisation ranges from 0.5 to 5.4. I therefore only present the marginal effects of number of parties and ideological distance over these values of polarisation.
