![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
This study investigates strategic voting for small parties in proportional representation systems, in previous work sometimes referred to as threshold insurance voting (Cox, 1997). Starting from theories of rational voting (Downs, 1957), three conditions for threshold insurance voting are developed: the voter considers potential government outcomes, votes for a party at risk of falling below an electoral threshold, and votes for another party than his or her most preferred one. The conditions are tested on the case of the 2010 Swedish general election. Using extensive data material and a conditional logit model of vote choice, the results show that in this election voters cast strategic votes for at least one of the small parties, the Christian Democrats which was included in the incumbent government coalition.
Notes
1. The data file was received from Henrik Ekengren Oscarsson, Department of Political Science, University of Gothenburg, 4 February 2012.
2. If the party is too weak, voters may on the other hand be discouraged to vote for the party not to waste their vote. This psychological threshold should vary between different political systems, as well as between individuals.
3. If the strategic considerations on the other hand happen to coincide with first party preference, the vote could be considered as a straightforward vote (see Farquharson, Citation1969: 30).
4. Respondents in the Swedish National Election Study perceived that among the parliamentary parties, the Christian Democrats was most at risk of falling below the threshold: 22% of the respondents believed they would fall below threshold, whereas 10% believed that the Centre Party and the Left Party would fall below it (figures based on Swedish National Election Study, pre-election sample, Question 20).
5. According to van der Eijk and Franklin (Citation2009: 106) it is common that parties and political leaders emphasize strategic aspects during election campaigns.
6. Of the voters in the pre-election study, 83% completed the post-election postal questionnaire.
7. Age is controlled for in the multivariate models, and a private income variable based on register data does not alter the main results.
8. The respondent's participation in the election has been controlled for vis-à-vis the register of voters.
9. Using vote intention as dependent variable did not alter the main results.
10. This can be done using interaction-effects with the alternatives in a conditional logit model (Long Citation1997; Long & Freese Citation2006: 307), or via an asclogit-command (alternative specific conditional logit) in Stata. In this study the asclogit-command is used. The asclogit model fits McFadden's (Citation1974) original conditional choice model.
11. A characteristic of multinomial and conditional logistic regression is that the relative chances of choosing one alternative before another do not depend on the other alternatives available (Alvarez & Nagler, Citation1998; Long, Citation1997; Long & Freese, Citation2006). This is known as the Independence of Irrelevant Alternatives assumption (IIA). In a party choice context, IIA assumes that the relative chances of choosing one party before another do not depend on the other parties available. For example, if one of the parties to the right is excluded from the election, the relative vote share of another (similar) right-wing party in relation to a party at the left side is supposed to stay the same. This would be the case if the voters who supported the excluded party abstain or vote evenly for the other parties. If the excluded party is at either side of the political left–right scale, it is reasonable to believe that the majority of this party's supporters would vote for a similar party. Then the relative vote share for this party in relation to a left-wing party would increase, and IIA would be violated. In the forthcoming conditional logit analysis, a Hausman test is used to check the IIA assumption. The Hausman test investigates whether the IIA is violated by excluding alternatives (parties) from the model (see Long, Citation1997; Long & Freese, Citation2006: 244). However, the validity of the Hausman test has been debated, and different tests of the IIA assumption often generate different results (Long & Freese, Citation2006: 243). Therefore, the Hausman test is only used as a point of reference for assessing IIA.
12. In the pre-election sample 608 respondents answered the question on threshold insurance. The distribution of answers was the following: “absolutely one of the most important reasons” 9%; “a fairly important reason” 15%; “not a very important reason” 21% and “not at all an important reason” 55%.
13. For the Green Party this variable measures the attitude towards one of the two leaders, Maria Wetterstrand.
14. The measurement is built on questions 29A, 29B and 29C. Those who have declared to be adherents of a specific party (29A) have also answered 29B: “Which party do you like best?” The same respondents also answered 29C on strength of party identification: “Some people are strongly convinced adherents of their party. Others are not so strongly convinced. Do you yourself belong to the strongly convinced adherents of your party?” Those who responded “No” were coded as 0, “Yes” coded as 1 for the party they had declared they liked best (29B). Those who answered that they were not adherents of any party at all (29A) coded as 0. This measurement thus distinguishes voters who strongly identify with a party from voters who do not strongly identify with the specific party. In the pre-election Swedish election study sample, 18% identified strongly with a specific party.
15. The measurement is built on question 21AA: “Now a few questions about how the Alliance parties and the opposition parties have performed since the 2006 election. You may answer using the scale on this card.”(scale -5 to +5)
“How do you think the four Alliance parties, i.e. the Centre Party, People's Party Liberals, Moderate Party and the Christian Democrats, have performed as a government since the 2006 election?”
16. The education-variable is a dummy, coded 0 if the individual has no higher education than high school (gymnasium), 1 if the individual attends or has attended some higher education. The reason for this is that a political socialization process and orientation towards rational thinking is supposed to take place when attending higher education, which should affect vote choice in general and strategic voting in particular (see Dalton, Citation2008).
17. Class is not included in the analysis as class in Sweden today is a concept which is difficult to operationalize, and its importance for vote choice has decreased over the last few decades (Oscarsson & Holmberg, Citation2013). Including subjective class belonging as a dummy, where working class is coded 1 and others as 0, did not alter the impact of the relevant variables for this analysis. The class variable is also linearly related to the education variable, which is why there should be problems with multicollinearity if including both variables in the model.
18. Within the conditional logistic analysis it is not possible to limit the analysis to voters who voted non-sincerely and compare them with those who voted sincerely, as the number of voters who voted non-sincerely and have responded to the threshold and coalition related questions is too small to conduct a conditional regression analysis with vote choice as dependent variable. Still, all variables that are needed to identify sincere or non-sincere voting are included in the model (first party preference and vote choice).
19. To check the IIA assumption (see footnote 11) eight Hausman tests were performed. The Hausman test is built on an equation that excludes one party (alternative) from the model, and reveals if the relative chances to choose any of the remaining alternatives is altered. A significant value of the test indicates that IIA has been violated (see Long, Citation1997; Long & Freese, Citation2006: 244). Due to the complexity of the test, only the choice-specific variables were included. The tests show that IIA is violated if the Moderate Party or the Green Party is excluded. However, if the Christian Democrats is excluded, IIA is not violated. This may be another indicator that this party has fluctuating voters. The limited number of cases in the pre-election sample may decrease the validity of the Hausman test (Long & Freese, Citation2006: 243). The Hausman test is therefore only used as a point of reference for detecting a potential problem with the model. Additional models have been conducted to check the robustness of the major results. A mixed logit model which relaxes the IIA assumption (see Glasgow, Citation2001) confirms the main results: threshold insurance incentives had a significant impact in voting for the Christian Democrats. However the mixed logit model requires that at least one of the variables is specified as random (Glasgow, Citation2001; Hensher & Greene, Citation2003). In this analysis it is not evident which variable or variables should be treated as random. Mixed logit also requires better data quality than less complex MLE models (Hensher & Greene, Citation2003). In this study the sample is relatively small. Therefore the conditional logit model was chosen.
20. Other variables held at their median values. Holding the other variables constant at their mean values generates similar result, significant at a 0.1-level.