Cognitive dissonance and political attitudes: The case of Canada

Abstract

While scholars accept that attitudes have an impact on behavior, cognitive dissonance theory asserts that behavior can, in turn, affect attitudes. The theory suggests attitudes may be transformed by the simple act of voting. Informed by dissonance theory and employing election study survey data from three Canadian federal elections, this article considers the impact of cognitive, affective, and behavioral factors on changes in party evaluations, arguing that elections serve as a significant stimulus for attitude change. Dissonance theory is found to be compatible with observed attitude changes between pre- and post-election questionnaires. Findings have implications for the study of attitude formation and change, the effects that campaigns and elections have upon voters, and survey design.

Keywords:

• Elections and attitude change
• Cognitive dissonance theory
• Election studies

Acknowledgements

The author would like to thank Cameron Anderson, Clive Seligman, Laura Stephenson, Robert Young and this Journal's anonymous reviewers and editor for their feedback on earlier versions of this article. Any remaining errors and omissions are the sole responsibility of the author.

Notes

1 Cognitions are defined as thoughts, pieces of knowledge, or beliefs.

2 Other approaches include making cognitions unimportant or adding consonant cognitions.

3 Bem (1967) posits that the impact of behavior on beliefs is not caused by cognitive dissonance, but by individuals inferring opinions from actions. This position has been discounted as an explanation for dissonance phenomena, however, as evidence has amassed that dissonance does cause an unpleasant state of arousal (Hogg and Cooper, 2003; Zanna and Cooper, 1974).

4 The exceptions are independent candidates and candidates from regional parties such as the Bloc Quebecois.

5 The CES did not have a pre-election component until 1988, so previous elections cannot be used to study attitude change. The 2008 CES only contains data on post-election party evaluations for a fraction of participants, and the 2011 had no post-election data on this variable. Elections from 1993 to 2000 are excluded as they were contested by five major parties. From 2004 onwards, the old three party system returned outside of Quebec, so results from 1988 are comparable to those from 2004 to 2006.

6 Some non-voters report that having voted (Silver, Anderson, and Abramson, 1986). The inclusion of such individuals biases results against H1. These people are classified as voters, but are not subject to pressure to increase ED, so the data provide an underestimation of the relationship expected by H1. This type of misreporting is not considered problematic.

7 The same effect should be observed for sincere and insincere voters. The latter includes strategic and protest voting (M^cGregor, 2012). Insincere voting causes attitude-behavior dissonance (Bølstad et al., in press). Regardless of sincerity, evaluations of the party voted for relative to the alternatives should increase.

8 Election study data are available from the Canadian Opinion Research Archive queensu.ca/cora/ces.html. Election results are available from Elections Canada (elections.ca).

9 While the percentage of CES respondents who report not voting is smaller than the actual turnout rates from these elections, there remain enough such individuals to yield statistically significant results when this group is compared to voters.

10 Elinder focuses only on support for Sweden's largest party, the Social Democrats, rather than considering ratings of multiple parties.

11 Quebec's party system differed from that of the rest of Canada in 2004 and 2006 because the Bloc Quebecois only ran candidates in Quebec, so only data from outside the province are considered for these years.

12 Dissonance theory predicts that attitudes towards the two parties not voted are influenced in the same manner. The presence of more than two alternatives “adds very little complexity to the analysis of the dissonance which exists after the decision is made” (Festinger, 1957, p. 36).

13 Δ ED = (vote_post − opponents_post) − (vote_pre − opponents_pre)

14 An instrumental variable is used as a proxy for a potentially endogenous regressor to derive consistent estimates. Instruments are correlated with the endogenous explanatory variable but not with the error term. An instrument is also independent of the dependent variable.

15 This question is found in the post-election questionnaire in 2004, so responses may be influenced by election results or respondents’ voting behavior. In 2006, however, the question is part of the pre-election questionnaire. This difference may explain differences in the observed significance levels for this variable in the 2004 and 2006.

16 The fewer than 20% of respondents who did not see free trade as the most important issue are excluded.

17 Disloyal partisans are grouped with non-partisans. Voters do not have to justify long-standing attachments to a party they do not vote for.

18 Cases with invalid TOVD responses are removed, after being identified using Fournier et al. (2001) method.

19 While inconsistent voters are omitted here, the conclusions from this analysis do not change if they are included.

20 The debates were held on Oct. 24 (French) and Oct. 25 (English) in 1988, June 14 (French) and June 15 (English) in 2004 and Dec. 15 (English) and Dec. 16 (French) in 2006.

21 Average ΔED is higher for individuals interviewed prior to the debates for all three elections. However, only in the 2006 election does this difference near statistical significance, and only at the 90% confidence level. The relationship between voting and ΔEDis not influenced by pre-election interview date.

22 Descriptive statistics are found in Appendix B.

23 While dissonance exerts an upward pressure upon ΔED, H1 holds in instances where ΔED decreases for voters, provided that it decreases more for non-voters. Dissonance is only one of many factors that may influence party ratings, and there may be instances where attitudes towards all parties change after an election. Thus H1 is supported if the non-voter coefficient in Table 1 is negative.

24 F-statistic values are 111.9 and 49.3 for 2004 and 2006, respectively, indicating these values are relevant in the relationship between the potentially endogenous variable and the instrument. Values above 10 are strong instruments. Weak instruments provide little improvements in consistency above OLS (Stock et al., 2002).

25 Initial ED is on a scale from -100 to 100, while all other variables range from 0 to 1.

26 Caution is taken when comparing the impact of this factor from one election to the next, as this variable is operationalized differently in 1988 than it is in 2004 and 2006.

27 The interaction of national and local loser term produces statistically insignificant results.

28 Results not shown but available from author upon request.

29 Results not shown but available from author upon request.