Does Public Opinion Respond to Shifts in Party Valence? A Cross-National Analysis of Western Europe, 1976–2002

Abstract

Recent empirical research has demonstrated that, in addition to policy positions, parties’ electoral support is also affected by their character-based valence attributes such as competence, integrity, and unity. Thus far, however, research into the effects of parties’ character-based valence attributes has not examined how such party attributes affect public opinion. The article examines whether changes in parties’ character-based valence attributes motivate shifts in public opinion – specifically, whether public opinion shifts leftwards when right-wing parties’ character-based valence attributes suffer relative to leftist parties, and vice versa. It presents the results of pooled time-series analyses of the relationship between parties’ valence attributes and shifts in public opinion for nine European polities. The findings suggest that changes in parties’ character-based valence attributes do motivate shifts in public opinion as hypothesised, and the effects are substantively large. These findings have implications for party strategies and for our understanding of the factors which shape public opinion.

Notes

1. The data used for the measure of valence must also be comparable cross-nationally as this allows for the possibility of pooling data across countries, since in any single country the universe of elections that can be studied is too small for any credible attempt at statistical analysis. No subjective measures of parties’ competence, integrity, or unity currently exist, though opinion polls do occasionally ask respondents to rate parties’ competence and integrity. However, such subjective measures are problematic in time-series cross-sectional analysis since the frequency, quality, and comparability of survey research data covering the valence dimensions of evaluation studied here is surprisingly poor. Further, survey-based measures of valence are subject to the problem of respondent rationalisation where respondents’ perceptions of political elites’ character-based traits are coloured by respondents’ policy preferences.

2. A recent search of JSTOR’s political science journal articles returned over 670 articles, many appearing in top-tier journals, that have drawn on Keesing’s as an authoritative source for data. For example, Keesing’s has been employed to create the ‘Terrorism in Western Europe Events Data’ dataset (Engene 2007), to provide governing coalition data (Tavits 2008), and for data on party seats (Horowitz and Browne 2008).

3. A similar scoring scheme for content analysis of news stories is employed in Dalton and Duval’s study of the effects of British public opinion on European integration (see Dalton and Duval 1986: 124–5).

4. The coding guidelines in the supplementary appendix are available from the author and provide additional details regarding these decision rules. Using column space to evaluate an event’s importance is typically the weakest guideline. Of greater import is who was involved in the particular event (a high-ranking party official such as a minister, or a rank-and-file party member), the language used to describe the event in question, its context, and consequences. For example, if a senior party official was involved in an event described using language such as ‘serious blow’, ‘highly embarrassing’, ‘damaging’, ‘perceived negligence’, ‘widespread condemnation’, ‘major scandal’, ‘resigned following the disclosure’, etc., then the affected party would most likely be given a score of –2 in the appropriate category.

5. The negative bias is not surprising. For instance, if a prominent politician has an extramarital affair this is likely to attract media coverage, but if the same politician stays happily married this does not typically attract equally extensive media coverage. Similarly, the public are most likely far more concerned when a politician abuses their official powers of office than when they do not. The Keesing’s reports reflect that bad political behaviour draws far more attention than good behaviour. Consequently, the percentage of positive codings is negligible.

6. To classify parties as being either left-wing or right-wing, I rely on the Comparative Manifesto Project’s classification of party families. Members of the Communist, Social Democratic, and Green party families were coded as left-wing, while members of the Conservative, Christian, and Nationalist party families were coded as right-wing (see Appendix A of Budge et al. 2001).

7. My thanks go to Ronni Abney for her assistance with these additional codings.

8. The standard wording of the left–right ideological self-placement survey item asks, ‘In political matters, people talk of “the left” and “the right”. How would you place your views on this scale?’

9. See also Adams et al. (2004, 2006) on this point. Relatedly, Dalton, Farrell, and McAllister (2011) find that very high percentages of European publics can correctly identify parties as being either leftist or rightist.

10. Economic data was drawn from the World Bank’s ‘World Data Bank’ which includes time-series data for a number of important economic indicators. The data is readily available from the World Bank’s website.

11. To measure changes in parties’ policy positions I rely on data compiled by the Comparative Manifesto Project, which includes measures of parties’ overall left–right policy emphases. For my purposes I calculate the difference between the average of all right-wing parties’ policy positions at the time of the current election at t as compared to the previous election at t–1, and do the same for all left-wing parties.

12. The countries included in the study are all those for which pertinent data regarding parties’ character-related valence attributes is currently available (Italy, Britain, France, Germany, Spain, Greece, Portugal, Denmark, and the Netherlands). The time period covered by the study are those years for which I have both valence-related data and data regarding shifts in mean public opinion. Data from Italy is excluded post-1992 due to the radical shake-up the Italian party system experienced, which saw a number of established parties disappearing from the political landscape. Public opinion data for Greece and Spain is only available from 1981 and 1986 respectively.

13. The results of the F-test for fixed effects were F_{8, 22} = 1.52, p = 0.21, and for the likelihood ratio test for random effects, X₁² = 0.14, p = 0.36.

14. One plausible interpretation of Figure 1 is that much of the significance of party valence occurs at low levels of change. However, as Figure 2 illustrates, this could also be due to the presence of relatively few observations occurring at high values of change. That opinion shifts occur at relatively low levels of change in party valence certainly makes intuitive sense since it may only take involvement in one incident (a notable scandal, for example) for the electorate to evaluate that party differently and to shift, on average, away from it. However, an alternate interpretation relates not so much to the lack of observations occurring at high values of change, but that changes in party valence may produce diminishing responses. In other words, changes in party valence are likely to motivate shifts in public opinion to a point, but thereafter increasingly negative views of parties’ character valence cease to have a great impact on motivating opinion shifts i.e. a ‘saturation effect’ occurs. To adequately test these propositions thoroughly requires a greater number of observations than is currently available but remains fertile ground for future study.

15. Unfortunately, including economic variables in the analyses notably lowers the N due to missing data in the World Bank’s time-series data. Searches for additional, and comparable, economic data did not reveal an appropriate alternative data source.

16. An anonymous referee suggested that the number of parties in a given system may plausibly affect the results. To test this proposition, I ran an analysis which also controlled for the ‘effective number of electoral parties’ using data drawn from the ‘Comparative Political Data Set I’ available online from the Institute for Political Science at the University of Bern. The results also continue to support those reported in Table 2, and the coefficient for the ‘effective number of parties’ variable was not statistically significant suggesting that the findings hold regardless of the number of parties in a given party system. My gratitude to the referee for making this suggestion.

17. My thanks to an anonymous referee for suggesting this particular analysis. The results of analyses run using an unweighted measure of party valence, excluding a lag of the key independent variable (change between t–1 and t–2), as well as employing a lag of the key independent variable but dropping the non-lagged version of this variable (change between t and t–1) also broadly support the findings reported in Table 2. Results available from the author on request.

18. I thank an anonymous referee for suggesting an exploration of the relative impact of valence on different subconstituencies.

19. The exogenous measure of valence used in the analyses presented here contrasts with research by Serra (2010), who presents a model in which candidate valence is assumed to be endogenous in that candidates can work to improve their valence-related images by furthering their education, and gaining relevant political experience that should improve their ‘on the job’ competence and so forth.