Partisan Dealignment, Party Attachments and Leader Effects

Abstract

This paper takes issue with the widespread assumption of constancy in the meaning and measurement of party identification over time and space. It holds that the partisan dealignment common to established democracies has eroded the validity of the Michigan scale’s assumption that voters can identify with only one party, arguing instead that dealignment can encourage secondary attachments to other parties. To persist with an exclusivist conceptualization of party identification under these conditions is shown to risk the misspecification of voting models since the effects of these secondary attachments can be misleadingly attributed to short‐term influences, like leaders, associated with those parties. The dynamics of the voting decision can thereby be seriously distorted. The force of this argument is demonstrated through a comparison of general election outcomes in relatively aligned Australia and relatively dealigned Great Britain.

Acknowledgements

I would like to thank the journal’s editors and the three referees, all of whom made insightful comments on earlier drafts that made for an improved final product.

Notes

1. Good reviews of the various issues and debates surrounding the notion of party identification are Budge et al. (1976), Miller and Shanks (1996, pp. 117–150), Clarke and Stewart (1998), and Green et al. (2002).

2. It is not new to draw attention to this particular potential weakness of the Michigan party identification scale. Weisberg (1980, p. 36), for example notes that while “The American Voter … treat[ed] the concept of party identification in very broad terms, … the traditional survey questions do not make allowance for the possibility of multiple identification.” Our interests diverge at this point, however. His is with the dimensional structure of party identification, and particularly with the accommodation of a notion unique to the United States, political independence. Mine is very different and lies with the potential for the partisan exclusivism inherent in the Michigan scale to produce mis‐specified models of vote choice, a potential that is greater the more dealigned the party system.

3. To avoid confusion, the general notion of an enduring loyalty to a party will be termed “partisanship” or “partisan attachment” and the terms “identification” and “affect” will be used for the specific models using the Michigan scale and thermometer scores respectively.

4. An exemplar is found in the multi‐nation Comparative Study of Electoral Systems project. The format of its partisanship question is “Do you usually think of yourself as close to any particular political party?” If respondents answer in the affirmative, they are asked “Which party is that?” This is followed by “Do you feel close to any other political parties?” If their answer is yes, the follow‐up question is “Which other parties do you feel close to?”

5. The same observation of unwarranted exclusivity applies at least equally to a third conceptualization of party identification that appears in the 2005 British Election Study. It asks: “Some people think of themselves as usually being a supporter of one political party rather than another. Do you usually think of yourself as being a supporter of one particular party or not?” The same question was not asked in Australia and so it is not considered further in this analysis.

6. The argument here is that there is a relationship between partisan dealignment and the risk of model misspecification, but not that it is necessarily deterministic. The circumstances of particular elections can always conspire to influence it. For example, the party leaders in a particular election may, by dint of their own personalities, have little or no electoral appeal so that their impact will be trivial regardless of the level of partisan alignment in the party system.

7. The data used are the 2004 Australian Election Study, which is available from the Social Science Data Archive at the Australian National University and the post‐election component of the 2005 British Election Study, which is available at ⟨http://www.essex.ac.uk/bes/2005⟩. Election studies in both countries have traditionally depended all but exclusively on Michigan‐types measures of party identification. The British analysis is based on a weighted, cross‐sectional sample for Britain as a whole. Both national teams are to be congratulated for their fine studies..

8. Australia and Britain differ, of course, with regard to the presence of compulsory voting in the former. This difference might work to our analytical advantage, however, since being forced to vote likely helps to explain the sustained higher levels of overall partisan alignment in Australia in Figure 1 (Converse, 1969).

9. This is only to argue, of course, for comparability on the grounds of criterion validity. Their comparability in terms of construct validity is another matter. One way of improving the latter might be to switch the thermometer scale anchor points to wording capturing feelings of closeness to a party since its substantial overlap with the Michigan definition of party identification has been demonstrated (Barnes et al., 1988). Thus, for example, a thermometer scale question along the lines of “Generally speaking, do you usually feel very close to the Labour (Conservative, etc.) party, or not at all close to it?” would have the twin benefits of remaining faithful to the spirit of the original Michigan conceptualization and being repeatable for individual parties. Given the argument developed in this paper, it should also improve the validity of models of voting choice the more dealigned the party system.

10. The notion of political independence has no meaning in either country. Thus, the Australian question is: “Generally speaking, do you usually think of yourself as Liberal, Labor, National or what?” In Britain, it is slightly different: “Generally speaking, do you think of yourself as Conservative, Labour, Liberal Democrat, Scottish Nationalist Party, Plaid Cymru, Green Party, or what?”

11. A related problem, of course, is potential endogeneity in the leader–vote relationship; it may be that leader evaluations do not “cause” the vote, but rather that causality is the other way around. Since Clarke et al. (2004, p. 118), among others, have demonstrated that “feelings about party leaders are weakly exogenous to electoral choice”, I follow the common convention of assuming that the flow of causality is from the leaders to vote choice.

12. This procedure describes the coding of the British data. The Australian procedure was slightly different in that the survey asked first across a series of 12 discrete issues “whose policies – the Labor Party’s or the Liberal–National Coalition’s – would you say come closer to your own views on each of the issues separately”. Respondents were then asked “which of these issues was most important to you and your family during the election campaign?” I created party‐specific dummies by taking respondents’ most important issue, identifying the party that came closest to their own views and, ignoring the issue itself, scored the respondent 1 if their closest party was Labor (or Liberal‐National) and 0 otherwise. A third dummy was created for those who said that no party came closest to their own view or who didn’t know which party came closest.

13. Experimentation with other election‐specific issues common to the two countries, particularly terrorism and the war in Iraq, proved not to disturb the overall pattern of results in Table 2. Especially since these issues were measured very differently in the two national surveys, it was decided to exclude them from the analyses.

14. Dummies are included for subjective working class and subjective middle class. The majority of voters in both countries identifying with neither class are the excluded category.

15. Casting a first‐preference vote for either coalition party as opposed to the ALP is the dependent variable in the Australian analysis. The question assessing the issue competence of the parties does not distinguish between the two parties, and so takes the form of a coalition response in Table 2. The leader affect variable, however, is different in that it refers only to the Liberal leader since he was the sitting prime minister and would unquestionably continue in that role if the Coalition were to win the election. If the two parties are disaggregated, there is no independent electoral effect for the National party leader.

16. The overall pattern of socio‐demographic effects is largely unsystematic. The one exception is that the subjectively working class in Britain are more likely than those with no class identification to vote against the Conservatives in both the Conservative–Labour and Conservative–Liberal Democratic equations. The full tables are available from the author on request.

17. Another explanation of their limited impact is might be that they are highly endogenous to the vote (see van der Brug et al., 2007, p. 22). Whatever the explanation, though, their limited impact means that, unlike the other short‐term force, party leaders, this impact cannot shrink further when secondary attachments are controlled.

18. This scenario also applies in the British 2001 general election. Multinomial logit analysis shows the Conservative and Liberal Democratic leader to have a significant electoral impact in an exclusivist voting model. Their effect disappears, however, once secondary partisan attachments are taken into account in an affect‐based voting model.

19. This exercise cannot be repeated for Britain since matching data there are not available prior to the 2001 general election.

20. The odds ratios in Table 5 are calculated from exactly the same set of variables making up the Australian and British equations in Table 3. Full details of the 1996, 1998 and 2001 Australian equations are not presented in the text partly for reasons of space and partly because they are very similar in outcome to the 2004 affect and identification models in Table 3. Thus the 2004 model in Table 3 can be taken as a reliable guide to the general pattern of Australian results. The full equations for 1996, 1998 and 2001 are available from the author on request.

21. It is also consistent with the distinctive pattern of media coverage of the 2001 election. A computer content analysis of Australia’s leading newspapers for the 30 days before the 1998, 2001 and 2004 elections showed that the mentions of John Howard relative to the Liberal party jumped from a value of 2.31 in 1998 to 5.62 in 2001 and fell back to 3.93 in 2004. The matching figures for ALP leader, Kim Beazley are 2.25 in 1998 and 3.09 in 2001. Mark Latham replaced Beazley for the 2004 election and the figure for him is 3.56.