The Rise of the Volatile Voter
The Rise of the Volatile Voter
Abstract and Keywords
In this chapter we show how the twin processes of partisan dealignment and party system fragmentation have underpinned the increase in electoral volatility. Fragmentation creates volatility because smaller parties consistently lose a much higher proportion of their voters between elections than the major parties. Partisan dealignment matters because there is a strong and consistent relationship between a voter’s level of partisanship and the likelihood of them switching parties at the next election. While this accounts for a substantial proportion of the trend in volatility, it is less clear why partisan identification has itself declined. We show a clear pattern of generational replacement in partisan identification, with newer cohorts entering with lower levels of partisanship and remaining relatively stable over time.
Politicians and journalists have long obsessed over identifying the pivotal voters in elections; the ‘swing voters’ who might actually change which party they vote for. Each party, it is said, must win the swing vote in an election to have a chance of winning a majority. Swing voters have been thought of as a narrow segment of the electorate, those who might be won over by different parties given particular circumstances. Election strategists form profiles of the kinds of people they believe to be swing voters, such as ‘Essex Man’, ‘Worcester woman’, or ‘Mondeo Man’. They are often thought to be the small but moveable part of an otherwise loyal electorate; a small island of active switchers among an ocean of habitual supporters who parties can count on election after election.
When Butler and Stokes looked at panel data of vote choice in the 1960s, they observed that only around 13 per cent of those who voted in both elections switched their vote choices between elections (Butler and Stokes 1969). They concluded that differential turnout and cohort replacement were the major drivers of electoral change. Slightly over a decade later, Särlvik and Crewe (1983) saw sufficient change (21 per cent of voters switched) that they considered there to have been a ‘decade of dealignment’. However, even the switching seen there seems modest when compared to the levels seen in 2015 (43 per cent) and 2017 (33 per cent). Across the four elections from 2005 to 2017, around 60 per cent of voters switched parties at least once.1 Far from being the minority of the electorate, swing voters—defined as people who switch their support to different parties between elections—now comprise the majority of the modern British electorate.
In Chapter 3 we argued that the growth in the number of these ‘volatile voters’ has increased the potential for electoral shocks to have a significant impact. That is to say, the more voters are prepared to move to another party, the more unstable the party system becomes. In this chapter we focus on the long-term changes that have helped to generate this more volatile electorate. We identify several potential influences on levels of vote-switching between elections. Some of these cannot account for the long-term increase in volatility because they do not follow the same trend over time. One—the increasing ideological similarity between the (p.51) main parties—does help to account for increasing volatility, but only to a modest extent. There are, however, two processes that can account for the substantial increases in volatility over the long term: partisan dealignment and party system fragmentation.
High levels of partisanship—or party identification—are expected to create a stable basis for vote choice and therefore limit voter volatility (Campbell et al. 1960; Butler and Stokes 1969). Consistent with this idea, we show that long-term partisan dealignment—the weakening of the attachments between voters and political parties—has a very strong impact on the level of electoral volatility. Additionally, and for a variety of reasons (which include institutional advantages and the electoral system), the larger parties tend to retain a greater share of their supporters from one election to the next. This means that the increased fragmentation of the party system in recent years has resulted in more voters switching their vote choices between elections. The success of smaller parties at one election tends to lead to lower overall levels of vote retention at the next election, increasing the amount of switching between elections as the votes for minor parties have increased over time, until, that is, the 2017 General Election.
Partisan dealignment and party system fragmentation only go so far in explaining increases in voter volatility. There remain unexplained increases in volatility in recent elections that cannot be understood without also taking specific events—electoral shocks—into account. In subsequent chapters we therefore focus on the electoral shocks that have led to an increasingly volatile electorate changing its electoral choices to an even greater extent, contributing to further volatility and dramatically shaping the outcomes of the 2015 and 2017 General Elections.
4.1 Partisan dealignment
The British two-party party system was relatively stable in the post-war era, with high levels of partisan identification associated with processes of political socialization and the strong class-based links of the two main parties. This stable system existed with strong party loyalties leading to stable patterns of vote choice. However, one of the best documented trends in British politics has been the decline in the number of people identifying with a political party, and the decline in the proportion of identifiers who have a strong attachment (Särlvik and Crewe 1983; Crewe, Särlvik, and Alt 1977; Dalton 1984).
In the BES, party attachment is measured using questions that ask respondents to say which party they feel closest to, followed up with a question about how strong those feeling of attachment are. Table 4.1 shows the question wordings used in the BES and British Social Attitudes (BSA) surveys. Note that the partisan strength wording changed slightly between the earliest waves of the BES and later surveys. The two series complement each other. The BES surveys provide a (p.52) time-series covering each election back to 1964. The BSA surveys only started in 1983, but are conducted every year, allowing us to analyse change in the electorate’s partisan attachments between elections. The BSA questions consistently receive lower levels of respondents reporting a partisan attachment than the BES questions, likely due to the filter question which invites respondents to state a lack of identity.
Table 4.1 Party identification question wordings in the British Election Study and British Social Attitudes surveys
[Q1] Generally speaking, do you usually think of yourself as Conservative, Labour, Liberal, or what?
[Q1] Generally speaking, do you think of yourself as Labour, Conservative, Liberal Democrat, (Scottish National/Plaid Cymru) [in Scotland/Wales], or what?
DO NOT PROMPT
[Q1] Generally speaking, do you think of yourself as a supporter of any one political party?
[Q2 if no at Q1] Well, do you generally feel a little closer to one of the parties than the others?
[Q2 if ‘none’ at Q1] Do you generally think of yourself as a little closer to one of the parties than the others?
[Q2 if no at Q1] Do you think of yourself as a little closer to one political party than to the others?
[Q3 if yes at Q2] Which party is that?
[Q3 if yes at Q2] Which party is that?
DO NOT PROMPT
[Q3 if yes at Q1 or Q2] Which one?
DO NOT PROMPT
[Q4 if party given at Q1 or Q3] Well how strongly [party] do you feel: very strong, fairly strong, or not very strongly:
[Q4 if party given at Q1 or Q3] Would you call yourself very strong [party], fairly strong, or not very strong?
[Q4 if party given at Q3] Would you call yourself very strong [party], fairly strong, or not very strong?
(p.53) As can be seen in Figure 4.1, the combined size of the ‘not very strong’ and ‘none’ categories has increased steadily over the fifty years of British Election Studies. This long-term fall in party identity has mainly resulted from falls in levels of identification with the two major parties.3 The proportion of the electorate reporting a very strong party identification has also plunged from 45 per cent in 1964 to only 10 per cent in 2005, with a particularly noticeable drop in strong identification in the 1970s.4 On this measure, partisan identity had reached its nadir by the 2005 election and remained at a similar level since, with the two most recent general elections, 2015 and 2017, witnessing a small increase in partisan identification.
It is clear, then, that the linkage between the major parties and the electorate has weakened in Britain, as it has in many other advanced democracies (Dalton 1984; Dalton 2012b; Scarrow 2004; Clarke and Stewart 1998). If voters do (p.54) not identify with a political party, and if they lack a strong sense of attachment, it is much easier to switch support to another party or not to vote at all (Huddy 2013). It is not surprising, therefore, that partisan dealignment and vote-switching are closely connected phenomena (Blais et al. 2001; Dassonneville, Hooghe, and Vanhoutte 2012; Farrell, McAllister, and Broughton 1994; Johnston 1987; Dalton, McAllister, and Wattenberg 2000; Rattinger and Wiegand 2014).
The relationship between party identification and vote-switching can be understood through the ways in which party identification stabilizes a voter’s loyalty to a political party. As we discussed in Chapter 3, partisanship can be viewed as a form of social identity that provides a lens through which voters evaluate politics—a ‘perceptual screen’ (Bartels 2002; Campbell et al. 1960; Huddy 2001; Butler and Stokes 1969).5 Because the psychological motivations to maintain one’s existing identity are very strong (Lodge and Taber 2013), partisans form judgements about political parties (and leaders, the economy, policy positions, etc.) in line with their prior attachments. Partisans are also less likely to seek out information that challenges their existing viewpoints, less likely to be exposed to contrary views in family and social networks, and less likely to accept information that is contrary to their existing preferences (Lodge and Taber 2013; Zaller 1992). Therefore, an electorate comprising fewer people holding strong partisan identities should lead to greater responsiveness to political events and competition. Not only are people less positively biased towards a preferred party, they are also less negatively biased against another. The result is a greater willingness to switch parties (Rattinger and Wiegand 2014).
Figure 4.2 demonstrates the level of vote-switching for respondents of different strengths of partisanship for each of twelve pairs of consecutive elections (labelled according to the second of each pair) starting with 1964–66 and ending with 2015–17. It replicates Figure 2.4 in Chapter 2, but for each level of partisan identification.
Figure 4.2 reveals three key relationships. First, across all of these pairs of elections, the more someone identifies with a political party, the less likely they are to switch their vote choice between elections. Second, the relationship between party identification strength and vote-switching—if we focus on the gap between the level of switching for each group—has remained broadly consistent over the past fifty years.6 Third, and importantly, switching is higher in recent elections (p.55) within each category of party identification strength. The first two observations, in conjunction with the increase in non-identifiers and weak identifiers we reported earlier, suggest that partisan dealignment may account for at least part of the long-term rise in volatility. The third observation, however, suggests that increased volatility in recent elections cannot be explained by declining partisanship alone: partisan dealignment had, to a large degree, bottomed out by 2005, yet volatility continued to increase and did so even within the different levels of identification, at least until the most recent election in the data series in 2017.
How dealignment works: The role of generational replacement
The key process behind these changes is the replacement of older generations with strong party ties by younger generations with weak or no party ties. Party identification has been described as a long-term attribute of voters that is socialized at an early stage of development (Campbell et al. 1960; Butler and Stokes 1969). If the party identification of parents becomes weaker, so will that of their children, as parents cease to provide partisan cues (Martin and Mellon 2018; Dinas 2013). To put it another way, we would expect the children of the 1960s to be exposed to much more partisan socialization than the children of later decades, simply because the electorate in the 1960s had higher levels of party identification. There is, then, likely to be a ‘ratchet effect’ in partisan dealignment, with each generation being less likely to be socialized into partisanship than the one before.
(p.56) To examine this process, we need to separate cohort replacement from within-cohort change.7 For this we need data that are measured more often than once per election cycle. The British Social Attitudes Survey data provides a useful source, as it is conducted yearly, although the data only go back as far as 1983 rather than all the way to 1964. Using BSA data, Figure 4.3 shows that each new political generation (since those entering the electorate prior to 1964) has entered with lower levels of party identity than the political generation before. Most of these generations have maintained relatively stable levels of party identity once they have entered the electorate. The only exception to this stability is the most recent political generation: those who entered the electorate under the Conservative government since 2010 (we do not show them on the chart as they have only a handful of observed years). This newer generation displays unstable levels of partisanship that at times are higher than the preceding generation. This may be because these voters have been newly enthused by politics in the 2014–16 era, or it may simply be an artefact of the small sample size of voters in this age range.
The significance of generational replacement becomes clear when we decompose the change in party identity into within-cohort change and cohort replacement. In other words, we consider: what portion of change can be attributed to the (p.57) differences between the average person entering and exiting the electorate, versus changes in the attitudes of people already in the electorate?8 Figure 4.4 shows the trends in levels of party identity in the BSA, decomposed in this way. The total line (left-hand-side figure) shows the overall difference in party identification in the BSA compared to 1983, the first year the survey was conducted. The trend starts at zero, so the y-axis refers to the cumulative change that has taken place. Party identification declined by around 20 percentage points since 1983 up to 2009 followed by a partial recovery. By 2017, the total fall in party identification was only 9 percentage points compared with the first BSA survey in 1983.
The right hand panel of Figure 4.4 shows the percentage point change that can be attributed to within-cohort change and cohort replacement, respectively. The figure tells us that the increase in non-identification since 1983 has come almost entirely from cohort replacement—that is resulting from the difference between cohorts entering and leaving the electorate. Overall, the difference in partisanship (p.58) between incoming and outgoing cohorts has contributed 15.8 percentage points towards dealignment since 1983.
From 2014 to 2017, the long-term decline in party identification driven by cohort replacement was somewhat offset by increasing levels of party identification within existing generations, but has so far been insufficient to reverse the overall trend. Within-cohort change shows large but not trending fluctuations year-to-year. In particular, we see large increases in rates of party identification in election years. This fits with the findings of Michelitch and Utych (2018) who found that, across eighty-six countries, levels of partisanship vary by 12 percentage points across the electoral cycle. Interestingly, the last four years of the available BSA data (2014–17) all show higher levels of party identity. This likely reflects the series of high-profile political events (2014 Scottish referendum, 2015 General Election, 2016 EU referendum, and 2017 General Election that took place in this time period). Nonetheless, given the higher partisanship of older cohorts, either existing cohorts will need to become much more partisan, or new cohorts entering the electorate will need to attain the far higher partisanship levels of their grandparents to maintain current levels of partisanship.
Why has partisan identification declined?
There is still the question of how this generational cascade of weakened political socialization was initiated. Previous researchers have suggested a number of possible causes, including the decline of class divisions in British party politics, the ideological convergence of parties, and the rise of a more informed and educated electorate. None of these provide convincing explanations of dealignment.
It has been claimed that class voting has declined since at least the 1970s (Crewe, Särlvik, and Alt 1977; Clark and Lipset 1991; Franklin and Mughan 1978), although this was disputed for many years on the basis that the claim conflated the absolute size of class-aligned voting with the relative propensity of classes to vote for Labour or the Conservatives (Heath, Jowell, and Curtice 1985; Evans 1999b). The evidence of class dealignment became much clearer after the rise of New Labour in the 1990s. Under Tony Blair’s leadership, Labour shifted to the centre (Bara and Budge 2001) and focused less on appealing to the working class (Fairclough 2000; Evans and Tilley 2017, ch. 6). This was accompanied by a dramatic decline in MPs from working-class backgrounds (Heath 2015) and a similar fall in the extent to which Labour was perceived as a party that represents the working class. These changes were in turn accompanied by a large decline in differences between the working and middle classes in voting for Labour versus the Conservatives (Evans and Tilley 2017; Heath 2015). However, it is difficult to explain the decline in partisanship as a result of a decline in levels of class voting. A substantial amount of partisan dealignment preceded the onset of the most (p.59) pronounced period of class dealignment, from 1997 onwards. Also, the partisan dealignment among the working class during the New Labour years does not explain a great deal of the more general downward trend in partisan identification.
Similarly, ideological convergence between the Conservative and Labour parties appears unlikely to explain the long-term decline of partisanship. Political polarization in Britain did not uniformly decline over the time period we are examining. In the late 1960s and 1970, Labour and the Conservatives appeared to be close together, in both their manifesto content and in the perceptions of voters.9 Polarization then peaked in the 1980s before returning to levels similar to those in the 1960s. While the 2017 election saw a modest increase in the proportion of voters perceiving ‘a great difference between the parties’, the level has not returned to anywhere near that seen in the 1980s.10 It is highly improbable that a curvilinear trend can explain a more or less linear decline.
Another influential explanation for the decline in voters’ attachments to parties is the growth of a more educated, informed, and critical electorate. The theory of cognitive mobilization predicts that partisan cues should be more important for less educated citizens. Higher levels of education will therefore reduce partisanship because highly informed voters do not require the heuristic or shortcut of party labels (Dalton 1984). The average level of tertiary education has substantially grown among BES respondents since 1964, increasing from around 10 per cent to more than 35 per cent of the population in 2017, a trend which is certainly consistent with this idea. However, the evidence for cognitive mobilization as a cause of partisan dealignment is limited. Dassonneville et al. (2012) find that the aggregate patterns of education and partisan dealignment in Germany align closely, but the individual level relationship is absent or even reversed. Similarly, Berglund et al. (2005) find that the relationship between education and partisanship is not stable over time, and the relationship disappears in some cases once age is controlled for.
These factors do not explain a large portion of the over-time decline in party identity. This can be seen when we model the decline in party identity over the eleven elections between 1964 and 2017 using (pooled) post-election BES cross-sectional surveys. We estimate the impact of convergence (perceived difference between the major parties), cognitive mobilization (educational attainment), and (p.60) socialization (parental party identification) on whether a respondent has no party identification. We also control for other variables that, according to the literature, may be linked to party identification, including age, sex, marital status, class, religiosity, union membership, and region. A fixed-effect for each year is used to measure the trend in dealignment before and after adding the explanatory variables.11 Our model shows that having no party identity is strongly predicted by perceiving Labour and the Conservatives to be similar, by lacking a religion, and by having a parent who lacked a party identity when the respondent was growing up. However, these factors explain relatively little of the over-time trend in partisan dealignment. The effect of education on non-identification is minimal at the individual level and explains none of the over-time trend in dealignment.
Figure 4.5 shows the increase in non-identification, compared with 1964, for each year before and after we account for differences in the explanatory variables. The solid line can be interpreted as the raw (unadjusted) increase in the proportions with no party identification compared to 1964. The dashed line is the remaining difference in each election year after differences in our explanatory variables are taken into account. Figure 4.5 shows that even after all of the potential influences described above are included, we can explain relatively little of the decline in party identity. In the 1980s, when parties were perceived as more (p.61) polarized, voters were actually more dealigned than would be expected given the perceived levels of difference between the parties.
While our cohort analysis demonstrates that partisan dealignment has been driven primarily by generational replacement, the reason for this generational change is unclear. Even with the benefits of fifty years of BES data, and drawing on variables that represent the most plausible explanations of the decline in partisan dealignment, we cannot account for that trend. That is to say, we know that new cohorts are becoming less attached to political parties, but this is not explained empirically by party convergence, by cognitive mobilization, or by parental socialization. The cause of partisan dealignment, which is making the electorate more volatile and vulnerable to electoral shocks, is not something we can explain. We can, however, highlight its important consequences. Partisan dealignment is connected with volatility and also with a further source of increasing volatility, party system fragmentation, to which we now turn.
In Chapter 2 we described how, alongside increased voter volatility, there has been a decline in the two-party vote share. The corollary of this has been a sharp rise in the share of smaller parties, and an increase in the effective number of electoral parties, calculated using Laakso and Taagepera’s (1979) formulation. The two-party share of the vote has declined since the heyday of the two-party system in the 1950s and 1960s, with the lowest two-party shares recorded in 2010 and 2015. This was followed by a sharp reversal in 2017 and a drop in the effective number of electoral parties (see Figure 2.2).
Fragmentation is important for volatility because smaller parties do not typically retain voters to the extent that large parties do. A share of the vote for smaller parties in one election should increase the expected numbers of switchers in the following election. There are a number of reasons why we expect minor parties to struggle to hold onto their voters. They often campaign on a narrower set of issues than major parties, which means a voter may defect when they no longer see one of those issues as salient; their voters tend to have weaker partisan identification; and they often have fewer resources (campaign funding, quality candidates, media coverage, campaigners). While minor parties have been improving on these over time, they still have far less access to resources than major parties in the British system. Most importantly, in a majoritarian system such as that in the UK, there is the danger that minor party votes will be perceived as a wasted vote if the party fails to be competitive locally. There is also a subtle mathematical effect that makes it easier for major parties to retain voters. If we imagine an over-simplified model of voting where voters choose randomly, the larger parties would retain a higher proportion of their voters simply by chance. The Conservatives (p.62) received 42.4 per cent of the vote in 2017, which means that they would only need to be twice as good at attracting their own previous voters as they are at attracting voters in general in order to reach 80 per cent retention. The Liberal Democrats, by contrast, received 7.4 per cent of the vote and would therefore need to be nearly eleven times better at attracting their own previous voters than voters in general, in order to maintain the same levels of retention.12 To put it another way, the baseline likelihood that a voter does something common (voting Labour) two elections in a row is relatively high, compared to the probability that a voter does something rare (voting Green) two elections in a row. Insofar as any voters tend to make a new decision at each election (rather than sticking with their old vote choice by default and then deciding whether to defect), this will increase the observed retention rates for major parties and reduce the observed retention rates for minor parties.
What is the evidence that minor party voting contributes to electoral volatility? Figure 4.6 shows that the defection rate of Liberal Democrats and other minor party voters has been consistently higher than that of the major parties. Across our twelve election pairs, the Liberal Democrats lose an average of 44 per cent of their voters from the previous election, and ‘other’ parties have lost 50 per cent of their previous voters. Conservatives and Labour, by contrast, lose an average of 18 per cent and 17 per cent respectively (although this has increased somewhat over time).
Although minor parties are relatively more likely to rely on new recruits who have already shown a proclivity for switching, minor party volatility is not entirely driven by this. Looking across three sets of three elections, minor party voters who had voted for the same minor party in the two previous elections were still more likely to switch parties than major party voters who were voting for the major party for the first time.13 Defection rates are even higher for new minor party voters, of whom at least three-quarters defected at the next election. This means that a strong minor party performance at the previous election greatly increases the expected level of volatility at the subsequent election. However, minor party voting is not independent of party identity, which is the other main driver of volatility. That is, minor party voters, on average, have considerably lower levels of party identity.
Sources of fragmentation
Fragmentation is linked to electoral volatility, but then we also need to ask the question of why has the party system become more fragmented?
The most obvious (tautological) answer is that major parties have failed to maintain their appeal to supporters. There are a number of reasons this may have occurred. First, political parties have long competed around economic issues since the emergence of the class cleavage following the industrial revolution (Lipset and Rokkan 1967). In Britain, party politics was organized around economic issues of left and right for many years (Heath, Jowell, and Curtice 1985; Evans, Heath, and Lalljee 1996). However, by the late 1990s and early 2000s, non-economic issues such as crime, immigration, and the environment have become more salient. The rise of a second ideological dimension was originally thought to be linked to increases in the prominence of ‘post-material’ values (Inglehart 1981), which focused on the increased importance of issues such as the environment. However, much of the rise in non-economic issues in Britain is the result of the increased salience of issues at the conservative end of the spectrum, such as crime and immigration (Green and Hobolt 2008). As we argued in Chapter 3, new issues can become salient not just though value change but as a result of a shock: we demonstrate this in Chapter 5 with respect to the increased importance of immigration and Europe.
Whatever the root cause, the rise of new issues is challenging for the major parties because it leads many voters to feel cross-pressured: that is, some voters will prefer one party on economic issues but another party on social or cultural issues. This trade-off is exacerbated by the fact that all British parties and their (p.64) candidates show a strong correlation between their liberal–authoritarian and economic left–right views (r = 0.70) while there is no such correlation for voters (r = 0.04).14 It is not just the major parties which fail to offer left–authoritarian or right–liberal choices to voters. Even UKIP, the Liberal Democrats, the SNP, and the Green party all offer either left-liberal or right-authoritarian positions. The difference with these smaller parties is that they downplay their economic message while emphasizing the second-dimension issues that are their focus. For example, immediately after the 2015 election, 46 per cent of BESIP respondents were unable to place UKIP on the economic redistribution scale, but only 25 per cent were unable to answer about UKIP’s position on the EU integration scale. The rise of new issues has therefore opened up opportunities for new parties to compete around non-economic issues including the environment (the Greens), immigration and Europe (UKIP), and national self-determination (SNP and Plaid Cymru).
A second crucial factor in the failure of major parties to maintain their support is the decline of party identification that we discussed above. As we have already demonstrated, party identifiers are more likely to stay loyal in terms of their vote choices. British voters had strong attachments to Labour and the Conservatives in the 1960s and voted for them in high numbers. Consequently, the decline in partisanship has tended to hurt the major parties more than the smaller parties—simply because they started from a position of relative strength. As we saw in Figure 4.2, higher partisanship tends to reduce vote-switching at the subsequent election. In the 1960s, this protected the votes of Labour and the Conservatives, but subsequent dealignment weakened this protection. However, unlike some of the relationships we demonstrate in this chapter, the relationship between partisan dealignment and fragmentation is a contingent one. As Butler and Stokes (1969) observed, strong inherited partisanship tends to maintain existing electoral alignments. In the 1960s, strong partisanship protected the high vote shares of Labour and the Conservatives, but dealignment removed that protection. However, if smaller parties started with a base of strong partisans, dealignment could just as easily have led to the consolidation of the party system. A third influence on fragmentation is likely to be the range of parties on offer to voters and how viable they are. The average number of parties standing in constituencies has been increasing, but it is not clear that the simple number of options is the most relevant measure. The mere presence of small parties does not meaningfully increase electoral choices if voters are not interested in them, do not know anything about them, or do not perceive them as having any chance of success. Electoral viability is likely to be crucial. To gain representation in a (p.65) first-part-the-post system like Britain’s, new and smaller parties must overcome the fear of electors wasting their votes on parties which have little apparent chance of winning seats in Parliament (Duverger 1954; Franklin, Niemi, and Whitten 1994). In other words, small parties suffer because their supporters strategically vote for larger parties (Ferland 2014). One way that small parties can overcome this is by demonstrating viability by performing well in second-order elections. The Liberal Democrats in Britain have used local elections to demonstrate their electoral viability in particular areas (Russell and Fieldhouse 2005). Similarly, more permissive electoral rules at European Parliament elections lower the cost of electoral coordination, enabling smaller parties to overcome some perceived problems of viability (Prosser 2016b). European Parliament elections have been described as serving as ‘midwives’ to the birth of new parties in Europe, which subsequently start to play a significant role in first-order elections (Curtice 1989; Ysmal and Cayrol 1996). Examples of this include the French Front National who caused a major surprise when they won 11 per cent of the vote at the 1984 European Parliament election and then went on to win 9.6 per cent of the vote and their first seats in the National Assembly in 1986 (Ysmal and Cayrol 1996); and the German Greens (Muller-Rommel 1993) who gained representation in the European Parliament of 1983 (with 5.6 per cent of the vote) and on that basis, one year later, were able to enter the German federal parliament with 8.2 per cent of the vote.
The increasing number of second-order elections in Britain—in particular for devolved institutions and European Parliament—have provided additional opportunities for small parties to establish an electoral foothold, leading to increased small party vote share and fragmentation. The most dramatic example is undoubtedly UKIP’s success in 2015 following their first-placed finish in the UK’s European Parliament elections, but SNP and Plaid Cymru success was also built on strong performances in devolved elections; and Green success in 2015 was built on the back of strong European performances in 2009 and 2014.
A further important aspect of whether a party can be considered to be genuinely cognitively available to a voter is whether the party is regularly mentioned in the media (Hopmann et al. 2010). We collected mentions of ten parties15 in nine national newspapers16 to create an effective number of media parties measure, which is calculated in an equivalent way to the effective number of electoral parties measure (Laakso and Taagepera 1979), substituting shares of media mentions for shares of total votes cast. Figure 4.7 shows that the effective number of media parties has been steadily increasing over this whole time period, indicating (p.66) that the electorate’s media diet includes significant coverage of parties other than Labour and the Conservatives. It is worth noting, however, that while minor parties are receiving more coverage in total, the media attention that each minor party receives is still well below that received by each of the major parties. An upward trend does not prove that supply has an effect at the individual level, but the trend is at least consistent with supply having some role in increasing fragmentation.
4.3 Explaining volatility
We have considered how we can explain partisan dealignment and the fragmentation of the party system. We now show how each of these—partisan dealignment and fragmentation—have contributed to the over-time increase in electoral volatility in British elections.
To understand how much of the increased level of switching can be explained through the trends we described, we model the predictors of switching in the British Election Study inter-election panels over the previous five decades. We run a pooled logistic regression model of switching across the eleven inter-election panels for which we have the relevant variables, and include a dummy variable for each election pair.17 As well as variables testing the impact of our two key factors, (p.67) we include a measure of the ideological convergence of the major parties and a number of controls including education, sex, and marital status. To reflect the complex interplay between our key factors we include a number of interactions between these variables which are discussed below.
First, consider dealignment. Based on the evidence above, we expect an increase in the number of people who do not identify with a party (or identify only weakly) to account for some of the increase in vote-switching. However, we would not expect this mechanism to work for people who have a different party identity from the party they voted for at the previous election. In that case, a strong party identity will be pushing people away from their vote choice rather than attracting them to it. There has consistently been around 10 per cent of the British electorate who vote for one party but have an attachment to a different party, so it is important to interact the effects of party identity with an indicator of whether the voter identified with the same party they voted for previously (to avoid dampening the effect of party identity). To assess the impact of dealignment on volatility, we therefore include strength of party identification, whether or not a person voted for the same party as their party identification, and the interaction of these.
Figure 4.8 shows the relationship between party identity strength and vote volatility, by consistency of party identity and vote choice in the prior election (controlling for other factors). Voters who have no party identity at the first election have an average 35 per cent chance of switching to another party by the second election. Among consistent voters, switching falls to just 13 per cent for those with a strong party identification. In contrast, cross-pressured voters, those who voted for a party other than the one they were attached to, are more likely to switch their vote choice if their party identity is stronger (although that difference is not significant). In other words, the impact of partisan dealignment is conditional on consistency of vote and party identification.
To capture the effect of fragmentation, we include a dummy variable representing whether a respondent was a major party voter (Conservative or Labour) in the previous election. Our model tells us that even after accounting for party identification and other predictors of vote-switching, a voter who voted for another party in the first election was 23 percentage points more likely to switch parties in the subsequent election, compared to a major party voter.18 The fact that people have increasingly voted for parties other than Conservative and Labour is therefore a substantial contributor to overall volatility. To illustrate this, take (p.68) two hypothetical elections where the major and minor party retention rates are 83 per cent and 50 per cent, respectively (the average values across all elections). In the first election, the major party share of the vote is 90 per cent (typical in the 1960s) and in the second it is 65 per cent (close to the level in 2010 and 2015). Based on just the different two-party vote shares, we would expect the first election to see 20 per cent of all voters switch parties and the second election to see 29 per cent of voters switch. In other words, a large proportion of the difference in individual-level volatility between the 1960s and 2010s can be attributed to the size of the minor party vote share.
We might expect that ideological convergence has some direct effect on voter volatility for two reasons. First, a reduced distance between two parties reduces the space that a voter has to jump from one to the other, and may therefore allow them to choose between the parties on other grounds than ideological position, such as performance (Green 2007). Second, a reduced distance between the two major parties increases the likelihood that a voter sees neither party as adequately representing their preferences and therefore switches to a minor party. Both of these mechanisms are explanations for why major party voters would be more likely to switch in the presence of convergence. However, neither mechanism would apply to non-major party voters, so we include an interaction of convergence with major party voting, so as not to obscure its effect among major party voters. To explore this we included a variable measuring the perceived difference (p.69) between the major parties, and interacted this with whether or not the respondent voted for Labour or the Conservatives.19 We find that while a perception of clear difference between the major parties reduces the likelihood of vote-switching, the effect is absent or even reversed for people who voted for other parties. A Conservative or Labour voter who perceives a great deal of difference between the parties is 9.9 percentage points less likely to switch parties at the next election than a Conservative or Labour party voter who perceived not much difference between the parties (Figure 4.9). However, if the voter supported another party at the first election, then seeing a great deal of difference between Labour and the Conservatives is associated with a 4.9 percentage points higher likelihood of switching parties at the next election.20 In other words, convergence increases volatility for major party voters but has no effect, or perhaps even an opposite effect, among smaller party voters.21 This makes sense spatially, as major party convergence leaves more space at the extremes for minor parties to compete. However, this means that as the level of support for minor parties has increased, (p.70) the total effect of convergence on vote-switching has become weaker, dampening the negative effect of convergence on volatility.
We have seen that, over this period, fragmentation and dealignment in combination with party convergence affect the chance of voters switching allegiance between elections. But how far do these factors account for the sharp rise in volatility we documented in Chapter 2? To test this, we use a pooled model predicting party switching across the eleven pairs of elections we model between 1964 and 2017. This model simultaneously accounts for all the factors we have discussed: party system fragmentation, partisan identity, convergence, occupational class, education, and other demographics. We also include separate dummy variables for each election, so that we can estimate how much extra switching we see compared with the base category of the 1964–66 election pair. We then compare this residual level of switching to the actual level of switching that took place. If our model has explained the time trend then this residual level of switching should be substantially lower than the observed increase in switching. Figure 4.10 shows the time trend in vote-switching (measured as the percentage point increase in vote-switching compared with the 1964–66 elections) before and after modelling the variation.22 The dark line shows the actual increase in switching since 1964 and 1966 and the lighter dashed line shows the residual increase in switching not explained by the model. We can clearly see that the variables included in the (p.71) model explain a substantial portion of the increase in vote-switching since 1964–66. For instance, in 2015 the level of vote-switching was 29 percentage points higher than the level of switching between 1964 and 1966. However, once we account for how the predictors of vote-switching changed between the two elections, the residual increase in vote-switching falls to 17 percentage points. This means that we have explained around 40 per cent of the difference in vote-switching that was observed in 1964–66 and 2010–15. Thus, the dramatic result seen in 2015 can be partially explained by the long-term trends that have driven British politics, but we also need to look to election-specific factors or shocks to explain the extremely high level of volatility seen in that election.
To see how much each of the separate factors explains the trend over time shown in Figure 4.10 we calculate the percentage reduction in the mean of the marginal effect of all of the election year dummy variables compared to 1964–66, for a series of models. Each bar in Figure 4.11 represents the reduction in the mean marginal effect of the election dummy variables for models which include each factor separately. In other words, it shows what percentage of the area under the solid line in Figure 4.10 that can be explained by each factor. In the full model (model 4), we reduce the average increase in volatility by 43 per cent.23 Figure 4.11 show that dealignment and fragmentation are the key factors in explaining (p.72) the increase in volatility, with convergence actually making the increase more anomalous.24 Despite the importance of partisan dealignment and party system fragmentation, however, we clearly see several elections (1974, 1997, 2017, and especially 2015) that have large unexplained increases in volatility. As we show throughout the rest of the book in 2015 and 2017, these spikes in volatility, which are not explained by secular trends, are largely attributable to specific electoral shocks.
In the last fifty years, we have seen important long-term trends that have made voters more likely to switch parties between elections. In this chapter we have shown how the twin processes of partisan dealignment and party system fragmentation have underpinned this increase in volatility.
Whilst we have found a strong and consistent relationship between a voter’s level of partisanship and the likelihood of them switching parties at the next election, and that this accounts for a substantial proportion of the trend in volatility, it is less clear why partisan identification has itself declined. We offered tests of the most plausible explanations of partisan dealignment and found that it is difficult to explain the downward trend in partisan attachments. Beyond generational replacement, the causes of partisan dealignment are somewhat elusive. We find little support for the cognitive mobilization theory that partisanship has declined because more educated voters have less need for partisan cues; and only a weak link to class dealignment, despite the reduction in class voting. Moreover, we do find substantial effects of party convergence on levels of party identification, but this does not account for the decline in partisanship. Partisan dealignment remains a hugely important trend, and yet one it is not yet possible to explain empirically. We have, however, shown a clear pattern of generational replacement in partisan identification; with newer cohorts entering with lower levels of partisanship and remaining relatively stable over time. This is important as it implies that voter volatility it is likely to stay with us for some time, as younger cohorts of voters with lower levels of party identification move through their voting lives. However, there are signs that young voters entering the electorate in the last few years may be beginning to break the downward trend in party identification.
The other main factor driving volatility is fragmentation. We have shown that smaller parties (other than Labour and the Conservatives) consistently lose a much higher proportion of their voters between elections than the major parties. This means that a higher share for these minor parties increases volatility. (p.73) Fragmentation is explained by the rise of cross-cutting issues and, relatedly, the supply of viable choices.
Nonetheless, despite the strong relationship between fragmentation, dealignment, and voter volatility, these factors do not fully explain the upward trend in volatility. The factors included in our model accounted for some of the rise in the number of ‘swing voters’, but we saw that, even adjusting for all these factors, there are still large unexplained election-specific spikes, most notably in 2015. To understand these we must return to the implications of the theory we set out in the third chapter of this book: voter volatility is a product not only of long-term secular trends but is the result of unanticipated and unexplained electoral shocks which act as a catalyst for vote-switching, especially among an electorate who, for the reasons explored in this chapter, have become less fixed in their voting habits.
(1) We calculate this in two ways. First, by taking British Election Study Internet Panel (BESIP) panellists who took all four post-election surveys and voted each time. This gives a stable voters figure of 39 per cent but is based on a sample of just 562 voters. The alternate approach supplements the panel data with vote recall data obtained as soon after the election as possible. This gives a much larger sample of 19,189 voters and a stable voter figure of 45 per cent.
(2) The party identification strength question wording was changed following the first three BES post-election panels in 1964, 1966, and 1970 and the second set of panels covering 1970 and 1974. This gives us an approximate picture of how much difference the wording makes. Using the original wording, 1970 has strong party identification of 47 per cent, compared with 44 per cent in 1966 and 30 per cent in 1974. Using the revised wording, the 1970 strong party identification figure is 42 per cent. This means that the old wording somewhat overstates the 1970–74 drop in party identification, although both wordings agree that the drop was large (17 percentage point versus 12 percentage points). We further conducted a survey experiment randomizing the two formulations and found lower levels (8 percentage points) of strong party identification using the post-1966 wording.
(3) Other parties have increased their levels of identification substantially in recent elections as their vote share has increased, but this accounts for a tiny share of all respondents.
(4) The extent of this sharp drop needs to be treated with caution. The strength of party identity was only asked of Labour and Conservative identifiers in the two 1974 elections, which will have had the effect of reducing the number of strong identifiers slightly. Miller, Tagg, and Britto (1986) find a less pronounced drop between 1970 and 1974 in their analysis of surveys conducted by the Conservative Party. In their study, very strong identifiers fell from 42 per cent in 1966 to 37 per cent in 1970, to 33 per cent in February 1974.
(5) This view is not unchallenged even among scholars who agree that party identity is best characterized as a form of social identity. Green, Palmquist, and Schickler (2002) argue that apparent evidence for perceptual screens is better characterized as evidence for partisans holding genuinely different values in how they evaluate political events and strong priors about their preferences. Follow-up studies have tended to confirm the perceptual filter model (Bartels 2002; Druckman, Peterson, and Slothus 2013; Gaines et al. 2007; Lodge and Taber 2013), especially for low salience issues (Carsey and Layman 2006).
(6) Vote-switching between 2015 and 2017 was similar for strongly and fairly strong identifiers. This may be statistical variation or may reflect the cross-cutting importance of the EU in 2017 (see Chapter 9).
(7) By generational replacement we mean the change attributed to the differences in levels of partisan identification between cohorts leaving the electorate (dying off) and those entering as they come of age. Within-cohort change captures the extent to which the level of partisan identification of those in the electorate changes over time. We define a political generation according to the government in power when a person came of voting age.
(8) For detail of the equations we use to define within- versus between-cohort variation see the appendix to Chapter 4. Because we define the age cohorts narrowly (in single birth years) and we observe nearly all years sequentially in the BSA, this algebraic decomposition need only assume that there is no within-cohort change for the newest age cohort between the time when they entered the electorate and were interviewed. This approach is similar to that proposed by Firebaugh (1997; 1990). Firebaugh’s approach has been criticized for not distinguishing age and period effects (Glenn 2005; Rodgers 1990) in its measure of cohort replacement. However, as Firebaugh (1997; 1990) argues, this critique conflates cohort replacement effects with cohort effects. Although our analysis is not an age-period-cohort (APC) analysis, APC models specified in line with Grasso et al. (2017) find large cohort effects in party identification.
(9) Across this period, the perceptions of BES respondents on the differences between Labour and the Conservatives closely track left–right positioning measured by the Comparative Manifestos Project (Volkens et al. 2015).
(10) We should note, however, that despite the similarity in policy programmes in the 1960s, the two parties were seen as being very different in terms of whose interests they represented. The Labour Party was seen to clearly represent the working class and the Conservatives the middle class—the backbone of political competition at the time. After 1997, when the Labour Party abandoned its distinctive role in representing the working class, the electorate saw it as no longer representing the interests of the working class per se (Evans and Tilley 2017). The role of convergence with respect to social group representation—as opposed to ideology—in the decline of partisanship remains worthy of further investigation.
(11) See Table A4.1 in the appendix for details of the model. The unexplained difference could be a combination of omitted variables and changes in the relationships between variables and dealignment.
(12) This follows the same logic as a pool effect in intermarriage rates. Blau, Blum, and Schwartz’s (1982) study showed that much stronger preferences for in-group marriage are needed to maintain intermarriage in small social groups than large social groups.
(13) The three sets of triplets of elections with connected panel data we examine are February 1974–October 1974–1979; 2005–2010–2015; and 2010–2015–2017.
(14) Correlations based on data from the British Candidate Survey 1992–2015 and BESIP 2014–2017. Left–right and liberal–authoritarian positions of candidates and voters measured using graded response item response theory (IRT) models to generate latent variables for each dimension.
(15) Conservatives, Labour, Liberal Democrats, Plaid Cymru, Scottish National Party, UKIP, Green Party, Referendum Party, British National Party, and Respect. For details of the search terms used, see Table A4.3 in the appendix.
(16) The Mail, Express, Telegraph, Times, Sun, Mirror, Guardian, Independent, and Star.
(17) We weight each panel to contribute the equivalent of 1,000 cases, so that the larger recent panels do not overly influence the pooled effects. We cluster the standard errors at the election-pair level, but this either hardly affects the standard errors or in some cases shrinks them. This is promising, given that a substantial inflation of clustered standard errors compared with unclustered can indicate model misspecification (King and Roberts 2015). For the full regression tables of these models see Table A4.2 in the appendix.
(18) This figure is based on the marginal effect of voting for a minor party at election 1 on voting for a different party at election 2, accounting for all the other effects in our model.
(19) The pooled model predicting vote-switching therefore contains the main effects of having voted for a minor party at time 1, perceptions of major party convergence at time 2, and the interaction of those two effects.
(20) Note that this difference is not statistically significant.
(21) It is not possible to conclusively say whether this is a causal effect or instead reverse causation where identifiers are strongly motivated to see differences between ‘their’ party and their rivals.
(22) 2005–2010 vote-switching is not included in these models because the difference between Labour and Conservatives question was not asked as part of the online panel in 2010.
(23) As with the dealignment models, the remaining unexplained variation could be some combination of omitted variables and changes in the relationship between independent variables and vote-switching.
(24) Interacting fragmentation with convergence also reduces the proportion of the time trend explained by fragmentation even though the overall model fit improves.