Embed Size (px)
Citation preview
1
Why aren’t we obsessed with image?
Party image and electoral choice in Britain
Robert Johns
Department of Government
University of Essex
Wivenhoe Park
Colchester
CO4 3SQ
UK
Tel: +44 1206 872508
Email: [email protected]
Paper prepared for the panel on ‘Candidate and party image’ at the 2012 Annual Conference
of the EPOP Specialist Group of the Political Studies Association, Oxford, 7-9 Sep.
.
Acknowledgements
The data analysed in this paper come from the long-running British Election Study series.
Thanks are due to the Economic and Social Research Council for funding these studies and to
any team of principal investigators that has attempted to measure party image. I would also
like to thank Pablo Fernandez-Vazquez and David Denver for useful comments on earlier
drafts of this work.
2
The concept of ‘party image’ has long been neglected by those studying electoral
behaviour. This is surprising given that several developments – partisan dealignment,
ideological convergence and the resulting predominance of valence concerns – imply a
powerful influence for such broader images. In this paper, I use survey data from
recent British elections to investigate the impact of party images. General views of the
parties – as ‘united’ rather than ‘divided’, ‘in touch’ rather than ‘out of touch with
ordinary people’, and ‘keeping’ rather than’ breaking promises’ – had a pronounced
impact on party choice, even holding constant party identification and a huge range of
other predictors. Moreover, these images are less strongly coloured by partisanship
than are other core predictors, such as leader evaluations. I conclude that party image
– both concept and measurement – warrant much closer attention.
1. Introduction
The valence or ‘performance politics’ model has become the dominant framework for
explaining voting behaviour – and, in particular, election outcomes – in Britain (Clarke et al.,
2004, 2009, 2011; Green, 2007). In an era when the parties are broadly agreed about the
goals of policy, voters are left to choose between parties according to which is likelier to
achieve those goals – that is, which will do a better job in office. It is not in serious dispute
that voters are more likely to support a party that they likely to govern effectively. However,
the antecedents of perceived performance are less clear. How do voters judge a party and its
competence for office?
The assessments of party identifiers are obviously likely to be swayed by partisan
sentiments; indeed, the judgements of strong identifiers might be said to be predetermined
rather than formed afresh at each election. However, the result of partisan dealignment is that
many British voters lack such ‘standing judgements’ of the parties. Such voters are often
inclined instead to use a convenient short cut: their evaluations of party leaders (Clarke et al.,
2004; Evans and Andersen, 2005). This is an enticingly simple task since we can all react to
people without necessarily knowing much about what parties have been doing or might do in
future. Party leaders are also prominent figures, widely recognized even by less politically
attentive voters. Not only that, but, in the 2010 and probably subsequent British general
elections, these leaders were placed squarely in the electoral shop window via televised
debates. The prominence of leaders’ personalities – both in election campaigns and
increasingly in models of party choice – is one reason why modern politics has been seen as
increasingly image-dominated.
But it is not just party leaders that have images. Parties themselves are familiar
objects on the political landscape, and thus conjure up mental associations for voters as soon
3
as they hear the party named. Moreover, many voters lack detailed or up-to-date knowledge
about parties’ records or policies. Short of such information, voters can be guided by their
broader images of the party – as moderate or extreme, as united or divided, as ‘in touch’ or
‘out of touch’, and so on. These are obvious means of judging the likely performance of a
party in office. Yet, despite its conceptual centrality to the ‘performance politics’ model, the
notion of party image has been almost totally neglected in recent studies of British electoral
behaviour. The purpose of this paper is to test whether that neglect can be justified.
It begins with a discussion of the definition and content of party images. Then I set
out the case for supposing that these images have an independent impact on voting behaviour.
Having briefly described the data to be used, I show that this potential is realised in recent
British elections, and discuss the implications not only for analysts but also for designers of
election surveys.
.
2. What is party image?
Party images are the mental pictures that voters have of the political parties. As defined by
Butler and Rose (1960: 17), ‘a party image is nothing more than a party as it appears to the
public, the picture left by its surface characteristics’. These definitions are as broad as they
are simple. As discussed below, there are numerous sources of party images and so the
contents of these mental pictures could vary considerably across parties, across the electorate
and across time. Even an individual voter may well have a vague or inconsistent picture of a
given party. Nonetheless, voters’ inclination to simplify their political thinking means that
their images of parties will tend to be dominated by one or two key mental associations. And
these associations are often persistent over time and shared by large proportions of the
electorate. Such prominent associations are the core of a party’s image.
Considering that politics in Britain (and elsewhere) are often characterized – usually
pejoratively – as having become image-dominated, it is surprising that the term ‘party image’
is largely absent from recent studies of British elections. It is not as if the notion is new to
electoral analysis. Party images were first discussed by Graham Wallas, in a (1910) book
entitled Human Nature in Politics, in which he pointed out that a political party was no
different from any other object in people’s lives – when they heard its name mentioned, they
would automatically call up whatever mental image they had formed of that party. Party
image was also a prominent concept in early election survey work, notably Milne and
MacKenzie’s (1954, 1958) studies of voters in Bristol North East at the elections of 1951 and
1955. Academic attention probably peaked in 1959 at an election which also saw a wider
4
preoccupation with these images, not least because it was widely regarded as the first
‘television election’ (or, at least, the first at which a large majority of British households had
a TV set). Trenaman and McQuail (1961: 36), in a study of the role and impact of television
in 1959, noted that ‘the term “party image” has been on the lips of the party leaders, the
newspaper columnists, and of the many political commentators who have been discovering
the reasons why Labour lost’. Interest in the concept persisted during the 1960s and
questions measuring party image were included in the first BES surveys, Butler and Stokes
(1974: 338) concluding that such images did much to shape voters’ behaviour.
Since then, however, the notion of ‘party image’ has been shelved by most electoral
researchers. This is partly just a matter of changing terminology. Many features of parties
identified in later work as important influences over electoral choice could reasonably be
described as aspects of image. It is also the result of class dealignment. Not surprisingly,
given that many political parties arose out of social divisions, voters’ overriding image of a
party is often its association with a particular social group. The relevant group will obviously
depend not only the party but also on those social divisions that have been most prominent in
a country’s politics. Thus, for example, American voters’ images of their parties include
associations with race and (increasingly) religion, in addition to the traditional differences in
socioeconomic status between Democrats and Republicans (Brewer, 2009). In Britain, where
all else was ‘embellishment and detail’ compared with the dominant class cleavage (Pulzer,
1967; Butler and Stokes, 1974), early studies of party image assessed how closely voters
associated the Conservatives and Labour with their traditional class bases (e.g. Benney, Gray
and Pear, 1956). A party’s image and its class image were treated as more or less
synonymous.
These studies also revealed two persistent tendencies. Second, voters associated
Labour with the working class more closely than they associated the Conservatives with the
middle class. This difference was often invoked as one reason for the Conservatives’
electoral success in the 1950s. While Labour was seen – including by its own supporters – as
a party battling for the working class against the middle class, the Conservatives were more
likely to be seen as a ‘one nation’ party, concerned with improving prosperity across the
board rather than fighting the corner of one class against the other. This helps to explain why
the party was able routinely to win substantial chunks of the working-class vote (see
Nordlinger, 1967; McKenzie and Silver, 1968).
As the power of social class to predict party choice has declined, so has interest in the
class images of parties. One consequence is the lack of up-to-date survey data bearing on this
5
issue. However, there is a question that was asked in most BES surveys during the 1970s,
1980s and 1990s. Respondents were asked, for both Labour and the Conservatives, to say
whether they would describe that party as ‘good for one class’ or ‘good for all classes’.
Typically, more than half of the electorate chose the ‘one class’ option, suggesting that each
party retains a class image in many voters’ minds. The glaring exception is New Labour in
1997, when many voters responded to the party’s explicit aim of reaching out to middle- as
well as working-class voters. This may be due partly to a popularity effect: parties which are
out of favour anyway tend also to be condemned as good for only one class. Nonetheless, the
scale of the difference between Labour’s 1992 and 1997 results also signals a considerable
blurring of the party’s class image. There were corresponding large shifts on other questions,
asking about how closely the parties look after the interests of class-related social groups.
Labour in 1997 was seen as somewhat less concerned than in 1992 with the unemployed and
the trade unions, and more concerned than previously with the interests of big business
(Heath, Jowell and Curtice, 2001, ch. 7). Yet, while these shifts brought Labour’s image
rather closer to that of the Conservatives, there remained pronounced differences with the
Conservatives clearly seen as pro-business, anti-union, and oriented towards middle-class
rather than working-class interests. Despite decades of dealignment and a studious avoidance
of class rhetoric by the parties, voters have not wholly abandoned their sociological images of
the parties.
Even in the era of alignment, however, there was a good deal more to party images
than social class. Survey questions asking what voters liked or disliked about a party, or what
they thought that party stood for, elicited a wide variety of responses about that party’s issue
positions and priorities, its leadership, its trustworthiness, its competence, and so on (Milne
and MacKenzie, 1958, ch. 9; Trenaman and McQuail, 1961, ch. 3; Butler and Stokes, 1974,
ch. 16). Consistent with the claim that ideological thinking is the preserve of a minority
(Converse, 1964; Luskin, 1987), few voters spontaneously used ideological terms or spatial
reasoning when describing the parties. Instead, in identifying a party’s basic orientation or
values, voters tended to cite its position on one or two core issues. A prominent stance on an
important issue, such as being hostile to the European Union, can become central to a party’s
image. As Milne and MacKenzie (1958: 130) argue, enduring relevance is more significant
than current salience: ‘Party images, then, are symbols; the party is often supported because it
is believed to stand for something dear to the elector. It matters little that the ‘something’
may be an issue no longer of topical importance; the attachment to the symbol, and to the
party, persists’. The durability of these issue images is not necessarily beneficial to a party.
6
In the 1950s and 1960s, Labour profited from its association with the quest for full
employment but lost out because it was also widely associated with the increasingly
unpopular policy of nationalization. More recently, the Conservatives’ long-standing image
as a party of lower taxes and spending proved an electoral advantage in 1992 but, then and
since, has cost them support among those fearing the impact of cuts on public services like
health and education.
As these examples illustrate, a party is on safer ground if its image involves
associations with valence rather than position issues. According to Butler and Stokes (1974:
23), ‘the parties have achieved their prominence as political actors in the public’s mind by
linking themselves to goals that matter to the electorate whose support they seek’. While
some voters may have doubts about the state intervention required for full employment and
investment in schools and hospitals, these are broadly agreed objectives and so it was
advantageous for Labour to be clearly associated with them. Most of all, parties seek to
connect themselves in voters’ minds with the purest valence issue, economic prosperity. One
reason for the preoccupation with party image at the 1959 election was a widespread
perception – borne out by voter surveys – that the Conservative Party had become associated
with economic success and improved standards of living. This image had been cultivated by
then Prime Minister Harold Macmillan in a famous 1957 speech reminding voters that most
of them had ‘never had it so good’, and was hammered home by the campaign slogan ‘Life’s
Better with the Conservatives. Don’t Let Labour Ruin It.’ Trenaman and McQuail’s (1961)
analysis suggests that the Conservative image as a party of prosperity was the key to their late
surge in support. Electoral history before and since indicates that parties are indeed virtually
unbeatable if associated with economic prosperity – and more or less doomed to defeat if
associated with economic failure. Moreover, the parties’ economic images and reputations
are at least as important as objective economic indicators in determining voting behaviour.
These arguments demonstrate the central role played by party image in the
‘performance politics’ approach to voting behaviour set out by Harold Clarke and his BES
colleagues (Clarke et al., 2004, 2009). While they do not use the term ‘party image’, they
place heavy emphasis on a party’s reputation for competence and delivery on core valence
issues, especially the economy. When such a reputation – whether good or bad – is clearly
established in electors’ minds, it becomes a central element of that party’s image. It is also
likely to become a general reputation rather than one tied to any single issue. This point was
noted by Butler and Stokes (1974: 339) in their assessment of party images:
7
[s]ome qualities of party image, such as strength or modernity or reliability,
are so broad that they could be linked to almost any set of government
outputs. A party may be seen as trustworthy or as bound to make a mess of
things without any necessary reference to the area in which it can be trusted
or in which it is bound to make a mess. Indeed, some image qualities have
much more to do with ‘intrinsic’ values of party, which are not related to
outputs of government at all.
Butler and Stokes refer to these qualities as ‘valence’ aspects of party image because, just as
no voter wants economic decline and rising crime, no voter wants a governing party that is
weak or incompetent.
Beyond attributes such as strength and competence, there are other elements of party
image that are also related – if more indirectly – to performance politics and might well
influence voters’ willingness to support that party. For example, Labour’s electoral success
in the 1960s was often credited to their image as a more progressive and modern party.
Against a Conservative Party perceived as ageing and backward-looking, Labour was better
placed to capture the mood of youthful excitement and reaction against tradition (Butler and
Stokes, 1974, ch. 16). During the 1980s, Labour and the Conservatives shared an image
problem in that both were widely seen as extreme. When asked to describe each party as
‘extreme’ or ‘moderate’, more than half of BES respondents in 1983 and 1987 chose the
former option. In these polarized conditions, it is not surprising that the SDP/Liberal
Alliance achieved the highest post-war vote shares for the centrist option. Perceived
moderation is clearly not a necessary condition for electoral success given that Thatcher’s
Conservatives won both 1980s elections. Nevertheless, it is hard not to suspect that New
Labour’s advance was owed partly to a dramatic fall in the proportion of voters perceiving
the party as extreme.
Perhaps concluding that ideological convergence had rendered the extreme/moderate
question redundant, BES researchers have dropped it from recent surveys. However, they
have included questions about four valence aspects of party image. Respondents were asked
whether they see parties as ‘united’ or ‘divided’; as ‘in touch’ or ‘out of touch with ordinary
people’; as a party that ‘keeps’ or that ‘breaks its promises’; and as a party ‘capable’ or ‘not
capable of being a strong government’. Table 1 shows responses to these questions for the
three main parties at each of the past three elections.
8
Table 1
Four valence aspects of party image, 2001-2010
2001 2005 2010
United
Conservatives 9 24 61
Labour 70 33 25
Liberal Democrats 88 85 49
In touch with ordinary
people
Conservatives 17 28 36
Labour 54 48 43
Liberal Democrats 68 65 54
Keeps its promises
Conservatives 30 28 34
Labour 43 25 24
Liberal Democrats 70 56 29
Capable of being a strong
government
Conservatives 35 46 71
Labour 84 67 42
Liberal Democrats 47 32 30
There is the obvious popularity effect noted earlier. The Conservative results show that an
improvement in a party’s overall standing tends to boost its image across the board.
Conversely, as Labour lost electoral ground, all aspects of its image suffered at least
somewhat. Nonetheless, party images are obviously more than simply a reflection of a
party’s general popularity – which gives them at least the potential for an impact on electoral
choice. Labour retained most of its credit, and its lead over the Conservatives, in terms of
being in touch with ordinary people. And the Conservatives’ advance was evidently not
achieved by persuading voters that it would keep its promises. The big swings between the
two major parties were on the (probably related) matters of party unity and capacity for
strong government. On the surface, at least, these look the more electorally potent aspects of
party image.
The fact that the various party image measures move in the same direction perhaps
implies that they are multiple indicators of a single latent image variable. On the other hand,
some of the items look more closely linked than others: unity might help a party to keep its
promises and to act as a strong government, but is not obviously related to being in touch. It
is an empirical question – to be addressed in this paper – whether party image is better
measured in separate dimensions or through the more parsimonious approach (taken by
Andersen and Evans, 2003) of constructing an overall index for a party’s image.
9
The results for the Liberal Democrats illustrate two final points about the content of
party images. First, these impressions can be rather vague. The Liberal Democrats projected
a strikingly favourable image in 2001 and 2005 but, with the party having neither held nor
seriously contended for power for decades, it is difficult to imagine voters having a clear
picture of its capacity to govern or even the value of its promises. Second, if impressions of a
party are rather superficial, they can change quickly. The positive image of the Liberal
Democrats had largely worn off by 2010 – especially when it came to trustworthiness and
unity (in a survey conducted following the party’s decision to join the Conservatives in a
coalition government) – and their ratings look much more like those of the other two parties.
As the example of Black Wednesday illustrates, it is not only the Liberal Democrats whose
reputation is precarious. Compared to the durability of social class associations, parties’
valence images are much more transient. As a consequence, while sociological images may
be quite powerful in explaining individual voting decisions, valence images have greater
potential to explain electoral change and election outcomes.
3. Party images and party choice
It is well established that most voters lack much political knowledge (Zaller, 1992; Delli
Carpini and Keeter, 1996). Harder to show directly but almost as well established is that
voters simplify the cognitive task of voting, using one or two simple decision rules to obviate
the need for detailed information and arduous processing (Popkin, 1991; Sniderman et al.,
1991; Clarke et al., 2004). Party image can serve as such a simplifying device or heuristic.
Rather than scrutinising a party’s policy proposals and past record across a range of issues,
voters can use their general impression of that party – does it seem united, sensible and
competently led? Party images are particularly useful in the case of opposition parties, where
no recent record is available. They are an economical means by which to judge likely
performance. And the fact that these images may be rather vague does not preclude their
usefulness – voters are attracted as much by the economy and accessibility of a heuristic as
by its clarity or accuracy. The notion of voting decisions being based on hazy impressions
rather than precise information is entirely consistent with what we know about voter
psychology.
Parallel arguments have been adduced to explain why leader or candidate images are
important influences on voting behaviour (Clarke et al., 2004). In some ways, it is easier to
form images of leaders than of parties. We are used to judging people and will
(automatically) make a number of politically relevant judgements – about a leader’s
10
intelligence, honesty, competence, and so on – based not only on what they say and do but
also on their faces, voices and body language. In contrast, it is much harder – especially for
the less politically aware – to judge complex abstractions like issues and parties. One
consequence is that leader images may be a source of party image – particularly when a party
has been out of power for a while. For example, the widespread perception that Labour had
by 1997 become a plausible party of government probably owed a lot to impressions of Tony
Blair as a strong and capable leader. Rather than scrutinizing the party’s policies or
personnel, many voters instead judged it by its leader.
Yet there are strong grounds to regard party images as having an independent effect
on party choice, rather than merely mediating the effect of leader images. Elections in
parliamentary systems are first and foremost a competition between parties, and so the voting
decision will call party images to mind. And some important facets of party image, like
unity, are collective characteristics that cannot be inferred from a leader’s image. In any
case, projection can run in the other direction: voters may infer a leader’s characteristics from
his or her partisan affiliation. Finally, most parties in mature democracies have been around
a lot longer than their leader at a given election, and so their images are more familiar and
deeper-seated.
Of course, there is a major constraint on the electoral influence of party (or leader)
images: partisan predispositions. The argument was stated clearly in a British electoral
context by Butler and Stokes (1974). Having noted the strong tendency for those who
evaluated a party’s image positively to report voting for that party, they observe: ‘We do not
from this suppose that these aspects of party images are pre-eminent in determining party
choice; the extent of agreement between the two is largely due to electors’ making their
images of the parties fit their pre-existing preferences’ (p. 347). This was already clear from
the earlier studies of class images (Benney et al., 1956). Party identifiers were prone to see
their party as serving the country as a whole while denouncing its opponent as narrowly
focused on the interests of a particular class. The extent of bias should not be overstated,
however, and for two reasons.
The first is the process of partisan dealignment that has taken place in Britain and
almost all other Western democracies in the past fifty years or so. In Britain today, around
half of the electorate is either non-partisan or only weakly aligned with one of the parties.
These voters have more clear-eyed images of the parties and, since they are also more likely
to be undecided in the run-up to an election, there is more scope for party image to influence
their choices. The results in Table 1 confirm that image ratings are not simply reflections of
11
partisanship. Labour in 2001 and the Conservatives in 2010 enjoyed much bigger leads in
terms of unity and strong government than they enjoyed in terms of partisanship or electoral
support at those elections. The pattern can only be explained if, in the reasonably objective
view of a substantial proportion of the electorate, Labour’s image deteriorated while that of
the Conservatives improved. It does not stretch credulity to suggest that this contributed to
the considerable Labour to Conservative swing between those elections.
Second, on Fiorina’s (1981) alternative – and increasingly popular – reading of party
identification as a running tally of voters’ perceptions and impressions of the major parties,
party image looks more like an antecedent than a consequence of partisanship. To say that a
voter is predisposed to support a party because she has found it to be reasonable, capable and
in touch is to put party image in the causal driving seat. This raises questions about how to
specify models testing for an independent effect of party image. On a Michigan reading of
the concept, partisanship should be controlled because it predisposes some people both to
vote for a party and to have a particularly flattering view of it. On a Fiorina reading, that
would be to over-control because the partisanship variable already includes some of the
effects of party image. There is clearly no scope here to adjudicate between rival models of
partisanship (and, in any case, both are probably valid but for different subsections of the
electorate). More importantly, we lack the longitudinal data through which the complex
causal relations between partisanship, leader images and party images might be disentangled.
But there is the possibility of assessing the extent of partisan bias in party image measures.
And we can draw a qualified interim conclusion: if party image variables have a significant
effect on party choice in a multivariate model that also includes partisanship, then they have
passed a particularly stiff test of their independent effect.
While countless such models have been run using BES data, party image variables
have been almost entirely excluded. One exception, and a spur for the update in this paper, is
the analysis reported by Andersen and Evans (2003). While their main focus was on leader
effects, they also show that an overall party image variable – consisting of the
‘united/divided’, ‘keeps/breaks promises’ and ‘capable/not capable of strong government’
questions from Table 1 – was a powerful predictor of party choice in 2001, even with an
array of other variables held constant. Other previous studies have been based on more
meagrely specified models, but also suggest that party images have an independent effect on
electoral choice. Butler and Stokes (1974: 416) found that popular images of the parties were
an important reason for vote-switching between the elections of 1964, 1966 and 1970.
Trilling’s (1976) analysis of US elections showed that party images were not only
12
independent of partisanship but could even override it, especially in cases of relatively weak
party identification. Finally, perhaps the most direct evidence of image effects comes from
Milne and MacKenzie’s (1958) early study. When they asked voters in Bristol ‘What made
you finally decide to vote the way you did?’, party image was mentioned more often than any
other kind of reason. They concluded simply: ‘it seems that images are much more important
in determining voting behaviour than are issues’ (p. 159).
Aside from the arrival of television, there is another possible reason why the late
1950s saw a preoccupation with party image. This was the era of the post-war (‘Butskellite’)
consensus, which gave voters limited scope to choose between parties on the basis of
ideology or policy. This enhances the importance of the valence aspects of party image, as is
neatly demonstrated by a political journalist’s remark about the 1959 election, ‘I am sure that
people care much less about future programmes just now than about choosing which side will
give the better and friendlier service under roughly the present rules. Or to put it another
way, which side has the nicest competent fellows’ (quoted in Butler and Rose, 1960: 28). In
the much more polarised 1980s, by contrast, voters were choosing not just between parties
and their images but between widely different political outlooks and programmes. Or it might
be said during that period that a party’s ideological stance – both its direction and its
extremity – was a dominant facet of its image. Either way, we would expect that subsequent
ideological convergence (see Bara, 2010) has once again brought valence aspects of party
image – like competence, unity, and trustworthiness – to the fore (Green, 2007), which makes
all the more noteworthy the neglect of party image in studies of 21st-century British elections.
To recap: there are strong conceptual grounds to regard party image as an important
component of the ‘performance politics’ model, and the limited empirical work so far also
points in that direction. In the next section, I describe the data required for a more detailed
inspection of the electoral impact of party images.
4. Data and measures
In order to measure the impact of party image on electoral choice, it is necessary first to find
out what these images are. Aside from the focus group approach often used by the political
parties themselves, two main methods have been used. In the open-ended approach, survey
respondents are asked to describe in their own words what they like and dislike about the
parties (or perhaps – again in their own words – what they see the parties as standing for).
This has been the standard approach in the American National Election Surveys and was also
used in early studies of party image in Britain. As Trenaman and McQuail (1961: 23) put it,
13
if there are common elements in people’s open-ended responses about a party, ‘one might
reasonably describe this commonality as the party image’. The alternative closed approach is
illustrated by the results in Table 1. Respondents rate the parties on various aspects of party
image that have been decided in advance by the survey researcher. In effect, open questions
ask voters for their images of a party as it actually is, while closed questions specify what that
party should be and ask voters about how closely it conforms to that ideal.
Both approaches have pros and cons. Open questions reveal great diversity in the
content and clarity of party images. Some attribute might dominate one voter’s thinking but
go apparently unnoticed by others. Different parties call different kinds of attributes to
voters’ minds. The contents of party images can also vary considerably across elections and
this variety is suppressed by simply asking for ratings of each party on the same standard list
of attributes at each election. Yet open questions also show that some themes – like
competence, unity and trustworthiness – recur at more or less every election. Moreover, if
the aim is to assess the overall images of the parties at election time and to relate these
assessments to voting behaviour, closed question ratings are much easier to use than a mass
of open-ended materials. That has been the approach taken by BES researchers and hence
closed questions are the measures available for this paper.
The specific measures have already been indicated in Table 1. Precise wordings and
formatting can be seen in the Appendix, a screen grab from the 2010 BES questionnaire
which included the four items for each of the three main parties. The 2010 study is our main
focus of analysis but the core regressions are also run on the 2005 survey. The replication
provides not only a validation test but also some initial indication of whether the importance
of party image variables depends on electoral context, the 2010 election having been
altogether a more hotly-contested and less predictable election. Further details of sampling
and fieldwork in these two BES data collections are available at http://bes2009-10.org/ and
http://www.essex.ac.uk/bes/2005/Survey%20Documents/bes%20technical%20report%20plus
%20internet%202007.pdf (for 2010 and 2005 respectively).
While recent BES have also involved a major internet component, the on-line surveys
have not included the party image variables and so analysis here is based on the ‘traditional’
face-to-face component. However, the party image items are not just excluded from the
internet survey but also exiled to the self-completion supplement which respondents are
asked to mail back following their post-election face-to-face interview. This has obvious
disadvantages (and, indeed, one subsidiary aim of this paper is to suggest that these items are
given higher priority in future). It rules out any attempt to tease out causal effects of party
14
image on other explanatory variables, notably party identification. It means that the party
image questions are asked of a smaller and less representative sample, because those
respondents willing to complete an additional questionnaire are not a representative but a
disproportionately engaged and motivated subsection of the target sample. And it means that
party image responses are coloured by events since the election, which obviously biases any
attempt to use self-completion responses to ‘predict’ or explain a party choice that was both
made and reported earlier.
However, an important clarification is needed here. Immediate post-election surveys
are subject to a kind of rationalisation bias, in which respondents seeking to appear consistent
bring their perceptions and opinions into line with the vote just cast or reported. This
consistency bias inflates the apparent effect of those perceptions on vote choice. By contrast,
the kind of ‘events bias’ described above tends if anything to weaken the apparent effect of
the variables measured some time after the election. The point is well illustrated by the 2010
election. As noted earlier, the Liberal Democrats’ decision to join the Conservatives in
coalition hugely dented the former’s reputation for keeping promises. Yet those post-election
dismissals of the Liberal Democrats as untrustworthy cannot account for the choices made on
polling day itself; indeed, many of those who then lost faith in the Liberal Democrats’
promises will have done so having voted for the party. All of this means that the analyses
here provide, if anything, a conservative test of the effects of party image on voting
behaviour, because the source of bias is not that vote choice but subsequent events. Ideally,
of course, the party image variables would have been taken from a pre-election wave of the
survey. Fortunately, we can at least take that route with most of the other explanatory
variables used in the multivariate modelling of party choice.
The party image variables are sold short not just in terms of questionnaire location but
also in terms of response scale. While the large majority of BES attitude items involve four-,
five- or eleven-point scales, party image is measured using dichotomies along with a “don’t
know” option. There is no obvious conceptual reason to regard these variables as
dichotomies rather than continua, and empirically it is hard to believe that these measures do
justice to the shades of public perceptions of their parties (see Krosnick and Fabrigar, 1997).
Also, in the absence of a midpoint, we might reasonably assume that at least some of the
“don’t know” responses reflect ambivalence rather than indifference: that is, for example,
those who regard the Conservatives as neither wholly in touch nor wholly out of touch but
somewhere in between. For this reason, and following the example of Andersen and Evans
15
(2003), several of the analyses are run with a three-category version of the image variables –
that is, with “don’t know” recoded to a middle value.
5. Results
The main purpose of this section is to estimate the impact of party image perceptions on
voting behaviour in the 2005 and 2010 elections. Before reporting the multivariate models
required for that task, two prior steps are worth taking. The first is to examine just how
driven are these variables by partisan predispositions, since this places an upper limit on their
potential influence over electoral choice. Then I look at the question of the dimensionality of
party image, assessing whether the vote models should include separate measures or a
composite index for each party.
Partisanship and party image
Clearly, we would expect a healthy positive correlation between liking for a party and
evaluations of its image. The key question here is whether that correlation is so strong as to
raise suspicions that party image is little more than a proxy for partisan predispositions.
Since the question “how strong is too strong?” is a difficult one to answer, a yardstick for
comparison is useful, and an obvious choice here is leadership evaluations. Attitudes to party
leaders have been shown to be quite strongly conditioned by partisanship but nonetheless to
have an independent impact on party choice (Bartle, 2005; Andersen and Evans, 2003; Clarke
et al., 2004). In this analysis, then, I investigate the bivariate relationships between party
identification and both leader affect and party image variables.
Comparison between the leader and party image results will be distorted if the two are
measured on very different response scales, and so the leader affect items have been
collapsed from eleven-point scales to match the format of the image dichotomies. As a
robustness check, I run the comparison with both three- and two-category versions of the
image items – in the latter case excluding those who gave a midpoint or “don’t know”
response on the leader affect scales. Since party identification is a nominal variable (with
five categories: none, Conservative; Labour; Lib Dem; other), an appropriate measure of
association is Cramer’s V (which is based on the χ2 statistic from the relevant crosstabulation,
adjusted for sample size). Each cell of Table 2 represents a V statistic denoting the
association between party identification and the party or leader evaluation in question. The
final column of the table gives, for each evaluation, a simple arithmetic average of the
associations across the parties.
16
Table 2
Bivariate associations (Cramer’s V) between party identification and party image
variables/leader evaluations
Conservative/ Labour/ Lib Dems/ Average
Cameron Brown Clegg
3-category variables
Party image
In touch 0.36 0.32 0.19 0.29
United 0.23 0.24 0.16 0.21
Promises 0.36 0.32 0.19 0.29
Strong 0.30 0.37 0.18 0.28
Leader evaluation 0.37 0.36 0.24 0.32
2-category variables
Party image
In touch 0.51 0.42 0.25 0.39
United 0.28 0.32 0.18 0.26
Promises 0.53 0.45 0.32 0.43
Strong 0.41 0.49 0.29 0.40
Leader evaluation 0.52 0.54 0.32 0.46
Most of the associations in the table are indeed quite strongly positive. The exceptions tend
to be in the Liberal Democrat column, understandably given that there are fewer Lib Dem
identifiers and that – at least at the time of the 2010 election – that party and its leader tended
to excite less antagonism. More important for present purposes is that the party image
variables look if anything less biased by partisanship than are the leader evaluations. The
differences are often narrow but, whether the three- or two-category measures are used, the
average association between partisanship and leader evaluations is stronger than the
corresponding average for any of the party image variables. Judgements of party unity were
the most ‘objective’ – that is, the least predictable based on partisanship – which perhaps
indicates that internal disharmony is the hardest facet of image to fake. In any event, the
bottom line here is that the party image variables do not appear so contaminated by partisan
bias that they have no potential for independent influence over electoral choice.
The dimensionality of party image
I assess the dimensionality of party image by entering the four items for each party in 2010
into a principal components analysis. Since the party image variables are dichotomous (or at
best trichotomous), the PCA is based on a matrix of polychoric correlations between the
17
items. These correlations are generally strong, mostly between 0.5 and 0.7, which already
points towards a single-factor solution. A more rigorous test is available from the results in
Table 3, which show the eigenvalues and proportions of variance explained by the first two
components in each case. Again, the analyses are run with both the three- and two-category
versions of the image items.
Table 3
Summary results from principal components analyses of four party image items
Conservative Labour Lib Dems
3-category variables
Eigenvalues
Factor 1 2.80 2.90 2.28
Factor 2 0.55 0.54 0.65
% variance explained
Factor 1 70.1 72.5 57.0
Factor 2 13.7 13.6 16.1
N 1767 1758 1748
2-category variables
Eigenvalues
Factor 1 3.21 3.14 2.88
Factor 2 0.44 0.44 0.52
% variance explained
Factor 1 80.2 78.4 72.0
Factor 2 11.0 11.1 13.1
N 1140 1379 672
The message from these results is clear: for all three (and especially the largest two) parties,
party image can reasonably be treated as a unidimensional concept. In all cases, the first
component accounts for comfortably over half of the variance, dwarfing the explanatory
contribution of the second and all subsequent components. The same conclusion is equally
clear if we instead apply Kaiser’s criterion and extract only factors with an eigenvalue greater
than 1. The message about unidimensionality is even louder and clearer when using the
‘cleaner’ dichotomous measures of image (i.e. omitting “don’t know” responses rather than
treating them as a middle category). For the sake of completeness, it is worth noting that an
oblique rotation tends to uncover a consistent pattern: the ‘in touch’ and ‘keeps promises’
items tend to load on the same factor, while ‘unity’ and ‘strength’ retain their own separate
components. Nonetheless this looks like embellishment and detail. A single-dimensional
18
solution is reasonable as well as pleasingly parsimonious. In the concluding section, I discuss
the implications of this result. Meanwhile, it justifies the construction of party image indices
for the upcoming regressions.
Party image and party choice
The basis for this section is a series of multinomial logistic regressions predicting party
choice in the 2005 and 2010 elections. Analysis is confined to voters who chose one of the
three major parties, Conservatives, Labour and Liberal Democrats, partly due to cell size
issues but also because we have party image measures only for those three parties.
Conservative voting is the reference category of the dependent variable.
The models are well specified, including a wide range of socio-demographic, identity,
attitudinal and perceptual variables shown in previous research – notably the models in
Clarke et al. (2004, 2009, 2011) – to predict electoral choice in Britain. The full list of
controls is shown in Table 4 (with the reference category for categorical variables shown in
parentheses). Precise codings and wordings can be found at the BES web addresses cited
earlier. I will not discuss the effects of these other variables in detail; they have received
attention elsewhere but here are of interest only as yardsticks against which to assess the
relative impact of party image. The image items are therefore the main variables of interest.
In the first set of analyses, guided by the principal components analysis results, I use a simple
summative image index for each party. Since there were numerous “don’t know” responses
for the Liberal Democrats on some questions (especially in 2005, when voters might
reasonably not have considered whether the party was capable of strong government), I avoid
using the dichotomous image items because they would drastically reduce sample size – a
particular problem with so many predictor variables.
19
Table 4
Multinomial logistic regressions of party choice (base = Conservative) in 2005 and 2010
2005
2010
Labour
Lib Dem
Labour
Lib Dem
B s.e.
B s.e.
B s.e.
B s.e.
Age (18-34)
35-44 -.15 .43
.33 .40
-.94 .72
-.71 .63
45-54 -1.10** .45
-.44 .41
-.77 .73
-.39 .64
55-64 -.60 .48
-.09 .43
-1.04 .73
-.92 .65
65+ -1.10** .47
-.45 .43
-1.05 .73
-.42 .65
Gender (female) .53* .28
.19 .25
-.22 .38
-.26 .33
Ethnicity (non-white) .71 .81
.23 .79
-.29 .75
-.16 .70
Religion (none)
Church of England -.43 .31
-.52* .28
.13 .43
-.23 .36
Church of Scotland -1.22** .49
-.91** .43
-.15 .73
-.30 .64
Catholic -.84* .50
-.99** .48
-.46 .63
-.65 .58
Non-conformist -.86 .74
-1.07 .69
-.28 .82
-.99 .73
Non-Christian -.49 .57
-.49 .51
.51 .86
.27 .79
Social class (not classed)
AB .28 .49
.22 .44
.61 1.16
.49 .89
C1 .28 .48
.48 .44
.65 1.16
.45 .89
C2DE .51 .51
.95** .47
.99 1.18
.55 .93
Degree -.35 .38
.39 .33
1.05** .46
.47 .39
Private sector .14 .27
-.00 .24
-.08 .36
-.19 .30
Own home -.70** .35
-.47 .33
-.04 .48
-.04 .42
Newspaper (none)
Right-wing -.18 .29
-.20 .25
-.32 .37
-.97*** .32
Left-wing .24 .44
-.07 .42
1.19* .66
1.35** .60
Party ID (none)
Conservative -1.41*** .41
-1.10*** .32
-1.96*** .63
-1.24*** .45
Labour 2.05*** .41
1.09*** .40
2.84*** .54
1.79*** .53
Lib Dem .64 .50
1.81*** .43
.04 .64
1.68*** .50
Other 1.10** .54
.86* .50
.79 .88
1.60** .70
Leader affect
Conservative -.17** .07
-.09 .06
-.47*** .13
-.62*** .12
Labour .21*** .06
.04 .05
.42*** .09
.06 .07
Lib Dem -.07 .07
.03 .06
.43*** .13
.93*** .12
Party best on MIIa
Conservative -.99*** .37
-.74** .29
-.00 .48
-.73* .39
Labour .59 .41
.19 .40
.52 .79
-.08 .80
Lib Dem 2.42*** .81
2.70*** .74
-.75 .92
-.68 .72
Party best on economy
Conservative -.29 .37
-.31 .30
-.95* .53
.16 .47
Labour .42 .38
-.27 .35
.51 .79
.86 .81
Lib Demb -- --
-- --
.27 .96
1.72** .83
Tax-spend position .10 .07
.04 .06
.01 .10
.04 .09
Crime position .01 .06
.04 .05
-.10 .07
-.06 .07
20
EU approval -.25* .14
-.21* .12
.20 .18
.30* .16
War approvalc -.07 .10
-.14 .09
-.36** .16
-.38*** .14
Outcome evaluations
Economy .08 .16
.28** .14
-.42** .16
-.37*** .14
Crime .30* .15
.24* .14
-.02 .19
.07 .17
Immigration -.02 .13
-.01 .12
-.36* .19
-.03 .16
NHS .20* .13
-.04 .11
-.12 .18
-.26 .16
Party image
Conservative -1.10*** .13
-.90*** .11
-.95*** .19
-.67*** .17
Labour .91*** .13
.33*** .13
.71*** .17
.36** .16
Lib Dem -.10 .15
.70*** .13
.00 .19
.10 .17
Cox & Snell pseudo-R
2 0.60
0.70
N 1553 1223
* p<0.1; ** p<0.05; *** p<0.01
a ‘MII’ refers to ‘most important issue. Respondents are asked to name that issue and then to name the party (if
any) best equipped to deal with that issue; the second question is the basis for this variable. b In 2005, the Liberal Democrats were not offered as an explicit response option to the question of which party
would best handle economic difficulties. c Reflecting the changing prominence of the two conflicts, this variable concerns Iraq in 2005 and Afghanistan
in 2010.
Amid the welter of detail in Table 4, the key result is that party image does have a clear
independent effect on party choice. Even controlling for an array of other variables, nine of
the twelve party image effects are statistically significant and in the expected direction. And
two of the exceptions are predictable enough: we probably would not expect voters’ image of
the Liberal Democrats to influence their choice between the two major parties. Not only are
the Conservative and Labour image effects significant across the board, but the coefficients
look substantial in size relative to their standard errors. But the strength of relationships in
logistic regression is not easily discerned from logistic coefficients; it is more informative to
look at the effects of a variable on predicted probabilities of the dependent variable. Here, I
use the Stata command ‘prchange’ to calculate the change in predicted probability associated
with a one standard deviation change in each independent variable: from half a standard
deviation below to half a standard deviation above its mean. With dummy independent
variables, where means and standard deviations are less meaningful, I instead calculate the
effect of moving from 0 to 1.1 In this multinomial context, there are two effect sizes for each
predictor (reflecting the two coefficients: Labour versus Conservative and Lib Dem versus
Conservative) and so the changes in predicted probabilities are averaged across the two.
1 Since this amounts to the full range of that independent variable, this approach tends to bias effect size
calculations in favour of dummy variables. A common alternative, illustrating the effects of a full-range change
in all variables (e.g. Clarke et al., 2004), instead inflates the apparent effect of variables – like leader effects –
with longer response scales.
21
Those changes are illustrated for both 2005 and 2010 elections in Figure 1. Two effect sizes
are calculated, reflecting the earlier discussion about whether party identification should be
controlled when calculating image effects. The ‘net’ effect is calculated from the models in
Table 4 which include party identification; the ‘gross’ effect is from models (not shown in
full) from which partisanship is omitted.
Figure 1
Mean standardized effects of independent variables on party choice probabilities
a) 2005 b) 2010
0 0.1 0.2 0.3
Con leader
Lab leader
LD leader
Con MII
Lab MII
LD MII
Con economy
Lab economy
EU approval
War approval
Economy
Crime
Immigration
NHS
Con image
Lab image
LD image
Average change in probability for a 1 s.d. change in X
'Gross'
'Net'
0 0.1 0.2 0.3
Con leader
Lab leader
LD leader
Con MII
Lab MII
LD MII
Con economy
Lab economy
EU approval
War approval
Economy
Crime
Immigration
NHS
Con image
Lab image
LD image
Average change in probability for a 1 s.d. change in X
'Gross'
'Net'
For the main parties at least, the absolute size of party image effects is quite large and
reasonably consistent across the two elections. A standard deviation change in perceived
image of the Conservatives is associated with an average change in party choice probabilities
22
of around 0.15. Surprisingly, the Liberal Democrats’ image mattered less in 2010 than 2005,
despite the party’s ostensibly stronger challenge in the former election. One likely
explanation, supported by the results for party leader effects elsewhere in the graph, is that it
was their leader, Nick Clegg, rather than the party itself that was uppermost in voters’ minds
in 2010. This brings us to consideration of the relative importance of party image, and this
does differ quite markedly across the two elections. Party image looks the most important
factor in 2005 – especially considering that the comparison with ‘most important issue’ is
biased in favour of the latter (see note 1). In 2010, however, the effects of image are
surpassed by those of leader evaluations – (an obvious but speculative explanation is the
introduction of televised debates) – and matched by the effects of the highly topical variable
asking about the party deemed best able to cope in times of economic difficulty.
Nonetheless, even in 2010, the images of the major parties remain among the stronger drivers
of electoral choice, comfortably more influential than the attitudinal variables and the
evaluations of performance in various policy domains. Moreover, and crucially, this impact
is barely dented by controlling for partisanship (i.e. shifting from the ‘gross’ to ‘net’ effects).
While those who support a party are naturally inclined to view it more favourably, this cannot
account for the subsequent tendency for those who view a party more favourably to vote for
it.
Given these conclusions about effect size, we would expect the inclusion of image
items to improve the overall predictive power of models of party choice. The point is tested
in Table 5, which show fit statistics – Cox & Snell’s pseudo-R2 and AIC2 – for three models:
i) the Table 4 model but without any image items; ii) the model exactly as it is in Table 4; and
iii) a model including the individual image items rather than the composite indices. Bearing
in mind the number of variables (including partisanship) controlled in the first base model,
the boost in explanatory power – a 0.06-point increase in pseudo-R2 and a large fall in AIC -
delivered by the party image indices in 2005 is striking. The corresponding effect in 2010 is
weaker, partly because – as indicated in Figure 1 – variables in the base model did a better
job of predicting party choice in that year, leaving less scope for party image to have an
impact. Nonetheless, the smaller AIC for Model 2 indicates that the image indices are a
worthwhile addition in 2010 as well. Since the indices are based on the four image items,
replacing the former with the latter cannot reduce pseudo-R2, but AIC can tell us whether
2 The Akaike Information Criterion (AIC) is a basis for model selection that takes account of parsimony as well
as predictive accuracy, penalising models that deliver little explanatory power for a large increase in the number
of variables included. Lower values of AIC signify a better model.
23
any increase in explanatory power is worth the loss of parsimony. The answer varies across
the two elections. In 2010, AIC is comfortably higher in Model 3 than Model 2, indicating
that the indices are the preferable approach – in line with the PCA results in Table 3. In
2005, the individual items generate a perceptible increase in pseudo-R2 and the AIC falls
slightly. This implies that party image was less resolutely unidimensional in 2005 and thus
that there was more scope for the different facets of party image to exert distinct effects.
Table 5
Summary statistics from multinomial logistic models of party choice
i) Without party
image
ii) With composite
image indices
iii) With individual
image items
2005 (N=1553)
Cox & Snell pseudo-R2 0.54
0.60
0.62
AIC 1706.4 1505.9 1500.9
2010 (N=1223)
Cox & Snell pseudo-R2 0.68
0.71
0.71
AIC 1001.5 957.1 981.2
It is therefore worth a final analysis, running Model 3 and using the same effect size approach
as in Figure 2 to show the relative importance of the different aspects of party image. These
models are not reported in full and so statistical significance is shown graphically: where
neither of the coeefficients (Labour vs. Conservative or Lib Dem vs. Conservative) was
significant at the p<0.05 level, the effect size bar has been made transparent.
24
Figure 2
Mean standardized effects of image items on party choice probabilities, 2005 and 2010
0 0.02 0.04 0.06 0.08 0.1
Con in touch
Lab in touch
LD in touch
Con united
Lab united
LD united
Con promises
Lab promises
LD promises
Con strong
Lab strong
Lib strong
Average change in probability for a 1 s.d. change in X
2010
2005
A first point worth noting is that each of the four facets of image is significant – statistically
and often substantively – in at least one case. This vindicates the choice of image measures
in the BES. However, there are differences in relative importance, both across aspects and
across elections. Perceptions of party unity are relatively unimportant, although they
mattered for the Conservatives in 2010 which, as per Table 1, will have worked in the party’s
favour. Faith in the major parties’ promises matters consistently (while faith in Liberal
25
Democrat promises barely mattered, understandably given the party’s severely limited scope
to act on them prior to 2010). The asymmetries in the ‘strong government’ results may be
explained by the shifting focus of voters’ doubts concerning that criterion: in 2005, the
Conservatives’ competence was more in doubt and so their image mattered more; by 2010,
the Conservative opposition was more widely seen as competent than the Labour incumbents,
and so it was judgements of Labour that mattered most. In contrast, the asymmetries on the
‘in touch’ variable defy simple explanation. The fact that the Conservatives are most often
tested on that criterion makes sense given that their social origins are furthest from ‘ordinary
people’, but there is no obvious reason why that variable would have mattered for the other
two parties in 2005 but not 2010. A more detailed analysis of these differences should
involve more elections (and, ideally, more aspects of image). For the moment, perhaps the
bottom line is that significant effects outnumber non-significant effects. That reinforces the
potential for party image to explain party choice.
6. Conclusion
The key question for this paper was: can the overlooked variable of party image make an
independent contribution to explaining electoral choice in Britain? The answer is a clear
‘yes’. Even in well-specified models controlling for an array of other variables, party image
variables had effects that were not only statistically significant but comparable in size to
those of variables – like leader evaluations and assessments of the parties’ economic
competence – that are cornerstones of the valence or ‘performance politics’ model. But party
image is not a rival or even a complement to that model. There are clear conceptual links
between valence politics and aspects of party image like unity and trustworthiness. The case
made here is not only for the routine inclusion of party image in models of British voting
behaviour, but also for the incorporation of image as a key element within the currently
dominant valence framework.
Another thing that the party image measures have in common with leader evaluations
and competence ratings is that they are inevitably subject to endogeneity bias (Wlezien et al.,
1997; Evans and Andersen, 2006). Partisan predispositions will tend to inflate the
association between party image and vote. Moreover, this bias is only mitigated and not
eliminated by controlling for party identification, because there are those who are not party
identifiers but nonetheless go into an election campaign with inclinations for and against
certain parties. Yet endogeneity cannot simply explain away these effects. For one thing, we
saw that perceptions of party image were more independent of party identification than were
26
leader evaluations. For another, the endogeneity thesis cannot explain why, among various
predictors that prime partisanship equally strongly, some have much larger coefficients than
others. The fact that party image was consistently important, often more so than variables
such as evaluations of a party’s economic competence, clearly indicates that image counts at
the ballot box.
The neglect of party image in recent analysis of British voting behaviour raises a
chicken-and-egg question: are these items exiled to the self-completion questionnaire because
they have (wrongly) been deemed unimportant, or are they omitted from analysis because it
is unsatisfactory to predict post-election reported vote using perceptions reported in a later
mailback survey? Either way, the results here make a compelling case for giving these items
equal billing to those whose place in the valence model goes unquestioned but whose
empirical performance is no better than that of party image. Attention might also be given to
the measures themselves. If even these rather crude dichotomies yield quite strong effects, it
is worth examining whether more refined response scales would reveal an even stronger
impact. As for the content of the measures, there is a case – at least in some sort of pilot
study – for the kind of open-ended approach used in the ANES as well as the earliest British
examinations of party image. Even if such open-ended material is hard to deploy in
quantitative analysis, it could guide the selection and design of new forced-choice items that
would better capture the mental images that British voters today have of their parties.
27
References
Andersen, R. and Evans, G. (2003) ‘Who Blairs Wins? Leadership and Voting in the 2001
Election’, British Elections & Parties Review, 13, 229-47.
Bara, J. (2010) ‘The 2010 Manifestos: Was it only “the economy, stupid”?’ Paper presented
at the Elections, Public Opinion and Parties conference, University of Essex, 10–12
September.
Bartle, J. (2005) ‘Homogenous models and heterogeneous voters’, Political Studies, 53: 653-
75.
Benney, M., Gray, A. P. and Pear, R. H. (1956) How People Vote (London: Routledge &
Kegan Paul).
Brewer, M. (2009) Party Images in the American Electorate (New York: Routledge).
Butler, D. and Rose, R. (1960) The British General Election of 1959 (London: Macmillan).
Butler, D. and Stokes, D. (1974) Political Change in Britain, 2nd edn (London: Macmillan).
Clarke, H., Sanders, D., Stewart, M. and Whiteley, P. (2004) Political Choice in Britain
(Oxford: Oxford University Press).
Clarke, H., Sanders, D., Stewart, M. and Whiteley, P.. (2009) Performance Politics and the
British Voter. Cambridge: Cambridge University Press.
Clarke, H., Sanders, D., Stewart, M. and Whiteley, P. (2011) ‘Valence politics and electoral
choice in Britain, 2010’, Journal of Elections, Public Opinion and Parties, 21(2), 237-53.
Converse, P. E. (1964) The nature of belief systems in mass publics. In David Apter, ed.,
Ideology and Discontent (pp. 206-69). New York: Free Press.
Delli Carpini, M. X. and Keeter, S. (1996) What Americans Know About Politics and Why It
Matters. New Haven, CT: Yale University Press.
Evans, G. and Andersen, R. (2005) ‘The Impact of Party Leaders: How Blair Lost Labour
Votes’, in P. Norris and C. Wlezien (eds), Britain Votes 2005 (Oxford: Oxford University
Press), 162–80.
Evans, G. and Andersen, R. (2006). The political conditioning of economic perceptions.
Journal of Politics, 68, 194-207.
Fiorina, M. (1981) Retrospective Voting in American National Elections (New Haven, CT:
Yale University Press).
Green, J. (2007) ‘When Voters and Parties Agree: Valence Issues and Party Competition’,
Political Studies, 55, 629–655.
Heath, A., Jowell, R. and Curtice, J. (2001) The Rise of New Labour: Party Policies and
28
Voter Choices (Oxford: Oxford University Press).
Krosnick, J. A and Fabrigar, L. R. (1997), ‘Designing rating scales for effective measurement
in surveys’. In L. Lybert et al (eds.) Survey Measurement and Process Quality (New
York: Wiley), 141-64.
Luskin, R. C. (1987) ‘Measuring Political Sophistication’, American Journal of Political
Science, 31 (4), 856-99.
McKenzie, R. and Silver, A. (1968) Angels in Marble (London: Heinemann).
Milne, R. S. and MacKenzie, H. C. (1954) Straight Fight (London: Hansard Society).
Milne, R. S. and MacKenzie, H. C. (1958) Marginal Seat (London: Hansard Society).
Nordlinger, E. (1967) Working-Class Tories (London: Macgibbon & Kee).
Popkin, Samuel L. 1991. The Reasoning Voter. Chicago: Chicago University Press.
Pulzer, P. G. (1967) Political Representation and Elections in Britain (London: George Allen
& Unwin).
Sniderman, P. M., Brody, R. A., and P. E. Tetlock (1991) Reasoning and Choice:
Explorations in Political Psychology. Cambridge: Cambridge University Press.
Trenaman, J. and McQuail, D. (1961) Television and the Political Image (London: Methuen).
Trilling, R. J. (1976) Party Image and Electoral Behavior (New York: Wiley).
Wallas, G. (1910) Human Nature in Politics (London: Constable).
Wlezien, C., Franklin, M. and Twiggs, D. (1997). Economic perceptions and vote choice:
Disentangling the endogeneity. Political Behavior, 19(1), 7-17.
Zaller, John R. (1992). The Nature and Origins of Mass Opinion. Cambridge: Cambridge
University Press.
29
Appendix: Wording and formatting of 2010 BES party image items
