The Distinctiveness of Referendum Voting? The Cases of the ...€¦ · 1997 Scottish Devolution,...
Transcript of The Distinctiveness of Referendum Voting? The Cases of the ...€¦ · 1997 Scottish Devolution,...
1
The Distinctiveness of Referendum Voting?
The Cases of the 1975 EEC Membership, the
1997 Scottish Devolution, and the 2011
Electoral Reform referendums
Paper presented at the Elections, Public Opinion and Parties (EPOP) conference,
University of Exeter, September 2011
DRAFT. This is a draft of a paper that is likely to be extensively revised prior to publication.
Please do no reproduce without the author’s consent.
SAMANTHA LAYCOCK
(Goldsmiths College, University of London and Brunel University)
2
Abstract
The study of voting in referendums follows much the same format as for studies of general
election voting, probably because referendums are treated as a subset of elections. There are
good reasons, however, to treat these elections differently. The purpose of referendums is to
decide issues, while general elections are to decide which party or parties will form the
government. Parties are often divided on referendum issues, hence the reason for letting the
voters decide, and referendum issues tend to be held on low salience issues. General
elections, on the other hand, are contested by long-established parties, on highly salient
issues. We might reasonably expect, therefore, that influences on vote differ significantly
between referendums and general elections. In this paper, a number of claims about the
differential effects of party identification, issue positions, government evaluations and
campaign effects, and, more generally, the relative impact of long-term and short-term
factors, are investigated using British Election Study panel data on the same sample of voters
in contiguous general elections and referendums. The evidence suggests that the same voters
place different weights on party identification and issue positions in deciding how to vote in
these elections, and tend to give greater weight to more short-term considerations in
referendums. There is little evidence that government evaluations or campaigning have
differential effects between the election contexts.
Key words: referendums, general elections, electoral context, voting behaviour, issue-voting,
party identification, government evaluations, campaign effects.
3
Dunleavy asserts that ‘people systematically vote differently at different types of election’
(1990: 463). This claim was prompted by findings from Northern Ireland elections, where the
fortunes of political parties vary between Westminster, local, and European elections. He
goes on to speculate that differences in the institutional features of these elections, such as
their timing in the parliamentary term and the nature of the campaign, may influence how
people vote. By extension, he claims that the calculations voters make in reaching their vote
decisions depend on the electoral context. For each type of election voters are influenced by
considerations that are distinctive to it. The logic of Dunleavy’s claim is strikingly simple but
widely ignored in the voting behaviour literature.
There are several reasons why we should apply Dunleavy’s logic to referendum voting.
First, ‘only a modest amount of attention has been given to empirical analyses of voting
behaviour in referendums’ (LeDuc and Pammett 1995: 3). For the most part, we have detailed
country specific analyses of referendums and, to a lesser extent, comparative studies of
referendums (Hug and Sciarini 2000: 3-4). However, the country-specific literature (e.g.,
Cronin 1989; Kobach 1993; Sinnott 1995) and the comparative studies (e.g., Butler and
Ranney 1978 and 1994; Gallagher and Uleri 1996) tend to give highly descriptive and/or
historical accounts. Thus, they often fail to provide a rigorous analysis of voting.
Second, attention is rarely ‘focused on voting in contexts other than that of
candidate/partisan elections’ (Magleby 1984: 166). LeDuc and Pammett (1995: 5) claim that
referendums are ‘devoid of some of the long-term partisan and social anchors’ evident in
elections – in particular, party attachments and social group memberships – and that their
outcomes are even more dependent on the short-term elements of a campaign. Hence, they
claim ‘referendums are subject to greater volatility and uncertainty than that typically found
in ordinary parliamentary elections’. Some studies have taken account of the distinctive
nature of referendums when examining voting in them (see LeDuc and Pammett 1995; Clarke
4
and Kornberg 1994) but these are few and far between, and, as far as the author is aware,
there are no studies comparing influences on voting across referendums and general
elections.
A third reason is that there are no general models of referendum voting. The
development of such models seems to have been hampered by the relative infrequency of
referendums, and perceptions that they are ‘one-off’ events, in which voting is simply an
expression of policy preferences (Brown, McCrone, Paterson, and Surridge 1999: 115).
Those working in the field of EU referendums (Szczerbiak and Taggart 2004; Hobolt 2009)
and Swiss referendums (Hug 2002) have come closest to developing general models. The aim
of the research here, therefore, is to develop and test a set of propositions about referendum
voting that takes account of the distinctive characteristics of the referendum.
The paper is organised as follows. To begin with, the contextual differences between
referendums and general elections are explored. From these a number of hypotheses are
developed as to how some of the influences on vote in general elections differ in
referendums, namely party identification, issue positions, government evaluations, and
campaign effects. The more general hypothesis that short-term influences count more in
referendums than in general elections is also developed. Then, these hypotheses are tested on
data from the 1974-9 and 1997-2001 British Election Study (BES) panel surveys, the BES
2010 Continuous Internet Panel Survey (CIPS) and the BES 2011 Alternative Vote
Referendum Study (AVRS). The conclusion considers the evidence of differential effects on
voting between general elections and referendums and considers how referendum surveys
may be adapted to better take account of these differences.
5
Elections v. Referendums
At its simplest, an election is a means of making a choice by voting for a preferred option. In
this respect, a referendum is clearly an election. However, for most people, most of the time,
an election is a choice between party candidates competing for political office. Voters may
take account of policies but this is just one among several considerations when deciding how
to vote. A referendum, on the other hand, does not influence the distribution of political
power, but, rather, decides a matter of policy to be implemented by those already in power.1
In this sense, voting in a referendum is more akin to legislative voting, except that voters can
only accept or reject a referendum proposal; they cannot initiate a vote or amend a proposal
as legislators may do.
Referendums also differ in the types of issues they address. Often the issues –
territorial, constitutional, and moral – are unrelated to the cleavage structure underlying party
politics. As such, they are unfamiliar, and thus, quite probably less important to voters than
partisan issues.
Another way in which referendums differ from elections is in the rules governing
their conduct (Uleri 1996: 8). They are often ad hoc and much is left to the discretion of the
politicians, such as the timing of the poll and the status of the result, especially in the UK
where there are no constitutional safeguards regarding referendums. Elections are always
mandatory and their results are always binding. They may be held, as with referendums, at all
levels of the political system and practice does vary from country to country, but, crucially,
their conduct and the status of the result are rarely negotiable.
1 Sartori (1987: 237) makes a clear distinction between the nature of the voting calculus in elections and
referendums: ‘in [elections], electorates at large choose a person or party that becomes, in turn, entitled to
make the decisions for them. Thus, while the electoral results are in themselves zero-sum, the voting act is a
voting into office that projects itself into processes that may become (in parliament, but especially in their
committees) positive-sum. In short, the voting act is not a final, self-contained act. Instead, referenda (sic) are
final. In this case, voters at large do not choose choosers but decide, and thereby close, an issue. Thus,
referenda are definitely zero-sum and cannot lead, in any sense, to a cooperative game’.
6
Finally, referendums differ from elections in their frequency and predictability. Practice
varies enormously across the world, with Switzerland and some US states holding
referendums frequently and other countries such as the UK hardly calling them. In total, ten
referendums have been held in the UK2, although others have been promised, but only two of
these have been national (1975 and 2011). Most UK voters have little, if no, experience of
voting in referendums. Elections, conversely, happen at frequent and predictable intervals and
with the advent of European and regional elections UK voters have had more opportunities to
vote than ever before.
How referendums change the vote calculus
‘Voting behaviour is traditionally seen as a function of party identification, predispositions,
and ideological preferences’ (de Vreese 2006: 584). It is widely recognised that referendums
differ from general elections in three important respects – the purpose of the election, the
nature of the vote decision, and the regularity with which these elections are held. Many
studies of voting behaviour in referendums acknowledge these contextual differences but do
not compare their effects with partisan elections. Two broad approaches have emerged in
studies of the numerous recent EU referendums between characterisations of these
referendums as ‘issue’ elections or as ‘second-order’ elections (Szczerbiak and Taggart 2004:
564; Hobolt 2009: 29-31) but this is not a comprehensive examination of the differences
between referendum and party elections. In the rest of this section, these differences are
clarified.
It is clear that the referendum context differs from that in elections in three significant
ways – in what is at stake, in the rules governing them, and in their frequency and
2 Referendums held in the UK include: the 1973 Northern Ireland Sovereignty referendum, the 1975 EEC
Membership referendum, the 1979 Devolution referendums in Scotland and Wales, the 1997 Scottish and Welsh
Devolution referendums, the 1998 Good Friday Agreement referendum in Northern Ireland, the 1998 Greater
London Authority referendum, the 2004 North-East Regional Assembly referendum, the 2011 Welsh
Devolution referendum and the 2011 Alternative Vote referendum.
7
predictability. Two of these – what is at stake and their frequency and predictability – are
likely to affect what influences the vote decision.
In referendums, issues, rather than parties or candidates, are at stake. This would suggest that
evaluations of parties or emotional attachments with them, as encompassed by the concept of
party identification, are likely to be less influential in referendum voting (Hobolt 2005: 89;
De Vreese and Semetko 2004: 700-1; Schuk and de Vreese 2008: 104). Party identification in
general elections is both a source of information about the parties and a motivation to support
a party or to behave strategically to undermine the prospects of a disliked party. In
referendums, a different kind of compact exists between voters and parties – one that centres
on whether the party’s position accords with that of the voter on the referendum issue. In this
sense, referendums are a form of second-order election (Reif and Schmidt (1980), a less
important election because it does not determine the distribution of political power. Indeed,
referendums might be considered the least important of all second-order elections (regional,
local, European and by-elections) because there are no seats of any kind at stake. A common
feature of all second-order elections is the punishment of governing parties and a boost in
support for minor parties. Similarly, referendums are opportunities for voters to punish
governing parties by rejecting their referendum proposals but may be even more vulnerable
to the punishment trap (Schneider and Weitsman 1996) because there are no immediate
consequences to governments if their proposal is rejected. The beneficiaries of many of the
shock defeats of EU referendums in recent years have been the smaller, more extreme
Eurosceptic parties, as predicted by the second-order model.
The politics surrounding referendum issues conspire to reduce the influence of party
identification. Referendums are frequently called because of internal party divisions over an
issue. Parties are often unable to project clear unambiguous cues to their supporters. The
yes/no nature of the debate leads to parties collaborating under the auspices of umbrella
8
groups, which serve to further confuse voters as to the parties’ stances. Referendum issues sit
uneasily within the dominant left-right cleavage of normal party politics (for example
electoral reform or European integration) (Hobolt 2007: 161), which is why parties may be
divided and why governments cannot resolve these issues through normal parliamentary
means. The lack of clear party ownership of the referendum issue signals to voters a
departure from normal party politics and suggests to voters that defection is permissible.
After all, if parties cannot command the loyalty of their representatives, are willing to
campaign with rivals and voters cannot easily identify the referendum proposition with any
particular party, then why should mere supporters remain loyal?
Balanced against these arguments of the reduced effects of party identification on
referendum voting are claims that party recommendations act as heuristics or cues which help
voters to navigate the complexities of deciding on referendum issues (Jenssen and Listhaug
1999: 7-8; Hobolt 2009: 35-7). Rational choice theory suggests that information costs on
political issues are always high (relative to the benefits or the weight of one vote). Voters
lack the time, interest, and cognitive capacity to access information about the referendum
issue and struggle to evaluate the likely consequences of accepting or rejecting the proposal
(Maddens 1996: 54; Borges and Clarke 2008: 437; Clarke, Kornberg, and Stewart 2004:
346). Party identification also helps voters to determine what information is trustworthy and
which parties’ recommendations to follow (Lupia and McCubbins 1998). LeDuc (2005: 186)
goes so far as to suggest that party positions ‘provide one of the strongest available
information cues to voters’. Yet, there appears to be an inherent contradiction in the role of
party cues in referendum voting. While commentators agree that voters are most in need of
cues in referendums, parties face significant barriers in persuading voters to adopt their
positions. The empirical evidence demonstrates that large numbers of referendum voters do
not adhere to the party line (Pierce, Valen, and Listhaug 1983; Siune and Svensson 1993;
9
Kobach 1994; Magleby 1994; Trechsel and Kriesi 1996; Sciarini and Listhaug 1997; Midtbo
and Hines 1998; Hobolt 2006; de Vreese 2006; Hobolt 2009: 135-60) and that party
identification ranks some way below issue positions in influencing referendum preferences
(Hobolt 2009: 75-9).
It would seem obvious that if issues are the subject of referendums then issue
positions assume greater importance in referendums than they do in general elections. After
all, the issue at stake is defined by the referendum question, whereas as in general elections
there may be several issues of interest. Voting on issues, even on a single issue, however, is
arguably more demanding of voters than voting on parties/candidates, especially when
referendum issues are not prominent in the party political debate or on the news agenda.
Ironically, the cognitive difficulties of voting on single issues might explain why partisan
influences persist in referendums, albeit in a somewhat weaker form than in general elections,
when we might expect them to disappear altogether.
Expectations about the importance of issue preferences are not well developed in the
referendum literature. Some reference is made to the complexity of referendum questions
(Aitkin 1978: 131; Wright 1978: 150-9; Tonsgaard 1992: 301; Clarke and Kornberg 1994:
945; Neijens, Minkman and Slot 1998: 301), and there is recognition of the generally low
levels of knowledge of referendum issues (Tonsgaard 1992: 141; Neijens et al. 1998: 302).
The high level of knowledge about the issue in EU referendums stands out as the exception
rather than the rule (Siune and Svensson 1993: 106; Hobolt 2009: 135-60). Apart from these
few examples, there is little theorising about the place of issues in referendum voting.
The infrequency and unpredictability of referendums renders them and the issues they
address, which tend to be one-offs, unfamiliar. Familiarity with elections and their issues is a
source of pragmatic knowledge which can be built up with experience and enable the
development of a voting record. Referendum voters have no such short cuts to fall back on. A
10
vote in a referendum, therefore, might be considered a ‘step in the dark’ (Pattie et al. 1999:
307). Consequently, short-tem influences such as government evaluations, newspaper
readership and exposure to the campaign are liable to be as, if not more influential, than they
are during elections.
According to instrumental theories of voting behaviour, judgements of a
government’s record should be irrelevant in a referendum; the issue at stake is not the overall
performance of the government. However, for those voters with low levels of interest in, and
knowledge of, the referendum issue, government evaluations may be one of the few familiar
cues they can rely on to judge the merits of the referendum issue. If the government has
performed well, this may incline voters to support the referendum proposal; equally, if the
government’s performance is poor, they may be more likely to reject the proposal (Hug and
Sciarini 2000: 7). Voters find it somewhat easier to identify the government’s position on the
referendum proposition, especially if it is the author of the proposal (LeDuc 2002:148 in
Farrell and Schmidt). Governments also enjoy certain privileges during the campaign which
make its position more readily identifiable – official information leaflets may be skewed in
favour of the ‘yes’ option; government figures are better known; and the government’s
position may carry greater weight from the mere fact of being the government. Voters will
expect the government to have expertise on the issue and, therefore, its recommendations
might be thought of as more trustworthy (Lupia and McCubbins 1998: 57-8). As Franklin,
Marsh and McLaren (1994: 102) argue in relation to EU referendums, ‘referenda conducted
in the context of national party politics, with the government of the day urging ratification of
a treaty they have themselves negotiated will inevitably be contaminated by popular feeling
towards the government’. Aside from those referendums that are constitutionally required or
are held in response to popular initiatives, the motivations of governments in calling a
referendum also may be an important consideration. All of the referendums held in the UK
11
have been for reasons of political expediency – whether to resolve a divisive issue (1975 EC
referendum), to entrench a decision that has already been taken (1997 Scottish devolution
referendum), or as a condition of government formation (2011 Alternative Vote referendum).
There is no research relating to the impact of perceptions of government motivations in
calling referendums, although this would seem like an obvious line of enquiry. As to whether
government evaluations carry more weight in general elections than in referendums is
unclear. At the very least, it would appear that government evaluations are as likely to
influence whether someone votes ‘yes’ or ‘no’ in a referendum as it is to influence party
choice in a general election.
In recent years, attention has turned to referendum campaigns and their effects on
referendum voting (LeDuc 2002; de Vreese 2004; de Vreese and Semetko 2004; de Vreese
2007; Schuk and de Vreese 2008; Hobolt 2009). It is claimed that referendum campaigns are
likely to be more influential than general election campaigns (although there are no direct
comparisons of general election and referendum campaign effects) because of greater voter
volatility, cue uncertainty, issue unfamiliarity and complexity than in general election
campaigns (de Vreese and Semetko 2004: 700; de Vreese 2007: 1; Schuk and de Vreese
2008: 103-4; Hobolt 2009: 88-9; LeDuc 2002: 145). In a comparison of opinion polling
evidence of vote intention, LeDuc found 50 per cent more volatility in referendum than
general election campaigns. Greater volatility is to be expected if party cues are more
difficult to follow because of the increased likelihood of party divisions and the presence of
umbrella campaign groups. Referendum propositions also tend to be on constitutional
matters, involving abstract principles. Such issues are distant from voter’s everyday
experience and are usually the preserve of elites. Assuming that people choose not to cast
their votes in ignorance, campaigns are likely to be particularly important in referendums, as
they provide valuable information (Popkin 1991:70; Holbrook 1996:16). The trouble is that
12
political scientists are undecided about the effects of campaigns (Schmitt-Beck and Farrell
2002:1; Hobolt 2009: 86-8).
None of the above claims about the differential effects of party identification, issues,
government evaluations and campaigns in general election and referendums are new. Apart
from the occasional observation that long-held predispositions are unlikely to be of use in
one-off votes on abstract, unfamiliar and, what many voters perceive to be, inconsequential
issues, there seem to be no comparisons of long-term and short-term effects in partisan
elections and referendums. Invoking Miller and Shanks (1996) funnel of causality metaphor,
we can classify party identification as a long-term influence and issue positions, government
evaluations and campaign effects as short-term influences. Following this reasoning, it would
seem logical to suggest that:
H1: Short-term influences have more influence over support for referendum
propositions then long-term influences,
In particular:
H1A: Party identification is a weaker influence on referendum then general
election voting, especially if the parties are divided and/or they campaign
under the auspices of umbrella groups.
H1B: Issue positions have more influence on referendum than general election
voting.
H1C: Government evaluations are as important in referendums as general
elections.
H1D: Campaigns and their media coverage have more influence on referendum
than general election voting.
13
The case studies
We are fortunate in the UK that three of the ten referendums held so far – the 1975 EEC
Membership Referendum, the 1997 Scottish Devolution Referendum, and the 2011 Electoral
Reform Referendum – have taken place within months of a general election. We are also
fortunate in other ways. These referendums were held in quite different circumstances, which
will provide robust tests of the hypothesis. The 1975 referendum was called to contain the
potentially explosive divisions with the governing Labour Party (King 1977). Prominent left-
wingers, such Tony Benn, opposed continued membership of the EEC on the grounds it was
a capitalist club, despite Harold Wilson’s renegotiation of the terms of membership when the
Labour Party came into office in February 1974. Collective ministerial responsibility was
suspended to allow dissenting members of the Cabinet to campaign against membership
during the referendum campaign. Unlike many referendum issues, membership of the EEC
would have been familiar to voters, as the prospect of UK membership had been debated
since the early 1960s and a previous attempt to join had failed.
The 1997 Scottish Devolution Referendum was held in somewhat different
circumstances. This time the pro-devolution parties – Labour, the Liberal Democrats and the
SNP – were unified and campaigned as part of the Scotland FORward umbrella group. The
Conservatives were the only party opposed, but like their rivals remained united. The
decision to hold a referendum was taken by Tony Blair, the Labour Party leader, in 1996,
while Labour was still in opposition. He did this to defuse concerns that devolution would
lead to tax hikes, as claimed by the Conservative Party, and to prevent a recurrence of the
1979 debacle, when the last Labour government had attempted Scottish devolution (Taylor
14
and Thomson 1999: 22; Denver, Mitchell, Pattie, and Bochel 2000: 42-3).3 Blair calculated
that the promise of a referendum would remove devolution as a general election issue and
reduce opposition to a devolution bill during its passage through parliament; instead,
opponents of devolution would focus their activities on the referendum campaign (Taylor and
Thomson 1999: 22; Denver, Mitchell, Pattie, and Bochel 2000: 42-3). The referendum was
also called to gain popular approval for an important constitutional reform and to entrench the
decision against the possibility that a future Conservative government would repeal it. Like
EEC membership, Scottish devolution had a long lineage. It had been on the political agenda
since the 1960s and voters had rejected a weaker form of devolution in the 1979 referendum.
The most recent referendum, on Electoral Reform, arose out of the deal struck to form
the Conservative and Liberal Democrat coalition government in the aftermath of the 2010
general election. The Liberal Democrats have long advocated change to the electoral system
in the hope that a more proportional electoral system would deliver a closer approximation of
Westminster seats to their share of the national vote. This was, therefore, a necessary
precondition of the coalition deal. While there were differences of opinion over the
desirability of electoral reform amongst those in the Conservative and Labour parties, and
some Liberal Democrats felt that the Alternative Vote proposal was not proportional enough,
all the main party leaders pledged their support for holding a referendum. Although there has
been a vocal minority campaigning on the issue for many years, there is little evidence that it
is an issue that many voters are familiar with, or, indeed, care much about.
We have, therefore, three referendums held under very different conditions – the EEC
Membership referendum seeking to heal government divisions, the devolution referendum
entrenching the policy against a future Conservative government that might attempt to
reverse it, and the Electoral Reform referendum held as a pre-condition of a coalition
3 Labour rebels in Parliament slowed the progress of the legislation through Parliament and eventually forced
the Labour government to hold a referendum.
15
agreement between one party in favour of electoral reform and another opposed. There is
somewhat less difference in the types of issues. All of them relate to issues that are distant
from people’s everyday experiences but two of them – EEC membership and devolution –
were well-known and one – electoral reform – rather less so. The referendums on the more
familiar issues of EEC membership and devolution resulted in overwhelming majorities in
their favour (see Table 1 for the official figures and the reported votes in these referendums),
while the Alternative Vote referendum failed by a wide margin.4
---Tables 1 and 2 about here---
Data
The data in this paper are drawn from four British Election Study panel surveys, which allow
us to directly compare influences on vote between referendums and general elections on the
same sample of voters. This removes any possibility that differences in the influences on vote
are attributable to sample bias.
The 1974-9 survey draws on the sample that participated in Butler and Stokes 1970
election survey. Of the 1,816 who responded to the earlier survey, the Essex University
survey team were able to contact 1,096 individuals. The sample was refreshed in February
1974, raising the sample size to 1,740. Respondents were interviewed face-to-face on three
occasions – in the months immediately after the 1974 (February and October) and 1979
elections. The June 1975 referendum survey was completed by post.
BEPS 2 consists of annual face-to-face interviews conducted each spring-summer
(around the time of the local elections) with the same respondents between the 1997 and 2001
general elections. Telephone or postal interviews were conducted in-between times (BEPS 2
4 To complement the referendum results in Table 1, the official results for, and the reported vote of the BES
respondents in, the comparator general elections – October 1974, 1997 and 2010 – are shown in Table 2.
16
Note for Users 1999: 1-2). The data used in this study were collected in the first two waves of
the BEPS 2 panel - May and September 1997 – which relates directly to the period of the
1997 general election and referendum. The panel contains a booster sample for Scotland,
yielding 882 Scottish respondents, which is weighted to remove respondents who were not
registered to vote and to compensate for the effects of panel attrition. This leaves a sample
size of 841.
The 2010 BES CIPS and the 2011 BES AVRS, although separate studies, draw on the
same internet sample from 2010. Each study consists of a pre- and post-election wave. In
CIPS, these waves are supplemented by daily rolling surveys throughout the campaign
period. In the AVRS, the pre-election wave consists of daily rolling surveys. The sample size
for both election studies is very large: N=13,356 for the general election and N=18,556 for
the referendum. To compensate for panel attrition, the data were weighted.
Methods
The empirical analysis compares vote in the devolution and general election context.
Conceptually and statistically this may be problematic. Conceptually, the vote choice is
different in the two contexts: akin to comparing apples with pears. The comparison is
warranted, however, on the grounds that our theories of election voting inform our
expectations about the influences on referendum voting.
Statistically, the comparison is complicated by the incompatibility of the dependent
variables. In referendums, electors usually face dichotomous choices – to either accept or
reject the referendum proposition. In general elections, there is usually a choice between
multiple parties. To make valid comparisons, the form of one of the dependent variables must
change to make it compatible with the other. Logically, since the subject is the distinctive
nature of voting in referendums, the form of the referendum choice determines how general
17
election party choice is conceptualised. Party choice, therefore, becomes dichotomous by,
restating the dependent variable as vote for one party versus the rest. As Labour was the
incumbent party for two of the three referendums, general election vote is reformulated as
Labour vote versus vote for the other parties. In this way, general vote most closely
approximates that in referendums, where voters have all possible choices are presented as a
dichotomy.
To ensure that the dependent variables are truly comparable, the analysis is limited to
those who voted in both the general election and the referendum. Thus, we can be fairly sure
that any differences in the relationships between the independent variables and vote choice
are not due to differential turnout. Unfortunately, limiting the analysis to those who voted in
both contexts has a detrimental effect on sample size for the EEC and devolution referendum
case studies.
The hypothesis and its derivatives are tested by a series of binomial logistic
regressions. Acceptance or rejection of the sub-hypotheses depends on whether the size and
significance of the coefficients of the key variables differ between the referendum and
general election models. Summary statistics – the chi-squared of the log likelihood function
and pseudo R-squared5 – are used to judge the overall hypothesis that short-term influences
on vote are of greater importance in referendums than in general elections.
Measures
A selection of variables are included in the analysis to represent the social background of
voters, their party identifications, issue positions, evaluations of the government, and
campaign influences. The social variables consist of the usual suspects found in vote analyses
5 Model chi
2 is analogous to the F-test in OLS regression (Menard 1995: 21). It tests the statistical significance
of the regression model. In logistic regression, pseudo R2 is used to reflect the non-linear assumptions imposed
by a binary dependent variable. There are several pseudo R2 measures to choose from and no consensus as to
which is best (Pampel 2000: 50). Nagelkerke is used here simply because it is the most prevalent in the
political science literature.
18
– age, gender, class, education, housing tenure and trade union membership. National
identity, as measured by the Moreno scale6, was included in the 1997 vote models as
Scottishness is closely associated with preferences for devolution and independence (Brown,
McCrone and Paterson 1998: 211). Party identification is measured by the long-established
question, ‘Generally speaking, do you think of yourself as Conservative, Labour, Liberal
(Liberal Democrat) or what?’ In the Scottish Election Study, from which data for the 1997
devolution referendum is drawn, the standard question also makes reference to the SNP.
Positions on a number of issues are included in the analysis. These vary with the
comparator pairs of elections and referendums but reflect what was on the political agenda at
the time. Some of these have been left as individual variables where only one measure was
available in the dataset. Where several measures are available, these have been scaled. In
1974/5 elections, the analysis contains measures of positions on nationalisation, social
services, wage agreements, trade union power and the EEC. For 1974/5, there is also an
issues scale consisting of attitudes towards NHS spending, comprehensive schools,
repatriation of immigrants, land ownership, the size of foreign aid, penalties for crime,
pollution control, workers’ control, curbing communists, spending on poverty, redistribution
of wealth, decentralisation and the preservation of the countryside. These variables loaded
onto a single factor with an eigenvalue of 2.51 and the resulting scale produced a Cronbach’s
alpha of 0.64. On the 1997 dataset, there are single measures of positions on devolution and
NHS spending. The dataset already contained a scale of the trade-off between the control of
inflation and tackling unemployment and it was possible to construct scales of attitudes
towards taxation and education. The taxation scale consists of five items relating to taxation
levels and the trade off between taxation and public spending. The five items loaded on to a
single factor with an eigenvalue of 2.28 and a Cronbach’s alpha of 0.70. Four items
6 Based on the scale developed by Luis Moreno; see Decentralisation in Britain and Spain: The Cases of
Scotland and Catalonia (1986, Ph.D. thesis University of Edinburgh).
19
comprised the education scale – government spending on education, abolition of private
education, competition between schools and grammar schools. They loaded on to a single
factor with an eigenvalue of 1.735 and a Cronbach’s alpha of 0.58. The 2010 general election
and the AV referendum contained measures of positions on taxation and spending,
Afghanistan, crime, and electoral reform. Two of these, taxation and spending and crime, are
0-10 scales. On Afghanistan, respondents were asked a single question about approval of
Britain’s involvement in the war against the Taliban. Two measures on electoral reform were
combined, relating to preferences for single party majorities and single party government.
Government evaluations were derived from measures of the government’s
performance in handling strikes and inflation in 1974. In 1997, evaluations of the outgoing
Conservative government on its handling of unemployment, inflation, taxation, the NHS,
crime, education, personal standard of living and the economy were used in the analysis of
general election vote but Labour government evaluations were used for referendum voting.
These consisted of Labour’s performance on reducing class sizes, NHS waiting lists and the
numbers of people on benefit, along with its record on devolution, public spending, taxation
and sleaze. The 1997 Conservative government evaluations scale had a Cronbach’s alpha of
0.71 and the Labour government scale was 0.72. A similar battery of items constituted the
Labour government evaluations scale for 2010. This government was judged on its record on
education, immigration, NHS, the financial crisis, the economy, Afghanistan, and taxation.
These evaluations loaded on to a single factor with an eigenvalue of 4.867 and a Cronbach’s
alpha of 0.93. No such battery is to be found on the AVRS. Instead, we have a single measure
of the Conservative government’s handling of the financial crisis.
Campaign effects can be measured in two ways – through direct personal contact with
the campaign (being canvassed, reading campaign literature, attending meetings) and through
indirect mediated channels (consumption of media coverage of the campaign) (de Vreese and
20
Semetko 2004: 705; Schuk and de Vreese 2008: 109-10; Hobolt 2009: 92-3). There are also
measures which may be indicative of campaign effects but are not measures of direct effects
(timing of the vote decision, opinion volatility over the campaign (LeDuc 2002: 151), the
level of don’t knows (de Vreese and Semetko 2004: 707), and the size of issue effects (Garry,
Marsh and Sinnott 2005: 215). Finding suitable measures of campaign effects that are
replicated for contiguous general elections and the referendums on the BES panel datasets
proved difficult. For the 1975 referendum, there are measures of personal exposure to the
campaign through readership of the ‘yes’ and ‘no’ campaign leaflets but there are no
equivalent measures of personal campaign exposure for the 1974 general election. There is,
however, a measure of indirect mediated effects with newspaper readership. For 1997, there
is only newspaper readership. We are a little better served in the 2010 and 2011 elections. In
addition to newspaper readership, we have measures of exposure to canvassing by political
parties and referendum campaign groups. For all the case studies, however, there is the
indicative measure of campaign effects of the timing of the vote decision.
Analysis
Party identification, issue position, government evaluation and campaign effects are
compared between the specimen general elections and contiguous referendums with reference
to the size of the coefficients, the direction of the relationship with vote and by their
statistical significance. These are illustrated in Tables 3, 4 and 5. The fits of the long-term
and short-term models are compared in Table 8 with reference to the model log likelihoods,
pseudo R2s and percentage correct predictions.
There was no expectation that the effects of social characteristics would vary
systematically between electoral contexts and so it proved in the vote models. There is the
odd anomaly. Older voters rather than younger voters were more likely to favour EEC
21
membership in the 1975 referendum, but this had no significant effect on general election
vote (both long-term and short-term models). Foremen and technicians were more likely to
vote Labour in October 1974 (0.81) and less likely to support continued EC membership in
1975 (-0.57), although the effect was somewhat smaller in the referendum (long-term model
only). Education appeared to have a greater influence in the 1975 referendum (1.02) than
1974 general election vote (-0.81) and this effect persisted even when controlling for issue
positions, government evaluations and campaign variables. In 1997, however, education was
only significantly related to general election vote (long-term model only). Younger (-0.01),
female (0.24), semi-skilled and unskilled (0.26), and trade union member (0.30) voters were
more likely to vote Labour in the 2010 general election but it was younger (-0.01), male (-
0.36), middle class (-0.28, -0.35, -0.47) and better educated (0.48) respondents who tended to
vote for the Alternative Vote. Most of these effects dropped out when controlling for short-
term effects. Significant gender (0.36) effects, however, persist for 2010 general election and
2011 referendum vote and strengthen for trade union membership (-0.36) in the general
election.
---Tables 3, 4 and 5 about here---
Our real interest in the long-term models, however, is in the expectations of Hypothesis 1A
relating to the relative impact of party identification across election contexts. The findings of
a simple crosstabulation of party identification and vote, shown in Table 6, suggest that party
identification is more strongly related to general election than referendum vote. In all cases,
party identifiers were more likely to follow the party line in the general elections than the
referendums. There is some variation between groups of party identifiers. Defection amongst
Labour party identifiers was particularly high in the EEC and AV referendums, down by over
20 points compared with the previous general election. In the Scottish devolution referendum,
22
SNP identifiers were more likely to vote in favour of Scottish Parliament than vote for the
party in the general election. Yet, the larger Cramer’s Vs for the general elections (0.634,
0.802, 0.601) supports Hypothesis 1A that party identification is a weaker influence over
referendum vote.
---Table 6 about here---
There is mixed support for this hypothesis in Tables 3, 4, and 5. In 1974 (Table 4),
identification with all parties has a significant bearing on whether people voted Labour or
not. In the EEC referendum, only Conservative party identification (0.95) is significantly
related to voting ‘yes’ to the proposition to stay in the Community, and the size of the
coefficient is smaller than for the general election (-2.61). In 1997, Conservative and Labour
party identification is significantly related to both general election (-1.72, 4.48) and
referendum (-1.34, 1.66) vote but the coefficients, as for the 1975 referendum, are smaller in
the referendum. SNP identification runs counter to this trend by becoming significant for the
Scottish devolution referendum vote (-0.35, 3.60), and strongly so. In 2010/11, party
identification effects are evident in both the general election and the referendum but, for the
most part, they are weaker in the referendum. There are, however, some anomalous findings.
Liberal Democrat identification appeared to exert a stronger effect on voting in favour of the
Alternative Vote electoral system (1.59, 2.04) than on whether to vote Labour (-1.15, -1.78).
Controlling for factors likely to influence voting closer to polling day, contrary to the pattern
found in Tables 3 and 4, the effect of Labour and Liberal Democrat identification increased
on both general election and referendum vote in Table 5.
Figure 1 illustrates further the differential effects of party identification on general
election and referendum vote. It shows the effects of party identification on vote probability,
holding the ordinal and interval variables at their mean values and the dichotomous variables
23
at their mode values. The probability of Conservative identifiers voting Labour in October
1974 falls by 0.41, but increases by less than half that (0.15) in the 1975 referendum. Labour
party identification increases the probability of voting Labour in 1997 by 0.74 but again by
less than half as much (0.34) on devolution. Conservative and SNP identification, on the
other hand, have a smaller impact on general election (-0.15, 0.05) than referendum vote (-
0.29, 0.47) in 1997. In 2010, Labour identification has a much bigger influence on the
probability of voting Labour (0.51) than on voting in favour of the Alternative Vote (0.03) in
2011. However, the effect of Conservative and Liberal Democrat identification is between
one half and two-thirds less in the 2010 general election (-0.16, -0.10) than in the 2011
referendum (-0.33, 0.35). Overall, in four (Conservative and SNP 1997, Conservative and
Liberal Democrat in 2010/11) of the seven pairs of party identification comparisons the
effects were greater in the referendums than in the general election. This was contrary to the
expectations of Hypothesis 1A.
---Figure 1 about here---
Why Conservative, SNP and Liberal Democrat party identification should more strongly
influence support for devolution and the Alternative Vote might be explained by the ease
with which these parties could be identified with either the ‘yes’ or ‘no’ campaigns. Despite
its clear preference for independence, the SNP joined the Scotland FORward campaign
arguing that a vote for devolution cleared the way for independence (Denver et al. 2000: 59-
63). The Liberal Democrats have long championed electoral reform and made a referendum
on it a precondition of their membership of the coalition government. In 1997 and 2011, the
Conservatives were distinctive as the only mainstream party opposed to devolution and
electoral reform. In 1975, voters would have struggled to distinguish between the main
24
parties on the referendum issue. All of them campaigned for continued EEC membership, but
the existence high profile defectors, such as Tony Benn and Enoch Powell (see King 1977
and Butler and Kitzinger 1996), would have raised doubts in voters’ minds about party unity
on the issue. In 2011, senior figures in the Labour party, such as John Prescott and Jack Straw
openly expressed their reservations about the Alternative Vote despite their leader, Ed
Milliband’s decision to adopt it as the party’s position.
These findings accord with those from elsewhere, which suggest there are large
variations in the success rates of parties in referendums in getting their supporters to follow
the party line. The strong influence of the SNP in the Scottish devolution referendum and the
Liberal Democrats on AV, and Labour and the Conservatives’ relative difficulty in
marshalling their supporters in favour of EEC membership also supports de Vreese’s (2006:
589) claims that the most successful parties tend to be smaller, strongly ideologically profiled
and those that campaign with a consistent message. A higher number of dissidents tend to be
found amongst supporters of larger centrist parties with high profile dissidents, which was
evidently the case in 1975.
Choosing suitable issue variables to test the effect of issues in general elections was
more challenging than for the referendums, where the obvious choice was the issue at stake.
Single issues do not dominate general elections as they do in referendums, since parties are
judged on a package of policies. I have attempted to measure the effects of those variables
that were topical for the chosen general elections. According to Gallup respondents in 1974,
the most popular responses to the question about the most urgent problems facing the country
were the cost of living, unemployment, other economic issues, housing and strikes (King
2001: 265). Strangely, the European Community hardly registered. In 1997, Scottish voters
were concerned about health, education, welfare, unemployment, taxation, and devolution in
that order (MORI and ICM 1997). In 2010, the economy was, by far, the greatest concern,
25
followed some way behind by immigration, unemployment and crime (Ipsos-MORI Issues
Index 2010).
Despite the pollsters’ claims that these issues were important to voters, for the most
part, they are not significantly related to general election vote once we control for party
identification. Looking at the short-term models in Tables 3, 4, and 5, we find that only
attitudes towards nationalisation and wage agreements were significantly related to 1974
general election vote and devolution to 1997 general election vote. In 2010, it was attitudes
towards taxation and spending and electoral reform. Those who favoured more
nationalisation (0.39) and who agreed that voluntary wage agreements were effective (0.92)
were significantly more likely to vote Labour in 1974. Attitudes towards the EEC had no
significant effect on general election vote. The higher up the devolution scale, the less likely
respondents were to vote Labour in 1997 (-2.70).7 Those favouring higher public spending
over tax cuts (0.78) were more likely to vote Labour in 2010, while supporters of change to
the electoral system (-0.29) were less likely to do so. As predicted by Hypothesis 1B,
positions on issues, especially the issue at stake, had a stronger effect on referendum vote.
EEC membership (1.50), devolution (5.65), and electoral reform (2.00) attitudes were
strongly, positively and significantly related to 1975, 1997 and 2011 referendum vote. Some
other issue positions were also significantly related to referendum vote – wage agreements
(0.71) in 1975 and crime (0.79) in 2011.
A straightforward comparison of the effects of individual issues across electoral
contexts is unrealistic, so in Figures 2(a), (b), and (c) the combined effects of the issue
positions significantly related to vote in at least one of the contexts are compared. In Figure
2(a) the effects of nationalisation, wage agreements, and EEC membership positions on
general election and referendum vote probability are compared. In Figure 2(b) we see the
7 The negative relationship between positions on devolution and Labour vote might be considered odd
considering devolution was Labour’s policy but those at the highest end of the devolution scale were supporters
of independence, and, therefore, more likely to vote SNP.
26
effects of devolution positions on vote probability as the issue position values change from
the lowest to the highest, while in Figure 2(c) the effects of positions on tax and spend,
Afghanistan, crime, and electoral reform are compared. In Figure 2(a) it is evident that issue
positions hardly change the probability of voting Labour in 1974 but increase the chances of
voting ‘yes’ to continued EEC membership by 0.57 from the lowest to the highest issue
position values. The probability of voting Labour only decreased by 0.25 in the 1997 general
election, while the probability of voting ‘yes’ to Scottish devolution increased by 0.44 in
Figure 2(b) from support for the Union to devolution to support for all out independence.
Likewise, moving through the range of positions on the salient issues in 2010/11, it is evident
that these issues had greater effects on referendum vote. The probability of voting ‘yes’ to the
Alternative Vote increased by 0.80, while it increased by the more modest 0.30 in the general
election. Clearly, as far as Hypothesis 1B is concerned, issues carried more weight in the
1975, 1997 and 2011 referendum than in the 1974, 1997, and 2010 general election voting.
---Figures 2(a) to 2(c) about here---
Some of the referendum literature contends that government evaluations are important
influences on referendum voting and there is some debate as to whether such evaluations
rival issue effects. Nowhere in the literature, however, is it considered whether these
evaluations differ across electoral contexts. Equally persuasive arguments can be advanced
that government evaluations are both more and less influential in referendums than in general
elections. The evidence from Tables 4, 5, and 6 is equivocal. Government evaluations
significantly influenced Labour vote in 1974 but not the EEC membership vote. They had no
significant influence over either general election or referendum vote in 1997. Yet, in 2010
and 2011, government evaluations were a significant consideration, but more so in the
27
general election (2.26) than in the referendum (-0.70). Those who positively evaluated
Labour’s performance in government were more likely to vote for the party in the 2010
general election. However, it was those who rated the Conservative-Liberal Democrat
coalition negatively who were more likely to vote for the Alternative Vote. Figure 3(a)
illustrates this finding more clearly. The change in the probability of voting Labour ranged
from 0.10 for those with the most negative perceptions of the Labour government to 0.90 for
those with the most positive impressions. In the referendum, the change in vote probability
ranging from the most negative to the most positive evaluations was from 0.60 to 0.30 – less
than half that for the general election. Hypothesis 1C – that government evaluations are
equally as influential over vote decisions in general elections as in referendums – is neither
supported nor disproved. However, the evidence suggests that in 2010/11, at least,
government evaluations weighed more on general election than referendum vote
considerations and that support for the Alternative Vote was, as figures such Peter Mandelson
hoped, was something of a protest vote against the Con-Lib coalition.
---Figure 3(a) about here---
The relative effects of campaigning are difficult to discern when measures are not replicated
across general elections and referendums in panel studies, as they are not in the specimen
cases. Of the few measures available, timing of the vote decision can provide impressionistic
evidence of the likely effect of the campaign, as well as being a possible independent variable
in the analysis. Table 7 shows that in the specimen general elections the majority of voters
reached their vote decision in the weeks and months prior to the start of the official campaign
(1974 78%, 1997 76%, 2010 60%). Most also decided before the referendum campaigns
(1975 72%, 1997 46%, 2011 59%). It is interesting to note that a majority of Scottish
28
devolution referendum voters reported that they decided their vote during the campaign. This
is at odds with claims that the vote for a Scottish Parliament reflected the ‘settled will’. It is
also interesting to note that almost as many voters in 2010 decided their vote during the
campaign as did in the Alternative Vote referendum. In fact, given that the AV referendum
was sprung upon voters, at least in comparison with the EEC and devolution referendums
which had been long foreshadowed, that most of them were able to decide before the
campaign suggests that they were more knowledgeable about the issue than we may have
given them credit for.8
Of course, there is a problem in making clear-cut distinctions between campaign and
non-campaign periods. The reality of politics is that campaigning never stops – rather it
waxes in the run-up to polling day and wanes in the interval between elections. Even so, it is
reasonable to assume that the closer to the polling day people decide how to vote the more
likely that vote has been influenced by campaigning – whether that be reading campaign
literature, being canvassed, or watching televised debates. The evidence from Table 7 is of a
higher proportion of late deciders in the referendums as compared with the general elections.
This suggests that the referendum campaigns, at least in 1975 and 1997, were somewhat more
important than in the general elections.
---Table 7 about here---
In Tables 3, 4, and 5, treating the timing of the vote decision as an independent variable only
appeared to significantly affect vote in 2010/11. Those who decided their vote during the
8 A clear majority of the pre-referendum wave AVRS respondents could accurately identify the Conservative
(64.6%) and Liberal Democrat (70.7) parties’ positions on the Alternative Vote, as they could for these parties’
leaders – 71.4% identified David Cameron as being against a change to the electoral system, while 75.8%
correctly identified Nick Clegg as being in favour. Respondents were much less knowledgeable about the
position of the Labour party and its leader, Ed Milliband. Responses were evenly split between Labour being
pro-AV (23.4%), against (16.5%) and divided (30.6%). They were a little better informed about the leader’s
position. Most said he was pro-AV (46.2%) but they were not the majority. Respondents were even less
knowledgeable about the smaller parties’ positions and those of their leaders.
29
2010 general election campaign (-0.57) were less likely to vote Labour than those who
decided earlier. Voters who decided during the AV referendum campaign were significantly
more likely to vote ‘yes’ (0.64) than earlier deciders. Figure 4(a) compares the effect of
deciding during the campaign between 2010 and 2011. The greatest change in vote
probability is found with general election voters. Their probability of voting Labour fell by
44 points if they decided during the general election campaign while it only increased by 16
points if they decided during the referendum campaign. Contrary to expectations, that general
election campaign appears to have had a greater effect on voters than the following year’s
referendum campaign.
---Figure 4(a) about here---
The lack of statistically significant relationships between the timing of the vote decision and
party vote in the 1974 and 1997 general election vote should come as no surprise. Late
deciders are unlikely to behave differently from early deciders, if, as in these general
elections, a Labour victory was a foregone conclusion. The lack of any statistically
significant relationship with referendum vote, however, does suggest that these referendum
campaigns were no more important than preceding general election ones. That may be,
however, because of familiarity with the referendum issues and the low intensity of these
campaigns. Butler and Kitzinger (1996: 273) claim the ‘yes’ vote in 1975 ‘was patently a
foregone conclusion’. The balance of messages in the 1975 and 1997 campaigns reflected
this consensus and was very much in the ‘yes’ camp’s favour, hence the positive coefficients.
Campaigning was muted in the 1975 (Butler and Kitzinger 1996) and 1997 referendums
(Denver et al. 2000: 111-4,116-9), in part because many felt the outcome was predictable but
also because local party activists were difficult to mobilise for a non-partisan cause. Those
30
who decided later, during the campaign, were as likely, therefore, to vote ‘yes’ as early
deciders because that was the dominant message and the referendum campaigns did little to
sway them.
Newspaper readership is another of the few available comparative measures of
campaign effects in UK general elections and referendums. Most newspapers took a position
on which party they wanted to win in the general elections and whether they supported the
referendum propositions. In general elections, historically, it has been the Conservatives who
commanded the most support. In referendums, newspapers tend to follow the line of their
party. In the 1975 referendum this meant that all newspapers, excluding the Morning Star,
were in favour of EEC membership (Butler and Kitzinger 1996: 214-45). In the Scottish
devolution referendums, Conservative newspapers such as the Scottish Daily Mail urged their
readers to vote ‘no’. Labour newspapers such as the Daily Record and the Sun strongly
backed devolution (Denver et al. 2000: 86). In 2011, the Mirror and Guardian newspapers
followed the Labour leadership in backing AV, while the Liberal Democrat leaning
Independent exhorted its readers in a front page spread on the day of the poll to ‘Just Say
Yes’. In Table 4, newspaper readership appears to have no significant effects on either 1974
general election or 1975 referendum vote. This is hardly surprising given the overwhelming
newspaper consensus for continued membership. In Table 5, newspaper readership had no
significant effects on general election vote but readership of the Mail (-3.06) and the
Guardian (-4.35) newspapers had strong and statistically significant effects on referendum
vote. Readers of these newspapers were less likely to vote in favour of devolution than those
of other newspapers. The Mail result is as expected. It was critical of the devolution policy
and its readers seemed to agree. That Guardian readers were less likely to support devolution
may have reflected the Englishness of this newspaper. Unlike other national newspapers,
there is no Scottish edition. As an English newspaper it would have carried relatively little
31
devolution news and largely held a neutral position on the issue. Scots choosing to read a
newspaper that carried predominantly English news would have been unlikely to be
supporters of devolution. In Table 5, readership of the Mirror (0.52) significantly increased
the likelihood of voting Labour in the general election, and readership of the Mail (-0.94) and
the Sun (-1.46) significantly reduced the likelihood of voting in favour of AV. There is,
however, no discernible pattern of difference between newspaper readership effects between
these general elections and referendums.
There is one other comparative measure of campaign effects, which is only available
to us in the 2010/11 datasets – exposure to canvassing. In Table 5, we see that being
canvassed by the Labour (1.72) increases the likelihood of voting Labour and decreases it if
canvassed by the Liberal Democrats (-0.69). Being canvassed by the ‘yes’ campaign
appeared to reduce the likelihood of voting ‘yes’ to AV (-0.54). In Figure 4(b), it is evident
that being canvassed by Labour and the ‘yes’ campaign had similar effects on vote
probability but being canvassed by the Liberal Democrats, clearly, had a greater effect. These
differences suggest that context is less important than which party or group is doing the
canvassing.
---Figure 4(b) about here---
The over-arching hypothesis for this paper is that predispositions such as party identification,
which have been labelled as long-term influences on vote, are less important to referendum
than general election voting. Conversely, issue preferences, government evaluations and
campaign effects (short-term influences), which exert their influence on vote decision closer
in time to polling day have more effect in referendums than general elections. In Table 8, the
model fit statistics (2 log likelihood, pseudo R-squared and percentage correct predictions) of
32
the long-term and short-term models of general election and referendum voting are shown.
For all the long-term models, the log likelihoods are smaller for the general election than for
the referendum models, indicating a better model fit. Amongst the short-term models, the log
likelihoods are not necessarily a better fit for the general elections than for the referendums.
Indeed, for the 1997 and 2011 referendums the log likelihoods are smaller. The improvement
in model fit between the long-term and short-term models is greater for the 1997 (61% v.
48%) and 2011 (91% v. 78%) referendums and almost equal to that for the 1974 general
election (48%) in the 1975 referendum (46%). The improvement in the pseudo R-squared and
the percentage correct predicted are all greater between the referendum long-term and short-
term models than for the general election models. They increase with each referendum, with
the 2011 referendum exhibiting the greatest difference. While the evidence is not definitive, it
seems to suggest that accounting for short-term influences adds greater explanatory power to
referendum than general election models, making them almost as good, if not a better fit than
such models.
Conclusions
At the beginning of this paper, I reasoned that referendum voting is distinctive from that in
general elections. This premise is supported on the effect of party identification, although not
necessarily in the way hypothesised, on issue preferences and on the relative impact of long-
term and short-term influences. The jury is out on government evaluations and campaign
effects, due to the equivocal findings on government evaluations and the paucity of
comparable measures of campaign effects across the electoral contexts.
The findings on party identification were somewhat contradictory. The evidence from
Table 6 suggests that across all voters the effects of party identification are weaker in
33
referendums than in general elections. In Tables 3, 4, and 5, however, party identification
appeared to exert a stronger influence on referendum voting amongst for identifiers with
smaller parties (e.g., the SNP in 1997 and the Liberal Democrats in 2011) with clearly
defined positions on the referendum issue and for identifiers with parties with distinctive
positions (e.g., the Conservatives in 1997 and 2011). For the larger parties (e.g., the
Conservatives in 1975 and Labour in 1997 and 2011), however, the influence of party
identification in referendums was notably weakened. The two sets of results can be
reconciled, nonetheless, if we take account of the fact that there are many more identifiers
with the Labour and Conservatives parties, than with the Liberal Democrats and the SNP.
Party identification effects are bound to appear stronger on general election than referendum
voting in aggregate. What the individual level analysis of Tables 3, 4, and 5 reveals is that the
effects of party identification in referendums vary according to the ability of parties to unify
around a clear and distinctive position.
These findings echo some of the sentiments found in the referendum literature that if
only parties could get their act together they would have more influence over referendum
outcomes (de Vreese 2006: 595; Neijens and van Praag 2006). This ignores, however, a
central truth about referendums. Many of them are called precisely because governing parties
are divided amongst themselves. They delegate the decision to the electorate in the hope of
finding a consensus that they cannot produce. But this seems a tall order. If professional
politicians cannot reach agreement within the confines of party discipline what hope for the
unsuspecting electorate. If long-established political parties cannot enforce discipline, we
should not be surprised if large numbers of party supporters, whose loyalties to their parties
are more tenuous than elected politicians, ignore the party’s wishes and go their own way.
The findings on the effects of party identification also suggest that Hypothesis 1A
needs to be revised and extended. A distinction perhaps should be drawn between identifiers
34
of larger mainstream parties and smaller more ideological parties. A reformulated Hypothesis
1A might read as follows:
Party identification is a weaker influence on referendum than general
election voting amongst identifiers with mainstream parties, especially if
these parties are divided. Party identification may be a stronger influence
on referendum than general election voting amongst identifiers with
smaller parties with clear unambiguous positions on the referendum issue.
The centrality of issue preferences in referendum election voting seems so self-evident that
for them not to have a greater effect on referendum than general election voting would be
unthinkable. The analysis provided patent evidence of the importance of issue preferences to
referendum voting. In the light of this evidence it might be tempting to downgrade partisan
attachments in referendum vote models and seek to explain referendum outcomes almost
entirely according to voters’ positions on the issues. This would ignore, however, the
important role that parties play in referendums. Voters may not loyally follow the party line
but they undoubtedly react to the way in which parties frame referendum issues and their
perceptions of the parties’ motivations in adopting the positions that they do. After all, parties
are the prism through which referendum issues are refracted. Marsh (2007:80-1) hits the nail
on the head when he writes,
future research on referendums should move beyond asking whether people use
party or issue cues to explore the nature of issue-related cues in much more
detail, asking how the referendum question itself was framed by the different sides
in the campaign and how far voters responded to particular frames. This implies a
closer link between studies of the content of campaigns and studies of voter
35
choice. To understand why people vote as they do ... it is vital to understand what
people think they are voting about.
Some referendum vote surveys include questions relating to the understanding of the
issue at stake but this is by no means routine. The Scottish devolution referendum
studies9 and the Alternative Vote Referendum Study are exceptional in that they contain
questions relating to knowledge of the referendum proposals, knowledge of party
positions on the issue at stake and, in the case of the 1997 referendum, predictions about
the likely effects of devolution. In many referendums, however, we simply do not know
how the issue is perceived and by what criteria the proposition is judged.
Furthermore, issue effects may not be what they appear. The inclusion of issue
preferences as independent variables in vote models assumes a direct effect on vote decisions.
Hobolt (2009) makes a persuasive argument that the size of issue effects is dependent upon
the intensity of campaigns – the more intense they are the more voters learn about the
referendum issue and the better able they become to relate their predispositions with their
vote preferences. If so, issue preferences may, in part, be an indicator of campaign effects.
This suggests the intriguing possibility that if campaign effects are properly accounted for in
vote models then issue effects might be reduced.
The comparison of the model fits in Table 8 certainly suggested that more of the work
in explaining referendum rather than general election vote was being done by short-term
influences. Unfortunately, there were limited opportunities to compare the effects of
campaigns with the panel data and the analysis was suggestive that their effects are no more
important in referendums. In recent studies of campaign effects in referendum much has been
9 There were three vote surveys of the Scottish devolution referendum – BEPS 2, the Crest team survey and
Denver’s survey.
36
made of their greater importance but it remains to be substantiated. It would be helpful if
referendum vote studies contained more items relating to media and campaign exposure. In
addition to newspaper readership, it would be useful to have measures of the impact of other
media. It is widely reported that television is the most important source of information on
politics but there tend to be few questions about television news and political programming.
Measures of campaign exposure are to be found on referendum vote surveys, such as
exposure to door-to-door canvassing, campaign advertising, leaflets, and broadcasts, receipt
and awareness of government information leaflet, but, at best, this is patchy and frequently
only some of these items are included.
Much progress has been made in recent years in the study of referendum voting and,
in particular, in the effect of campaigns but vote models remain indebted to those applied to
general elections. This paper has illustrated how these models only take us only so far in
understanding voters’ motivations. We need to push the boundaries and explore more fully
the interplay between parties and the framing of referendum issues and to devise more
comprehensive batteries of individual level campaign effects to adequately test their impact
on vote.
37
References
Aitkin, Don (1978), ‘Australia’, in David Butler and Austin Ranney (eds.), Referendums: A
Comparative Study of Practice and Theory. Washington, D.C: American Enterprise
Institute for Public Policy Research.
Borges, Walter and Harold Clarke (2008) Cues in context: Analyzing the heuristics of
referendum voting with an internet survey experiment, JEPOP 18 (4), pp. 433-48.
Brown, Alice, David McCrone, and Lindsay Paterson (1998), Politics and Society in Scotland
(2nd edn.). Basingstoke: Macmillan.
Brown, Alice, David McCrone, Lindsay Paterson, and Paula Surridge (1999), The Scottish
Electorate: The 1997 General Election and Beyond. Basingstoke: Macmillan.
Butler, David, and Austin Ranney (eds.) (1978), Referendums: A Comparative Study of
Practice and Theory. Washington, D.C: American Enterprise Institute for Public Policy
Research.
Butler, David, and Austin Ranney (eds.) (1994), Referendums Around the World: The
Growing Use of Direct Democracy. Basingstoke: Macmillan.
Butler, David, and Uwe Kitzinger (1996), The 1975 Referendum (2nd edn.). Basingstoke:
Macmillan.
Clarke, Harold, and Allan Kornberg (1994), ‘The politics and economics of constitutional
choice: Voting in Canada’s 1992 national referendum’, Journal of Politics 56 (4): 940-
62.
Clarke, Harold, Allan Kornberg and Marianne Stewart (2004), ‘Referendum voting as
political choice: The case of Quebec’, British Journal of Political Science 34, pp. 345-
55.
Cronin, Thomas (1999), Direct Democracy: The Politics of Initiative, Referendum, and
Recall. Cambridge, Mass.: Harvard.
de Vreese, Claes (2006), ‘Political parties in dire straits? Consequences of national
referendums for political parties’ Party Politics 12 (5), pp. 581-98.
de Vreese, Claes (2007), ‘Context, elites, media and public opinion in referendums: When
campaigns really matter’, in Claes de Vreese (ed.), The Dynamics of Referendum
Campaigns: An International Perspective. Basingstoke: Palgrave Macmillan.
de Vreese, Claes and Holli Semetko (2004) ‘News matters: Influences on the vote in the
Danish 2000 Euro referendum campaign’, European Journal of Political Research 43,
pp. 699-722.
38
Denver, David, James Mitchell, Charles Pattie, and Hugh Bochel (2000), Scotland Decides:
The Devolution Issue and the Scottish Referendum. London: Frank Cass.
Dunleavy, Patrick (1990), ‘Mass political behaviour: Is there more to learn?’, Political
Studies 38 (3): 453-69.
Rudiger Schmitt-Beck and David Farrell (2002), ‘Studying political campaigns and their
effects’, in David Farrell and Rudiger Schmitt-Beck (eds.) Do Political Campaigns
Matter? Campaign Effects in Elections and Referendums London: Routledge.
Franklin, Mark, Michael Marsh, and Lauren McLaren (1994), ‘Uncorking the bottle: Popular
opposition to European unification in the wake of Maastricht’, Journal of Common
Market Studies 32 (4): 455-72.
Gallagher, Michael, and Pier Uleri (eds.) (1996), The Referendum Experience in Europe.
Basingstoke: Macmillan.
John Garry, Michael Marsh and Richard Sinnott (2005) “ ‘Second-order’ versus ‘issue-
voting’ effects in EU referendums: Evidence from the Irish Nice Treaty referendums’,
European Union Politics 6: 201-21.
Hobolt, Sara Binzer (2005) When Europe matters: The impact of political information on
voting behaviour in EU referendums, JEPOP 15 (1), pp. 85-109.
Sara Binzer Hobolt (2006), ‘Direct democracy and European integration’, Journal of
European Public Policy 13 (1), pp. 153-66.
Hobolt, Sara Binzer (2007), ‘Taking cues on Europe? Voter competence and party
endorsements in referendums on European integration’, European Journal of Political
Research 46, pp. 151-82.
Hobolt, Sara Binzer (2009) Europe in Question: Referendums on European Integration.
Oxford: Oxford University Press.
Holbrook, Thomas (1996), Do Campaigns Matter? Thousand Oaks, California: Sage
Publications.
Hug, Simon (2002), Voices of Europe: Citizens, Referendums, and European Integration.
Lanham, Maryland: Rowman and Littlefield.
Hug, Simon, and Pascal Sciarini (2000), ‘Referendums on European integration: Do
institutions matter in the voter’s decision’, Comparative Political Studies 33 (1): 3-36.
Jenssen, Anders Todal, and Ola Listhaug (1999), ‘Voters’ decisions in the Nordic EU
referendums of 1994’. Paper presented at the Conference on Maturity and Malaise? The
Growing Use of Referendums in Liberal-Democratic Societies. Centre for the Study of
Democracy, Queens University, Kingston, Canada, May 14-16.
39
Kobach, Kris (1993), The Referendum: Direct Democracy in Switzerland. Aldershot:
Dartmouth.
King, Anthony (1977), Britain Says Yes. Washington, D.C.: American Enterprise Institute for
Public Policy Research.
King, Anthony (ed.) (2001), British Political Opinion 1937-2000: The Gallup Polls. London:
Politico’s Publishing.
Kobach, Kris (1993), The Referendum: Direct Democracy in Switzerland. Aldershot:
Dartmouth.
Kobach, Kris (1994), ‘Switzerland’, in David Butler and Austin Ranney (eds.), Referendums
Around the World: The Growing Use of Direct Democracy. Basingstoke: Macmillan.
Holbrook, Thomas (1996), Do Campaigns Matter? Thousand Oaks, California: Sage
Publications.
LeDuc, Lawrence (2005), ‘Saving the pound of voting for Europe? Expectations for
Referendums on the constitution and the Euro’, JEPOP 15 (2), pp. 169-96.
LeDuc, Lawrence, and Jon Pammett (1995), ‘Referendum voting: Attitudes and behaviour in
the 1992 constitutional referendum’, Canadian Journal of Political Science 28 (1): 3-
34.
LeDuc, Lawrence (2002), ‘Referendums and elections: How do campaigns differ?’ in David
Farrell and Rudiger Schmitt-Beck (eds.) Do Political Campaigns Matter? Campaign
Effects in Elections and Referendums London: Routledge.
Lupia, Arthur, and Matthew McCubbins (1998), The Democratic Dilemma: Can Citizens
Learn What They Need to Know? Cambridge: Cambridge University Press.
Maddens, Bart (1996), ‘Directional theory of issue voting: The case of the 1991
parliamentary elections in Flanders’, Electoral Studies 15 (1): 53-70.
Magleby, David (1984), Direct Legislation: Voting on Ballot Propositions in the United
States. Baltimore, Maryland: The John Hopkins University Press.
Michael Marsh (2007), ‘Changing what people think or changing what they think about?’ in
Claes de Vreese The Dynamics of Referendum Campaigns: An International
Perspective. Basingstoke: Palgrave Macmillan.
Midtbo, Tor, and Kjell Hines (1998), ‘The referendum-election nexus: An aggregate analysis
of Norwegian voting behaviour’, Electoral Studies 17 (1): 77-94.
Miller, Warren, and J. Merrill Shanks (1996), The New American Voter. Cambridge,
Massachusetts: Harvard University Press.
Neijens, Peter, Mark Minkman, and Jeroen Slot (1998) ‘Opinion formation in referendum
campaigns: A study of the Amsterdam referendums’, Acta Politica 33 (3): 300-16.
40
Peter Neijens and Philip van Praag (2006), ‘The dynamics of opinion formation in local
popular referendums: Why the Dutch always say no’, International Journal of Public
Opinion Research 18 (4): 445-62.
Pattie, Charles, David Denver, James Mitchell, and Hugh Bochel (1999a), ‘Partisanship,
national identity and constitutional preferences: An exploration of voting in the Scottish
devolution referendum of 1997’, Electoral Studies 18 (3): 305-22.
Pierce, Roy, Henry Valen, and Ola Listhaug (1983), ‘Referendum voting behaviour: The
Norwegian and British referenda on membership in the European Community’,
American Journal of Political Research 27 (1): 43-63.
Popkin, Samuel (1991), The Reasoning Voter: Communication and Persuasion in
Presidential Campaigns. Chicago: University of Chicago Press.
Reif, Karlheinz, and Herman Schmitt (1980), ‘Nine second-order national elections – a
conceptual framework for analysis of European election results’, European Journal of
Political Research 8 (1): 3-44.
Sartori, Giovanni (1987), The Theory of Democracy Revisited. Chatham, N.J.: Chatham
House Publishers.
Schmitt-Beck, Rudiger and David Farrell (2002), ‘Studying political campaigns and their
effects’, in David Farrell and Rudiger Schmitt-Beck (eds.) Do Political Campaigns
Matter? Campaign Effects in Elections and Referendums London: Routledge.
Schneider, Gerald and Patricia Weitsman (1996), ‘The punishment trap: Integration
referendums as popularity contests’, Comparative Political Studies 28 (4), pp. 582-607.
Schuk, Andreas and Claes de Vreese (2008), ‘The Dutch No to the EU Constitution:
Assessing the role of EU scepticism and the campaign’, JEPOP 18 (1), pp. 101-28.
Sciarini, Pascal, and Ola Listhaug (1997), ‘Single cases or a unique pair? The Swiss and
Norwegian ‘No’ to Europe’, Journal of Common Market Studies 35 (3): 407-38.
Sinnott, Richard (1995), Irish Voters Decide: Voting Behaviour in Elections and
Referendums since 1918. Manchester: Manchester University Press.
Siune, Karen, and Palle Svensson (1993), ‘The Danes and the Maastricht Treaty: The Danish
EC referendum of June 1992’, Electoral Studies 12 (2): 99-111.
Szczerbiak, Aleks and Paul Taggart (2004), ‘The politics of European referendum outcomes
and turnout: Two models’, West European Politics 27 (4), pp. 557-583.
Taylor, Bridget, and Katarina Thomson (eds.) (1999), Scotland and Wales: Nations Again?
Cardiff: University of Wales Press.
41
Tonsgaard, Ole (1992) ‘A theoretical model of referendum behaviour’, in Peter Gundelach
and Karen Siune (eds.) From Voters to Participants, Aarhus: Politica.
Trechsel, Alexander, and Hanspeter Kriesi (1996), ‘Switzerland: The referendum and
initiative as a centrepiece of the political system’, in Michael Gallagher and Pier Uleri
(eds.), The Referendum Experience in Europe. Basingstoke: Macmillan.
Uleri, Pier (1996), ‘Introduction’, in Michael Gallagher and Pier Uleri (eds.), The
Referendum Experience in Europe. Basingstoke: Macmillan.
Wright, Vincent (1978), ‘France’, in David Butler and Austin Ranney (eds.), Referendums: A
Comparative Study of Practice and Theory. Washington, D.C.: American Enterprise
Institute for Public Policy Research.
42
Tables and Figures
Table 1: Official and reported referendum results
1975
%
official
1975
%
reported
1997
%
official
1997
%
reported
2011
%
official
2011
%
reported
Yes vote 64.5 72.4 74.3 79.5 32.1 38.9
Source: Butler and Butler (2000), the Electoral Commission, BES 1974-9, BEPS 2 1997-8, BES CIPS 2010, and
2011 BES AVRS
Table 2: Official and reported general election results
Party 1974
%
official
1974
%
reported
1997
%
official
1997
%
reported
2010
%
official
2010
%
reported
Conservative 35.8 36.1 17.5 14.9 36.1 35.6
Labour 39.2 40.1 45.6 52.7 29.0 26.0
Liberal/Liberal
Democrat
18.3 17.6 13.0 7.9 23.0 27.1
SNP -- -- 22.1 20.7 -- --
Other 6.7 6.2 1.9 3.4 11.9 11.4
Source: Butler and Butler (2000), the Electoral Commission, BES 1974-9, BEPS 2 1997-8, BES CIPS 2010, and
2011 BES AVRS
43
Table 3. Long-term and short-term influences on general election and referendum vote in
1974/5
Long-term model Short-term model
Variables General election Referendum General election Referendum
Constant -0.77 (0.53) 0.31 (0.45) -5.49 (1.30)*** -0.80 (0.92)
Age 0.01 (0.01) 0.02 (0.01)*** 0.00 (0.01) 0.02 (0.01)*
Female -0.10 (0.24) -0.10 (0.18) -0.35 (0.35) -0.23 (0.26)
Class
(base=salariat)
Routine non-manual 0.17 (0.35) -0.30 (0.28) 0.46 (0.51) -0.43 (0.38)
Foremen and
technicians
0.81 (0.36)* -0.57 (0.29)* 0.75 (0.53) -0.36 (0.40)
Semi-skilled and
unskilled workers
0.50 (0.38) -0.72 (0.30)* 0.62 (0.55) -0.49 (0.42)
Self-employed -1.00 (0.48)* 0.16 (0.34) -0.51 (0.70) 0.30 (0.47)
Left school at 17+ -0.81 (0.38)* 1.02 (0.32)** -0.90 (0.57) 1.02 (0.43)*
Homeowner -0.24 (0.21) 0.12 (0.16) -0.44 (0.30) 0.12 (0.22)
Trade union
member
0.65 (0.23)** -0.09 (0.18) 0.47 (0.32) -0.32 (0.25)
Party identification
(base=none)
Conservative -2.61 (0.40)*** 0.95 (0.31)** -1.55 (0.57)** 0.08 (0.48)
Labour 2.53 (0.30)*** -0.53 (0.28) 2.06 (0.48)*** -0.18 (0.46)
Liberal -0.90 (0.35)* -0.08 (0.32) -0.65 (0.51) -0.07 (0.50)
Other -1.25 (0.64)* -0.99 (0.52) -1.35 (0.98) -0.35 (0.85)
Issues
Nationalisation 0.39 (0.18)* -0.18 (0.14)
Social services 0.22 (0.15) -0.14 (0.11)
Wage agreements 0.92 (0.35)** 0.71 (0.26)**
Trade union power -0.35 (0.39) 0.23 (0.29)
Issues scale 0.41 (0.65) -0.23 (0.49)
EEC -0.07 (0.21) 1.50 (0.17)***
Government
evaluations
0.97 (0.26)*** -0.24 (0.19)
Campaign
influences
Decided vote during
campaign
-0.04 (0.32) 0.29 (0.24)
Newspaper
readership
Express -0.37 (0.48) 0.17 (0.34)
Guardian -0.32 (0.93) 2.03 (1.19)
Mail -0.52 (0.61) 0.25 (0.44)
44
Mirror 0.38 (0.38) 0.52 (0.29)
Sun 0.25 (0.47) 0.27 (0.36)
Telegraph -0.70 (0.86) 0.96 (0.55)
Times 1.07 (1.33) 0.10 (1.37)
Summary statistics
-2 log likelihood 685.872 1068.046 359.842 581.673
Pseudo R2
(Nagelkerke)
0.71 0.20 0.78 0.40
% correct 88.9 74.7 90.2 79.3 Total N=1740 Missing=595/703/1009/1067
*p < 0.05, **p < 0.01, ***p < 0.001
Standard errors in brackets
Source: BES 1974-1979 (unweighted data).
45
Table 4. Long-term and short-term influences on general election and referendum vote in
1997
Long-term model Short-term model
Variables General election Referendum General election Referendum
Constant 1.32 (1.14) -0.07 (0.93) -0.96 (2.60) -1.53 (2.45)
Age -0.02 (0.01) 0.00 (0.01) -0.03 (0.02) 0.04 (0.02)
Female -0.35 (0.41) 0.15 (0.33) -0.31 (0.58) -0.56 (0.66)
Religion
(base=none)
Catholic -0.22 (0.63) -0.01 (0.55) 1.14 (1.03) 2.10 (1.22)
Presbyterian -0.35 (0.49) -0.73 (0.41) 0.42 (0.79) 0.77 (0.73)
Other -0.82 (0.63) 0.35 (0.52) 0.28 (0.94) 1.81 (1.08)
Class
(base=salariat)
Routine non-manual -0.66 (0.57) 0.44 (0.44) -1.33 (0.84) -0.57 (0.93)
Foremen and
technicians
0.63 (0.64) 0.63 (0.61) 1.54 (1.05) 0.43 (1.55)
Semi-skilled and
unskilled workers
-0.23 (0.52) -0.15 (0.44) 0.63 (0.78) -0.74 (1.03)
Self-employed -1.12 (1.07) -0.75 (0.79) -2.20 (2.38) -0.90 (0.77)
Left school at 17+ -0.86 (0.43)* 0.03 (0.38) -0.83 (0.71) 1.28 (0.76)
Homeowner -0.18 (0.43) -0.13 (0.38) -0.62 (0.63) -1.14 (0.77)
Trade union
member
0.01 (0.44) -0.01 (0.36) 0.28 (0.64) 0.45 (0.74)
Party identification
(base=other/none)
Conservative -1.72 (0.83)* -1.34 (0.54)* -0.66 (1.21) -1.54 (1.14)
Labour 4.48 (0.69)*** 1.66 (0.57)** 4.98 (1.10)*** 1.39 (1.14)
Liberal -1.28 (0.80) 0.59 (0.57) -1.22 (1.10) 0.09 (1.18)
SNP -0.35 (0.70) 3.60 (1.15)** -1.63 (1.15) 2.60 (1.58)
National identity -1.07 (0.84) 1.38 (0.59)* -0.45 (1.61) 0.60 (1.40)
Issues
Devolution -2.70 (1.25)* 5.65 (1.20)***
Tax scale 0.35 (0.41) -0.03 (0.45)
NHS spending 0.58 (1.20) 0.10 (1.10)
Education scale 0.17 (0.42) 1.13 (0.48)
Job/prices scale -0.44 (0.36) 0.30 (0.39)
Government
evaluations
-1.05 (0.57) 0.40 (0.31)
Campaign
influences
Decided vote during
campaign
1.24 (0.65) 1.11 (0.63)
46
Newspaper
readership
(base=other)
Express -0.54 (1.38) -3.72 (1.30)
Mail 1.13 (1.53) -3.06 (1.46)*
Mirror 1.84 (1.07) -1.67 (1.15)
Star -17.16
(16593.53)
-2.23 (3.82)
Sun 2.45 (1.26) -2.93 (1.47)
Telegraph 1.36 (3.29) 0.34 (2.45)
Financial Times -18.05
(41483.50)
21.56
(41483.50)
Guardian -0.81 (3.07) -4.35 (2.08)*
Independent -16.44
(23944.60)
20.34
(23116.33)
Times 0.08 (2.45) -4.84 (3.18)
Scot 1.53 (1.32) -2.07 (1.56)
Herald 0.89 (1.33) -2.74 (1.42)
Aberdeen -0.42 (1.28) -2.27 (1.27)
Summary statistics
-2 log likelihood 235.91 311.61 122.26 122.06
Pseudo R2
(Nagelkerke)
0.78 0.44 0.84 0.71
% correct 92.1 86.7 94.6 91.7 Total N=493 Missing=36/36/201/204
*p < 0.05, **p < 0.01, ***p < 0.001
Standard errors in brackets
Source: BEPS 2 1997-8 (weighted data).
47
Table 5. Long-term and short-term influences on general election and referendum vote in
2010/2011
Long-term model Short-term model
Variables General election Referendum General election Referendum
Constant -1.21 (0.18)*** 0.56 (0.12)*** -1.46 (0.52)** -1.86 (0.67)**
Age -0.01 (0.00)** -0.01 (0.00)*** -0.00 (0.00) -0.00 (0.01)
Female 0.24 (0.07)*** -0.36 (0.05)*** 0.36 (0.15)* -0.36 (0.18)*
Class
(base=salariat)
Routine non-manual 0.11 (0.08) -0.28 (0.05)*** 0.05 (0.18) -0.15 (0.20)
Foremen and
technicians
0.05 (0.10) -0.35 (0.11)** -0.01 (0.24) 0.04 (0.38)
Semi-skilled and
unskilled workers
0.26 (0.11)* -0.47 (0.06)*** 0.28 (0.27) 0.18 (0.24)
Self-employed 0.07 (0.18) -- 0.01 (0.42) --
Left school at 17+ -0.06 (0.07) 0.48 (0.05)*** -0.00 (0.17) -0.08 (0.20)
Homeowner -0.11 (0.08) -0.10 (0.05) -0.28 (0.18) -0.10 (0.19)
Trade union
member
0.30 (0.07)*** -0.00 (0.05) 0.49 (0.15)** -0.17 (0.18)
Party identification
(base=none)
Conservative -2.69 (0.17)*** -1.85 (0.08)*** -2.14 (0.33)*** -1.02 (0.31)**
Labour 2.27 (0.08)*** 0.10 (0.06) 1.32 (0.21)*** 0.57 (0.29)*
Liberal Democrat -1.15 (0.13)*** 1.59 (0.09)*** -1.78 (0.32)*** 2.04 (0.36)***
Other -0.69 (0.12)*** 0.14 (0.08) -0.62 (0.28)* 0.61 (0.32)
Issues
Tax and spend 0.78 (0.37)* 0.06 (0.41)
Afghanistan 0.12 (0.11) -0.10 (0.12)
Crime -0.45 (0.29) 0.79 (0.34)*
Electoral reform -0.29 (0.09)** 2.00 (0.12)***
Government
evaluations
2.26 (0.20)*** -0.70 (0.17)***
Campaign
influences
Decided vote during
campaign
-0.57 (0.14)*** 0.64 (0.16)***
Newspaper
readership
(base=other)
Express 0.09 (0.35) -0.55 (0.52)
Mail -0.03 (0.24) -0.94 (0.41)*
Mirror 0.52 (0.26)* -0.32 (0.42)
48
Star 1.14 (0.62) -0.99 (0.66)
Sun 0.04 (0.27) -1.46 (0.43)**
Telegraph -0.40 (0.35) -0.53 (0.45)
Financial Times 0.36 (0.85) -1.15 (1.32)
Guardian 0.22 (0.24) 0.37 (0.45)
Independent 0.18 (0.37) 0.92 (0.50)
Times -0.44 (0.27) -0.06 (0.43)
Canvassed by the
Conservatives
-0.11 (0.17) -0.04 (0.18)
Canvassed by
Labour
1.72 (0.18)*** -0.13 (0.19)
Canvassed by the
Liberal Democrats
-0.69 (0.16)*** 0.18 (0.18)
Canvassed by yes
campaign
-0.54 (0.19)**
Canvassed by no
campaign
-0.02 (0.17)
Summary statistics
-2 log likelihood 6943.12 12759.57 1511.71 1135.20
Pseudo R2
(Nagelkerke)
0.54 0.29 0.71 0.63
% correct 85.2 70.3 90.7 85.4 Total general election N=12180 Missing general election=2228/8916
Total referendum N=16881 Missing referendum=4127/14750
*p < 0.05, **p < 0.01, ***p < 0.001
Standard errors in brackets
Source: BES CIPS 2010, BES AVRS 2011 (weighted data).
49
Table 6: Comparison of partisan voting in general elections and referendums
Party identification General election Referendum % difference
1974/5
Conservative 84.2 88.8 +4.6
Labour 86.9 58.9 -28.0
Liberal 67.8 74.1 6.3
Total difference -17.1
Cramer’s V 0.634 0.309
1997
Conservative 76.8 66.3 -10.5
Labour 94.1 92.4 -1.7
Liberal Democrat 85.5 77.5 -8.0
SNP 82.4 98.9 +16.5
Total difference -3.7
Cramer’s V 0.802 0.571
2010/11
Conservative 87.5 88.1 +0.6
Labour 68.5 45.9 -22.6
Liberal Democrat 82.5 79.9 -2.6
Total difference -24.6
Cramer’s V 0.601 0.418 Note: all chi
2 significant
Source: BES 1974-9, BEPS 2 1997-8, BES CIPS 2010, BES AVRS 2011.
50
Table 7. Reported timing of the vote decision
Election year
Percentage who decided vote
during general election
campaign
Percentage who decided vote
during referendum campaign
N
1974/5 21.8 27.8 1737/1545
1997 23.9 54.0 585
2010/11 40.3 41.2 12042/15184
Source: BES 1974-9, BEPS 2 1997-8, BES CIPS 2010, BES AVRS 2011.
Table 8. Comparison of model fit of long-term and short-term effects models
1974 GE
Log
likelihood
1975 Ref
Log
likelihood
1997 GE
Log
likelihood
1997 Ref
Log
likelihood
2010 GE
Log
likelihood
2011 Ref
Log
likelihood
A. Intercept +
demographics
+ PID (long-
term factors)
685.9 1068.0 235.91 311.61 6943.12 12759.57
B. Intercept +
demographics
+ PID + issues
+ government
evaluations +
campaign
effects (short-
term factors)
359.8 581.7 122.26 122.06 1511.71 1135.20
A-B
(% change)
326.1
(48%)
486.3
(46%)
113.65
(48%)
189.55
(61%)
5431.41
(78%)
11624.37
(91%)
R2
change +7 (0.78-0.71)
+20 (0.40 – 0.20)
+6 (0.84-0.78)
+27 (0.71-0.44)
+17 (0.71-0.54)
+34 (0.63-0.29)
% correct
change
+1.3 (90.2-88.9)
+4.6 (79.3-74.7)
+2.5 (94.6-92.1)
+5 (91.7-86.7)
+5.5 (90.7-85.2)
+15.1 (85.4-70.3)
51
-0.41
-0.15
0.76
-0.05
-0.16
0.51
-0.1
0.15
-0.29
0.34
0.47
-0.33
0.03
0.35
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Con 1974/5
Con 1997 Lab 1997 SNP 1997 Con 2010/11
Lab 2010/11
LD 2010/11
E
f
f
e
c
t
s
o
n
v
o
t
e
p
r
o
b
a
b
i
l
i
t
y
Figure 1. Effects of party identification on vote probability
General election
Referendum
52
0
0.2
0.4
0.6
0.8
1
1.2
1 2 3 4 5 6
%
P
r
o
b
a
b
i
l
i
t
y
Issue categories
Figure 2(a). Vote probabilities by issue positions in 1974/5
General election
Referendum
53
0
0.2
0.4
0.6
0.8
1
1.2
0 0.25 0.5 0.75 1
%
p
r
o
b
a
b
i
l
i
t
y
Devolution
Figure 2(b). Vote probabilities by positions on devolution in 1997
General election
Referendum
54
0
0.2
0.4
0.6
0.8
1
1.2
Low Medium High
%
p
r
o
b
a
b
i
l
i
t
y
Issue positions (tax and spend, Afghanistan, crime, and electoral reform)
Figure 2(c). Vote probability by issue positions in 2010/11
General election
Referendum
55
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Negative Neutral Positive
%
p
r
o
b
a
b
i
l
i
t
y
Government evaluations
Figure 3. Vote probability by government evaluations in 2010/11
General election
Referendum
56
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Before campaign During campaign
%
p
r
o
b
a
b
i
l
i
t
y
Timing of the vote decision
Figure 4(a). Vote probability depending on the timing of the vote decision
in 2010/11
General election
Referendum
57
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Not canvassed Canvassed
%
p
r
o
b
a
b
i
l
i
t
y
Figure 4(b). Vote probability by exposure to canvassing in 2010/11
Canvassed by Labour (GE)
Canvassed by Liberal
Democrats (GE)
Canvassed by 'yes' campaign
(Ref)