Love and marriage in election times Inconsistency and types of floating voters

Full text

1
Love and marriage in election times
Inconsistency and types of floating voters
Stefaan Walgrave
Peter Van Aelst
stefaan.walgrave@ua.ac.be
Paper prepared for delivery at the General ECPR Conference
9-10 September 2005, Budapest
INTRODUCTION
In the heat of an election campaign pollsters often warn us that nothing is sure yet:
their poll results are only a snapshot and, more important, a large segment of the
electorate has not made its mind up yet. The ballot will be decided, pollsters
invariably say, by these undecided voters that tend to decide for which party to cast
their vote in the very last moments before entering the polling booth. Apart from
the fact that pollsters have every interest to state that the election game is not over
yet – if not, who would care to devote attention to their next prediction - the so-
called floating voters are still poorly understood. There are different kinds of
floating voters during an election campaign and the present approaches do not
really grasp these nuanced differences. The central claim of this paper is that some
voters who are usually considered to be completely volatile and floating are, in
reality, less floating; at the same time, a considerable segment of the electorate is
not considered as floating, yet they do float during the campaign. In other words:
we contend that there are different types of floating voters during the campaign.
The paper’s primary aim is to identify these types and to explain their behavior.
The paper’s secondary aim is empirical: testing the classic theories about electoral
volatility for the first time on Belgian campaign evidence, a country not really
washed over with loads of campaign research.
After reviewing the relevant literature and discussing our methods, we construct
types of floating voters based on inconsistency between subjective measures of
decidedness and objective measures of (intended) behavior. Quite some voters
maintain in the beginning of the campaign that they have definitely made their
mind up; yet they do change party preference during the campaign. Others claim at
the beginning of the campaign they do not know which party will get their ballot;
but throughout the campaign they stick with one and the same party. These
inconsistencies are no coincidence. They reflect voters’ uncertainty and, therefore,
can be used to construct different types of floating voters. Using the love and
marriage metaphor we coin four voter types: loyal voters, bachelors, adulterous
voters and silent admirers. In a second empirical section of the study we attempt to

2
account for these four types employing multivariate methods. Doing so, we will
test several theories about voter dealignment and electoral volatility.
To do all that, the study relies on a web-based panel survey in Belgium during the
2004 regional election campaign. The panel surveyed a considerable group of
Belgian (Flemish) voters in five separate waves in the run-up to the June 2004
elections. Although this is not a representative survey, its size, detail, and design are
well-suited for an explorative study.
THEORY, HYPOTHESES AND OPERATIONALIZATION
That voters change their party preference forms the core of democratic theory.
Only if people change preferences now and then can government be held
accountable. Consequently, explaining party loyalty, or electoral behavior in general,
has always been at the core of modern political science. Bye and large, scholars
agree that voters’ changeability has increased during the last three decades (Dalton
and Wattenberg 2000; Norris et al. 1999). Voter volatility comes in two guises. On
the one hand, voters change opinion between two elections. As elections are typically
years apart, people have ample time to make up their mind and to become used to
the idea to vote for another party. This slack change has often been explained by
deep-rooted trends of structural change with altering socialization patterns,
increased levels of education, and silent value shifts (Inglehart 1990). In short,
modernization would lead to a more critical citizen, scrutinizing the offer on the
political market skeptically, and by no means unconditionally loyal to a party. Party
dealignment is the consequence of modernization (Dalton, McAllister, and
Wattenberg 2000). Yet on the other hand, some voters change preference during the
campaign in the few weeks before the polls. Swift changing voters switch party
allegiance at some point in the election campaign, sometimes just hours before they
cast their ballot. Consequently, election campaigns became of more interest to
political scientists and the campaign literature boomed. Naturally, both trends, slack
change and swift change, are associated but analytically they are different things
(Lachat 2004). That people change opinion between elections does not make them
change opinion (again) during the campaign. Based on German and Swiss data,
Lachat demonstrated that the causal structure underlying both types of volatility is
partially different and that their long-term evolution is diverging. Well aware of the
fact that, empirically, both types of volatility are associated this paper’s focus is on
preference change during the campaign
Remarkably, while a lot is known about the floating voter and about changing
electoral loyalty in general, intra-campaign volatility has received relatively few
scholarly attention (McAllister 2002). Few studies have focused specifically on the
voters that shift during a campaign and even the basic facts – how many people
change preference during a typical campaign? - are not readily available (Blais

3
2004). Blais reviewed the existing panel survey evidence in five countries – the US,
UK, New Zealand, Canada, and The Netherlands – and concluded that between
8% (US) and 30% (New Zealand) of the voters switch allegiance during the final 30
days preceding the polls. In the countries under study, he notices a slight increase
in intra-campaign switches during the last decades. Note that Blais only takes voters
into account who indicated a vote intention in the campaign. Yet, it is unclear
whether this vote preference was preceded by a filter question asking people
whether they had made up their mind yet, which is the strategy we will follow in
this study. To our knowledge, the most comprehensive study of intra-campaign
switching has been undertaken by Granberg and Holmberg (1991) comparing long-
term time series in the US and Sweden based on evidence about 20 different
elections. The number of what they call ‘switchers’, people with a pre-campaign
preference that changed, slowly grew from 4% to 10% in Sweden (1956-1988) and
remained stable at 3% in the US (1952-1988). Finally, in his study on electoral
dealignment, Lachat (2004) compares intra-campaign volatility in 17 elections in
Germany and Switzerland. He concludes that intra-campaign volatility increased
starkly in Germany: from 17% (1970) to 33% (1990). For Switzerland, the story
was entirely different: intra-campaign mobility did not move at all but remained at
28% between 1979 and 1999.
In the electoral volatility literature, more attention has been devoted to the
supposed delaying of voting decisions. But, electoral volatility, changing preferences,
and deciding late, which can be shifting as well as confirming an earlier preference,
are two different things. A certain confusion has pervaded the literature in this
respect (Latimer 1987). Not all people who decide late actually change preference;
but all people who changed preference during the campaign are late deciders. Late-
deciding is a necessary condition for intra-campaign change. In this contribution we
focus on shifting preference and not on vote timing, although, both being
indicators of voters’ uncertainty they are most likely explained by roughly the same
determinants (Lachat 2004). What are the major findings in the research literature
about decision timing? Dalton and colleagues (2000) compiled comparative
evidence based on post-electoral recall data in twelve countries. They show that,
during the last decades, in all but one country there is neat tendency towards later
voting decisions: little by little people tended to decide later. McAllister found the
same in Australia, the US and the UK (McAllister 2002). Yet, Granberg and
Holmberg’s (1990) evidence does not completely support these find

4
Returning to the focus of this contribution, what does the research literature tell us
about the reasons for intra-campaign change? What variables can account for
people changing their preferences in the few weeks before an election? Analytically,
three different types of theories or rather types of variables might account for intra-
campaign volatility: socio-structural factors, attitudinal-behavioral factors, and
campaign context factors.
Socio-structural accounts
In the older electoral literature, the Michigan school ascertained that especially age
is a crucial predictor of electoral flexibility (Campbell et al. 1960; Converse 1969).
Older voters’ preferences have stabilized; they are much less responsive to new
events than the younger voting cohorts. Therefore, we expect age to be a predictor
for intra-campaign volatility. Modernization theorists, at the other hand, consider
especially education to be crucial. Rising educational levels enhanced political skills
and political interest (Inglehart 1990; Topf 1995). So we expect education to play a
role as well, although McAllister (2002), who probably made most headway with
dissecting late deciders (and supposedly intra-campaign changes), did not find any
net effect of education on late decision-making in the US, UK, and Australia. More
in general, modernization theory postulates that floating voters are to be found
among the so-called new middle class (white collar employees) (Zelle 1995). Lachat
has found in his longitudinal study for Germany and Switzerland that the saliency
of the class cleavage is not diminishing but still determines voter stability (Lachat
2004). Similarly, turning upside down the Lipset and Rokkan (1967) approach
which tries to account for electoral stability and not for change, volatility might
depend on the position a voter takes on the major cleavages organizing the party
system (see also: Schoen 2000). Apart from class position (left-right), the Belgian
party system anno 2004 is characterized by other cleavages. One of these
cleavages, though withering through secularization, is the catholic versus non-
catholic rift. Therefore we expect religiousness, more specifically Catholicism, to
be a predictor of non-volatility. Finally, although we are not aware of any
theoretical reasons why women and men should differ in intra-campaign volatility,
we need to control for sex.
Attitudinal-behavioral accounts
The attitudinal-identification approaches cannot be completely separated from the
socio-structural approaches. Education, for example, generates political interest,
which is a crucial variable according to the attitudinal school (Dalton and
Wattenberg 2000). Lachat (2004) even combines education and political interest in
a single measure of what he calls political sophistication. The modernization
theorists turn around the original argument of Lazarsfeld and colleagues who
described the original floating voter as being utterly uninterested in politics
(Lazarsfeld, Berelson, and Gaudet 1945). More political interest, or ‘cognitive
mobilization’ (Dalton 1984), leads to more volatility, the modernization theorists
claim. Yet this relationship probably is indirect, as cognitive mobilization works

5
‘via’ partisan identification (Lachat 2004). Another complication in terms of
political interest is that, according to Zaller (1992), the relationship between
political interest, or political sophistication, and volatility is non-linear. Especially a
middle level of political interest leads to frequently changing preferences (Lachat
2004). Granberg and Holmberg (1990) tested the effect of political interest
comparatively for 17 separate elections in the US and Sweden and found that, in
the US, political interest leads to significantly less volatility while, in Sweden,
interest in politics is not related to volatility at all. So, political interest can work
both ways, depending on the country and political context. As the Belgian polity is
more alike the Swedish and differs a lot from the American, we are unsure what to
expect about political interest and volatility in the present study. Similarly, scholars
have turned upside down Lazarsfeld’s findings for the recent period claiming that
media use determines volatility. The more a voter relies on the media for his
political information, the more it is likely that he/she is likely to change opinion
(Latimer 1987). This is caused by the fact that unsure voters, being confronted with
a decision to make, undertake greater efforts to find information. The most
obvious predictor for intra-campaign change, of course, is the Michigan’s school’s
main variable: partisan identification (Campbell et al. 1960). The closer one feels
to a party, the less chance that one will change preference before or during the
campaign. McAllister (2002), for example, substantiated this point regarding late
deciding in the US, the UK, and Australia; Lachat did the same for Germany and
Switzerland (Lachat 2004). More generally, as good as all studies about voter
volatility underscore the link between party identification and volatility. Zelle (1995)
contended that especially political frustration can explain floating votes: floating
voters have less trust in political parties; they are dissatisfied with the political
system in general. Finally, if we want to gauge intra-campaign volatility and claim
that this kind of volatility during the campaign differs from volatility between
elections, we need to control for inter-election volatility. Of course, we expect
inter-election volatility to be a very strong predictor of intra-campaign volatility.
Yet if other variables would remain significant predictors although inter-election
volatility is controlled for, this would imply that these variables specifically predict
intra-campaign volatility beyond the inter-election volatility effect.
Specific campaign context
All the factors above, the socio-structural as well as the attitudinal-identification
factors, are to a certain extent stable. Political frustration, for example, is unlikely to
change strongly between subsequent elections; people do not switch partisan
identification overnight… Intra-campaign changes, though, can also be determined
by aspects that are typical for a certain campaign; by things that vary from
campaign to campaign. Intra-campaign shifting differs from campaign to campaign
depending on the concrete political context. All long-term studies showed that,
although there is a secular trend towards more intra-campaign volatility, this trend
is certainly not linear. Extreme volatile campaigns are followed by stable
campaigns. On a micro-level too, some voters change opinion during a certain

6
campaign, but they do not in another campaign (Lachat 2004). That is one of the
reasons why general electoral volatility and specific intra-campaign change must be
distinguished from each other; they are close but, ultimately, different things.
McAllister (2002) showed that late decision-making, and supposedly intra-campaign
volatility, depends on incumbency. If voters are not familiar with the candidates,
especially when the previous incumbent is no longer a candidate for office, they are
more uncertain and postpone their voting decision. In a multi-party system with
strong parties, as in Belgium, the incumbency factor is less important and almost
impossible to measure. That is why we will ignore this factor in our analyses below.
More in general, intra-campaign switches differ between parties. Some parties run a
successful campaign, while other do not. As a consequence, in a certain campaign,
some parties’ voters tend to switch more than others. Controlling for previous
vote is a good strategy to test this. Zelle found that party preference is a very
strong predictor of volatility dwarfing all other variables (Zelle 1995). He also
substantiated that electoral volatility is not dispersed evenly over the whole political
spectrum. In Germany, for example, the green and social democrat parties share
the same electorate and some voters often switch between both parties. An
alternative control for previous voting behavior is the left-right self-positioning
of voters. There is no theoretical reason to expect that left-leaning voters would, in
principle, be more easily seduced to switch parties than right-leaning voters. Yet,
we know that in some campaigns switching happens especially at the right-side of
the electoral spectrum and sometimes at the left side. In the 2004 Belgian campaign
switching most likely happened at the left side, more precisely between the green
and socialist party. So, for this specific 2004 campaign in Belgium we expect left-
wing voters to have been more volatile than right-wing voters. Another reason to
expect left-wing volatility in this particular election is that the number of available
alternative parties for the right-leaning voter diminished in 2004. In fact, two
center-right parties, CD&V and NV-A, formed a cartel reducing the number of
potential changes; for the left-wing voter nothing changed and the offer on the
political market remained the same. The goal of a cartel is exactly to reduce
competition among like-minded. Finally, the ideological distance between a party
and its closest contender might be a useful campaign-related predictor of intra-
campaign change. This variable is campaign-specific as parties reposition
themselves before and during the campaign. The ideological distance factor mirrors
the argument put forward by Peter Mair (1997) who claims that voters not really
became more volatile in the recent decades. They may change actual preferences
but in fact they only shop for the party that is ideologically closest to their initially
preferred party. Granberg and Holmberg tested a version of this argument and
found that, over the years, in Sweden and the US, the amount of what they call
‘inter-bloc switchers’ is small. People change opinion but they merely shift their
vote to neighbouring parties (Granberg and Holmberg 1990; 1991). Lachat came to
similar findings for Germany and Switzerland (Lachat 2004). In stead of
distinguishing two different kinds of volatility, between neighbouring parties or
beyond neighbouring parties, and running different analyses as previous authors

7
did (Granberg and Holmberg 1991; Lachat 2004; Zelle 1995), we will test only one
model for intra-campaign volatility but control for ideological distance.
DATA AND METHODS
In this section we will do two things: explain our unusual web-based panel survey
method and introduce the Belgian case in general and the 2004 regional election
campaign in particular. We draw upon the University of Antwerp Web-based
Electoral Panel 2004 (UAWEP-2004). This is a five-wave, non-representative pre-
and post-electoral panel of 11,486 voters in Belgium (Flanders) in March-June
2004. The panel was devised with five consecutive waves: two pre-campaign waves
(W1 and W2), two campaign waves (W3 and W4), and one post-electoral wave
(W5). We consider the six weeks before the ballot to be the campaign period, well
aware of the fact that the long period before the elections is affected by the
campaign heat and can be considered as a ‘long campaign’ (Norris et al. 1999). In
Belgium, the 2003 national elections fell exactly one year before the 2004 regional
elections; the country was affected by generalized election fever for more than a
year.
Although also within academia gradually more web-based surveys are conducted
(Best and Krueger 2004; Dillman 2000), most web-panels are set up for commercial
reasons and it certainly is not a common research technique for electoral surveying
yet. We give some more details than usual about design and procedure. The main
problem with web-surveys is, of course, representativity. How can you get a
representative sample of the population with all members of the population having
an equal chance to be among the respondents? The strategy we followed to
circumvent that problem for the present study was very simple: we tried to
maximize diversity. We reconciled ourselves from the beginning with the fact that
our study could only be explorative and that we would not be able to get a true
random sample of the Belgian (Flemish) population. Self-selection is too big a
problem, as we could only ask people to participate but could not control in any
way for non-participants or for people that were not reached by us. That being
said, we tried to put together a panel that was as diverse as possible and that drew
respondents from all corners of society and with all walks of life. For that purpose
we recruited panel participants via banners on websites of popular radio stations, of
soccer teams, of associations of the elderly, of women’s organizations… Students,
taking a class in methods, distributed leaflets inviting people to participate at train
stations, on the streets, in bars... We also relied on snow ball sampling asking
participants to invite other people they knew. Additionally, via email we
recontacted all participants (N=3,500) that participated in a previous electoral panel
set up for the previous 2003 elections. For that previous panel, we used the same

8
recruiting strategies as for the 2004 panel survey1. All these sampling strategies
generated internal diversity, as we wished, but they also contributed to further
skewing the panel. For example: working via elderly organizations yielded, as
expected, quite some elderly participants; but it probably generated an
overrepresentation of organized elderly.
Recruitment existed in asking people to go to a website and to answer an html-
based questionnaire. People had to provide their email address so they could be
recontacted for the next waves. Only people participating in the first wave were
recontacted for the following waves. No new respondents were allowed after
closing the first wave. Response was mostly immediate with an enormous amount
of completed questionnaires in the first hours and days just after opening another
wave. Return rates dropped sharply after three days; then we sent a reminder which
temporarily spurred answering activity again; after which the response petered out
completely. As, once the campaign had started, we wanted to gauge more punctual
time-linked campaign effects we gradually reduced the time slot for filling in the
questionnaire as the campaign evolved. Table 1 contains the response rates.
Response rates go down gradually. Sometimes respondents decided not to
participate in, for example, the third wave, but then they stayed on board and
participated again in wave 4 and/or 5. The panel was substantially skewed, we will
come back to that immediately, but the drop out from W1 till W5 did not really
worsen its substantial initial skewness. For example, drop outs came from all
political leanings, were not less educated, not more female than male etc…
Table 1: Design of the University of Antwerp Web-based Electoral Panel 2004
(UAWEP-2004)
Date N ‘Response rate’
Pre-campaign W1 2-12 March 11,486 -
Pre-campaign W2 20-30 April 8,824 77%
Campaign W3 17-25 May 8,419 73%
Campaign W4 6-10 June 7,906 69%
Elections 13 June
Post-campaign W5 15-21 June 7,917 69%
Our design has advantages and drawbacks. The main drawback is that we do not
dispose of representative data for the Belgian (Flemish) population. Our main goal
is, however, not to put forward definitive statements about the Belgian (Flemish)
population and it’s electorally floating behavior in 2004. Rather, we want to develop
a typology of floating voters and try to account for the different types. The main
1 In W2 of UAWEP-2004 we asked respondents how they became involved in the UAWEP-2004
panel: most participants (42%) participated after being contacted via email by the organizers
(teachers and students); the participants that came from the 2003 panel were the second largest
group (36%); one sixth of the participants were recruited via websites (15%), 1% via leaflets and
6% via other (non-specified) channels.

9
question is whether we have enough diversity and variation in our sample to test
for relationships and effects between variables. Indeed, if variables do not vary, or
hardly, we cannot test associations for that variable. In what respect is our sample
most skewed? Comparing our panel participants with the population at large, three
features strike the eye: our participants are younger, they are higher schooled and
they have more interest in politics than the average Belgian (Fleming). These biases
are linked to two characteristics of the sampling procedure: students and university
teachers mobilized in their respective social environments; people were mainly
contacted via electronic media. The available data on internet use confirm that, in
Belgium, internet use is still concentrated in the younger and affluent segments.
Because we did not want to create the impression that our data are representative,
and because of unwanted effects of weighing procedures, we decide not to weigh
our data. Because of our tilted database, we will be extra-cautious when analyzing
and drawing conclusions about biased variables.
The main strength of the study is that the number of respondents is high. The
UAWEP-2004 contains 7,413 usable respondents, which answered at least one pre-
campaign (W1 or W2) and the post-electoral wave. Most electoral studies simply do
not contain enough volatile voters to systematically contrast faithful with changing
voters and, especially, to construct types of volatile voters (Latimer 1987). This
restriction applies even more to intra-campaign switchers as this group of voters
typically is even smaller. Limited numbers, most likely, are the main reason why
intra-campaign change has received little empirical scholarly attention. Moreover,
we dispose for a large amount of voters per party; for all major parties we have at
least 1,000 respondents in our database.
A second strength is the fact that due to the panel design we can avoid working
with questionable recall data. Most of the research into electoral volatility is based
on cross-sectional post-electoral surveys (Schoen 2000). After an election, people
are asked about their electoral behavior in the previous election and the elections
before that; often, voters are also asked about the exact timing of their voting
decision. Sometimes, these post-electoral surveys take place months after the actual
elections which reduces reliability of the answers considerably. Recall questions
tend to overrate consistency of voters (Schoen 2000; Schuman and Presser 1981).
As we will show in what follows, people’s answers are often unreliable or at least
inconsistent. What people say they did and what they actually did is not always the
same. In fact, our typology of voters is even based on the distinction between
words and deeds. Only panel studies tapping electoral preference several times
during the campaign are fit to gauge intra-campaign change.
Is our sample skewed towards more voter volatility? In other words: is the
UAWEP-2004 biased in terms of the dependent variable? If our respondents are
much more, or much less, inclined to shift preference during the campaign this
might endanger the validity of our conclusions. In our panel, we found 18% of

10
people changing preference during the six week campaign period: they voted for
another party on June, 13th than the one they indicated to have their preference in
W2 (April, 20th-30th). This figure is perfectly comparable with Blais’ (2004) data
about intra-campaign shifting in the US, UK, Canada, New Zealand and The
Netherlands and is by no means an extreme value. In fact, Blais found that, on
average, about 17% of voters switch preference during the campaign. Also in terms
of inter-elections volatility, our dataset seems not at all eccentric. In our panel 27%
switched parties between the 2003 and the 2004 elections. Academic surveys not
yet available, the commercial post-electoral polls of the 2004 elections ascertained
that 24% of the Belgian (Flemish) voters gave their vote to another party than in
20032.
Our study is based on Belgian evidence. To what extent is Belgium a typical
country, and to what extent were the 2004 regional elections typical elections?
Belgium has a strong, fragmented party system; it is being considered as a text-
book example of a consociational democracy (Lijphart 1999). Until 1999 the
Christian democrats were the key player always being part of government (since the
1950s). Belgium is a federal country with two completely separated party and media
systems. Parties do not compete in other regions but only with parties speaking the
same language; they only run for office in their region. The June 13th 2004 regional
elections – disputed separately in the Flemish, Walloon, Brussels and German
regions - came only one year after the May 2003 national elections. As parties and
candidates are the same in national and regional elections – all national government
ministers and all party presidents ran for regional – national and regional elections
hardly differ. The same parties and the same candidates crossed swords about the
same issues. Although lots of competences are regionalized, during the 2004
regional election campaign regional and national issues were debated. Consequently,
national and regional elections are very similar. In 2004, regional elections
coincided with the European elections. European issues were as good as
completely absent from the campaign. Our study only applies to the Flemish,
Dutch-speaking part of Belgium, comprising about 60% of the Belgian population.
The stakes of the 2004 campaign were, at the left side of the political spectrum, the
survival of the green party and, at the centre-right side, the race to become the
biggest party (and the right to take initiative to form a new Flemish government).
The greens survived and booked a tight but clear victory leading to a defeat of the
socialist party. The governing liberal party of Prime Ministers Guy Verhofstadt
suffered a painful loss which made the CD&V-NV-A-cartel the biggest party. The
extreme-right Vlaams Blok3 won again many votes which reinforced the party’s
status as most successful extreme right-wing party in Europe. Analyses showed that
the change at the left-side of the political spectrum took place during the campaign,
while the change at the right side of the political spectrum took place before the
2 See the TNS-media poll reported in De Standaard, October 8th, 2004
3 The party’s name switched from Vlaams Blok to Vlaams Belang later, we will use the initial
name.

11
campaign started (Van Aelst and Walgrave 2004). Finally, let us briefly consider
Belgian (Flemish) electoral volatility in a somewhat longer time-perspective. In
general, Belgian voters cannot be considered as being extremely volatile; but
volatility was clearly on the rise during the last decades. In the 1980s on average
15% of the Belgian (Flemish) voters changed parties between elections
(Swyngedouw, Billiet, and Carton 1992). In the 1990s’ elections more than 30% of
the voters reported to have changed parties. Compared to other countries, these are
by no means extreme figures (Dalton, McAllister, and Wattenberg 2000).
TYPES OF FLOATING VOTERS
We propose a typology of voters and intra-campaign change based on two
variables. (1) Did people, before the campaign started (W2, 20-30 April) state that
they made a (definitive) decision about their vote4? (2) Did they actually change
party preference during the campaign5? We constructed this second variable by
comparing the W2 preference with the W3 preference, the W3 with the W4
preference, and the W4 preference with the actual vote in W5. If a voter switched
party preference during any of these three transitions, we considered him/her to
have changed party preference during the campaign (even if this person returned to
his original party at a later stage)6. Crossing both variables we have four types of
voters7.
Why this typology? We depart from the idea that inconsistency is a good indicator
of uncertainty. Voters are increasingly uncertain about their choice. When they are
confronted with survey questions regarding party preference their uncertainty can
translate in two kinds of answers: they can explicitly state that they feel uncertain
and have not made their mind up yet; or they can, in a subsequent poll, simply
indicate another party as their preferred party. The first measurement of
uncertainty is based on people subjective self-perceptions (what they think). The
4 The question wording was the following: ‘Did you already decide for which party you will vote
at the coming Flemish of June 13th?’ Answers: ‘Yes, I took a decision’; ‘No, I did not take a
decision yet’.
5 If people stated they already took a decision (see above) they were asked the question ‘If yes, for
which party will you vote?’ If they had said they had not decided yet, the following question was:
‘If not, which party has the highest chance to get your vote?’
6 To be more precise, if W2 information was lacking because the person at stake did not
participate in this wave we drew on information of W1 in which all respondents participated. Of
the 7,349 voters in our sample, this was the case for about 500 records.
7 Our typology comes close to Granberg and Holmberg’s typology of voters, yet it is different as
their categorization lacks our bachelors and operationalizes our secret admirers differently. Note
that our UAWEP-2004 figures are in absolute terms not that different from their evidence for the
US and Sweden. At the end of their research period (1988) they find 3% switchers and 44%
stable voters in the US and 10% and 67% in Sweden. Our Belgian data are almost perfectly
situated in between the US and Sweden.

12
second indicator of uncertainty is based on objective (intended) behavior (what they
(will) do). Crossing both creates two extreme categories in which self-perception
and (intended) behavior are consistent: people who say they are certain and act
accordingly by sticking to the same party all the way through; people who state they
are uncertain and act accordingly by switching parties in subsequent polls. If
people’s answers to the question whether they had reached a definitive decision
always completely corresponded with reality, we would only have cases in the
second and the third quadrant of our typology, which is not the case. Both
intermediary categories consist of voters whose responses are inconsistent: their
self-perception and (intended) behavior are contradictory. They say they have made
up their mind and, still, they switch parties; they say they are uncertain but they
stick to the same party throughout the whole campaign. These ambivalent or
inconsistent voters, we claim, are situated in between the well-known loyal voter,
on the one hand, and the completely detached voter, on the other. They form a
third and possibly even, a fourth type of voters, not stable and not totally loose. An
important research question is whether these in-between voters form one
intermediary type or two intermediary types. Are people whose self-perception is
uncertain and whose behavior is certain any different from people whose self-
perception is certain but whose behavior is uncertain? This is an empirical question.
Yet, at least in principal, it is possible that, when confronted with uncertainty about
their vote, some voters start by saying they do not know while others start with acting
as if they do not know. Some people might first change opinion and then behavior;
others might do it the other way around. One of the main aims of the empirical
section below is testing differences between both kinds of inconsistent voters: do
we have one or two types of intermediary floating voters?
This typology is innovative, we think, as it uses a question that is normally
considered as a (methodological) filter question to construct types. In fact, the
typology profits from voters’ inconsistency to devise a typology of voters. In stead
of considering response inconsistency as an annoying methodological problem, we
use it as an asset to improve our knowledge of volatility. Since our typology
essentially draws on inconsistency, it can be used only on panel data. Recall data
from post-electoral studies systematically overrate voters’ consistency, studies
showed, because most voters want to give a reliable impression. Only confronting
people with the same question a few weeks or months later can lay bare their
inconsistency; a lot of them have long forgotten what answer they gave in the
previous wave of the panel.

13
Table 2: Types of floating voters (N=7,349).
Made a decision
(in W2)
Did not make a
decision (in W2)
Changed preference during
campaign
6.5% (N=477)
Adulterous voters
15.7% (N=1,152)
Bachelors
Did not change preference
during campaign
59.0 % (N=4,333)
Loyal voters
18.9% (N=1,387)
Secret admirers
Let us turn to our empirical evidence. A substantial fraction of the voters (6.5%)
said they made up their mind yet they still switch allegiance afterwards: we call
them adulterous voters. They say they are loyal, but they are not. An even larger share
(18.9%) say that nothing is sure yet, but during the campaign they stick to the same
party all the time; these are the secret admirers. They have a firm preference and
repeat this over and over again, but they do not seem to want to admit that they
fancy this party. Both other categories are consequent in their answers. The loyal
voters state in the beginning of the campaign they have made up their mind and they
act accordingly: they always indicate the same party as their preference (59.0%).
Finally, there are the consequent disloyal voters, we call them bachelors; this is the
pure floating voter. They indicate that they are unsure about their preference and
this is exactly what happens during the campaign: they switch parties (15.7%). Note
that we do not claim that the size of these four groups corresponds with the size of
these groups in the Belgian population at large – we remind the reader of the fact
that our sample is not a representative one. We think it is safe, though, to state that
also in the population at large all four kinds of voters exist.
Now that we have constructed our dependent variable, let us check whether our
intra-campaign change types are associated with other classic measures of voter
volatility. First, there is a clear association with timing of decision. In W5, after the
polls, we asked respondents when they had made their final decision8. Decision
time is a classic indicator among volatility studies. As expected the (Spearman’s
Rho) correlation is .73 which is very high: bachelors decide late, loyal voters early,
secret admirers and adulterous voters score right in the middle. Granberg and
Holmberg (1991) for the US and Sweden, and Lachat (2004) for Germany and
Switzerland, found similar correlations. Another variable often used to gauge
volatility is split-ticket voting. As the 2004 regional elections coincided with the
European elections, we can test for split-ticket voting. Again, the correlation is
substantial and significant (Spearman’s Rho .34): bachelors practice split-ticket
voting much more than loyal voters; secret admirers and adulterous voters are
situated in between. Finally, we also constructed a scale of how many times a voter
switched preference during the campaign, in contrast to our main variable tapping
8 The precise wording of the question was as follows: ‘When did you decide for which party you
would vote for the Flemish parliament?’ Answer categories were: the day of the elections, a few
day before the elections, a few weeks before the elections, before the start of the campaign, or a
longer time before.

14
whether a voter changed opinion at least once during the campaign. Correlation is
high again (Spearman’s Rho -.63) meaning that changing once and changing often
are closely related. All these positive tests reinforce confidence that our intra-
campaign change types are sound and part of a more encompassing volatility
syndrome.
EXPLAINING FLOATING VOTERS
We constructed four types of voters - two extreme types and two intermediary
types - that appeared to be robust and sound when bivariately tested. Yet, really
establishing whether we are confronted with different types of voters supposes that
those types are also multivariately different. We are quite confident that both
extreme types, loyal voters versus bachelors, will differ from each other in many
respects; but the question is whether both newly constructed intermediary types
whose behavior and statements seemed to be at odds with each other, which we
interpreted as being an indicator of uncertainty, really differ from both extreme
types and whether they differ from each other. In other words: only if we can
discern all four types from each other in multivariate analyses, will we have
constructed a valid and useful typology.
Therefore, we estimated multinomial logistic regression models predicting voter
type drawing upon the factors put forward in the theories of voter volatility. As we
want to devise a typology of floating voters, we took the bachelor, this is the pure
floating voter who said he had not decided yet and who switched party preference
during the campaign, as the reference category and contrasted it with loyal voters,
silent admirers and adulterous voters. Our models are built up in steps. First we test
for the socio-structural variables only (model I); next we add the political (attitudes
+ behavior) variables (model II); finally we insert campaign specific factors (model
III). Because of the skewedness of our dataset we complement our model with one
variable to control for the non representativity of our sample, and that is party
membership; we have relatively more party members among our respondents than
among the population as a whole.
Model I’s explanatory power is very limited (R² = .037), Table 3 shows. This
means that, in terms of socio-structural variables, both intermediary types and the
loyal voters, on the one hand, do not differ a lot from the pure floating voter we
coined them bachelors, at the other hand. As expected, loyal voters differ far most
from the bachelors: they are significantly older, more male and more catholic than
their more floating counterparts. This confirms the Michigan school’s main thesis
that socialization (age) leads to party loyalty. It also underscores the fact that deep
cleavages, in this case the religious divide, still structure the party landscape and
counteract partisan dealignment. We do not have a sound class measure in
UAWEP-2004, so we cannot test for this other cleavage. Statistically spoken, secret

15
admirers cannot, in socio-structural terms, be discerned from the bachelors.
Adulterous voters, who pledge loyalty to a party but leave their party later during
the campaign, are more male (what a surprise!) and somewhat lower schooled than
bachelors. Overall, sex is the strongest predictor: women state more often than
men that they are not sure about their vote (bachelors or secret admirers); men
more easily say that they are sure (loyal and adulterous voters) but they betray their
party more often. The fact that education, the crown variable of all modernization
theorists, is hardly significant raises questions about their claims. Yet, this may have
been caused by the fact that our dataset is tilted especially in terms of education;
there might be not enough educational diversity to definitively test for this factor.
The main conclusions of Model I are that: (1) both extreme types – loyals versus
bachelors - can more or less clearly be discerned from each other; (2) both
intermediary types seem to be closer to the bachelors than to the loyal voters; (3)
both intermediary types cannot really be distinguished from each other on the basis
of a socio-strcutural analysis.
In Model II the political variables are added. The overall explaining power of the
model is more than satisfying (R² = .354): political factors are much more powerful
explainers than socio-structural factors. The limited power of the socio-structural
predictors is further reduced: age vanish completely. Education remains only
slightly significant with a lower education for adulterous voters. The model
confirms that sex is the socio-structural variable that matters most, followed by
religion. Let us first turn to the political variables that do not play a role. Political
interest, again contradicting the modernization theorists, is never significant:
floating or not floating, political interest has got nothing to do with it. This is
congruent with Granberg and Holmberg’s findings in Sweden (Granberg and
Holmberg 1990). Yet, again, our sample skewedness might be responsible; (too)
many of our panel members had a keen interest in politics. The squared variant of
political interest tapping whether people with middle levels of political interest are
most prone to change parties, this is Zaller’s argument, is not significant either.
Media use can be discarded as well. Whether one follows politics intensively via the
media or not, does not affect one’s tendency for change or stability. Our control
variable party membership proves to be very powerful. Loyal voters, secret
admirers and adulterous voters all are more party members than the bachelors,
which makes sense. The same applies to partisan identification. It yields substantial
effects underscoring again the Michigan school’s argument. Secret admirers,
adulterous voters and, naturally, loyal voters identify more with parties than pure
floating voters. Democratic satisfaction is significant, but not in the direction Zelle
supposed: dissatisfaction is not detaching people from parties but exactly the
opposite. Loyal voters, adulterous voters and secret admirers all are more
dissatisfied with democracy in Belgium than the bachelors. This can be explained
by the extraordinarily loyal behavior of the voters of the extreme right-wing party
Vlaams Blok (Walgrave and Deswert 2004). Deeply dissatisfied yet stable at the
same time, they determine the dissatisfaction coefficients to a large extent. In the

16
next step we control for vote in the previous election and, as a consequence, the
dissatisfaction effect completely disappears. Anyway, our findings challenge Zelle’s
argument and show, at least, that dissatisfaction and volatility are not always
positively associated. The two final variables in Model II are the best predictors.
They grasp inter-election volatility in the long-term and in the short-term (see
technical appendix for more details). Both effects go in the same direction for all
types. People acknowledging that they have voted in the past for several parties are
to be found more among the bachelors, less among secret admirers, adulterous
voters and, especially, among loyal voters. The same applies for people that
changed party preference between the previous 2003 elections and the beginning of
the 2004 campaign. Again the loyal electorate has far least changed preference in
the run-up to the campaign. All this makes perfectly sense; it may even sound
somewhat tautological. Yet, the fact that many other variables remain significant
although we powerfully control for inter-election volatility, means they specifically
predict intra-election volatility and not just electoral volatility in general. Model II
leads to the following conclusions: (1) both extreme kinds of voters are sound
distinct types also with respect to the political variables; (2) both intermediary types
are now visible with a distinct, but still largely common, profile diverging in
political terms from the loyal voters and from the bachelors; (3) differences
between secret admirers and adulterous voters are modest but seem to be are
slowly (significant differences on four variables).
In Model III, we inserted the campaign variables. This further boosts the
explanatory power of the model (R² = .439). Socio-structural variables are
weakened further but most political variables endure the campaign variables’ test
without any problem. Intra-campaign volatility, hence, is determined by political
and campaign-specific effects at the same time. Of all previous party votes, only a
vote for the extreme-right Vlaams Blok is (very) significant. The 2003 Vlaams Blok
voters are more adulterous and loyal voters: they (say they) know what to vote at
the beginning of the campaign. Of course, this effect might change in the future, as
the Vlaams Blok assumes governmental responsibility for example; the changeable
character of this variable is the reason why we consider it as a specific campaign
effect. The mirror image of that loyalty is the disloyalty of the left-wing voters as
captured in the left-right scale. As expected, left-wing voters are, not inherently but
at least in the 2004 campaign in Belgium, more volatile. They are to be found less
among adulterous and loyal voters. During the 2004 campaign the competition at
the left-side was tougher than at the right-side. The most powerful predictors of
Model III are both ideological distance indicators. The first of these indicators taps,
before the campaign (in W2), the distance (1-10) between, on the one hand, the
party that is perceived as being closest (the party that will probably get the vote)
and, on the other hand, the party that scores second. The second indicator (W4-
W2) brings in the campaign dynamics. It assesses the ideological distance difference
between the first and second choice party between the beginning (W2) and the end
of the campaign (W4). The logic behind this double use of distance variables is

17
precisely that during the campaign the distance between the parties is constantly
manipulated. Parties move closer or further apart hoping that the voters would
notice those changes. As the campaign evolves perceptions of voters change. We
expect, for example, that when an uncertain voter’s vote slowly crystallizes, the
distance between this voter’s first choice and her second choice party would
gradually increase. In W2, loyal voters, adulterous voters and secret admirers
perceive the distance between their preferred party and the second party of their
preference as being much bigger than the bachelors, which is logical. The W4-W2
distance measure indicates that differences between distances are becoming bigger
for secret admirers and loyal voters compared to bachelors. This indicates that
those two voter categories become more certain of their choice as the campaign
evolves. Their favorite party’s lead is getting bigger and bigger. We will come back
to this point later. From Model III we should retain the following: (1) introducing
the campaign variables distinguishes both intermediary categories now clearly from
both extreme categories; (2) adulterous voters are somewhat closer to loyal voters,
secret admirers are more alike bachelors; (3) both intermediary types now clearly
emerge as being distinct from each other as well (five variables are different).

18
Table 3: Predictors of types of floating voters
Model 1
(Nagelkerke R²=.037)
(Ref. cat.=Bachelors, N=1063)
Model 2
(Nagelkerke R²=.354)
(Ref. cat.=Bachelors, N=925)
Model 3
(Nagelkerke R²=0.439)
(Ref. cat.=Bachelors, N=908)
Adulterous
voters
Secret
admirers
Loyal
voters
Adulterous
Voters
Secret
admirers
Loyal
voters
Adulterous
voters
Secret
admirers
Loyal
voters
Socio-structural variables
Age (high) ns ns 1.01*** ns ns ns ns ns ns
Sex (male) 1.43** ns 2.05*** ns ns 1.60*** ns ns 1.52***
Education (high) .88* ns ns .82* ns ns .85* ns ns
Religiosity (Catholic) ns ns 1.11*** ns ns 1.09* Ns ns ns
Political variables
Party membership (yes) 1.52** 1.47* 1.80*** 1.51** ns 1.72***
Political Interest (high)
Political Interest² (high)
ns
ns
ns
ns ns
ns
ns
ns
ns
ns
ns
ns
Media use (high) ns ns ns ns ns ns
Partisan identification (high) 1.40*** 1.20*** 1.99*** 1.31*** 1.13** 1.53***
Democratic dissatisfaction (high) 1.33** 1.15* 1.41*** ns ns ns
Inter-election volatility
Change last election (yes)
Change in general (frequent)
.ns
.84*
.51***
.85**
.22***
.64***
ns
82**
.57***
.88*
.37***
.70***
Campaign variables
Previous vote Vl.Bl. (others =ns) ns ns 2.40*
Left-right scale (right) 1.10* ns 1.12***
Ideological distance
Distance W2 (large)
Distance W2 – distance W4 (large)
1.45***
ns
1.20**
1.65***
1.65***
2.11***
N 434 1275 4006 362 1132 3598 357 1120 3579
Note: The coefficients represent standardized betas (Exp(B)) and their significance in a multinomial logistic regression analysis models predicting
different types of voters (the pure floating voters are the reference category) as the dependent variable. Sig. ***=.001 **=.01 *=.05. Exp(B)s
larger than 1.0 indicate a positive effect, smaller than 1.0 a negative effect. Collinearity statistics were checked for the tolerance of all variables.
See the technical appendix for coding details of all the items. Source: UAWEP-2004

19
In terms of the goal of this paper, establishing floating voter types and testing the
existing floating voter theories on Belgian evidence, the analyses in Table 3 permit
clear conclusions. First, the analyses tend to support the Michigan school
arguments and to question the modernization theorists. The presented evidence in
particular suggests that specific campaign dynamics - elements differing from
campaign to campaign – and that have not been fully theorized yet, at least not in
the floating voter literature, seem to be most crucial. The floating of voters is not
that much determined by what they are or how they think, but rather by the
available offer on the political market and the competition dynamics that emerge
before and during the campaign. We ran separate multinomial regressions for
socio-structural variables, political variables and campaign variables. The R² of the
campaign model turned out to be slightly superior9. To further substantiate our
point that campaigns matter we would need evidence covering several subsequent
campaigns maximizing campaign differences.
Second, we validated our four voter types among which three distinct types of
floating voters. Never call a floating voter just a ‘floating voter’; they differ. The
pure floating voter is the one who has no fix preference before the campaign and
who does not seem to adopt a preference during the campaign either. These are
our bachelors. They have the most different profile from the loyal voters who,
when the campaign starts, say that they have made up their mind and act
accordingly. Socio-structurally, politically, and regarding the campaign these both
types differ. Our main finding is that there are two intermediary types - more stable
than the pure floating voter and less certain than the loyal voter - that can clearly be
distinguished from both extreme types. These two intermediary types also differ
from each other in important respects. The crucial difference between both types
of moderate floating voters is that they react different on the campaign: the
adulterous voters get more uncertain during the campaign while the secret admirers
gradually get more certain during the campaign. Their campaign behavior is thus
completely opposite. That is why the campaign variables in Model III load so
strong. It is intra-campaign dynamics that distinguishes both types of moderately
floating voters. To further our analysis and to substantiate the existence of four
types of voters, we estimated four binomial logistic regressions contrasting all types
two by two. The R²s can be found in Table 4.
9 To be more concrete, the R² of the socio-structural model was .037, the R² of the political
model .32, and the R² of the campaign model .34.

20
Table 4: R²s of binomial logistic regressions contrasting the four voter types
Loyal voters Adulterous voters Bachelors Secret
admirers
Loyal voters .25 .55 .35
Adulterous voters .18 .11
Bachelors .17
Secret admirers
The table produces logic results. The extreme types are most different (.55). Loyal
and adulterous voters are closer (.25) and so are secret admirers and bachelors (.17).
Both intermediate types differ from each other, but are rather close (.11). So, we
end with a nearly perfect continuum - note the almost linear differences between
the R²s - going from the most certain to the most uncertain voters: from loyal
voters over adulterous voters and secret admirers to bachelors.
OUR FOUR VOTER TYPES RE-EXAMINED
Do these floating voters behave sensibly during the campaign? In other words: can
we make use of these types to describe voters’ actions and thoughts in the
campaign? Is, in particular, the behavior of adulterous voters and secret admirers
through the campaign any different?
The point of departure of both types is different and leads to different empirically
gaugeable consequences. Adulterous voters are sure about their vote when the
campaign starts six weeks before Election Day. They made a pledge of allegiance
and planned to stick with it. Therefore, we expect them to become only gradually
more uncertain of their vote. Slowly, doubts about their initial preference start to
surface. The adulterous voter first starts to cultivate adulterous thoughts only till, at
a certain moment, these thoughts translate in (intended) behavior: he/she decides
to swap and give his/her (intentional) vote to another party. If this is true and
thoughts precede actions, we would expect that (1) the actual party switch happens
rather at the end of the campaign than at the beginning. After all, this voter just
said that he would definitively vote for a certain party and it takes time to settle
with the fact that he/she will not behave consistently. In operational terms, we
would expect party change among adulterous voters at the end of the campaign to
be bigger than at the beginning of the campaign. Also, we would expect that (2)
adulterous voters’ perceived distance between the first and second preferred party
would gradually become smaller during the campaign. Other parties become only
more attractive in stages while the appeal of the original party slowly but surely
withers. As Tables 5 and 6 show, compared to the other types of voters, this is
indeed precisely what happens with adulterous voters. In Table 5 the adulterous
voters show exactly the reverse pattern, more switching as the campaign evolves, as
the pattern of the bachelors who seem gradually settling for a party. Table 6 shows

21
how uncertainty nestles in the mind of the adulterous voter. Gradually he/she
considers the distance between his/her initial party and his/her second choice party
to become smaller and smaller which leads, as we can see in Table 5, to preference
change at the end of the campaign.
Table 5: Consequences of types of voters: late party preference change among
adulterous voters (in %)
Bachelors Secret
admirers
Adulterous
voters
Loyal voters
Between W2 and W3 53.2 0.0 36.7 0.0
Between W3 and W4 47.1 0.0 45.0 0.0
Between W4 and W5 42.0 0.0 51.4 0.0
Table 6: Consequences of types of voters: mean party distance of first preferred
party versus second party (0-10)
Bachelors Secret
admirers
Adulterous
voters
Loyal voters
Distance in W2 0.59 0.99 1.15 2.19
Distance in W3 0.56 1.03 0.84 2.08
Distance in W4 0.58 1.09 0.79 2.14
Table 7: Consequences of types of voters: uncertainty about vote among secret
admirers (in %)
Bachelors Secret
admirers
Adulterous
voters
Loyal voters
Uncertain in W2 100.0 100.0 0.0 0.0
Uncertain in W3 76.8 57.0 27.9 0.0
Uncertain in W4 55.9 31.1 26.0 0.0
The story of the secret admirers is precisely the opposite. They start the campaign
with no candid choice, only a weak preference. Gradually they start to like their
party better and better and ponder about officially expressing their secret love. This
is what happens as the campaign evolves and they finally admit that, after a trial
period, they are now sure that they have found what they have been looking for.
Their initial weak preference solidifies and turns into a firm choice. If this is true
and actions precede statements we would expect that (1) gradually during the
campaign more and more secret admirers dare to speak out and state they have
decided. We would anticipate this trend to be stronger among secret admirers than
among bachelors who should also display a solidifying trend. Also, we anticipate
that (2) secret admirers’ perceived distance between their first and second choice
party, in comparison to the other types of voters, would grow gradually as they
become more certain about their preference. Table 6 and 7 corroborate these
hypotheses. Table 6 shows that the distance with the second choice party grows in

22
every wave while the other voter types do not display such a pattern. Table 7
indicates that secret admirers start publicly expressing their loyalty much faster and
stronger than the bachelors.
Wrapping up, this section showed that both intermediary types of voters mean
something in reality. They grasp different types of behavior during the campaign.
In the campaign, secret admirers and adulterous voters act contradictory. The
choice of the first solidifies; the second only get more uncertain.
CONCLUSION AND DISCUSSION
As Granberg and Holmberg (1991) stated, there are different types and layers of
undecidedness. Few voters, they say, are really completely undecided about what
party they will vote for. Many people have narrowed their choice down to two or
three parties. In this paper we have tried to disentangle this complex phenomenon
of the floating voter during the campaign. We drew upon the distinction between
subjective feeling of certainty about the choice and objective (intended) behavior.
We operationalized uncertainty in terms of inconsistency. Not all people who say
they are undecided act accordingly and shop around; not all people who say they
are decided behave correspondingly and stand by their party. In our Belgian panel
study of the 2004 elections, around 25% of the voters behaved inconsistent. As our
panel is not representative we are unsure about the real proportions of each of
these types, but we feel confident that all four types of voters exist among the
Belgian population at large and most likely also in other countries. Maybe not all
associations in our multivariate models correspond with reality; our dataset is tilted
and might underrate the effects of age, education and political interest. Yet in
general, we think to have constructed a parsimonious and robust typology.
Intuitively these types make sense, they are associated with the expected factors,
they demonstrably differ from each other in many respects, and they, perhaps most
important, represent different logics of voters during the campaign. The usefulness
of this typology, indeed, lies in the fact that we devised a typology of floating voters
that better grasps the actual behavior of uncertain voters during the campaign.
We showed that, within a specific campaign, these types incorporate a sequential
logic: adulterous voters become more uncertain during the campaign; secret
admirers, in contrast, grow more certain; bachelors keep oscillating wildly
throughout the campaign; only loyal voters do not move at all. Yet, maybe the
same types can be useful to describe the more general phenomenon of partisan
dealignment and realignment. What happens in the campaign in a condensed
period of just a few weeks, growing versus diminishing uncertainty, might also
happen in reality on a more encompassing level and much slower. In other words:
our typology capturing intra-campaign volatility might also be useful to capture inter-
elections volatility. Some people remain loyal over the long-term, election after

23
election (alignment). Others slowly drift away from their preferred party and switch
(temporarily) to another party (dealignment). Still others gradually develop a more
enduring sympathy for a party that might be new to them (realignment). Some
people, finally simply do not attach themselves to any party and fluctuate
constantly. Perhaps the most intriguing question is whether both extreme types can
be considered as the stable types while both intermediary types of the moderately
floating voters are transitory types. The idea would be that loyal voters become
adulterous voters and then bachelors. Bachelors would become secret admirers and
finally loyal voters, although this second movement seems less plausible in the
present situation of growing dealignment. Other variants and transitions are of
course thinkable, but it makes no sense that a voter would immediately step over
from loyalty to pure volatility or vice versa; he/she probably goes through one the
intermediary types. Both intermediary types form a kind of limbo zone. In macro
terms, the question becomes in what direction the streams of voters go: from loyal
to pure floating or the other way around. Does realignment compensate for
dealignment? We think to have showed that, in Belgium in 2004, all types existed
and that the Belgian voters in our panel moved in all directions; the secret admirers
– on track towards realignment? – were even more common than the adulterous
voters – very busy dealigning. We would need to have many more data stretching
out over a longer time period to test this.

24
TECHNICAL APPENDIX: CODING AND MEASUREMENT
Socio-structural variables
Age (high) Years old (Age)
Sex (male) Male 1, Female 2
Education (high) 5-categories from lower 1 to higher 5
Religiosity (Catholic) 3-categories: 1=Non-believer (non-believer, free-
thinker); 2= Christian, but not catholic; 3=Believer
1(mainly Catholic, also Protestant and Other)
Political variables
Party membership (yes) 1=active or passive member; 2=non-member
Political Interest (high)
Political Interest² (high)
Scale from not interested in politics at all (0) to highly
interested (10)
Media use (high) Combined scale of use of newspapers, radio and
television for political information
Partisan identification (high) Score for the ideas of the favorite political party (0-10)
Democratic dissatisfaction (high) Are you satisfied with the working of democracy in
Belgium? 1=not satisfied at all, 4=very satisfied
Inter-election volatility
Change last election (yes)
Change in general (frequent)
Change of party between the last election (2003) and
Wave2, before the campaign started
“Do you usual vote for the same party every election or
mostly change party?” 1=always the same party;
2=mostly the same party; 3=mostly different parties
Campaign variables
Previous vote Vl.Bl.
(other parties were not signif. )
1=Voted for the extreme-right party Vlaams Blok in
2003; 2=not voted Vlaams Blok
Left-right scale (left) Classical left (0) – right (10) scale
Ideological distance
Distance W2 (large)
Distance W4 – distance W2(large)
The difference between the appreciation of the ideas of
the favorite party (0-10), versus the score for the
second party
The same difference at the end of the campaign

25
REFERENCES
Best, Samuel J., and Brian S. Krueger. 2004. Internet data collection, Quantitative
Applications in the Social Sciences. Thousand Oaks: Sage.
Blais, André. 2004. How many voters change their minds in the month preceding
an election? PS Political Science and Politics 38 (4):801-803.
Campbell, A., Philip E. Converse, W. Miller, and D. Stokes. 1960. The American
voter. New York: John Wiley.
Converse, Philip E. 1969. Of time and partisan stability. Comparative Political Studies 2
(2):139-171.
Dalton, J. Russel. 1984. Cognitive mobilization in advanced industrial democracies.
Journal of Politics 46:264-284.
Dalton, J. Russel, and Martin Wattenberg. 2000. Parties without partisans. Political
change in advanced industrial democracies. Oxford: Oxford University Press.
Dalton, Russel J., Ian McAllister, and Martin P. Wattenberg. 2000. The
consequences of partisan dealignment. In Parties without partisans, edited by R.
J. Dalton and M. P. Wattenberg. Oxford: Oxford University Press.
Dillman, Don A. 2000. Mail and internet surveys. The tailored design method. 2nd ed. New
York: John Wiley & Sons.
Granberg, Donald, and Sören Holmberg. 1990. The Berelson paradox. Intention-
behavoir changers in U.S and Swedish election campaigns. Public Opinion
Quarterly 54:530-550.
Granberg, Donald, and Sören Holmberg. 1991. Election Campaign Volatility in
Sweden and the United States. Electoral Studies 10 (3):208-230.
Inglehart, Ronald. 1990. Culture Shift in Advanced Industrial Society. Princeton:
Princeton University Press.
Lachat, Romain. 2004. A Heterogeneous Electorate. Individual-level differences in
the process of formation of voting choices in Germany and Switzerland.
PhD, Institut für Politikwissenschaft, Universität Zurich, Zürich.
Latimer, Margaret K. 1987. The floating voter and the media. Journalism Quarterly 64
(4):805-812.
Lazarsfeld, Paul, Bernard Berelson, and Helen Gaudet. 1945. The people's choice. New
York: Duell Sloan and Pearce.
Lijphart, A. 1999. Patterns of democracy. New Haven: Yale University Press.
Lipset, Seymour M., and Stein Rokkan, eds. 1967. Party systems and voter aligments:
cross-national perspectives. New York/London: MacMillan.
Mair, Peter. 1997. Party System Change: Approaches and Interpretations. Oxford:
Clarendon Press.
McAllister, Ian. 2002. Calculating or capricious? The new politics of late deciding
voters. In Do political campaigns matter?, edited by D. Farrell and R. Schmitt-
Beck. London: Routledge.
Norris, Pippa, John Curtice, David Sanders, Margaret Scammell, and Holli A.
Semetko. 1999. On Message. Commununicating the campaign. London: SAGE.

26
Schoen, Harald. 2000. Gründe für Wechelndes Wahverhalten. Helfen neue
Instrumente Licht in das Dunkel zu bringen? Politische Vierteljahresschrift 41
(4):677-706.
Schuman, Howard, and Stanley Presser. 1981. Questions and answers in attitide surveys.
Experiments on Question Form, Wording, and Context. New York: Academic
Press.
Swyngedouw, Marc , Jaak Billiet, and Ann Carton. 1992. Van waar komen ze? Wie
zijn ze? Stemgedrag en verschuivingen op 24 november 1991.
Topf, Richard. 1995. Beyond Electoral Participation. In Citizens and the state, edited
by D. Fuchs and H. Klingemann. Oxford: Oxford University Press.
Van Aelst, Peter, and Stefaan Walgrave. 2004. De twee verkiezingscampagnes van
2004. Samenleving & Politiek 11 (6):19-21.
Walgrave, Stefaan, and Knut Deswert. 2004. The making of the (issues of the)
Vlaams Blok. The media and the success of the Belgian extreme-right party.
Political Communication 21 (4):479-500.
Zaller, John. 1992. The nature and origins of mass opinion. Cambridge: Cambridge
University Press.
Zelle, Carsten. 1995. Social dealignment versus political frustration: contrasting
explanation of the floating vote in Germany. European Journal of Political
Research 23:319-345.