exploring uncharted territory: the irish presidential elections 1997
TRANSCRIPT
EXPLORING UNCHARTED TERRITORY:
THE IRISH PRESIDENTIAL ELECTION 1997
by
Wouter van der Brug
Cees van der Eijk
both: Department of Political Science, and The
Amsterdam School of Communications Research (ASCoR)
University of Amsterdam
and
Michael Marsh
Department of Political Science
Trinity College
Dublin
Submitted to BJPS January 1999
Final version delivered: October 1999.
1
Abstract
Elections to the Irish Presidency belong to the category of those in which
hardly any political power is involved. In elections such as this one,
according to second-order election theory, voter behaviour reflects mainly
preferences in the first-order political arena, where actual policy is made.
This theory fails, however, to explain voter preferences in the 1997 Irish
presidential elections. An alternative perspective of a popularity contest,
suggesting that voters’ preferences are largely unconnected to their
political opinions, and generally idiosyncratic in nature, also fails to fit the
evidence. Nevertheless, the information, framing and priming that voters
were subjected to during the 1997 campaign are shown here to result in
the development of a significant unidimensional cognitive and
preferential ordering of the candidates.
Introduction
The comparative study of electoral behaviour has demonstrated that our
understanding of particular elections —even the most important and familiar ones—
is enriched by the search for a general theory of electoral choice that encompasses the
many different forms and contexts in which elections take place. Elections in which
there is little ‘at stake’ are often incorrectly considered as poor subjects for study.
But the study of such elections gave rise to an important body of theoretical and
empirical literature about second-order elections, which has greatly influenced our
understanding of first-order elections where much more is at stake in political
terms.1 This paper extends this body of work by analyzing Irish voters’ preferences
for the candidates running in the October 30, 1997 Irish presidential election.
2
Unfortunately, the amount of relevant empirical information about elections is
strongly related to their political significance. Occasionally, however, we find that an
‘unimportant’ election has been well observed. One such case was this presidential
election. The data at our disposal not only relate to the final choice, but span
virtually the entire campaign period, allowing us not only to investigate the
preferences as expressed on the ballot, but also their development during the
previous weeks.
The 1997 presidential election
The Irish presidency is unusual in many ways.2 Though the post carries few
usable powers, it is filled by direct election. Elections are actually infrequent,
with several presidents returned unopposed.3 Nominations are quite tightly
controlled, candidates requiring the support of 20 members of the Oireachtas
(the two chambers of parliament), or of four county or county borough councils,
although this second channel had never been used prior to 1997. Five
candidates were nominated, more than ever before (beating the three who stood
in 1945 and 1990). More remarkably four were women, and none of them fitted
the typical profile of being a senior member of the Dáil (lower house of
parliament). The candidates were: Adi Roche (nominated by Labour, also
supported by Democratic Left (DL) and the Green Party), Mary Banotti (Fine
Gael (FG)) and Mary McAleese (nominated by Fianna Fáil (FF) and also
supported by the Progressive Democrats) along with Rosemary Scallon and
Derek Nally. These last two were independents who obtained local council
nominations. Only Banotti, an MEP, was a practising politician. McAleese was a
3
Northern Ireland academic who had strong links to FF and stood as their
candidate for the Dáil in a general election in 1987, but Roche, an anti-nuclear
campaigner who ran a charity in aid of the children of Chernobyl, had no links
with Labour at all. 4 Derek Nally was a former policeman who ran a victims’
charity. Rosemary Scallon (stage name Dana) was a one-time winner of the
Eurovision Song Contest, who was working for a catholic TV station in the
USA.
INSERT FIGURE 1 ABOUT HERE
The campaign was most remarkable for seeing the decline of Roche, who looked
like the eventual winner in the first poll but who finished a very poor fourth less
than six weeks later (see Figure 1). Seen by Labour strategists as a campaigning
president, her campaign failed to strike a chord. After a poor start, when several
people once involved in her charity made public criticisms of her leadership
style, her support declined. Her organisation appeared amateurish, the parties
supporting her failed to mobilize their activists, and her campaign failed to
promote her as a credible president. McAleese, the other initial front runner,
also ran into difficulties but used them to her advantage. Confidential briefing
documents were leaked from the Department of Foreign Affairs which
appeared to brand her as being much too close to Sinn Féin (SF), and she also
received the personal endorsement of the SF president, Gerry Adams5. This
episode lasted over a week in which attention focused very much on McAleese
whose performance under the spotlight demonstrated considerable aplomb
under pressure, and also underlined some of the details underpinning her
campaign promise to ‘build bridges’ between peoples, both in the North, and
4
between North and South. McAleese was also seen in some quarters as being
too close to the Catholic church, whom she had served as an advisor during
1980s. Though clearly opposed to abortion, she was far from being a
mouthpiece for the hierarchy. She favoured a much greater role for women in
the church, and soon after her election, ran foul of the bishops when taking
communion in a (Protestant) Church of Ireland service. Scallon was more
clearly aligned with religious conservatism. She was very clear that she wanted
another referendum on abortion, and this ‘single issue’ appeal was a main factor
in her ability to obtain the local council nominations in the first place. She ran as
a candidate of the outsiders, those whose voice was unheard: the 49.7% who
had voted “no” to divorce in the 1995 referendum - despite the fact that all Dáil
parties had supported the yes position.
The campaign promised to be extremely bland. Without executive or legislative
powers, the job is about being, rather than doing, so the criteria for recruitment
are perhaps what people are rather than what they have done or will do. In the
past, party considerations had been predominant but, as Michael Laver
suggested, the whole area of presidential election campaigns had become
‘uncharted territory’6 after Mary Robinson’s victory in 1990. Following Mary
Robinson’s successful campaign, in which she undertook to use her office to try
to create a more inclusive Ireland, the candidates for the first few days appeared
to be competing on their ability to empathise. At the same time, the candidates
claimed their ‘experience’ would help them do the job well. Banotti’s career as a
European politician, and McAleese’s legal training were highlighted. However,
the campaign acquired a more astringent taste with attacks on Roche by some
5
former co-workers, and later attacks from several quarters on McAleese which
accused her of having pro-republican sympathies.7
INSERT TABLE 1 ABOUT HERE
The election itself was notable for a very low turnout, 48%, down 14% on the
previous presidential election in 1990. Some attributed this to McAleese’s
comfortable lead in the several polls published in the last few days,8 although
falling turnout has been a feature of all Irish elections in recent years. 9 While
Banotti did quite well in Dublin, and actually out-polled McAleese in some of
the more middle-class areas of the city, McAleese won with 59% to Banotti’s
41% on the second count (see Table 1).10
What was the meaning of the vote?
The major question addressed by this paper is how the vote in the Irish presidential
election is to be interpreted. What was it that the voters expressed when they
returned their ballots? Two rather obvious theoretical perspectives can be brought to
bear on this. Both derive from the observation that hardly any power is invested in
the office of President of Ireland. As suggested above, it is more about being than
about doing – which implies that the election may be regarded as a kind of popularity
contest in which voters express what kind of person they want as their head of state.
To the extent that this interpretation is relevant, we would expect voters’ preferences
for presidential candidates to be largely unconnected to their political opinions, and
possibly to be even idiosyncratic in nature.11 The alternative interpretation of the
vote is that, for lack of any other relevant cues in a contest between unknown
6
candidates, voters largely use the criteria they rely on in national (Dáil) elections.
This interpretation regards the Irish presidential election as a second-order national
election, that is an election that (in contrast to the first-order national elections to the
Dàil) plays no role in deciding who governs the country, but that is nevertheless
mainly driven by national political cues and considerations. 12 This view leads to the
expectation that the preferences for candidates reflect national party preferences and
the factors that underpin them. Moreover, because these candidates were relatively
unknown (at least in their capacity as politicians) at the beginning of the campaign,
we would expect such a structure to emerge more strongly as the campaign
unfolded and voters became more aware of candidates’ party affiliations.
Our strategy for analyzing voter choices is to a considerable extent determined by
the data at our disposal. We have comparable data from no less than five surveys
that were conducted for the Irish media.13 These surveys are rich in terms of
information on the dependent variable. Amongst other things, respondents were
always asked to rank-order the presidential candidates from most to least preferred.
In addition respondents were asked to indicate their first party preference for
national parliament. In terms of questions that can be used as independent variables,
however, these surveys are relatively poor. This forces us to address our main
research question to a large extent (although not exclusively) by methods
—generically known as unfolding— that focus on the internal structure of the set of
ranked candidate preferences. Earlier research has demonstrated that such methods
can illuminate the motivational dimensions underlying voter choice.14
Intertwined through our primary focus on the motivational basis of voters’
preferences is an interest in campaign effects. By this term we do not refer to changes
7
in the standing of the various candidates in the polls (such as the precipitous decline
in support for Roche), but rather to the consequences of the campaign for voter
motivations. Campaigns provide voters with stimuli that may help them in several
ways. First of all campaigns provide straightforward information that is necessary if
voters are to express existing motivations effectively. If, for example, voters want to
support the candidate of a party they identify with or feel close to, they need to
know which candidates those are. In view of the fact that most of the candidates
were little known to the voters, we may expect the campaign to help establish these
linkages, thus enabling voters to apply partisan criteria in their choice if they wished
to do so. Second, campaigns may have (often intertwined) priming and framing
effects.15 The former alert and sensitize voters to criteria that become important
during the campaign for distinguishing candidates from one another and that
subsequently may be applied by voters in determining their own preferences. When,
in the early stages of the campaign, Roche’s former associates accused her of a
dictatorial style of leadership, this may have had the priming effect of encouraging
voters to take leadership style into consideration when looking at the candidates.
Framing refers to attempts to define specific events in a particular way, thereby
affecting the criteria that people will employ in making up their choice. Scallon’s use
of the presidential election as a suitable vehicle for furthering a new national
referendum on abortion can be seen as an attempt to frame what the election was
about. By informing, priming and framing, campaigns impinge on voters’
motivational bases, the consequences of which are then to be observed in the
structure of preferences for the various candidates. In view of the presence of
unknown candidates in an election where little if any power was at stake, we expect
the campaign to be of crucial importance in informing voters and in helping them
define what their vote was about. Therefore, we expect that, whatever motivational
8
aspects we uncover, voters’ preferences will increase in clarity and importance
during the campaign.
Unfolding models
We can detect shared voter motivations by analyzing the ‘hidden’ and underlying
structure in their candidate preferences. In order to detect such (common)
underlying structures - if they exist - we use unfolding methods. Before doing so, we
first briefly introduce these methods.
Methods that are designed for the purpose of detecting underlying common
structures in preference data are referred to as unfolding models.16 These methods
attempt to construct a single spatial representation of both persons and stimuli - in
our case voters and presidential candidates. The positions of persons and stimuli in
this so-called joint space (referred to by Coombs as the J-scale) should be such that
the distances between persons and stimuli reflect (inversely) the empirically
observed preferences of each person individually (the so-called I-scales): the more a
voter prefers a candidate, the smaller the distance between them in the spatial
representation. (In more formal terms this is referred to as the assumption of single-
peaked preferences.)17 These relations should hold to a satisfactory degree for all
voters and for all candidates. If such a space can be constructed, the position of
voters as well as candidates on the dimensions that define the space can be
calculated, and used to characterize them in further analyses. These positions can
also be used in the substantive interpretation of the dimensions, which is usually
done by taking into account other known characteristics of the stimuli and of the
9
subjects. These dimensions can be thought of as being both cognitive in character
(the J-scales indicate where voters perceive candidates as well as themselves) and
evaluative (the I-scales reflect distances between a person and the candidates on the
J-scale, which are the inverse of preferences). An interesting aspect of such
representations is that, if they can be constructed, they indicate that all voters
involved evaluate all candidates to a large degree on the basis of the same criteria.
Inability to construct such a spatial representation of the observed preferences can
indicate several things. It may signify that not all stimuli (candidates) are evaluated
on the same criteria; it may signify that voters do not have (to a sufficient degree) the
same perceptions of candidates; it may mean that not all preferences given by
individual voters are based on the same criteria.
While the number of candidates may be rather large when seen in the light of Irish
presidential elections, it is rather small for unfolding analysis. Therefore, it only
makes sense to investigate the extent to which preferences for these candidates can
be represented in unidimensional spaces.18 Various algorithms for unidimensional
unfolding exist, all of which are designed for specific types of data. Some are
designed for complete rankings (order all stimuli according to preference), some for
partial rankings (pick the k most preferred stimuli out of a pool of n), and yet others
for ratings (indicate for each stimulus the level of preference).19 Some can only
handle dichotomous scores; others can handle larger numbers of categories. In this
paper we will report results from the unfolding algorithm MUDFOLD.20 The
advantage of MUDFOLD over other available programs is that it provides a goodness-
of-fit measure — the H-coefficient— which can be used to compare the strength of a
scale over time. It also allows us to compare the strength of different scales
containing different sets of stimuli.21 H attains an upper limit of 1 if the constructed
10
scale represents the data perfectly, without any violations. If, on the other hand, a
proposed scale yields as many violations with empirical observations as would
occur in the case of statistical independence of the stimuli, H is 0.22 MUDFOLD may be
used in various ways: to test the scalability of an imposed set of stimuli in an
imposed ordering; to test the scalability of an imposed set of items without imposing
an ordering between them; and in an inductive fashion, where the program
determines which items from a given pool conform in which order to the criteria for
scalability. As will be clear from the following, we used all of these modes.
Results
We want to know three things from the unfolding analysis. First, is there a structure
hidden in people’s preferences? If there is, second, how does it evolve in the course
of the campaign? And third, how does it relate to the structure of preferences in the
first-order arena? To answer these questions we conducted a large series of
exploratory unfolding analyses on voters’ candidate preferences.23 The purpose was
to assess whether the entire set of preferences could be represented by an unfolded
scale, or if it could not, which subsets of candidates could be so represented. In
conducting these analyses, we also tried to take into account the possibility that not
all preferences were structured by common criteria - that the strongest ones were,
while weaker ones were not. In order to test for this possibility, the analyses were
conducted on four different transformations of the original preference rankings (see
also footnote 21). These transformations vary by how the complete set of preferences
is interpreted. The first one uses all information in the complete set of preferences
(the candidates’ rankings being analyzed as ratings in the MUDFOLD program). The
other three transformations are all based on simplifications of these data, made by
11
collapsing some of the original preferences. The second series of analyses
distinguishes the preferences for the three most preferred candidates, while the
preferences for the remaining two are assumed to be equal and coded to rank 4. The
third series of analyses carries this transformation further by distinguishing between
three categories: first preference, second preference, and finally third, fourth and
fifth preferences collapsed into one category. The second and third of our analyses
was each designed to allow for the possibility that a common underlying structure
determines which (two) candidates are preferred most. Our final analysis allows for
the possibility that distinctions between the two most preferred candidates, is also
determined more idiosyncratically. Hence, the fourth and final series of analyses is
based on a dichotomy in which the distinction between the two most preferred
candidates is disregarded, as both are coded 1, as are the distinctions between all
lower ranked candidates that are all coded 0.
A large number of exploratory unfolding analyses were carried out, using different
codings of preference orderings, as well as including different sub-samples of
candidates. In order to render the results of these analyses comparable over time, the
results for the best fitting scales of a subset of four candidates from the total of five
(four in the first survey) are presented here. Figure 2 shows graphically, for each of
the five surveys, the strength of the ‘best’ unfolded scale of four candidates that
could be formed for each of the four different transformations of the original
preference rankings.
INSERT FIGURE 2 HERE
12
The general patterns that emerge from Figure 2 can be summarized by two
observations. Firstly, there is very strong evidence of a single dimension underlying
voters’ preferences on polling day (the rightmost observations). The H values
indicate unidimensionality far in excess of what might occur though chance. The
dichotomous data transformations (in which the two most preferred candidates are
coded 1 and the others coded 0) and the trichotomous codings (in which the two
most preferred candidates are coded 2 and 1 respectively, and all others 0) fit the
assumptions of unidimensional unfolding much better than codings that also
distinguish between further preferences. Second, the unfolding structure strengthens
through time. The campaign effects a gradual increase in the strength (goodness of
fit) of the unfolded scales. This increase is most striking between the first and second
survey. This development applies to both higher and lower ranked preferences. This
means that not only are all voters' first candidate preferences increasingly based on
similar criteria, but so are their lower preferences as well.24
In each of the last three independently conducted surveys, a scale indicating the
ordering Roche, Banotti, Nally, Scallon turned out to yield the best fit. Because of the
robustness of this finding, we are confident that this scale represents a meaningful
ordering of the candidates that reflects a common cognitive and preferential
dimension. The probability that the same set of candidates would form unfolded
scales in three succeeding independently drawn samples when no common structure
existed in the evaluations of candidates is infinitely small. Hence we must conclude
that the set of candidates Roche-Banotti-Nally-Scallon are evaluated increasingly by
a single underlying criterion.25 However, it is not obvious what this criterion might
be. Nor is it possible to construct a satisfactory unfolded scale that reflects the
preferences for all five candidates. McAleese defies inclusion in one and the same
13
scale as the others. The addition of McAleese to the scale in the fifth survey (after
Scallon) does not really provide any clues either as to how it can be interpreted
substantively.
We will consider the significance of McAleese’s position later. Before that, we will
consider the substantive significance of the scale derived from the preferences
across the other four candidates. The existence of this scale makes it clear that voting
was not a wholly idiosyncratic exercise, but does not, in itself, tell us the nature of
the underlying structure. If a second-order perspective of Irish presidential elections
has any merit, we would expect to find the preferences for candidates structured in a
more or less similar way as preferences for parties. We already know that people’s
first preference seems to have had a clear party component. In the exit poll, 71% of
FF voters supported McAleese and 68% of FG voters Banotti. Moreover, we find this
linkage increases in strength as the campaign progresses. 26 Expressed in terms of
Cramer’s V (an association measure for nominal data) the strength of this
relationship increases as follows from the first to the fifth survey: 0.22, 0.22, 0.29,
0.31, 0.37, all of these values being highly significant, if not particularly strong. 27
Of course the unfolded scale obtained in the analyses reported above reflects not
only first preferences but lower preferences as well. What we are most interested in
here then is not the link between first party preference and first presidential
preference but between party and the unfolded scales. Analysis of this relationship
suggested there was at best a very weak link indeed. This was achieved in several
ways. First, we examined the relationship between voters’ position on the unfolded
scale 28 and their vote if there were a general election. Second, we examined the
relationship between voters’ position on the unfolded scale and their position on a
14
similar scale measuring evaluation of the parties. This scale could be constructed for
the first two of our five surveys since these contained questions about how
respondents’ evaluated the performance of the leaders of the five largest Irish
parties.29 These evaluations can satisfactorily be represented by a unidimensional
unfolded scale that reflects largely the left-right positions of the parties in the
following order: Democratic Left, Labour, Fine Gael, Fianna Fáil, Progressive
Democrats.30 The scores on this scale correlated with those derived from the
candidate preferences, but the correlations were uniformly low: eta=0.16 between
vote and unfolded candidate preference and r=0.11 between unfolded candidate
preference and unfolded party leader preference. Even when we confined the
analysis to the set of the three party-nominated candidates, Roche, Banotti and
McAleese, the same result held. Moreover, while voter assessments of party leaders
from the three parties initially responsible for nominating Roche, Banotti and
McAleese, Labour, FG and FF, unfolded clearly in both surveys (H=0.46), the three
candidates do not give rise to a scale at all. The respective H-coefficients for the five
surveys are 0.03, 0.03, -0.02, 0.01, and 0.03. All this means that the structure of
partisanship is at best only weakly related to the structure underlying preferences
across the presidential candidates. The notion that people’s choices were made as if
the election were a second-order election gets little support.
Apart from partisanship and its underlying latent structure, we also assessed the
possible impact of issues that were brought to the fore during the campaign. The
issues of abortion and Northern Ireland were clearly present. 31 In addition to this,
the acrimony at the start of Roche’s campaign might have primed voters to the
question of leadership competence. The possibility that such concerns influence
preferences for candidates can be investigated empirically for the first two of these
15
issues, since the fourth survey contains a number of items on respondents’ issue
preferences. By correlating these with respondents’ scores on the candidate unfolded
scale we can determine this influence. In neither case did the correlation even reach
statistical significance.32 Of course it could be argued that such issue considerations
were of specific importance for the preferences for some, but not for all candidates.
After all, only some of the candidates were very outspoken on each of these issues.
We may thus continue our investigation from the other end: does knowledge about
which candidates were outspoken on certain issues and which were not help explain
(parts of) preference patterns, and, if so, how did these patterns evolve during the
campaign?
We considered three different issue concerns. The first was Northern Ireland. Two of
the candidates took clear and opposite positions on Northern Ireland: McAleese was
known to support the nationalist position while of all candidates Nally was clearly
least supportive of a nationalist view of the Northern Ireland problem. The other
three candidates more or less avoided the issue and can therefore not be located
adequately anywhere on a nationalist/anti-nationalist distinction. To the extent that
this issue was an important concern for voters, the number of people preferring
McAleese and Nally as their first two preferences should be smaller than would be
expected under statistical independence. To the extent that this issue gained in
importance during the campaign, this discrepancy between observed and expected
frequencies should increase. For the abortion issue the same reasoning holds for
Banotti (of all candidates the closest to a pro-choice view) and Scallon (most
outspoken anti-abortion). A third possible concern could be the personal quality of
candidates that can be seen as ‘competence for the job’. Only Banotti and McAleese
had prior experience - political and legal respectively - which could be brought to
16
bear directly on the governmental aspect of the presidential role. If this was a
concern for voters, the number of voters preferring Banotti and McAleese as first and
second candidates should be above its expected frequency under statistical
independence. If this concern became more salient, this discrepancy should
increase.33 Table 2 presents for these three pairs of candidates the ratio of observed
frequencies of being jointly first and second preference, and expected frequencies
under the null model.
INSERT TABLE 2 ABOUT HERE
Table 2 suggests that the evaluations of presidential candidates were slightly
structured by the issues of abortion and Northern Ireland, as is evident from the
ratio being considerably smaller than 1. Moreover, the Northern Ireland issue
increased in prominence. On the other hand, the data do not indicate that the
personal competence of the candidates played an important role in voters’
evaluations of all candidates, at least insofar these were based on previous
politically-relevant experiences. However, if we inspect the main reasons that
respondents in the Lansdowne/RTE exit poll provided for their first preference
vote, we find that most voters for Banotti and McAleese stated that their candidate
was ‘most qualified for the job’. Other voters were more inclined to report issues as
the main reason for their first choice — Roche being often associated with
environmental concerns and Scallon with family, social, and religious issues.34 The
association between voters’ first choice candidate and whether the reasons given
were qualifications or issues is 0.37 (Cramer’s V). This suggests that voters employed
different considerations in the case of different candidates, but, again, this does not
17
provide much guidance as to how to interpret the common structure in evaluations
that emerged from the unfolding analyses.
All of this provides an interesting paradox. As the campaign proceeds, a single latent
dimension, except for preferences for McAleese, increasingly structured voters’
candidate preferences. The substantive meaning of this common structure remains
unclear. The fact that specific issues — Northern Ireland and abortion — did affect
some voters’ preferences does not help very much as the scale as a whole cannot be
interpreted in terms of either of these two issues, whose impact is evidently
restricted to some candidates only. If no unfolded scale had been found at all, it
would not have been difficult to ‘explain’ this. It could have been argued that the
symbolic character of the office precluded policy considerations from being of great
importance. It could have been asserted that virtually all candidates were unknown,
so that preferences contained a large idiosyncratic component. It could have been
concluded that, in the case of this particular set of candidates, the substantive
concerns raised in the campaign (such as the nationalist and abortion issues) were
each restricted in their relevance to only a few of the candidates. The obvious
problem with all these lines of reasoning is that we did find an unfolded scale, one
that was robust across all our surveys, and that increased in strength during the
campaign. If an unfolded scale had been found that reflected partisan preferences,
and their underlying ideological structure, it would not have been difficult to
‘explain’ this either. The second-order theory of elections could have been invoked to
explain the spill-over from the first-order political arena, while social-psychological
insights about cognitive cueing would elucidate why, when confronted with
unknown stimuli, people oriented themselves on the association of the candidates
with familiar objects, such as the national parties. However, these considerations,
18
particularly the second-order perspective, were inconsistent with our empirical
observations.
Finally, we checked whether demographic or socioeconomic factors could account
for differences in candidate preferences. Since Irish politics is often considered to
contain strong local concerns and identities, we estimated the extent to which voters’
position on the unfolded scale could be predicted by region.35 The association is
weak (eta = 0.22) but stronger than the relation with party preference (eta = 0.16).
Other voter characteristics are even weaker: eta for gender is 0.10, for age 0.18, and
for social class 0.11. So, regional differences seem to be related to candidate
preferences, although even here the differences are limited.
Consequently, in some ways the 1997 Irish presidential election remains an enigma.
To some extent the space in which the presidential candidates competed was multi-
dimensional, since the different issues, ideological preferences and other
considerations do not all translate into the same ordering of candidates. Given the
constraints imposed by the fact that there are ‘only’ five candidates multi-
dimensional unfolding is unfeasible. However, it is also redundant, because we
found that a single dimension emerges from the unfolding. In theory we may
conceive of this as one of the dimensions of a larger multi-dimensional space that
dominates other dimensions to such an extent that a unidimensional method yields a
perfectly acceptable result. We could hypothesize that the unfolded dimension
relates to a continuum of different political styles or personal characteristics but such
a hypothesis remains speculative as long as we are unable to specify and to test for
the existence of such styles.36 The different political styles or personal characteristics
that are connected with the opposite poles of the unfolded dimension are not
19
necessarily logically contradictory, yet empirically they may be negatively
correlated. In such cases it is possible that a candidate might combine the political
styles associated with the two opposite extremes of the scale. In our analysis such a
situation would result in non-scalability, since the respective candidate would be
preferred equally by voters at different positions on the continuum. But in terms of
competition for votes, it would yield a strong bonus for the candidate in question. In
contradistinction to the other candidates, for whom an electoral trade-off is involved
(more of a good thing for some voters yields less of an other good thing that is
preferred by other voters) the candidate that successfully combines the two political
styles will be successful among voters at different positions on the dimension. This
could be a (somewhat tentative) explanation for the McAleese's victory: she enjoyed
strong support from all voters irrespective their most preferred position on the
unfolded scale.
Discussion
In the context of a first-order election, power to govern is at stake and voters are
directed to the question as to which party they entrust with power to govern. In the
context of a second-order election little power is at stake, and voters face the
question as to what kind of ‘message’ they can sensibly express, unrestrained by the
real consequences of allocating political power. In this Irish presidential election,
however, neither power, nor sending a message to the first-order arena seems to
have been at stake. Maybe this is the consequence of a particular combination of
contextual factors that will not necessarily be repeated. Had the presidential
candidates been recruited from Dáil and government positions, had they strongly
emphasized their partisan ties, had the election taken place at the end (rather that at
20
the start) of a Dáil term, and had strong inter-party conflict reigned in the first-order
political arena, then this election might very well have turned out to be a second-
order one in the classic sense of the term. But these conditions did not exist: the
candidates were unknown; two had no ties at all with national parties; the other
three emphatically de-emphasized their party affiliation during the campaign; and
only four months had passed since the last Dáil election - so the parliamentary
election remained an accepted indicator of the current electoral strength of the
various parties.
In view of the large number of voters whose first candidate preference coincided
with their first national party preference it would be incorrect to term this
presidential election non-partisan. On the other hand, the total absence of the normal
(first-order) structure of partisan preferences in the set of preferences for candidates
casts doubt on the interpretation of these concurrences as partisan-motivated.
Perhaps the second-order election perspective is inappropriate for Irish-style
presidential elections, because of the inherently greater importance of personal
qualities of candidates in comparison to parliamentary elections and the concomitant
smaller role of partisan factors. Obviously, this suggests a need for further inquiry,
in which two aspects are of importance. One is the need for more insight into voters’
preferences for leadership styles that may help explain differences in candidate
preferences. The second is to assess comparatively the impact of such preferences for
leadership styles in presidential elections that allocate real political power (as is the
case in the USA, and, although to a lesser extent, also in semi-presidential systems
such as France and Finland) and in those that do not (as in the Irish or Austrian
cases).37
21
The second outcome of our study concerns processes of informing, framing and
priming in election campaigns. The Irish presidential election was interestingly
atypical in the degree to which the candidates were unknown at the start of the
campaign. Voters needed to acquire relevant information in a short period of time.
The costs involved in this in terms of time and energy make it rational for voters to
focus on readily available cues, such as party labels, ideological labels and the like.38
However, our analyses did not show clear signs of these information shortcuts. A
possible explanation is that the information that became available during this
campaign was not linked by political or media elites to any coherent new or pre-
existing frame of reference. Schema theory as well as theories of ‘on-line information
processing’ argues that people interpret new information within a frame of reference
that is provided by their current cognitions and integrate this new information into
their ‘working memory’.39 But such a common frame of reference was not readily
available; it seemed rather that the five unknown candidates played in different
arenas. Some took stands on Northern Ireland, while some did not. Others took
stands on abortion, while some did not. Positions on these issues coincided neither
with each other nor with left/right positions. Two candidates not being aligned to
any party and the others not emphasizing such ties undermined the relevance of
partisan cues. Consequently, what the election was about was not framed in terms of
a single definition, neither by the first-order political arena, nor by the media, and
consequently not by the voters. That may explain why priming or framing effects are
not visible in a structuring of preferences that holds for all candidates and all voters.
Yet we also saw that particularly with respect to the Northern Ireland and abortion
issues the campaign did succeed in affecting preferences for those candidates that
took more outspoken positions.
22
Finally, the established if enigmatic insight that campaigns matter because of the
information and orientation they provide has definitely been confirmed.40 First of all,
we observed a tremendous amount of net fluctuation of preferences during the
campaign, in individual (gross) terms this will be even more pronounced (Figure 1).
Moreover, we saw that the linkage between first candidate preference and first party
preference, although far from deterministic, clearly strengthens in the course of the
campaign. In addition, even in the short span of this specific campaign, a latent
dimension developed that increasingly structures preference for four of the five
candidates. As discussed above, in the absence of adequate survey instruments that
may be employed as independent variables, the nature of this dimension remains
unclear. Our story of how the Irish presidential elections unfolded thus
demonstrates (again) that voters are no fools: they are sensitive to what is at stake, to
the particular palette of options for choice, and to the information that they obtain in
the course of an election campaign. The terrain of Irish presidential elections is not
quite as uncharted as it was, but our inability as yet to explicate the nature of the
orientations that voters developed in the campaign makes any presidential election
very much an opportunity for further discovery.
23
1 K. Reif and H. Schmitt, ‘Nine Second-order National Elections. A Conceptual Framework for the
Analysis of European Election Results’, European Journal of Political Research 8 (1980), 3-44; P. Norris ‘
Second-order elections revisited’, European Journal of Political Research, 31(1997) 109-14; K. Reif,
‘European elections as member state second-order elections revisited’, European Journal of Political
Research, 31 (1997), 115-24; M. Marsh and M. Franklin, ‘The Foundations: Unanswered Questions from
the Study of European Elections, 1979-1994’, in C. v an de r Ei jk and M . Fr an kli n, ed s , C hoo s ing E uro pe ?
The Eu ro pe an Ele cto r at e a nd Na ti on al P o litic s in t he F ac e o f Uni on . (An n Ar bo r : The Uni ver s i ty o f
M ic h igan Pr es s , 1996), pp. 11-32; E.V. Opp enhui s , C. v an de r Ei jk and M . Fr an kli n ‘T he P arty Co ntex t:
Ou tco m es ’, in van de r Ei jk and Fr ank li n , C hoo s ing E uro pe ? pp. 287-306; C. van der Eijk, M . Fr ankl i n,
and E.V . Op penhui s , ‘T he S tr ategi c Co ntex t: Par ty C ho i ce’ , in v an de r Ei jk and Fr ank li n , C hoo s ing
Eu ro pe ? pp. 366-90; C. J. Anderson and D. S. Ward, ‘Barometer Elections in Comparative Perspective’,
Electoral Studies, 15 (1997), 447-60; M. Franklin, ‘Electoral Participation’, in L. LeDuc, R. Niemi and P.
Norris eds, Comparing Democracies: Elections and Voting in Global Perspective (London: Sage, 1996), pp.
216-35 ; C. van der Eijk, M. Franklin and M. Marsh, ‘What voters teach us about Europe-wide
elections: what Europe-wide elections teach us about voters’, Electoral Studies, 15 (1996), 149-66. In a
somewhat different sense see also E. A. Bakker and A. Lijphart. ‘A critical test of alphabetic voting:
the elections at the University of Leiden’, British Journal of Political Science, 10 (1980), 521-25.
2 M. S. Shugart and J. M. Carey, Presidents and Assemblies: Constitutional Design and Electoral Dynamics
(Cambridge: Cambridge University Press, 1992).
3 It was not at all clear that an election would be required in 1997. Had the incumbent Mary
Robinson chosen to run again it is likely that she would have been unopposed. Once she
declared her intention in March to take up a position with the UN, an election seemed likely,
although the possibility that John Hume, leader of Northern Ireland’s SDLP would be an
unopposed nominee remained a real possibility until early September. Only once he decided to
stay in NI politics was an election a certainty.
24
4 For more detail on the nominations and campaign see M. Marsh, ‘The Making of the Irish President
1997’, in M. Marsh and P. Mitchell, eds, How Ireland Voted 1997 (Boulder CO : Westview/PSAI Press,
1999)pp. 215-242; John Doyle, ‘The Irish Presidential Election’, Irish Political Studies Vol 13, 1998, pp.
135-144.
5 The Sunday Business Post published extracts from Department of Foreign Affairs documents on
October 12 and suggested these indicated that McAleese was a Sinn Féin supporter, although
McAleese denied this, saying her remarks had been taken out of context. A second set of leaked
documents was published by the same paper on October 19, and once again was interpreted as
indicating that McAleese had Sinn Féin leanings. On October 16 Sinn Féin President Gerry Adams
told a radio interviewer that he would vote for McAleese (if given the opportunity).
6 M. Laver, ‘Pr es i den cy can di dates f ace co nu ndr um’ , Ir is h Time s, 22 Oc to ber 1997. O n R ob in so n ’s
v ic to r y, s ee: M . Gallagher and M. Marsh, ‘The 1990 Presidential Election: Implications for the Future’,
in R.J. Hill and M. Marsh, eds, Modern Irish Democracy: Essays in Honour of Basil Chubb (Dublin: Irish
Academic Press, 1993), pp. 62-81
7 On September 21 thirteen former employees of Adi Roche’s organisation announced they would not
be supporting her, as that her style of leadership was too authoritarian and dictatorial. Attacks on
McAleese followed a story in the Sunday Business Post which alleged that she had republican
sympathies (see note 3) and came particularly from Derek Nally’s organisation and later from John
Bruton, leader of FG: see Marsh, ‘The Making of the Irish President 1997’ and Doyle, ‘The Irish
Presidential Election’.
8 Four polls were published in the last week of the campaign, all predicting a comfortable win for
McAleese. In addition to two IMS polls (see note 13) there were also two MRBI/Irish Times polls,
published on October 25 and 29. Several commentators blamed the polls for the low turnout and
there was media discussion, for instance in RTE’s The Politics Programme the following week, of the
value of banning polls in the last part of a campaign.
25
9 For details see J. Coakley and M. Gallagher eds, Politics in the Republic of Ireland (London:
Routledge/PSAI Press, 1999, 3rd edition), Appendix 2. See Marsh, ‘The Making of the Irish President
1997’ p. 229 for a discussion of Presidential election turnout.
10 One of the advantages of the transferable vote system is that we get some idea of the preference
orderings of voters from the manner in which support transfers between candidates. However, in this
instance Nally, Scallon and Roche were all eliminated together as the combined of the bottom two
candidates was insufficient to bridge the gap between third (Scallon) and second (Banotti) and hence
there are no details on whose vote transferred where. However, polls suggested supporters of Nally
and Scallon favoured McAleese, while Roche’s voters preferred Banotti. For a description of the single
transferable vote (STV) system in the Irish context see R. Sinnott, ‘The Electoral System’, in Coakley
and Gallagher, Politics in the Republic of Ireland pp.99-126.
11 The expectation of politically unstructured, idiosyncratic preferences for candidates derives from
the absence in the campaign of any reference to candidate characteristics such as masculine vs.
feminine, young vs. old, formal vs. informal, etc. All candidates attempted to outdo each other in
being empathetic.
12 Gallagher and Marsh and later Sinnott suggested this perspective was a useful one but neither
study developed the parallels very far. See Gallagher and Marsh, ‘The 1990 presidential election’; R.
Sinnott, Irish Voters Decide (Manchester: Manchester University Press, 1995).
Depending upon their timing in relation to first-order (Dàil) elections, second-order elections may
give rise to various kinds of strategic voter behaviour. This is unlikely to occur, however, when a
second-order election is conducted shortly after a first-order one, as is the case here: regular Dàil
elections were held in June, only a few months before the presidential election. In such situations
there are minimal incentives for strategic voting and maximal for sincere voting. See, e.g. Reif and
Schmitt, ‘Nine Second-order National Elections’; Marsh and Franklin ‘The Foundations’; Oppenhuis
et al. ‘The Party Context’; M. Marsh, ‘Testing the second-order election theory after four European
elections’, British Journal of Political Science, 28 (1998) 591-607.
26
13 Four of these were conducted within the six weeks prior to the election, and one is an exit poll. The
first four were conducted by Irish Marketing Surveys for Independent Newspapers Ltd, starting on
18 September, at the end of the week in which four candidates were nominated. Polls were then
conducted on 2 October, 23 October and 25 October. Each poll interviewed approximately 1100
respondents drawn from 100 sampling areas. Face-to-face interviewing took place in randomly
located homes, with selection of individuals according to socioeconomic quotas. Lansdowne Market
Research conducted the exit poll for RTE on 30 October and interviewed 2498 voters face-to-face at
150 polling stations in all 41 constituencies. Full details of all questions and responses, by various
subgroups, for all but the RTE poll may be found at the Irish Opinion Poll Archive website:
http://www.tcd.ie/Political_Science/cgi/.
The data sets used here are available at the Irish Election Data Archive website:
http://www.tcd.ie/Political_Science/elections/elections.html.
14 W.H. van Schuur, Structure in Political Beliefs. A new unfolding model with application to European party
activists (Amsterdam: CT Press, 1984); W.H. van Schuur, ‘From Mokken to MUDFOLD and back’, in
M. Fennema, C. van der Eijk and H. Schijf eds, In search of structure. Essays in social science and
methodology (Amsterdam: Het Spinhuis, 1993), pp. 45-62; J. Tillie, Party Utility and Voting Behavior
(Amsterdam: Het Spinhuis, 1995).
15 H. Semetko, ‘The Media’ in L. LeDuc, R. Niemi and P. Norris, eds Comparing Democracies: Elections
and Voting in Global Perspective, Sage, 1996, pp. 254-79.
16 C.H. Coombs A Theory of Data (New York/London: Wiley, 1964, 2nd edition, 1975).
17 Coombs A Theory of Data; van Schuur, ‘From Mokken to MUDFOLD and back’; van Schuur, Structure
in Political Beliefs.
18 Multi-dimensional spatial representations would pose so few restrictions on the data that very
different configurations would all fit the data perfectly, which implies that they would be trivial.
Unidimensional representations of 3 or more stimuli will not by necessity fit well, so that if they do
fit, a relevant empirical result has been attained.
27
19 A good overview of different unfolding models can be found in a special issue of the journal
Kwantitatieve Methoden 42 (Jan 1993) devoted to the topic.
20 For an introduction to MUDFOLD see Van Schuur, Structure in Political Beliefs; W.H. van Schuur,
‘Stochastic Unfolding’, in W. E. Saris and I. N. Gallhofer eds, Sociometric Research, Vol. I: Data Collection
and Scaling (London: Macmillan, 1988) pp. 137-58; van Schuur, ‘From Mokken to MUDFOLD and
back’.
21 Unfortunately the advantages of MUDFOLD do not come without cost: its algorithm cannot handle
full rank orders, but only partial rankings (pick k out of n) or preference ratings. Consequently, in
order to use this algorithm, we either have to transform the data so that they can be interpreted as
partial rankings, or we must treat them as if they were ratings. Transforming full rankings to partial
rankings is relatively simple. Partial orderings are derived from the full rank ordering of candidates
by giving a score of 1 to the k highest preferences, and 0 to all lower preferences (i.e.: pick 2 out of 5
yields for every respondent a string of 5 scores, one for each candidate, two of which are 1’s, the rest
being 0’s). By using different values for k (the number of ‘picked’ items) and comparing the resulting
unfolding analyses one can assess whether or not all distinctions in the original complete preference
ranking are equally informative in terms of a common structure. Interpreting rankings as ratings
involves an inversion of preference ranks: the first preference candidate is assigned the highest
preference score, and so on. Variations of such ratings can be obtained by collapsing categories.
Comparing unfolding results from these different recoded preferences helps to assess which
distinctions in the original complete preference ranking are structured in a joint scale and which are
not.
Because of the need to transform our data to analyze them with MUDFOLD, we performed all
analyses also with a different program for unidimensional unfolding, UNFOLD. As each of these
algorithms has their own particular pros and cons, we used both and compared their results.
UNFOLD can handle proper rank-order data, and thus requires no re-coding of the data. However,
because UNFOLD is a deterministic program, there is no way to test its results against a null-
hypothesis of random response, which is one of the strong points of MUDFOLD. The UNFOLD
approach has been documented by R. van Blokland-Vogelesang, ‘Unimodal Social Preference Curves
28
in Unidimensional Unfolding’, Kwantitatieve Methoden, 42 (1993), 19-38 and R. van Blokland-
Vogelesang and P. van Blokland, Unfold: Unidimensional Unfolding of Preference Data. Users' Manual.
(Groningen: Iec ProGAMMA, 1989).
Since the analyses with both packages yielded the same substantive results, there is no need to
present them both here. Both are displayed in the appendix of a conference paper: W. van der Brug,
C. van der Eijk and M. Marsh Unfolding the Irish Presidential Election 1997. Paper presented at the 1998
Annual Meeting of the American Political Science Association, Boston 3-6 September 1998.
MUDFOLD and UNFOLD are both distributed by ProGAMMA, P.O. Box 841, 9700 AV Groningen,
the Netherlands. See also http://www.gamma.rug.nl.
22 According to Van Schuur an H-coefficient smaller than 0.30 indicates insufficient structure in the
data to justify treating the individual items as indicators of a single latent dimension. H coefficients of
between 0.30 and 0.40 indicate a weak unfolding scale, between 0.40 and 0.50 a medium scale, and
above 0.50 a strong scale. See Van Schuur, Structure in Political Beliefs.
23 In the first survey, preferences for only four candidates were asked, as Nally had not yet entered the
race. As of the second survey, preferences pertain to all five candidates.
24 Although in the course of the campaign weaker preferences also become increasingly structured by
the same criteria as stronger preferences, they are nevertheless more determined by idiosyncratic
yardsticks. It would be wrong to regard the latter as not rational or not meaningful; they are rather
more individually relevant. As a consequence, such weaker preferences (in effect third and lower
rankings) will be less effective in generating political representation. One reason for this is that, when
elections are viewed as a means of mass-elite communication, the electoral ‘message’ can only be
effective if voters employ common criteria, as argued by P. E. Converse, ‘Public Opinion and Voting
Behavior’, in F.E. Greenstein and N. W. Polsby, eds, Handbook of Political Science (Vol. 4) pp. 75-170
(Reading: Addison-Wesley, 1975) and by W. van der Brug, Where's the Party? Voters' Perceptions of
Party Positions (Amsterdam: Dissertation University of Amsterdam, 1997). Moreover, when we view
elections within the ‘responsible party’ model, parties should promote public policies in accordance
with voters’ preferences. The possibility of doing so increases when the behaviour of voters and of
their representatives is structured by the same underlying ideological continuum (e.g., J. Thomassen,
29
‘Empirical Research into Political Representation: Failing Democracy or Failing Models’, in M. K.
Jennings and T. E. Mann eds, Elections at Home and Abroad: Essays in Honor of Warren E. Miller (Ann
Arbor: University of Michigan Press, 1994), pp. 237-64.
25 A colleague suggested that the increasing strength of the unfolded scales might be caused by the
fact that voters became increasingly indifferent between lower ranked candidates. This explanation
was ruled out on two accounts. First, indifference between lower ranked candidates does not affect
the analyses in which all lower ranked values are collapsed in one category (see footnote 14). Second,
indifference yields, almost by definition, randomness, which by necessity defies scalability.
26 Marsh, ‘The Making of the Irish President 1997’p. 232-3.
27 These coefficients were computed on contingency tables that included only respondents who stated
a first preference for any of the parties that formally supported one of the presidential candidates. If
we included those without party, and modeled their expected vote as one for a non-party candidate,
Cramer’s V would be a little higher.
28 We used a procedure proposed by W. van der Brug, ‘Determining Scale Values for Subjects in
MUDFOLD’, Kwantitatieve Methoden, 44 (1993), 9-20. In these and subsequent analyses we focus on the
trichotomous codings of candidate preferences (distinguishing between the first, the second, and all
subsequent preferences combined) since that transformation of the data yielded the strongest
unfolding scales.
29 The questions were generally phrased as follows: ‘Are you satisfied or dissatisfied with the way
[leader] is doing his job as leader of [party]?’ In the case of Bertie Ahern and FF the question was: ‘Are
you satisfied or dissatisfied with the way Bertie Ahern is doing his job as Taoiseach?’
30 The best fitting MUDFOLD scales in both surveys were identical and yielded the following ordering
of the national party leaders: Proinsias de Rossa (DL), Dick Spring (Labour), John Bruton (FG), Bertie
Ahern (FF), and Mary Harney (PD). In both surveys the strength of this unfolding scale is the same:
30
H=0.39. Although not overwhelmingly strong, such a value is generally considered acceptable (see
footnote 22). The unfolding scales of electoral preferences across these parties as asked in the
European Election Studies of 1989 and 1994 were considerable stronger: well over 0.50. We must keep
in mind, however, that the questions analyzed here tap party preferences in a less direct way, as they
refer to evaluations of current party leaders ‘as leaders of their particular parties’. These items thus
tap a combination of party preferences and evaluations of leaders’ performance. In spite of this
somewhat less satisfactory item content, we clearly find a structure that resembles that from earlier
studies.
31 On Northern Ireland see note 5. Scallon was well known for her pro-Life sympathies. She was asked
about this issue on numerous occasions, most notably in a set piece interview with all the candidates
on a major RTE TV programme, The Late Late Show. Alone of all the candidates she declared that as
President she would not sign any piece of legislation that made abortion easier to obtain.
32 Question wordings are as follows: Northern Ireland: ‘Articles 2 & 3 of our Constitution assert a
legal claim to the territory of Northern Ireland. It has been suggested that they should be amended to
confirm our acceptance that there will be no change to the existing constitutional status of Northern
Ireland except by peaceful means and with the consent of a majority in Northern Ireland. Do you
think Articles 2 & 3 should be left as they are or should they be amended as outlined?’ The position
against amendment is seen as the more nationalist one.
Abortion: ‘It has been argued that the Supreme Court decision in the “x” case resulted in an
interpretation of the Amendment to the Constitution which the electorate could not have foreseen
when they voted in favour of that amendment. Do you favour or oppose the idea of a further
referendum being held on the question of abortion?’ The position in favour of a referendum is seen as
the more strongly conservative.
33 The expected numbers under statistical independence were computed with an algorithm
incorporated in MUDFOLD.
34 For details see Marsh, ‘The Making of the Irish President 1997’ p.236-7.
31
35 10 regions were distinguished.
36 It must be noted that in order to understand the substantive meaning of a latent dimension such as
the one found in this study, one should not only be able to label it, but also to regard it as a
continuum of different ‘ideals’ of voters. Consequently, this unfolding dimension cannot be about
‘competence’ or ‘honesty’, as it is difficult to conceive that voters exist who would prefer an
incompetent or dishonest candidate.
37 For a review of work on these themes see I. McAllister, ‘Leaders’, in LeDuc, Niemi and Norris,
Comparing Democracies pp. 280-98.
38 A. Lupia and M. D. McCubbins, The Democratic Dilemma (Cambridge: Cambridge University Press,
1997).
39 See, e.g., S. Feldman and P. J. Conover, ‘Candidates, Issues, and Voters: The Role of Inference in
Political Perceptions’, Journal of Politics, 45 (1983), 812-39; P. J. Conover and S. Feldman, ‘How People
Organize the Political World: A Schematic Model’, American Journal of Political Science, 28 (1984), 95-
126. See in this context also theories of ‘on-line information processing’: M. Lo d ge, K. M c Gr aw an d P.
S tr o h. ‘An I mpr es s i o n-d ri v en M o del o f C andi date Ev aluati o n’, A me ric an Po litic al Scie nce R evie w 87
(1989) , 399-419.
40 A. Gelman and G. King, ‘Why are American presidential-election campaign polls so variable when
votes are so predictable’, British Journal of Political Science, 23(4) 1993, pp. 409-51.
1
Appendix
This appendix gives an overview of the results of Mudfold in which the best
unfolded scales are sought containing preferences for four (out of five) candidates.
These results are described briefly in Table A.1. The H-coefficient is the most
important diagnostic for deciding whether a set of items can be interpreted as
indicators of a single latent dimension. If the value of the H-coefficient is lower than
0.3 the items do not form a scale, values between 0.3 and 0.4 indicate a weak scale,
values between 0.4 and 0.5 indicate a moderately strong scale (W.H. van Schuur,
Structure in Political Beliefs. A new unfolding model with application to European party
activists (Amsterdam: CT Press, 1984)).
2
Table A.1: summary description of main results of MUDFOLD analyses
Values of the H-coefficients
codings of
candidate
preferences
permutations 1st
wave
2nd
wave
3rd
wave
4th
wave
5th
wave
full rankings:
0, 1, 2, 3, 4
DAEB
CADB
-.11
* -.08 .09 .13 .25
0 (twice), 1, 2, 3 DAEB
ACED
DACE
-.11
*
*
.02
.01 .10 .13 .40
0 (3 times), 1, 2 DAEB
DECB
DACE
.12
*
*
.37
.30 .43 .46 .42
0 (3 times), 1
(twice)
DAEB
ADCE
DACE
-.09
*
*
.26
.22 .43 .47 .40
(A = Banotti, B = McAleese, C = Nally, D = Roche, E = Scallon)
* = cannot be estimated because Nally was not then a candidate
3
FIGURES AND TABLES
Table 1: First and second count vote shares
Banotti McAleese Nally Roche Scallon
1st Count 29.3% 45.2% 4.7% 7.0% 13.8%
Transfers + 12.0% +13.5% -25.5%
2nd count 41.3% 58.7%
Source: Seán Donnelly, Elections ’97 (Dublin: Seán Donnelly, 1998) p.433.
4
Table 2: Number of respondents ranking two candidates first and second divided by
the expected frequency if preferences were statistically independent.
18 Sept 2 Oct 23 Oct 25 Oct 30 Oct
Northern Ireland
(McAleese vs. Nally)
Nally not yet
running 0.92 0.75 0.72 0.63
Abortion
(Banotti vs. Scallon) 0.78 0.74 0.73 0.65 0.73
Competence
(Banotti and
McAleese)
1.01 1.03 0.97 0.99 0.91
Source: calculations on the basis of the IMS and Lansdowne data detailed in note 13.
Note: cell entries indicate the number of voters who were ranked two candidates
first and second (or second and first) as a proportion of the number that would be
expected to do so on the basis of statistical independence of candidate preferences.
For example, in the survey conducted on the 2nd of October 77 respondents ranked
McAleese and Nally as the first two candidates (Nally first and McAleese second or
vice versa). The expected frequency under the null model of statistical independence
(as computed by MUDFOLD) was estimated at 83.3. The value 0.92 is the ratio of 77
and 83.3. The further the proportion from unity, the greater the apparent impact of
the issue on which the two candidates were divided (Northern Ireland, Abortion) or
united (Competence).
5
Figure 1: Campaign polls showing each candidate’s share of (first) preference votes.
0
10
20
30
40
500 2 4 6 8
10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44
Date
Nally
Scallon
Roche
Banotti
McAleese
18 Sept30 Oct
Supp
ort