Turnout and abstention at multi-level elections in Great Britain

FIFTH FRAMEWORK RESEARCH PROGRAMME (1998-2002)
Democratic Participation and Political Communication in Systems of
Multi-level Governance
Turnout and abstention at multi-level elections
in Great Britain
Colin Rallings and Michael Thrasher
Department of Politics and International Relations
University of Plymouth
Work in Progress
March 2003
Draft text not to be quoted without permission of the author(s).
Introduction
The principal objective is to examine differences of turnout and, by implication, abstention at multi-level
elections within Great Britain. We develop four basic approaches to the study of democratic participation at
parliamentary, local and European elections, and, for Scotland and Wales, elections to the devolved assemblies.
First, we describe for the period 1979 onwards, some general patterns and trends in turnout at four levels of
governance are outlined. Regression models covering three types of aggregate electoral data (parliamentary,
European and local) are constructed using percentage turnout as the dependent variable. Second, continuing the
analysis of aggregate voting data, the analysis utilises a method of ecological inference to determine both the
pattern of electoral movement between different elections and how different social groups may be responding to
the decision to vote or not at these elections. Third, the focus moves away from aggregate towards survey data,
initially examining the findings of Eurobarometer 52, which sought to highlight factors that may have impinged
upon electors decision to participate at the 1999 European elections. Fourth, remaining with survey data, we
examine the data for Great Britain produced by the Asia-Europe survey, conducted in early 2001. This survey
probed further behind the decision to vote, not just at European elections but also at parliamentary and local
elections. It offers a rare opportunity to compare how people respond to the different circumstances of
democratic participation in multi-level governance.
Before moving to the examination of these themes,
however, we provide a fuller description of the data to be used in the analysis and a brief overview of electoral
procedures and practices in that might have a bearing upon levels of electoral turnout.
1
Data
A range of aggregate and survey data is used to explore the determinants of turnout across different types of
elections. These data are general election results between 1979-2001; results from European elections 19791999; election results for English local authorities 1979-1999; elections to the newly devolved assemblies in
Scotland and Wales. Where appropriate, these data are augmented by national census information drawn from
surveys in 1981 and 1991. Although parliamentary and European constituencies are constructed from local
electoral wards neither the support for parties nor percentage turnout are normally provided at levels lower than
the constituency. In most cases, therefore, comparisons of turnout are provided at the constituency or higher
(e.g. regional) levels. However, it is possible, for the 1999 European Parliament elections only, to have electoral
and turnout data at the parliamentary constituency level.
In contrast with other countries of the EU analysis of aggregate electoral data for Great Britain has to contend
with difficulties that arise following periodic and large-scale adjustments to electoral boundaries. In the period
under consideration parliamentary constituency boundaries, for example, experienced two major and one minor
boundary reviews. In effect, this means that it is difficult to provide uninterrupted constituency level data across
a twenty year period. Similar effects also operate for European and local elections. Such alterations do not
simply provide problems for the analysis of election results and turnout. Additional problems are created when
trying to produce socio-economic data from the national census at the local ward level. Two censuses were
conducted in 1981 and 1991 but it has not always been possible to combine these data with local wards. This is
because either the census has been conducted in a new ward when elections are still being conducted using old
wards or while the census has been taken in existing wards a new ward has immediately been used for electoral
purposes. Despite such problems there remain a sufficient number of cases that aggregate data analysis has
proved possible.
Augmenting the analysis of aggregate data are two opinion surveys that are used to shed light on individual
level characteristics and motivations for voting. First, Eurobarometer 52, which surveyed electors shortly after
the June 1999 European elections and sought to identify differences in characteristics of voters and non-voters
as well as the impact on the voting decision of the European election campaign and media coverage. The
second survey, conducted in February 2001, comprised a multinational study covering 18 countries, including
Great Britain. This survey was broader in its coverage of voting than Eurobarometer 52 in that respondents
were asked questions regarding the frequency of voting at general, European and local elections.
2
Institutional facilitation and electoral procedures
Simple plurality, single ballot voting is used for local and general elections, a feature shared until very recently
with elections to the European parliament. Prior to the 1999 elections, however, PR-List voting was introduced
for the European contests. Instead of comprising some 84 single-member constituencies Britain was now
divided into nine separate regions with varying district magnitudes. Scotland and Wales each formed separate
regions with England divided into nine regions. The smallest English region was the North East (4 seats) while
London, North West (10) and the South East regions (11) formed the largest electoral areas. Despite pressure,
the Home Office opted for a closed list system. Northern Ireland’s three seats in the European parliament
continue to be elected by the Single Transferable Vote system. Elections in 1999 to the newly established
Scottish Parliament and Welsh Assembly were held using the Additional Member System (AMS). Voting for
all elections is held on a Thursday, although the count for the European elections is held on the Sunday
following when polling has ceased across all member states. For general, European and devolved assembly
elections polling takes place between 0700-22.00 hours but for local elections polling stations open at 0800 and
close at 21.00.
The number of voters using a postal ballot for general elections was consistent between 1979-1997, ranging
from between 2-2.4% of all votes cast.
Following the 1997 general election, however, the government
introduced legislation, which relaxed the previous restrictions for obtaining a postal vote and in effect now
permitted a postal vote on demand. Unsurprisingly, this had the effect that in 2001 no fewer than 5.2%, or one
in twenty voters, used this method to cast a ballot. Those that seek and obtain a postal vote are more likely to
vote than those that rely upon a visit to a polling station. Indeed, for general elections the turnout amongst those
with a postal vote has been consistently above 80% since the end of the Second World War. Although
comprehensive data for postal votes are unavailable for local elections they are published for European
elections. These show that, for the first direct elections in 1979, postal ballots accounted for 4% of the total vote
but since then the percentage has fallen, accounting for 2.1% in 1994 and 1.7% in 1999. Of course, for the 2004
European elections there will be postal voting on demand so we should expect to see an increase in the
proportion of voters casting a ballot by post. In order to boost electoral turnout further the government has
determined that the local elections, normally due in May, will be postponed for a month and held simultaneous
with the European elections.
There are significant differences between parliamentary and European elections both in the proportion of
electors seeking a postal vote and in the subsequent turnout amongst these electors. Putting aside the recent and
unusual case of the 2001 election it is generally the case that 2% of the electorate apply for a postal ballot prior
to a general election but with the single exception of 1979 the percentage for European elections barely rises
above 1%. Moreover, while more than eight in ten applicants for a general election postal ballot do then vote
the percentage for European elections falls from a high of 66.3% in 1979 to a low at the most recent election in
1999 of 57%. In short, as a proxy measure of interest in the election, the evidence from postal voting suggests
that a priori interest in European elections is substantially lower even before the official campaign has properly
got under way and that interest amongst the normally committed postal voters is correspondingly less since they
3
turn out to vote in lower proportions. Nevertheless, results from recent experiments conducted in 2000 and
2002 for local elections suggest that a greater use and flexibility in the use of all-postal ballots would
significantly and positively influence levels of turnout at European parliament elections.
Those electors that choose to vote in person would not notice any difference in the location and relative density
of polling stations for each type of election. In practice Electoral Administrators, located in each local authority,
are responsible for determining where, and how many, polling places to use. There are no strict regulations
regarding how many polling stations are required for a given population size. Polling station density is less for
rural and suburban areas than is the case for more densely populated areas. These polling places are then used
for general, European and local elections alike. The ease of voting, therefore, is not a factor that is significantly
affected by changing patterns of polling station location and density.
It was only recently for the 2001 general election that spending limits have been introduced for national
electoral campaigns. Before then parties were free to spend as much as they liked but there are strict limits to
the amount each candidate can spend fight his or her constituency. The candidates’ personal spending limit
imposed for general elections is higher than the equivalent for European election, suggesting that institutional
mobilisation is greater for the former than the latter. Moreover, although national campaign expenditures are
always estimated, parties spend considerably more fighting general than European elections.
Campaign
coverage, and by implication, impact, is certain to be affected by this pattern of differential expenditure.
Electoral participation and abstention across levels of governance
Turnout in Great Britain generally lags behind the rest of Europe and even from a relatively lower base have
been in decline over the period (see Figure 1). In 1979, for example, turnout at that year’s general election stood
at 76.2%, peaking at 77.9% in 1992 but then falling very rapidly to just 59.1% at the most recent contest in
2001. That represents a 17-percentage point decline in just two decades.
Local elections in England use a complex four-yearly cycle and to derive the underlying patterns of electoral
turnout a number of adjustments have been made. First, unusually high local turnouts were recorded in both
1979 and 1997 because local elections were fought simultaneously with general elections held in those years.
Omitting those exceptional cases from the analysis presents a clearer picture of the underlying trend. Second,
each year local elections are held in different parts of the country and there are sharp distinctions in socioeconomic composition of these authorities. Generally, more prosperous local authorities enjoy higher levels of
electoral participation than do poorer areas. A straightforward year on year comparison, therefore, suggests
some sharp rises and falls in turnout when in reality such differences largely mask differences in socioeconomic composition. Accordingly, we have generally used a moving four-yearly average to calculate annual
local electoral turnout, the exceptions being the two general election years when the average of the previous
three years has been used instead. The result is a smoothed, though notably declining, trend in local electoral
turnout.
4
Turnout at European Parliament elections has been lower in absolute terms than that found at local elections but
the trend line corresponds with the declining pattern identified for the other election types. For first direct
elections held in 1979 turnout across England measured just 31.8%, a figure repeated in 1984 before rising to a
peak of 36.8% in 1989. A modest decline in 1994 was then followed in 1999 by a precipitous decline to just
22.7%. In short, turnout at European elections has fallen by almost a third since MEPs were first directly
elected. In 1999 the first elections to the Scottish Parliament and Welsh Assembly took place. In Scotland,
where the parliament has some limited primary legislative powers, turnout was 58.2%, thirteen percentage
points lower than the turnout in that country for the 1997 general election but identical to that seen at the 2001
general election. Turnout for the Welsh Assembly elections at 46.4% was lower but its powers are considerably
less than that of its counterpart in Scotland.
Comparisons of Electoral Turnout
Parliamentary, European, Local and Devolved Elections
90
80
70
Turnout %
60
58.2% S.
50
46.4% W. S. Scotland
W. Wales
40
30
Local
20
European
10
1979
General
1983
1987
1991
1995
1999
Figure 1: Decline in turnout
Figure 1 suggests a pattern of declining turnout at the national level affecting each of the different types of
election. We can test further for this common pattern by looking at the relationship between turnout at lower
levels of aggregation, namely regional and parliamentary constituency level. If common factors are at work in
determining the level of turnout for different types of elections then we should expect to find positive and high
correlation for turnout at general, European and local elections. If, however, there were particular factors that
operated in some regions/constituencies but not in others and for different elections (for example, a relatively
high turnout in some constituencies or regions for local and/or European elections) then weaker correlations
might be expected.
5
Figure 2 shows turnout aggregated to nine Government Office English regions for the last two general elections,
the 1999 local and European elections.
No account is taken here of differences in the socio-economic
compositions of these regions. Although the absolute level of turnout varies slightly between regions within
regions there is a fairly consistent pattern for the different types of election. The South East of England, for
example, normally has a turnout marginally higher than the national average, regardless of the nature of contest
while the North East consistently has turnouts that are marginally below the national average. In aggregate
terms, therefore, voting at the regional level reveals no significant differences in turnout according to the type of
election, local, European or parliamentary, being fought.
80.00
?
?
?
?
?
?
?
>
60.00
>
>
>
>
>
>
>
50.00
40.00
K
K
K
K
K
K
K
K
K
K
30.00
K
K
&
&
&
West Midlands
&
South West
&
South East
London
Eastern
East Midlands
20.00
North West
&
&
North East
&
&
&
K
&
&
GO region
Figure 2: Voter turnout in English Regions
6
&
Yorks&Humber
Turnout
>
ge97
ge01
loc99
&
euro99
?
?
70.00
?
>
K
Error Bars show Mean +/- 1.0 SD
These findings are supported by a comparison of turnouts at the parliamentary constituency level, i.e. European
elections disaggregated, local elections aggregated). Figure 3 and Table 1 show turnout for different elections,
although varying in absolute terms, are highly correlated, suggesting that electors within constituencies behave in
consistent ways. Indeed, even for local elections, where sometimes the main parties may not present candidates,
the correlation with these elections and those for the national European parliament is .6 or higher. That suggests
that at least some of the factors influencing the decline in turnout at one type of election are also at work elsewhere.
Moreover, the relatively high correlation (0.849) between the European elections in 1999 and that of the general
election held two years later means that the difference in turnout between these two elections (some 35 percentage
points) should be explained by factors of a broad, rather than constituency-based, nature.
Scatterplot of LOGITs for 1997 G.E. and 2001 G.E.
t
t
1.00
Linear R egression
Linear
RIndividual
egressionPrediction
with
ith
95.00%
i 95.00%
t
l Individual Prediction
Interval
$
$
$
$
$
$ $$
$ $$ $ $$ $$
$ $
$$ $
$$$ $$
$
$$
$ $$ $ $$
$$ $ $
$
$ $ $$
$ $$ $
$
$ $ $$$$$$ $$$$ $$ $$$ $$
$ $$$ $ $$ $$$ $
$$ $ $
$
$
$
$
$ $$
$ $$ $$$$
$$$ $$$ $
$$$$ $
$$ $ $$ $$$$$$$$$
$
$
$ $
$$$ $$$ $$$$ $ $$$
$ $$ $ $$$$$
$ $$$
$ $
$
$$$ $$$$ $$$$ $
$ $$$$$$ $ $$$
$
$ $$$$$ $$$$ $ $ $$
$
$ $$$$$$$$$ $$ $ $
$ $$$$ $$$ $$$
$ $ $ $ $$$$$
$$
$$$
$$ $$ $$
$ $ $ $$$$$$$$$
$
$$$$ $ $ $ $$
$$ $$ $
$ $$ $
$
$ $ $ $$$$ $
$
$
$
$
$
$
$
$
$ $ $$$
$
$
$ $ $
$ $ $$ $$ $$$
$$$ $$ $$$ $
$ $ $$ $$$ $ $
$
$
$
$
$
$$$ $$
$ $ $ $ $$
$ $ $
$ $
$ $ $$$$ $
$$
$
$
$ $
$
$
$ $ $ $$ $$$ $ $ $ $ $
$
$$
$
$ $$ $$ $$$ $$ $
$$ $ $ $ $
$
$ $ $$ $
$$$$ $ $ $
$$ $ $$ $ $ $
$
$$ $ $
$$ $
$ $$ $
$ $ $ $$
$
$
$$
$ $
$
0.75
0.50
log
it0
1
0.25
$
0.00
$
$ $
$ $$
$
$
$
-0.25
$
$
$
$$
$
$$ $
$
$ $ $
$
$ $ $
$$
$
$ $$
$$
$
$
Scatterplot of LOGITs for 1997 G.E. and 1999 Euro turnouts
l ogit01 = -0.47 + 0.91 * logit97
$
$
-0.75
R-Square = 0.84
-1.00
log
it99 -1.25
e
$
-1.50
$
-1.75
$
$
-2.00
$
-0.50
$
$
$ $
$
$
$$
$
$$ $ $$$ $
$$$
$ $$ $ $
$$ $
$
$ $ $ $$$
$
$ $$$ $
$$ $$$ $ $ $
$
$ $$ $ $ $
$
$
$
$
$
$$$ $$$ $$$$ $$$$$$
$
$
$ $$
$$$$$$ $$
$$ $ $ $ $
$$
$
$ $$
$
$
$$$$$$$ $$$$$$ $$
$$ $$$$
$$ $ $$ $$$
$
$ $$$
$$ $
$
$$$$ $ $ $$$
$$
$
$
$$ $
$$$ $$ $$$$$$$
$$ $
$ $$$
$$$$ $$$$$$$$ $ $$$
$$
$ $ $$$ $$$ $$$$ $
$
$
$
$
$
$$ $$$$
$ $ $ $$$$ $
$ $
$
$
$ $
$$ $$ $ $
$$
$$$$
$ $ $
$
$
$ $ $ $$
$$ $$ $$$ $ $ $ $$
$
$$ $
$
$
$ $$ $
$ $$ $ $$
$ $ $$$$$$$ $ $ $$$$$$$ $$
$$$$$
$$ $
$
$$
$
$ $$
$ $$$
$$
$$$ $$$$ $ $
$
$ $ $ $
$
$ $ $$$ $$
$$$ $
$
$ $$$$$$ $ $ $$ $
$
$ $$$ $
$$$
$
$ $$$
$
$$
$
$
$
$ $$ $$$$
$ $
$ $
$
$ $ $$
$
$
$
$
$
$
$
$
$
$
$
$$$
$$
$
$$ $$ $ $$$ $ $$ $
$
$
$
$$ $$ $
$
$
$
$
$
$ $ $ $ $ $
$
$$$$$$ $ $ $$
$
$$$
$
$$
$
$
$
$
$
$
$$ $
$ $
$$ $
$ $
$
$ $
$
$
$$ $ $$
$
$$
$
$
$
$
$
$
$ $
$
$
$
$ $
$
$
$
logit99e = -1.97 + 0.77 * logit97
R-Square = 0.53
$
$
$
$
$
-2.25
$
0.25
0.50
0.75
1.00
1.25
0.25
logit97
1.00
$
$
$$
$ $
$$
$$ $ $ $ $
$ $
$ $
$
$
$
$
$$ $ $$$ $$ $
$
$
$
$$ $$$ $ $
$
$ $ $ $$ $$$$ $$$
$
$
$$ $ $$ $ $ $
$ $$
$$$ $$$ $$
$ $$$$ $$ $$$$$
$$
$ $ $
$$
$$
$$$ $
$$$$$$$$$$
$
$ $$$ $$$ $$$$$$
$
$$$$
$
$ $$$ $$$ $$$ $$ $
$$ $$$$ $$$$ $
$$$
$ $$ $
$
$$
$$$
$$
$ $
$
$$$$$$$$ $$
$$$$ $
$
$ $
$
$
$
$
$
$
$$ $$ $$$$
$$ $
$$ $$$ $$$$$$$$$$ $$
$$ $$$$
$$
$$$$$ $ $$ $$ $
$$ $
$ $$ $$$$$$$ $ $
$$
$ $ $$$$$$
$$ $
$$ $ $ $
$
$$
$$$ $ $
$ $$$ $
$ $ $$ $$$ $
$ $ $$$$$$$
$$ $$$$ $ $$
$
$$ $ $ $$ $$$$$ $$
$$
$
$
$$
$$$$$ $
$$ $$$
$$ $ $$
$ $ $ $$$ $ $ $$$
$
$
$ $ $ $$$$ $ $$ $ $
$
$$
$
$
$ $ $ $$
$
$
$
$
$ $$
$ $ $ $$ $ $ $$$$$ $ $ $ $
$
$ $$ $
$
$$
$$ $
$ $
$ $$
$
$
$ $$
$$
$
$ $ $ $$$
$ $
$ $$
$
$
$ $$ $$ $$
$ $ $ $ $ $$
$$
$$
$
$ $
$ $ $ $ $$ $
$ $
$
$ $
$
$
$
$
$$
$$
$
$
$
$
$
$
$ $
$
$
$
$
$
$
0.75
0.50
0.25
0.00
$
-0.25
$
0.50
0.75
1.00
1.25
logit97
Scatterplot of LOGIT s for 1999 Euro and 2001 G.E. turn
log
it01
Linear Regression with
Linear Regression
with
95.00% Individual
Predic
95.00% Individual
I t
l Prediction
Interval
-0.50
$
Scatterplot of 1999 Local and European Election Turnou
Linear R egression with
Linear R egression
95.00% Individual Prediction
ith
Interval 95.00% Individual Pr
I t
l
-0.25
logit01 = 1.37 + 0.80 *
logit99e
-0.50
Linear Regression
95.00% Individual
Linear Regression with
95.00% Individual Prediction Interval
R-Squ are = 0.72
-0.75
lo
git
99
e
-1.00
-1.25
-1.50
-1.75
$
$
-2.00
-0.50
$
$ $
$ $
$
$
$$
$$ $ $$
$ $$$ $ $
$
$
$ $
$ $ $$$
$ $ $$ $$ $$
$
$
$$ $$ $$$ $$$$$$$ $
$$$ $$$$$$$$$$$$$$$$$$ $
$$ $ $$
$ $$$ $$$$ $ $$$$$
$$
$$$$$$$$$$$$ $ $$
$ $
$$$$
$
$
$
$
$
$
$
$
$$ $ $ $$ $
$$
$
$ $ $$$$$$
$$$$$$
$ $ $$$$$$$$$$ $
$$ $$$$
$
$ $ $ $$ $$$$$$$$$$ $$$ $
$ $ $$$
$ $$$$$ $
$$ $ $$
$$ $$ $$ $ $
$ $$
$
$
$
$
$$ $$$$$$$ $$
$
$
$
$ $$ $$$$$$ $ $ $ $ $
$$$$ $$$$$$$$
$
$$$$$$ $$
$$$ $$$$
$$$ $$$
$ $ $$ $ $$$$$
$
$$$$ $ $
$$$ $$$$ $$$ $$$ $$ $
$
$$$$ $$ $ $ $ $ $$$$$$$$ $$ $ $ $
$$$$ $ $
$ $$
$ $$ $ $$$$$$
$
$$ $$ $
$
$$$
$
$
$
$$$ $ $
$ $$
$$$$
$
$ $ $$
$
$ $$$ $
$$
$$$ $ $$$ $$ $ $
$ $ $$ $ $$ $$$$$ $$ $
$
$
$$
$ $$$$ $$ $$$ $ $ $ $ $ $ $
$
$
$
$
$
$
$
$ $ $ $$
$
$$
$
$
$ $$ $
$$$ $
$
$$ $$
$
$
$
$$ $$
$
$
$$$$
$ $
$
$
$
logit99e = -0.68 + 0.78 * logit99l
R-Square = 0.67
$
$
$
$
-2.25
-2.00
-1.75
-1.50
-1.25
-1.00
-0.75
-1.50
logit99e
-1.25
-1.00
-0.75
-0.50
logit99l
Figure 3: Scatterplots of electoral turnout (logits) for different types of election
7
-0.25
0.00
0.25
Table 1:
Logits of Electoral turnout at General, European and Local Elections by Parliamentary
Constituency (Pearson correlation coefficients)
1997 G.E.
2001 G.E.
1999 local
2001 G.E.
Pearson Correlation
.916
Sig (2-tailed)
.000
N
529
Pearson Correlation
.731
.849
Sig (2-tailed)
.000
.000
N
529
529
Pearson Correlation
.607
.743
.820
Sig (2-tailed)
.000
.000
.000
N
529
529
529
1999 Euro
1999 local
Next, we show in Table 2, regression models (relating to two general elections 1997, 2001; two European
elections 1994 and 1999; local elections for 1999). Independent variables include socio-economic data on age,
class and deprivation drawn from the national census, closeness of party competition as well as dummy
variables for English regions with the South West as the reference category. The R2 values range between .83
for the 1999 European and 2001 general elections and a low of .67 for the local elections. Not all independent
variables appear significant in each of the equations, however, suggesting that differing influences are at work in
influencing the general level of participation at each election. For example, the abstention rate amongst younger
people is statistically significant for parliamentary elections but that relationship does not hold for other types of
election. By contrast, constituencies that have higher proportions of older voters (65+) are more likely to have
higher turnout for local elections. However, relative levels of affluence or deprivation appear to have more
consistent effects across the range of elections. Rural areas appear more associated with voting at European
elections as do the London and South West regions.
8
Table 2: Electoral turnout at parliamentary, European and Local elections
GE97
R2
79%
(77%
without
gocode)
(Constant) 73.2
GE01
LOC99
83%
(80%)
EURO99
67%
(57%)
EURO94
83%
(76%)
73%
.000
61.2
.000
21.3
.000
32.3
.000
35.8
.000
-.91
.000
-.89
.000
-.15
.251
-.53
.028
-.05
.953
-.11
.041
-.01
.877
.37
.000
.20
.005
.58
.035
V18_1 class1+class2
.05
.001
.12
.000
.19
.000
.20
.000
.15
.176
V11
-.26
.000
-.20
.000
-.08
.000
-.04
.097
-.24
.001
MAJ "majority"
-.08
.000
-.12
.000
-.01
.500
-.11
.000
-.06
.110
RURAL
-.42
.167
2.0
.000
2.8
.000
4.3
.000
2.1
.012
EMIDLAND
1.1
.029
.58
.278
.52
.184
1.5
.028
EASTERN
-.24
.628
-1.0
.040
.60
.109
-.71
.300
LONDON
-.99
.050
-.87
.095
2.3
.000
5.0
.000
NORTH_E
2.3
.000
2.4
.000
.32
.516
2.2
.013
NORTH_W
1.1
.017
-.99
.041
-1.2
.000
-.84
.191
SOUTH_E
-1.1
.015
-1.4
.004
-.01
.978
-1.5
.015
SOUTH_W
.57
.274
1.6
.002
2.4
.000
2.8
.000
YORK_HMB
-.54
.270
.06
.906
-1.0
.005
.18
.781
V14
Age 18-24
V17_2
Age 65+
no car
Note: We were unable to use the same regional codes for the 1994 European elections. Data are for England only.
Electoral participation and abstention: Ecol Analysis
We used data from four elections in order to compile voter transition matrices. These are two general elections,
1997 and 2001, the 1999 European Parliament elections with the fourth type referencing local elections.
Because of the different patterns of party competition we exclude data for Scotland and Wales. Additionally,
we use some socio-economic data to examine patterns of EU voting/abstention.
The unit for analysis is the parliamentary constituency and obviously no modification (aggregation or
disaggregation) of the data are required in the cases of the two general elections. There are a total of 529
parliamentary constituencies covering England for these years. Regarding the 1999 European election this was
9
the first occasion that PR-List voting was used. For this purpose England was divided into a total of nine
electoral districts covering the various regions. A total of 71 seats were determined in this manner with district
magnitude ranging from 4-11 seats. None of the European electoral districts cut across a parliamentary
constituency boundary.
However, we were able to obtain European voting data disaggregated to the
constituency level.
In the case of local elections the procedure for aggregating data into parliamentary constituencies proved
somewhat more complex, largely because of the nature of the electoral cycle. Local elections are fought in
wards but not all wards hold an election in any given year. Therefore, in order to obtain as comprehensive a set
of results as possible, data from three elections years, 1998, 1999 and 2000 were used for the purposes of
aggregation, with the majority of wards holding elections in May 1999, just one month before the EU elections.
General Election Voting, 1997 and 2001
The first Ecol analysis compares voting/abstention at the two most recent general elections held in 1997 and
2001 respectively (see Table 3). Our purpose here is to examine the implied movement of voters between these
two elections and eventually to compare those findings against evidence acquired from the British Election
Study, 2001. We employ four party categories (Conservative, Labour, Liberal Democrat and ‘Other’) and
abstention. The actual vote for minor parties grouped here into ‘Other’ was 4.8 per cent in 1997 and 4.0 per
cent in 2001. We should recall that the turnout (England only) in 1997 was 71.4 per cent, falling to just 59.1 per
cent in 2001.
The first part of Table 3 presents row percentages and suggests that 82 per cent of Conservative 1997 voters
remained loyal to the party at the 2001 general election, with fewer than one per cent defecting to Labour. Ecol
calculates that one in eight of Conservative supporters in 1997 abstained at the 2001 general election. Labour’s
support was less solid. Less than three-quarters of Labour voters in 1997 continued to support the party at the
2001 election (71.5%) with 27 per cent of its 1997 support choosing instead to abstain. The profile of the third
placed party, the Liberal Democrats, lies somewhere between the two main parties with 76% of former Liberal
Democrat voters continuing to support the party at the 2001 election and a further 17% instead abstaining. It is
calculated that no fewer than 86% of those that abstained in 1997 did so again in 2001.
Turning to examine the column percentages we find that the parties’ 2001 vote largely comprised voters that had
been supporters at the previous general election. The Conservative party, for example, received slightly less
than 95% of its 2001 vote from previous Conservative voters. The comparable figures for Labour and Liberal
Democrat are 90.7% and 84.8% respectively. This suggests that the 2001 general election was fairly stable in
terms of switching between parties and that no party was able to transform its support base. The column
representing 2001 abstainers is interesting, showing as it does that six out of ten abstainers in 2001 had also not
voted at the 1997 election whilst one in five had voted for Labour in 1997. Certainly, this is confirmed in some
part by the aggregate voting figures. Turnout in 2001 was the lowest at any general election since 1918, falling
most in Labour held constituencies. No major party received more votes in aggregate in 2001 than in 1997.
10
Incidentally, we repeated the Ecol analysis dividing the data by region but the findings suggest considerable
homogeneity in voting behaviour (Tables not reported here).
Survey data, drawn from the various British Election Studies support, in part, the ecological inference. The
2001 BES, for example, estimates a lower abstention rate amongst 1997 Labour voters (17.5%) which Ecol
estimates at 26.9% but the figures for Conservative supporters (16.6% of 1997 Conservative voter abstained in
2001) and Liberal Democrat voters (13.2% abstained in 2001) are directly comparable to the inferences obtained
from ecological data.
Table 3: Voting and Abstention at 1997 and 2001 General Elections
Voter
General
Con97
Lab97
LD97
Oth97
Abst97
Voter
General
Con97
Lab97
LD97
Oth97
Abst97
Mobility.
England.
Row
General:
Con01
Lab01
81.8
0.8
0.3
71.5
5.3
0.2
4.2
4.4
0.9
6.6
20.8
24.5
Pct.
LD01
4.1
0.1
76
4.4
2
11.5
Mobility.
England.
Column
Pct.
General:
Con01
Lab01
LD01
94.4
0.8
8.5
0.4
90.7
0.4
3.3
0.1
84.8
0.7
0.6
1.3
1.2
7.8
4.9
100
100
100
Oth01
1
1.2
1.5
9.1
4.2
2.3
Abst01
12.3
26.9
17.0
77.8
86.3
40.9
100
100
100
100
100
100
Oth01
10.5
15.8
8.5
13.4
51.7
100
Abst01
7.3
20.4
5.3
6.5
60.4
100
24.0
31.1
12.8
3.4
28.6
100
Voting at General and European elections
The 1999 European elections fell at the mid-point between the 1997 and 2001 general elections.
It is
interesting, therefore, to perform two comparisons using the Ecol method. First, in Table 4, we use the 1997
general election as our base year and compare voting and turnout then with that at the 1999 European election.
Second, in Table 5, the European election forms the base year, which is compared to the outcome of the 2001
general election. The 1999 election was good for the main opposition party, the Conservatives but Labour
struggled to re-capture the momentum that had taken it forward to a landslide victory at the 1997 general
election.
Ecological inference of voting and abstention between these elections appears to support this
impression. While just over half of Conservative supporters in 1997 chose to abstain two years later more than
eight out ten Labour supporters did so. Liberal Democrat supporters, made aware by their own party that the
electoral system would serve them better, abstained in relatively low numbers. Table 4 also appears to confirm
that some electors have become serial abstainers with more than 90% of 1997 abstainers continuing in 1999 to
11
boycott the electoral process. Confirmation of this pattern is provided in Table 5 where the vast majority (93%)
that had abstained from voting at the European election continued to do so at the subsequent general election.
Table 4: Voting and abstention at the 1997 General and 1999 European Elections
Voter
General
Con97
Lab97
LD97
Oth97
Abst97
Voter
General
Con97
Lab97
LD97
Oth97
Abst97
Mobility.
England.
Row
EU99:
Con99
Lab99
29.0
0.8
0.6
15.2
9.4
0.4
3.6
2.1
0.8
4.2
8.7
6.3
Pct.
LD99
2.8
0.2
15.5
1.6
0.8
3
Mobility.
England.
Column
Pct.
EU99:
Con99
Lab99
LD99
79.8
3.2
22.3
2.2
75.6
2.5
13.8
0.8
66
1.4
1.1
1.8
2.7
19.3
7.4
100
100
100
Oth99
9.2
0.8
12.3
5.3
1.4
4.6
Abst99
58.2
83.1
62.3
87.4
92.7
77.4
100
100
100
100
100
100
Oth99
47.7
5.2
34.3
3.9
8.9
100
Abst99
18.1
33.4
10.3
3.9
34.3
100
24
31.1
12.8
3.4
28.6
100
Oth01
0.5
1.1
1.1
2.9
2.6
2.3
Abst01
3.8
25.7
10.8
13.0
49.1
40.9
100
100
100
100
100
100
Oth01
1.7
3.1
1.5
5.8
87.9
100
Abst01
0.8
3.9
0.8
1.5
93.0
100
8.7
6.3
3
4.6
77.4
100
Table 5: Voting and abstention at the 1999 European and 2001 General Elections
Voter
EU99
Con99
Lab99
LD99
Oth99
Abst99
Voter
EU99
Con99
Lab99
LD99
Oth99
Abst99
Mobility.
England.
Row
General:
Con01
Lab01
84.5
1.5
2.3
69.8
19.8
1.6
49.6
3.5
13.4
25.6
20.8
24.5
Pct.
LD01
9.8
1.1
66.6
30.9
9.2
11.5
Mobility.
England.
Column
Pct.
General:
Con01
Lab01
LD01
35.5
0.5
7.4
0.7
17.9
0.6
2.9
0.2
17.5
11.0
0.7
12.4
49.9
80.8
62
100
100
100
12
Voting at Local and European elections
Comparisons between voting at the 1998-2000 local elections and the 1999 European Parliament elections. All
cases (N=529) were included in this analysis and adjustments were made in those cases where a local ward was
uncontested. In these cases it was assumed that if there had been an election, the turnout would have been
relatively low because of the likely large majority of the incumbent party. We further assumed that the votes
would massively favour the incumbent party. Turnout for such wards was estimated at the minimal level for
contested wards in the relevant authority and estimated vote was awarded to the incumbent party.
Examining the row percentages in Table 6 the analysis suggests that the main party of opposition, the
Conservative party, performed best of all in persuading its supporters to vote in both local and EU elections.
Only a third of those that had supported the party at the local elections decided to abstain for the EU contests.
This stands in marked contrast with Labour, the party of government, and the Liberal Democrats. Differing
explanations for these findings may be advanced.
Table 6: Voting and abstention at 1999 Local (adjusted) and European Elections
Voter
Local
Con99l
Lab99l
LD99l
Oth99l
Abst99l
Voter
Local
Con99l
Lab99l
LD99l
Oth99l
Abst99l
Mobility.
England
Row
EU99:
Con99e
Lab99e
51.2
0.7
0.8
36.9
7.5
0.9
6.1
2.0
3.5
3.0
8.7
6.3
Pct.
LD99e
2.2
0.3
18.4
3.2
1.8
3.0
Mobility.
England
Column
Pct.
EU99:
Con99e
Lab99e
LD99e
63.6
1.2
7.9
1.0
64.1
1.0
6.6
1.1
46.7
2.1
0.9
3.1
26.7
32.6
41.3
100
100
100
Oth99e
11.7
0.9
8.9
11.6
3.3
4.6
Abst99e
34.2
61.0
64.2
77.1
88.4
77.4
100
100
100
100
100
100
Oth99e
27.5
2.2
14.9
7.5
47.8
100
Abst99e
4.8
8.6
6.4
2.9
77.3
100
10.9
10.9
7.7
3.0
67.6
100
In Table 7 we included only those constituencies where every local ward held an election sometime between
1998-2000 – if any ward was unopposed then the entire constituency was excluded from the analysis. This has
the effect of reducing the number of constituencies available for analysis from 529 to 414. In our view this
procedure does not materially affect the results of the analysis, particularly if one examines the column
percentages for abstention at the European elections.
13
Table 7: Voting and abstention at 1999 Local (unadjusted) and European Elections (N=414)
Voter
Local
Con99l
Lab99l
LD99l
Oth99l
Abst99l
Voter
Local
Con99l
Lab99l
LD99l
Oth99l
Abst99l
Mobility.
England
Row
EU99:
Con99e
Lab99e
52.4
0.7
0.8
37.3
5.4
1.0
3.3
2.6
2.9
2.9
8.1
6.4
Pct.
LD99e
1.8
0.3
18.6
2.0
1.7
2.9
Mobility.
England
Column
Pct.
EU99:
Con99e
Lab99e
LD99e
68.5
1.1
6.6
1.1
66.3
1.2
5.1
1.2
49.6
0.8
0.8
1.4
24.5
30.6
41.2
100
100
100
Oth99e
13.5
1.4
7.4
6.7
3.0
4.3
Abst99e
31.6
60.2
67.6
85.4
89.5
78.3
100
100
100
100
100
100
Oth99e
33.0
3.6
13.1
3.1
47.2
100
Abst99e
4.3
8.8
6.6
2.2
78.2
100
10.6
11.5
7.6
2.0
68.3
100
Voting at European Elections and Socio-Economic characteristics
Next, we conducted two analyses that examined voting at the 1999 European elections and, using data from the
1991 census, social class and age. Prior research on voter turnout had identified these as key independent
variables and the Ecol procedure provides a simple but effective means for substantiating the findings obtained
from survey data. This analysis would also help to inform the selection of variables used in the second part of
this chapter.
Social class, as defined here, is derived from the 10% sample in the 1991 census of occupational categories,
normally referred to as the Registrar General's Social Class. This schema allows for six categories:
I Professional etc occupations
II Managerial and Technical occupations
III Skilled occupations
(N) non-manual
(M) manual
IV Partly-skilled occupations
V Unskilled occupations
Middle-class occupations are generally referred to as those falling in the categories I-IIIN, with the working
class defined as groups IIIM-V. There is also a sixth category that includes armed services personnel and those
for whom it was impossible correctly to classify occupation. Although there is considerable debate about the
most appropriate method for describing social class, for our purpose, which is designed to present a broad
14
picture of which social groups are abstaining more than others, this classification is suitable. The result is
shown in Table 8.
Table 8: Voting and abstention at the 1999 European election by Social Class.
Voter
Mobility.
euro99:
Social Class
class I
class II
class IIIN
class IVM
class IV
class V
No class
Voter
class I
class II
class IIIN
class IVM
class IV
class V
No class
..
Con
23.5
14.8
9.0
5.2
3.0
2.0
3.4
8.7
Mobility.
euro99:
Row
Lab
3.9
2.4
7.4
6.1
9.3
14.5
9.3
6.3
..
Con
16.5
46.6
11.2
15.1
4.2
0.9
5.5
100
Pct.
LD
8.6
5.0
2.7
1.4
1.1
1.1
2.1
3.0
Column
Lab
3.8
10.8
12.8
24.9
17.9
9.1
20.7
100
Oth
12.4
7.9
4.6
1.9
1.5
1.4
3.2
4.6
Abst
51.7
69.9
76.2
85.4
85.1
81.0
82.0
77.4
100
100
100
100
100
100
100
100
Oth
16.5
47.4
10.9
10.3
4.0
1.2
9.6
100
Abst
4.1
24.9
10.7
28.1
13.3
4.1
14.8
100
6.1
27.6
10.9
25.4
12.1
3.9
13.9
100
Pct.
LD
17.4
45.8
9.8
11.6
4.4
1.5
9.5
100
The row percentages show that while just over a half of professional and managerial workers (class I) abstained
from the European elections more than three-quarters of white-collar non-manual workers did so. For skilled
manual workers, partly- and unskilled occupations, however, electoral turnout fell below 20% across all three
categories. Similar analyses, conducted for general and local elections, controlling for class, produced virtually
identical patterns, though of course the absolute levels of turnout and abstention were different. The analysis
using a class variable, therefore, confirms findings from the research literature and from various British election
Studies, that white-collar occupations participate more than do blue-collar workers and that those drawn from
the professions and managerial occupations participate most of all.
The second of our socio-economic categories is age (see Table 9). We made no adjustments for the fact that the
census data were compiled in 1991 while the vote data refer to 1999, in the belief that relative positions between
and among constituencies would not vary. The row percentages show that the 18-24 age group abstained to the
largest degree while those in the 40-54 age group abstained least of all. Certainly, the former finding is
regularly confirmed by survey data. The largest category amongst EU abstainers in 1999 were, according to the
column percentages, the 25-39 age group, reflecting their larger numbers in the population as a whole.
15
Table 9: Voting and abstention at the 1999 European election by age
Voter
AGE
age18_24
age25_39
age40_54
age55_64
age65_
Voter
AGE
age18_24
age25_39
age40_54
age55_64
age65_
Mobility.
England
Row
EU99:
Con99
Lab99
2.3
9.5
3.4
10.0
17.4
2.7
9.3
4.4
9.9
4.3
8.7
6.3
Pct.
LD99
1.7
2.3
3.8
3.0
4.0
3.0
Mobility.
England
Column
Pct.
EU99:
Con99
Lab99
LD99
3.4
20.2
7.5
11.0
45.8
21.5
48.3
10.5
30.7
14.2
9.5
13.1
23.1
13.9
27.2
100
100
100
Oth99
2.3
3.3
5.3
4.5
7.2
4.6
Abst99
84.2
81.0
70.8
78.9
74.5
77.4
100
100
100
100
100
100
Oth99
6.7
20.7
27.8
12.9
31.9
100
Abst99
14.6
30.0
22.2
13.6
19.6
100
13.4
28.6
24.3
13.4
20.4
100
Individual facilitation and mobilisation
In this section we look first at the findings of Eurobarometer surveys and the pattern of voting/abstention at
European Parliament elections. We then analyse evidence about mutli-level voting and abstention drawn from
the Asia/Europe survey.
Two Eurobarometers (numbers 41 and 52) are examined, although the primary interest is in the more recent
survey that followed the previous election in June 1999. Eurobarometer 52 was conducted in autumn 1999, just
months after the European elections when respondents were asked whether or not they had voted at the election.
The responses showed that 30.1% claimed to have voted (actual turnout 24%), 64.7% abstained whilst the
remaining 5.2% either could not recall or refused to answer (responses to a question from Eurobarometer 41
show a similar pattern, 31.5% voted, 63% did not vote, 5.1% could not remember or refused). It is common for
self-reported turnout to exceed actual turnout and in this particular case the gap is sufficiently small that we can
safely ignore the differences in reported and actual turnout. No responses and Don’t Knows were excluded
from the analysis reducing the number of cases to 775 from Eurobarometer 52.
Age and education repeatedly figure as key explanatory variables of electoral turnout. Our preliminary analysis
of Eurobarometer data, therefore, focussed on the relationship between respondents’ demographic
characteristics and their attitudes towards the European Parliament as an institution and voting. Table 10 shows
responses about trust in the Parliament grouped by the respondent’s level of educational attainment and whether
they voted at the 1999 election. Those who tend to trust the Parliament are much more likely to have voted than
16
are those tending not to trust the institution. Additionally, non-voters were more likely to be uncertain about
trusting the parliament as evidenced by the larger proportion of ‘don’t knows’. The data also suggest a linear
relationship between educational attainment and trust with the more highly educated placing more trust in the
Parliament. A similar analysis, grouped according to age differences, however, shows that the youngest age
group (18-24), regardless of whether they voted in 1999, placed most trust in the Parliament, compared with all
other age groups. It is older people in Britain, aged 65 or over that tended to display higher levels of distrust in
the Parliament. The chi square test was significant at the .05 level.
Table 10: Education, Trust and Voting in European Elections
32
Tend not to
trust
38
32.7%
38.8%
28.6%
28
40
13
34.6%
49.4%
16.0%
40
23
17
50.0%
28.8%
21.3%
100
101
58
38.6%
39.0%
22.4%
20
85
58
12.3%
52.1%
35.6%
52
108
81
21.6%
44.8%
33.6%
29
44
28
28.7%
43.6%
27.7%
101
237
167
20.0%
46.9%
33.1%
Tend to trust
Voted
education
d8>0 & d8<=15
Count
% within education
d8>=16 &
d8<=17
Count
% within education
d8>=18 or d8=0
(still studying)
Count
% within education
Total
Count
% within education
Did not
vote
education
d8>0 & d8<=15
Count
% within education
d8>=16 &
d8<=17
Count
% within education
d8>=18 or d8=0
(still studying)
Count
% within education
Total
Count
% within education
DK
28
Turning to measures of mobilisation, Table 11 shows whether respondents noticed the most intense form of
electoral campaigning, a personal visit from party campaign workers, and whether the respondent’s educational
attainment had any bearing on the subsequent decision to participate in the election. The first point to note is
that the responses suggest that overall the 1999 election campaign was not particularly visible with just 8%
reporting a personal visit from party workers during the campaign. The second point is that, regardless of
educational attainment, those that voted were more likely to report mobilisation than were those that abstained.
However, a chi square test revealed these relationships to be insignificant (0.05 level). The campaign, as
measured by a single indicator of party canvassing, therefore, did not appear to have persuaded more electors to
participate. However, the questionnaire allowed for at least nine categories of campaigning and we cannot rule
out the possibility that participation is related more to the degree of campaign exposure rather than the type of
exposure.
17
Table 11 Campaign mobilisation, education and voting in European elections.
Yes
Voted
education
d8>0 & d8<=15
Count
% within education
d8>=16 &
d8<=17
Count
% within education
d8>=18 or d8=0
(still studying)
Count
% within education
Total
Count
% within education
Did not vote
education
d8>0 & d8<=15
Count
% within education
d8>=16 &
d8<=17
Count
% within education
d8>=18 or d8=0
(still studying)
Count
% within education
Total
Count
% within education
No
DK
16
78
4
16.3%
79.6%
4.1%
18
62
1
22.2%
76.5%
1.2%
7
72
1
8.8%
90.0%
1.3%
41
212
6
15.8%
81.9%
2.3%
15
135
13
9.2%
82.8%
8.0%
19
203
19
7.9%
84.2%
7.9%
6
86
9
5.9%
85.1%
8.9%
40
424
41
7.9%
84.0%
8.1%
Using logistic regression to help identify the characteristics of voter/non-voters in the 1999 European elections
we hypothesised that voter turnout at European elections would be influenced both by levels of individual
facilitation and mobilisation. Independent variables, therefore, included each respondent’s age and formal
education as well as indicators relating to their general interest in politics and knowledge about the European
Union. Social class variables were not used because of difficulties in arriving at a satisfactory definition using
the data. We also hypothesised that those participating in the election would express a positive view regarding
future expansion of the European parliament’s role within the Union. It was also important to measure the
impact of the electoral campaign upon respondents, particularly as the low turnout subsequently aroused
concern.
The results of the logistic regression are shown in Table 12. Overall, the education variable was significant,
with those receiving a longer formal education more likely to vote but respondents whose formal education
concluded at the end of secondary school less likely to vote. Increased age too made voters more likely to
participate in the European elections. The more respondents knew about the European Union the more likely
they were to turn out to vote. Unsurprisingly, those that wanted to see the role of the European parliament
increased were more likely to have voted at the 1999 election. It was certainly the case that respondents that had
been canvassed during the campaign, or at least acknowledged the campaign’s impact in some other way were
more likely to have voted than those for whom the campaign went unnoticed. The model correctly classified
75% of respondents. However, it appears that we are more successful at discovering why respondents did not
vote (86% correctly classified) rather than explaining the characteristics of those that did vote (53% classified).
18
Table 12: Logistic regression of electoral participation at the 1999 European elections
-2.87 (0.57)
Constant
Age
Scale from 18 towards
0.03 (0.01)**
*
Formal education
Younger than 15
-0.39 (0.27)
Between 15 and 18
-0.71 (0.24)**
Older than 18 or still studying
X
Discuss politics
Scale from ‘often’ to ‘never’
-0.29 (0.15)*
Knowledge about European Union
Scale from ‘Know nothing at all’ to ‘Know a great deal’
0.12 (0.05)*
**
Future Role of the European Parliament
More important
1.19 (0.27)**
Less important
0.18 (0.27)
The same
0.60 (0.29)*
DK
X
Electoral campaign impact
Scale from 0 towards 6
0.43 (0.05)**
Final -2 Log Likelihood
754.9
2
Pseudo R (Cox&Snell)
0.25
Pseudo R2 (Nagelkerke)
0.35
219.3 (9)**
Model Chi-square (d.f.)
75% (86, 53)
% correctly classified
775
Number of observations
*
- p < 0.05;
**
- p < 0.01;
X
- reference category for categorical independent variables;
Table entries are estimated logistic regression coefficients, with standard errors in parentheses.
We then replicated this analysis using data from Eurobarometer 41, which addresses the 1994 elections. We
used variables that were either identical to those used in the previous analysis or, as in the cases of some
attitudinal variables, close approximations. The results are shown in Table 13 and suggest that the same factors
have influenced voter participation at the last two European elections.
19
Table 13: Logistic regression of electoral participation at the 1994 European elections
-2.44 (0.36)**
Constant
**
Age
Scale from 18 towards
0.02 (0.00)
Education age
Younger than 15
-0.08 (0.29)
Between 15 and 18
-0.21 (0.20)
Older than 18 or still studying
X
**
Interest in politics
DK
-3.8 (7.8)
A great deal
1.5 (0.29)**
To some extent
0.89 (0.22)**
Not much
0.55 (0.22)*
Not at all
X
*
For or Against efforts being made to united European countries
DKt
0.10 (0.29)
For – very much or to some extent
0.40 (0.16)*
Against - to some extent or very much
X
**
Electoral campaign impact
Scale from 0 towards 6
0.22 (0.05)
Final -2 Log Likelihood
1186.0
2
0.14
2
0.18
Pseudo R (Cox&Snell)
Pseudo R (Nagelkerke)
142.8 (10)**
Model Chi-square (d.f.)
64% (61, 67)
% correctly classified
959
Number of observations
*
- p < 0.05;
**
- p < 0.01;
X
- reference category for categorical independent variables;
Table entries are estimated logistic regression coefficients, with standard errors in parentheses.
Overall, therefore, the findings from Eurobarometer surveys confirm that age and education are important
distinguishing characteristics that help set voters from non-voters apart. Once again, we are faced with the
paradox of voter turnout in western liberal democracies – despite evidence that individuals with a higher
education vote more than the average and the population becoming generally more educated, there has been a
20
secular decline in turnout. Additionally, European voters can be distinguished apart from non-voters by
referencing the extent to which people discuss politics, take an interest in the workings of the European Union
and believe that the institution for which they claimed to have voted should have its powers extended.
Mobilisation factors also appear at work, with voters more likely to have noticed some direct connection with
the electoral campaign. However, Eurobarometer 52 concentrated solely on voting/non-voting at one specific
European parliament election. As such, it does not help us with the rather more complex question that emerged
from our study of aggregate voting. Why is it that there are such large variations in the levels of electoral
abstention in England? If we can identify differences of a more subtle nature, between voters at general
elections and other types of voter who participate on a more regular basis at local and European elections, we
would be nearer our goal of highlighting those factors important for democratic participation in systems of
multi-level governance.
The Asia-Europe survey
The Asia-Europe survey (ASES) was conducted in 18 countries during 2001, half of which were in Europe,
including the United Kingdom. Since our analysis covers only voters in England respondents from Scotland
and Wales are excluded from the analysis. Using ASES data we first developed a multinomial model of
electoral turnout at general, European and local elections. Questions 406, 407 and 408 asked respondents about
their frequency of voting or not at general, local and European elections respectively. Possible responses fell
into one of five categories ranging from, ‘Voted in almost all of them’ to ‘Never voted in...election’ and ‘I am
not qualified to vote’. Those respondents that failed to answer at least one of the voting questions were
excluded from the analysis. Therefore, only those respondents from England who answered all three questions
about voting were included in the analysis, reducing the sample size to 845 from the original 1,014. The five
category variables from the three questions about voting habits were recoded into binary variables - comprising
individuals that stated that they voted at almost all elections of a particular type and individuals who only voted
occasionally, rarely or never at all. Although this may appear at first as a rather strict definition of voting we
found that the proportion of respondents now coded as ‘voting’ in European elections was 23% - virtually
identical to the actual level of voter turnout at the 1999 European elections. Those responding that they voted
‘almost always’ at general elections was 62% (compared with 59% at the 2001 election) with the figure of 46%
voting at local elections somewhat higher than the actual turnout. Since our purpose was to identify any distinct
characteristics of voters/abstainers at European elections this restricted definition of voter offered the greatest
purchase for this particular objective.
Initially, we developed a multinomial model of voter turnout. Four categories of voter type were identified:
•
Three level voters – respondents who voted ‘almost always’ at general, European and local elections.
(n=186)
21
•
Two-level voters – including respondents who voted at general and local elections (n=194);
respondents voting almost always at general and European elections (n=11); a respondent who voted at
local and European elections only (n=1).
•
One-level voters – respondents who voted at only one type of election – general (n=134), local (n=7)
and European (n=3).
•
Non-voters (n=309).
Non-voters were used as the point of comparison. A range of independent variables, both social characteristics
as well as attitudes and interests towards politics and participation, were included in the model. A small number
of these were significant (p < 0.05) but overall the model did not provide sufficient information other than that a
comparison of different groups of voters (voting almost always once, twice or three times according to the
electoral level) with electoral abstainers shows that three time voters differ from abstainers more significantly
than do one time voters from abstainers. We determined, therefore, to approach the problem in a different way,
this time using binary logistic regression.
As before our use of independent variables was directed by the existing research literature on voter turnout. The
variables comprised the following:
•
Age – recoded as a five category variable
•
Pattern of residence – four categories
•
Education attainment – five categories
•
Gender, male or female
•
Employment status – full-time, retired, other
•
Interest in politics – four categories ranging from ‘very interested’ to ‘not at all interested’
•
Party identification – those identifying with any political party were coded as ‘Yes’, with an additional
two categories ‘None/Don’t Know’ and ‘No response’
•
Ideological placement – four categories on a left-right continuum and a further category for ‘No
response’
•
Importance of ideology – three categories
•
Satisfaction with politics – three categories
•
Sense of belong – three categories
•
Citizen duty – three categories
•
Free-rider abstention –a variable constructed from a responses to a question that sought to discover
whether an individual would abstain from an election if he/she felt others were certain to vote. Three
categories ‘Agree’, ‘Neutral’, ‘Disagree’ were used.
We viewed the issue of multilevel voting as a nested set of decisions or voting levels (see Figure 3). Voting
level 1 divided respondents into one of two sets - the set of all respondents coded as ‘non-voters’ distinguished
apart from the set that voted almost always in at least one type of election (usually general elections). Voting
level 2 then distinguished between those voting at least at general and local elections and those that did not vote
22
in local elections. Finally, voting level 3 considered all those people that voted at general and local elections.
These were divided into one of two sets – those that only voted at general and local elections compared with
those that voted in addition at European elections. A series of logistic regressions were used to identify
characteristics of respondents that fell into the various sets and sub-sets. Two models are shown for purposes of
comparison. Model 2 contains all variables listed above and identified in the research literature as compelling
characteristics suitable for investigation. Model 2, however, contains a more complete range of variables based
on prior analysis of voter turnout. Table 14 shows the results of the conditional logistic regressions.
whole population
where
- people who voted at general election;
- people who voted at local election;
- people who voted at European election.
Figure 4: Levels of voter participation
23
Table 14: Logistic regressions of voter turnout and abstention at general, European and local elections
GB
Model 1
Model 2
Voting level 1
Voting level 2 Voting level 3 Voting level 1 Voting level 2 Voting level 3
1.20 (0.90)
Constant
2.05 (1.17)
0.01 (1.25)
0.04 (1.01)
2.46 (1.3)
-0.49 (1.38)
*
-1.83 (0.57)*
-0.44 (0.56)
*
*
*
-1.61 (0.52)*
*
*
Individual facilitation
**
**
**
**
Age
18-24 years old
-3.11 (0.47)
**
-1.29 (0.45)*
-0.37 (0.45)
-2.41 (0.57)
*
25-39 years old
-2.61 (0.45)
**
-1.12(0.37)*
-0.36 (0.36)
-1.93 (0.55)
*
-0.40 (0.48)
40-54 years old
-1.61 (0.45)
**
-0.49 (0.36)
-0.24 (0.31)
-0.92 (0.55))
-0.88 (0.49)
-0.22 (0.44)
55-64 years old
-1.40 (0.46)
**
-0.00 (0.39)
-0.47 (0.32)
-0.90 (0.51)
-0.14 (0.43)
-0.45 (0.39)
X
X
X
X
65+ years old
X
X
Pattern of residence
Alone
-0.02 (0.55)
-0.46 (0.75)
-1.44 (0.89)
-0.05 (0.56)
-0.61 (0.76)
-1.41 (0.91)
With partner
0.35 (0.52)
-0.08 (0.72)
-1.06 (0.86)
0.40 (0.53)
-0.21 (0.73)
-1.00 (0.89)
With other
-0.03 (0.55)
-0.20 (0.77)
-1.75 (0.91)
0.01 (0.55)
-0.34 (0.78)
-1.66 (0.94)
X
X
X
X
No response
Educational
attainment
No qualification
X
*
*
-0.80 (0.33)
*
X
*
-0.85 (0.40)*
-0.61 (0.34)
-0.81 (0.34)
*
-0.93 (0.41)*
-0.58 (0.35)
Below A-level
-0.35 (0.30)
-0.84 (0.37)*
-0.43 (0.32)
-0.30 (0.31)
-0.83 (0.37)*
-0.41 (0.33)
A-level
-0.41 (0.40)
-0.54 (0.50)
0.20 (0.49)
-0.43 (0.40)
-0.54 (0.51)
0.17 (0.50)
Degree
0.10 (0.38)
-0.04 (0.47)
-0.30 (0.40)
0.08 (0.38)
-0.01 (0.48)
-0.38 (0.40)
X
X
X
X
X
0.16 (0.18)
0.30 (0.23)
0.00 (0.23)
X
X
X
-0.11 (0.19)
0.03 (0.25)
-0.16 (0.25)
-0.44 (0.41)
-0.13 (0.37)
X
X
Higher degree
X
Gender
Male
Female
Employment status
Full time
Retired
0.91 (0.44)*
Other
X
Continued overleaf …/
24
…/ Table 14 continued
Individual
mobilisation
**
*
**
**
*
*
Interest in politics
Very interested
1.26 (0.38)
**
0.31 (0.44)
1.45 (0.43)**
1.21 (0.41)*
0.40 (0.47)
1.32 (0.46)**
0.36 (0.33)
0.52 (0.33)
-0.43 (0.31)
0.20 (0.34)
X
X
-0.31 (0.56)
-0.30 (0.51)
*
Fairly interested
0.95 (0.23)
**
0.32 (0.31)
0.66 (0.32)*
0.90 (0.24)*
*
Not so interested
0.53 (0.22)
Not at all interested
*
X
-0.37 (0.30)
0.27 (0.33)
X
X
**
0.52 (0.22)*
X
**
Party identification
Yes
0.97 (0.39)
None/DK
-0.11 (0.40)
-0.35 (0.55)
-0.20 (0.49)
-0.60 (0.58)
-0.30 (0.55)
-0.09 (0.41)
-0.55 (0.59)
-0.22 (0.57)
X
X
X
X
X
Left
-0.21 (0.35)
0.39 (0.45)
0.28 (0.45)
Centre-Left
-0.10 (0.27)
0.21 (0.34)
0.34 (0.36)
Centre-Right
-0.38 (0.29)
-0.35 (0.36)
0.33 (0.38)
Right
-0.50 (0.41)
-0.44 (0.46)
0.14 (0.47)
X
X
X
No response
*
X
0.95 (0.40)*
Ideological placement
No response
Importance of
ideology
Important
0.74 (0.36)*
-0.38 (0.49)
0.45 (0.54)
Unimportant
0.60 (0.29)*
0.08 (0.42)
0.21 (0.48)
X
X
X
Satisfied
0.19 (0.28)
0.15 (0.33)
0.64 (0.30)*
Neutral
0.08 (0.18)
-0.17 (0.23)
0.40 (0.23)
X
X
X
DK
Satisfaction with
politics
Dissatisfied
*
*
Sense of belonging
No
sense
of -0.44 (0.28)
0.62 (0.30)*
0.62 (0.33)
-0.43 (0.28)
0.47 (0.35)
1.00 (0.38)** -0.69 (0.32)
0.61 (0.30)*
0.67 (0.34)*
0.60 (0.36)
1.01 (0.39)**
belonging
Sense of belonging -0.71 (0.32)
No response
*
X
X
X
*
X
X
X
Continued overleaf …/
25
…/ Table 14 continued
**
**
**
**
Citizen duty
Agree
1.39 (0.30)
**
0.47 (0.49)
0.94 (0.59)
1.36 (0.31)*
0.59 (0.51)
0.81 (0.61)
*
Neutral/DK
0.21 (0.38)
Disagree
X
-0.89 (0.60)
0.59 (0.79)
0.23 (0.38)
-0.84 (0.62)
0.46 (0.80)
X
X
X
X
X
-0.11 (0.36)
0.04 (0.40)
**
**
Free rider abstention
Agree
-0.85 (0.22
**
-0.06 (0.35)
0.01 (0.39)
-0.87 (0.23)*
*
Neutral/DK
Disagree
Final -2 Log
Likelihood
Pseudo R2
(Cox&Snell)
Pseudo R2
(Nagelkerke)
Model Chi-square
(d.f.)
% correctly classified
Number of
observations
0.02 (0.26)
X
0.27 (0.36)
0.14 (0.36)
0.01 (0.27)
0.21 (0.38)
0.13 (0.38)
X
X
X
X
X
899.3
614.0
598.7
885.2
599.6
588.1
0.31
0.10
0.11
0.32
0.12
0.13
0.43
0.15
0.14
0.44
0.18
0.17
63.5 (22) **
52.2 (22) **
77.9 (33) **
62.8 (33) **
364.5 (22) **
378.7 (33) **
78 (61, 88)
77 (12, 97)
63 (59, 66)
79 (62, 88)
78 (17, 96)
65 (62, 68)
966
617
470
966
617
470
*
- p < 0.05;
**
- p < 0.01;
X
- reference category for categorical independent variables;
Table entries are estimated logistic regression coefficients, with standard errors in parentheses.
Model 1 consists of 8 predictors (explanatory variables) and Model 2 consists of 13 predictors.
Voting level 1: a binary logistic regression is used to determine what factors are significant for distinguishing between
people who do not vote at all and those who do vote at one elections at least.
Voting level 2: considering people who do vote at general elections, a binary logistic regression is used to determine what
factors are significant for distinguishing between people who do not vote at local elections and those who do vote.
Voting level 3: considering people who do vote at general and local elections, a binary logistic regression is used to
determine what factors are significant for distinguishing between people who do not vote at European elections and those
who do vote at all three type of elections.
26
Voting level 1
Voting level 1 represents an attempt to identify those broad characteristics associated with voters compared to
non-voters. The respondent’s age is clearly a significant factor in helping to identify whether someone is likely
to be a voter or not. Compared with those aged 65 or over, all other categories are likely to abstain more, with
the youngest age category abstaining the most. This is unsurprising given the strength of this characteristic in
prior analyses of voter turnout. The signs on the coefficients for educational attainment, also identified by
earlier research as a distinguishing feature of voters/non-voters, appeared to confirm such findings but none of
the categories were statistically significant in our model, nor indeed was the variable itself. What did appear
significant was the respondent’s stated interest in politics. Those ‘very interested’ and ‘fairly interested’ in
politics were more likely to vote than were those that said that they were ‘not at all interested’. Similarly, those
that confirmed a level of partisan identification were more likely to participate compared with those without a
sense of attachment towards a political party. Unsurprisingly, those agreeing with the statement that, ‘Citizens
have a duty to vote in elections’, were more likely than not to be those that voted almost always, at least at a
general election. Equally, voters, unlike non-voters, generally disagreed with the statement, ‘Since so many
other people vote in elections, it really doesn’t matter whether I vote or not’. The final logistic regression
equation correctly classifies 78% of cases. Summarising, compared with non-voters, voters are likely to be
older, interested in politics, have some sense of partisan attachment, possess a strong sense of citizen duty and
do not take account of how others might behave when deciding to participate at an election.
Voting level 2
Our analysis of aggregate voting data reveals a large and recurring gap in turnout between general and local
elections. Within the broad category of voters, could we distinguish apart those that were likely to vote in both
general and local elections and those likely to restrict their electoral participation to the parliamentary level?
Once again, younger age groups are likely to be reluctant participants compared with their older voters.
Similarly, respondents who possessed either no formal or the minimal educational qualification might turn out
to vote in general but not local elections when compared with the most educated voters. However, other factors
did not appear as statistically significant, suggesting that notions of civic duty and party identification although
useful for helping to distinguish differences between voters and non-voters, could not contribute further towards
a more fine-grained categorisation (according to type of election) of voters. Although overall the equation
correctly classifies 75% of cases the model’s performance appears skewed. While 96% of cases in the category
of ‘vote at general and also vote in local elections’ could be correctly classified, only 17% of those that voted at
general but abstained from local elections could be correctly classified.
Voting level 3
Perhaps the most intriguing question is whether there is anything distinctive about the group of voters that might
be described as habitual voters to the extent that not only do they vote in general and local elections but also
participate in those for the European parliament. In this third conditional logistic regression three variables
27
proved statistically significant. The first of these is pattern of residence. We had expected to find that those
living alone would be less likely to be participants at European elections since there would be less peer pressure
involved. However, our findings suggest that those respondents that lived with someone else other than a
spouse or partner were least likely to vote in European elections. A bivariate analysis with frequency of voting
showed that many respondents designated as ‘other’ were people living either with parents or with siblings.
Further analysis of this category of respondents, in terms of education, income, sense of economic well-being,
failed to identify a distinct social type. At the time of writing we are unable to put forward a suitable
explanation for this finding. More easily explained is the apparent finding that European voters are likely to
have a heightened interest in politics compared with their fellow citizens. Turnout might be low for European
elections but at least that do vote regularly are likely to be the most informed and/or interested in politics!
Perhaps the most intriguing finding relates to a respondent’s sense of belonging to a community, spatial,
religious or ethnic.
This categorisation specifically excludes ‘Europe’ as a category since that sense of
identification is the focus of another question (and where most respondents chose not to think of themselves as
European). Those that have, in addition to a sense of nationality, an identification with another community, do
appear likely to vote more than others in European elections. Among such people, of course, there may be a
stronger sense of social capital than is the case generally. Of the three voting levels considered here this
particular level had the lowest percentage of respondents correctly classified overall, just 63%, but this is not
entirely surprising given our target group was those that voted at all three levels.
The multinomial approach gave us no greater purchase on the complex issue of multi-level voting other than
stating the rather obvious conclusion that voters at three levels of election differ more significantly from
abstainers than do those voters restricting themselves to just one level of voting. Examining multi-level voting
as a series of conditional logistic regressions does appear to have shed more light on this issue, while not exactly
bathing it in sunlight! Habitual abstainers tend to be younger, are educated less, have little or no interest in
politics, possess an under-developed sense of civic duty and are at ease at letting others participate while they
themselves do not. As the aggregate data show, however, such attitudes are becoming more widespread for all
types of election. Those that restricted their participation to general and not local elections tended to be aged
below 40 years of age and were more likely to have little or no formal educational qualifications. Other than
those differences, however, there was little to separate one-level from two-level voters. The most committed
electoral participants, three-level voters, did appear to differ from two-level voters in terms of their interest in
politics and a sense of attachment to a community other than the nation state.
Conclusions
Examination of aggregate data reveals some fluctuation in turnout but over the long-term there has been a steady
decline in electoral participation for all levels of governance. In fact, our data show a positive and high
correlation amongst turnout for all three types of election. That suggests that there are some factors contributing
towards this general decline as well as additional factors that are specific to each type of election. The gap
between general, local and European election turnout has been remarkably stable over the twenty year period.
28
the general decline has occurred despite various public concerns being expressed about the problem of low
turnout and despite legislation designed to facilitate the act of voting. Voter turnout at elections to the European
parliament lags considerably behind that for general elections and is even lower than that encountered at local
authority elections. This pattern persisted for the 1999 European election, when PR-List replaced FPTP voting,
and it was widely anticipated that such a change might help stimulate turnout, particularly amongst supporters of
minor parties. This was not the case as turnout reached a record low, although it is debatable how many electors
knew of a change in voting procedures given the already low salience attached to the election. One useful
measure of interest in an election is the proportion of electors applying for a postal vote and the figures show
that applications at a general election are at twice the level recorded for European elections. Moreover, among
those that apply for a postal vote a lower proportion then use that vote at European than at general elections. It
is difficult not to conclude that generally electors are less interested in the election both before and during the
campaign proper.
The analysis using a method for ecological inference suggested many of those that abstain from voting at
general elections do so for other types of election. The reverse is clearly not the case, however, because of the
overall differences in turnout. The analysis also demonstrated that parties in government struggle most to
mobilise their supporters at both local and, to a greater extent, European elections. Comparisons of voting and
abstention between local and European elections showed a similar pattern to that between European and general
elections. Incumbent parties find it difficult to mobilise supporters, particularly when in England European
elections follow just a month after the local elections in early May. By contrast, a party that has been lifted by
success at local elections does not appear to suffer from supporters reluctant to turn out again so soon after that
initial victory. Once again, the aggregate data suggest that among those people that do not participate in local
elections a very large proportion, almost nine in ten, continue not to participate at European elections.
Survey data largely confirmed the picture emerging from the examination of aggregate data. Eurobarometer 52,
which asked people whether they had voted at the 1999 election was conducted only a short time afterwards.
There were some positive signs from this survey, which suggested that the election campaign, modest by the
standards of a general election, had at least registered with voters. The logistic regression highlighted the
importance of other variables, including age, education and the respondent’s interest in politics as important.
However, European elections, fixed as they are in the calendar, are always going to suffer, to a greater or less
extent, according to their position within the parliamentary electoral cycle. The 1999 election fell at the midpoint of the 1997-2001 parliament and some explanation for the dramatic collapse in turnout should be
attributed to that fact. Local electoral turnout rises and falls according to the influence of parliamentary
elections and there is no reason to suppose that their European counterparts are immune from that influence.
The ASES data, which asked about the frequency of voting at all three types of election, facilitated our search
for explanations for the differences in voter turnout. We used conditional logistic regressions in our effort to
capture some of the subtle forces at work when electors make their decision to participate or not at a given
election.
However, the further we probed this conundrum the less evidence there was by way of that
explanation. Notions of civic duty, interest in politics, partisan attachment and an unwillingness to free-ride on
the decision to vote of others, were all characteristics that separated voters from non-voters. Amongst the set of
29
voters by far the largest group were those that restricted their visit to a polling station for a general election. It
proved relatively straightforward to correctly identify the type of person that would vote at a general and a local
election but more difficult to explain why someone who did engage in voting should then restrict that behaviour
to just one type of election. These people might have made that decision on the basis of the relative functions
exercised at different levels of governance, of a keen appraisal of the costs and benefits from voting. Or it might
be that such people have not been mobilised to vote, either by political parties or of other agencies, such as the
media. A smaller sub-set of respondents did assert that they were what might be called habitual voters, voting at
general, local and European elections. This hard core of democratic participants have a great interest in politics,
so there habit is not simply borne out of a sense of civic duty but is mediated by an awareness of what is taking
place before them. Moreover, these are also people that possess a sense of community identity, separate from
that with their national identity.
30