Investigating differences in electoral turnout: the influence of ward ...

Environment and Planning A 2008, volume 40, pages 1250 ^1268

doi:10.1068/a39169

Investigating differences in electoral turnout: the influence of ward-level context on participation in local and parliamentary elections in Britain Scott Orford

School of City and Regional Planning, Glamorgan Building, University of Cardiff, King Edward VII Avenue, Cardiff CF10 3WA, Wales; e-mail: [email protected]

Colin Rallings, Michael Thrasher, Galina Borisyuk

Local Government Chronicle Elections Centre, University of Plymouth, Plymouth, Devon PL4 8AA, England; e-mail: [email protected], [email protected], [email protected] Received 10 May 2006; in revised form 15 August 2006; published online 29 August 2007

Abstract. Electoral turnout is an important measure of the health of a liberal democracy. Although research identifies factors that affect electoral participation, we still know little about how electors in a specific location respond to opportunities to vote for different kinds of local, national, and supranational institutions. This paper addresses this issue by analysing the relative rates of turnout at local, parliamentary, and European elections within three time periods for the London Borough of Brent. It uses turnout data for individual polling districts to investigate whether relative differences in turnout are sustained across time, whether polling districts perform consistently or not for different types of elections and whether variations in turnout are related to marginality. The results indicate that turnout at different types of elections is not stable even within tightly constrained time periods and that there are statistically significant differences in the relative rates of participation between polling districts. Geographically, the differences in relative rates of turnout appear to be spatially clustered, particularly with respect to local elections and this may reflect an increase in the concentration of party campaigning in marginal wards.

Introduction Electoral turnout has been extensively studied, especially in recent years with levels declining in many countries (Blais, 2000; Blais et al, 2004; Franklin, 1996; 2004; Pattie and Johnston, 2005). The majority of this research is restricted to national elections, despite voters being given many opportunities to vote. In the UK, for example, it is clear that many more people participate at a general election than turn out for a local council or European Parliament election. Indeed, this so called `turnout gap' is one of the largest for any liberal democracy: almost twice as many electors participate at a general election than do so at a local election. The gap is wider still between general election voting and that for the European Parliament. Britain is almost unique in the extent of the turnout gap. In some countries, notably the United States, turnout for all elections is comparatively low; in others, participation at subnational elections is similar to that found in national contests (see Blais, 2000, page 37, table 1.5). Evidence suggests, however, that common factors are at work in determining turnout to elections at all levels. Survey data, from both the UK and elsewhere identify age, education, and occupational status as well as an individual's strength of party identification and interest in the campaign as factors determining turnout (Blais, 2000; Franklin, 2004; Whiteley et al, 2001; Worcester and Mortimore, 2001). When individuals perceive real differences between competing parties then participation increases (Clarke et al, 2004; Heath and Taylor, 1999; Pattie and Johnston, 1998). Analysis of parliamentary results shows that turnout is related to the social composition and political marginality of individual constituencies (Curtice, 2005; Denver and Hands, 1997;

Investigating differences in electoral turnout

1251

Pattie and Johnston, 2005). The expectation is that the closer the likely result the higher the turnout. Studies of local electoral turnout, though less frequent, identify similar factors. Participation is lower in `safe' seats than it is in marginal ones (Rallings and Thrasher, 1997). The proportion of elderly residents and population stability are positively correlated with turnout (Rallings and Thrasher, 1990; Rallings et al, 2000). Limited survey data show that age, length of residence, sense of involvement in local politics, and strength of party identification are also related to turnout (Miller, 1988). The strength of an individual's sense of `civic duty', concern about who won the election, and engagement with the campaign all appear critical to the choice of whether or not to participate at a local election (Curtice et al, 2001; Rallings and Thrasher, 2003). Widespread low turnout at European elections led Reif and Schmitt (1980) towards a theory of `second-order' elections. Absolute levels of turnout were lower than in national elections because electors perceived the European contest as less important because no government was being elected. Moreover, many of those who did vote in European elections were expressing a protest towards the incumbent national government (Reif, 1984). While some disagree, citing attitudes towards the EU as more significant in determining whether someone votes or not (Blondel et al, 1998) other research (Franklin et al, 1996; Schmitt and Manheimer, 1991) seems to confirm the second-order thesis. Notwithstanding these differences, most observers of European election turnout agree that factors of age and education, interest in politics, party identification, and a sense of civic duty are as relevant to our understanding of the decision to vote in these elections as they are for parliamentary and local election turnout. That said, there is no doubt that European election turnout is consistently lower than for other types of election. Election timing can affect European turnout both positively and negatively. In 1999 the European contests followed a month after the local elections, perhaps contributing to `voter fatigue' and reduced turnout. By contrast, the 2004 elections were coincident with local elections in some areas and undoubtedly contributed to a rise in European turnout. When turnout is naturally low the ease of voting becomes critical to turnout (Blondel et al, 1998; Franklin, 1996) and certainly at the 2004 election the higher turnout in regions that used allpostal ballots supports this view. Other things being equal, the move towards a proportional representation system of voting in 1999 should have resulted in higher turnout (Blais and Carty, 1990) but in Britain it coincided with a record low turnout. Mattila (2003) argues that in the low-salience European elections people may not even investigate the nuances of changes to the electoral system. This leaves a puzzle to be solved. Those factors that influence turnout at general, local, and European elections appear to be substantially similar, but the numbers voting are markedly different. The objective of this paper is therefore to investigate the causes of the differences in relative turnout: in particular, we are interested in the effect of local campaigning and marginality upon turnout to national and local elections. In order to control for the social factors that may influence turnout, we are comparing elections for the national and European parliaments and local council that are held over a relatively short time period for turnout at the polling district level. Polling districts are the smallest unit used for all electoral purposes and equivalent to the US precinct. By keeping the time period between elections short, whatever social determinants of participation are at work in a polling district at one election will also be active for others. Thus other things being equal, we would expect that a relatively high level of turnout at the polling district for one type of election will result in a relatively high turnout in the same polling district for the other types of election within

1252

S Orford, C Rallings, M Thrasher, G Borisyuk

the short time frame; and vice versa. As will be argued later, we are using European election turnout as the base for comparison. European elections record the lowest levels of turnout and are, by common consent, regarded as the least affected by any local contextual factors: for example, marginality and campaigning (Butler and Westlake, 2000; 2005). By contrast, local and parliamentary elections are, potentially at least, more affected by such issues. The structure of the paper is as follows. We describe the turnout data available for analysis and provide some background information about the electoral politics of Brent, our chosen case-study area, which will aid interpretation of our findings. We outline the principal methods used in the analysis and report our main findings. In the conclusion we consider some of the implications of these for the understanding of electoral turnout and suggest how, using similar methods, we might develop this research initiative further. Case-study and context The focus for this case study is the London Borough of Brent, which has complete polling-district-level data on turnout for parliamentary, European, and local elections for the period 1979 ^ 2001. Polling districts are administrative areas that local authorities use for conducting elections. Each electoral ward is subdivided into a number of polling districts where people physically go to vote. As parliamentary and European elections are not distinctly ward-level elections, polling districts are a useful spatial scale to study and also provide an unusually fine-grained analysis of aggregate voting data (averaging about 1300 electors). Aggregate voting data have been used in previous research to demonstrate the importance of contextual effects: for example, the importance of party campaigning and the expected closeness of the outcome, in affecting turnout (Franklin, 2004; Pattie and Johnston, 2003). Aggregate polling district data have an advantage over more traditional survey data in this instance as surveys may overreport turnout because some respondents will be reluctant to admit electoral abstention while voting is viewed as a civic duty. However, the chief weakness is that survey data do not facilitate analysis of relative turnout. While the factors influencing general election turnout are widely known, these surveys do not simultaneously ask respondents about their participation in other types of election, notably local and European Parliament elections. Brent authority also supplied digitised boundary data for its polling districts. Boundary changes during the period were negligible(1) and Brent has the added advantage of encompassing three entire parliamentary constituencies (Brent East, Brent North, and Brent South), thus providing greater opportunities for direct comparisons. Brent wards and polling districts were subsequently redrawn for the 2002 London borough elections meaning that comparisons of election turnout data after this date are problematic. Figure 1 illustrates the electoral geography of Brent in 2001. The three time periods used in the study cover elections over a twenty-year period. Time period 1 relates to the 1979 parliamentary election, the 1979 European election, and the 1978 local election. Time period 2 relates to the 1992 parliamentary election, the 1989 (1) In time periods 1 and 3, each of the elections was administered using the same number of polling districts (114 and 115, respectively). In time period 2, the number of polling districts varied: 114 were used in the 1989 European election, 117 in the 1990 local election, and 118 in the 1992 parliamentary election. The increase in the number of polling districts in time period 2 was a result of subdividing existing polling districts. In order to compare polling-district-level turnout for each election in this time period, GIS was used to merge the new polling districts in the 1990 and 1992 elections in order to recreate the original polling district geography used in the 1989 European elections. Percentage turnout was then recalculated for the merged polling districts.


1253

Figure 1. Constituencies and wards in the London Borough of Brent, 2001.

European election, and the 1990 local election. Time period 3 relates to the 2001 parliamentary election, the 1999 European election, and the 1998 local election. As the duration of each time period is fairly short (no more than three years), changes in the characteristics of the polling-district population within each time period would not be significant enough to have any noticeable effect on the turnout at the three elections. However, the political context can change substantially between elections and there is a need to understand this when interpreting the impact of marginality upon turnout. To this end we have examined election results in Brent for the elections prior to those from which our polling district data are drawn in order to identify the possible impact of political competition and mobilisation on turnout levels. At the October 1974 general election Labour was returned in Brent East with a majority of 16.6% over the Conservatives and with one of 28.5% in Brent South. The Conservatives held Brent North by 19.8% over Labour. Given the anticipated (and realised) swing to the Conservatives in 1979 Brent East is likely to have been treated as a semimarginal by the parties with a consequent slight possible increase in mobilising activity. A similar pattern was observed before the 1978 local elections. In May 1974 the wards in Brent East had voted Labour by a majority of 11.2%; in Brent North there was a 27.5% Conservative majority; and in Brent South a Labour one of 23.4%. However, electoral activity is likely to have concentrated on the six out of thirty-one wardsöas it happens, two in each constituencyöthat had majorities in 1974 of less than 10%. The polling districts in these wards might therefore be expected to have a higher rank of turnout than at other types of election. The 1979 European Parliament election was the first of its kind. The whole Borough of Brent was part of the large London North West constituency and campaigning was conducted at a very broad level with no difference likely between polling districts. The London North West constituency returned a Conservative MEP by a majority of 24.2%.

1254


A decade later and Brent reflected the national parliamentary dominance of the Conservatives, allied with greater support for Labour at second-order elections. In 1987 the Conservatives won Brent North by 35.1%, and cut back Labour's majority in Brent South in 19.5% and in Brent East to a highly marginal 4.5%. At the 1992 general election Brent East was, accordingly, the only seat in the borough to register an increase in turnout compared with 1987. In local elections Labour had done reasonably well in 1986 helped by reaction to the abolition of the Greater London Council. It had a majority of 22.4% in East and 37.7% in South, and had slightly shaved the Conservative margin to 25.2% in North. Again there were just six wards throughout Brent with majorities of less than 10% at the time of the 1990 contests, two in North and four in East. The Conservatives had won London North West by only 4.5% at the 1984 European Parliament elections, but there is no evidence to suggest that electors in one part of Brent were mobilised any more or less than in any other at the 1989 contest. By the end of the 1990s the balance of politics in Brent had shifted sharply to the left. In line with this both Brent East and Brent South posted massive Labour majorities at the 1997 general electionö45% and 57.1%, respectively. Brent North was also won by Labour with a majority of 10.5%, following the biggest Conservative-to-Labour swing anywhere in the country (18.9%). This victory was a surprise gain for Labour, but meant that North became a key `defend' seat for the party in 2001, perhaps prompting comparatively high levels of turnout in its constituent wards. There had been little sign of such a Labour surge at the time of the 1994 local election. Brent North was `won' by the Conservatives by a margin of 33% and with no individual ward having a majority of less than 10%. There were just two wards in South (where Labour had an overall cushion of 16.9%) with a similar level of marginality. Most electoral activity in 1998 was likely, therefore, to be concentrated in East, where five out of the eleven wards had majorities of 5% or less. Two had been held by the Conservatives in 1994; two by Labour; and one had returned a councillor from each party. Although the strict analytical base for the 1999 European Parliament elections was the 1994 contests, the introduction of a new electoral system within a London regional constituency meant that there was no special ward-level or polling-district-level campaign activity in Brent. In each time period, therefore, it would seem likely that turnout variation between polling districts will be more apparent at parliamentary and local than at European election level. Investigating turnout and marginality through time

The objective of the paper is to investigate the impact of marginality on the differences in turnout to parliamentary, local, and European elections Although there are welldocumented large absolute differences in turnout to these three types of elections, we are assuming that the relationships in turnout within each tightly constrained time period are stable: that is, polling districts that have a relatively high turnout for one election will also have a relatively high turnout for the others even though the absolute differences in turnout may be large. We can also reasonably argue that European elections should have little or no local contextual variability within Brent because the electoral district involved is much larger than Brent and in time period 3 voting was also by proportional representation for a party list. Hence European election turnout should not be significantly influenced by any notion of marginality, and certainly not compared with local and parliamentary turnout. To this end we are treating participation in European elections as a measure of the `core vote' for each polling district electorate. By `core vote' we mean it represents people who almost invariably vote whatever the election or situation and are not influenced by factors such as marginality or local campaigning. This assumption requires some caveats which we will now outline.


1255

The first caveat concerns the strength of the relationship between European election turnout and turnout to local and parliamentary elections. If European election turnout is to be treated as a measure of the `core vote', then there should exist at least a moderately strong relationship between this and turnout to the other elections. Correlation coefficients in table 1 show such a relationship between European and parliamentary and local election turnout at polling-district level although the strength of this relationship is decreasing through time. With the exception of time period 1, the correlation between local and European rates of turnout is the stronger of the pair, although the differences are minor. Hence electoral turnout within each time period seems strongly related and this relationship is explored in more depth in a later section. Table 1. Correlation coefficients between European turnout (E) and local (L) and parliamentary (P) turnout in each time period.

Pearson's r

Time period 1

Time period 2

Time period 3

P&E

L&E

P&E

L&E

P&E

L&E

0.82

0.74

0.73

0.76

0.67

0.71

The second concerns population change over a period and the stability of the core vote. Figure 2 shows how relative turnout to European elections has remained quite stable. The most significant variation occurs between time periods 2 and 3 in the wards of Barham and Wembley Central in Brent South which experienced an increase in relative turnout. There is also some variation in Brent East but this is not particularly substantial. If European turnout is to be treated as the core vote, then any change in relative turnout should in part reflect population changes. Both Barham and Wembley Central saw above-average population migration in the year before the 2001 Census whilst the areas with stability in relative turnout to European elections are also stable with respect to population movement, with below-average migration in the year prior to the 1981, 1991, and 2001 Censuses. Overall, there is a moderately strong correlation between changes in relative turnout between consecutive time periods and population migration (r 0:61 for time periods 1 and 2 and r 0:52 for time periods 2 and 3). The third concerns the characteristics of the core voters and how these have changed through time. Table 2 shows correlation coefficients of several key voter characteristics at polling-district level against European turnout. Whilst there has been some change in the strength and significance of the relationship between the Table 2. Correlation coefficients for polling-district-level European turnout in each time period. Variable (%) Age 18 ± 29 years Retired Unemployed Economically inactive White Asian Managerial/professional occupation Degree qualification Local authority housing tenure Private rented housing tenure Three or more cars

1979

1989

1999

ÿ0.65** 0.32** ÿ0.66** 0.65** 0.45** ÿ0.11 0.70** 0.54** ÿ0.37** ÿ0.22** 0.64**

ÿ0.36** 0.22** ÿ0.44** 0.33** 0.28** 0.14 0.52** 0.37** ÿ0.35** 0.04 0.41**

ÿ0.33** 0.41** ÿ0.32** ÿ0.26* ÿ0.16 0.47** 0.03 ÿ0.05 ÿ0.28** ÿ0.08 0.41**

** Significant at the 1% level; * significant at the 5% level.

1256


(a)

(b)

Top quintile Middle quintile Bottom quintile

(c) Figure 2. Percentage European turnout classified into quintiles by Brent polling districts for the three time periods: (a) 1979; (b) 1989; (c) 1999.

variables and European turnout, particularly in the white and Asian variables and in the variable measuring managerial/professional occupation, there has also been some relative stability: for instance, in the variables relating to local authority housing tenure and car ownership, and to a lesser extent, in the age variable. These changes will in part reflect the changing demographics of the polling districts but will also reflect changes


1257

in voter participation in European elections. The critical thing to note is that, although the characteristics of the core voters may not have stayed the same over the duration of the study (indeed it is perfectly reasonable to assume that there will be some changes in their demographic composition, particularly in areas of population migration), it is the stability within time periods that is important. Hence, the socioeconomic characteristics of the core vote may vary between the three time periods, but we are assuming that they are relatively stable in the short duration within each time period. The fourth caveat concerns electoral participation and the occasional use of European elections in protest voting. In this instance, the 1989 European elections were characterised by a large number of people voting for the Green Party, apparently as a protest vote against the main parties (Curtice, 1989). Certainly, evidence from Eurobarometer 32 (1989) conducted after the European election does not suggest that a new wave of previous nonvoters was responsible for the increase in support for the Green Party. We are assuming that in this election, the core voters participated as usual, a significant proportion defecting from the main parties towards the Greens. By examining the relationships between local and European turnout and parliamentary and European turnout within each time period, the effects of marginality of wards and constituencies should become apparent given that the social and demographic characteristics of the voting population should remain substantially the same within a particular time period. Other factors such as the weather on polling day and the time of year may also cause variations in turnout but these are not controlled for in this study. However, it is assumed that these factors will have only a small impact and that they will not have a significant effect on our results. Two principal statistical techniques were used to investigate the effect of marginality upon turnout to local and parliamentary elections. First, spatial autocorrelation was calculated to measure the degree of spatial clustering in levels of turnout by polling district. Levels of voter turnout will exhibit some degree of spatial clustering due to voters in neighbouring polling districts sharing similar views and characteristics. However, if marginality is also a factor influencing turnout, then we may expect this spatial clustering to be stronger for polling districts within the same wards for local elections and polling districts within the same constituencies for parliamentary elections when compared with the spatial clustering of polling districts which ignore these membership ties. To this end, Moran's I index was calculated as a measure of spatial autocorrelation following Cliff and Ord (1981), with three definitions of neighbourhood. The first treats two polling districts as neighbours if they share a common boundary (that is, they are adjacent). As is usual, a row-standardised spatial weights matrix is used to define the geographical connectivity, corresponding to a form of spatial smoothing which simplifies the interpretation (Anselin, 1995). The second treats polling districts as neighbors only if they are located within the same ward, regardless of whether they share a common boundary. The third treats polling districts as neighbours only if they are members of the same constituency. If marginality is causing an increase in turnout, then we would expect measures of spatial autocorrelation based on the second and third definitions of neighbourhood to be stronger and statistically more significant than measures based on the first definition. The second statistical technique used to investigate the effect of marginality upon turnout is multilevel modelling. Since the organisation of elections is multilevelledöin the sense that people vote within polling districts that are nested within particular wards within constituenciesöwe may expect a complex geography of turnout to occur. Thus variation in polling-district turnout depends not only upon the core vote within a polling district but also upon the degree of marginality and local campaigning in the ward and also the characteristics of the constituency. Multilevel modelling is increasingly

1258


being used in electoral studies (eg Johnston et al, 2003) and we believe that this is the first time that it has been used to analyse the spatial variations in election turnout. Results and discussion Descriptive analysis of election turnout in the three time periods

Statistical summaries of percentage turnout at each election in each of the three time periods are reported in table 3. Graphical summaries indicated that percentage turnout approximates a normal distribution in nearly all of the elections across the three time periods. The standard deviation measures the variation in turnout between polling districts for each election. Within each time period the greatest variation in turnout occurs in local elections although the differences with parliamentary elections are rather small and for period 3 there is very little difference. Table 3. Descriptive summary of percentage turnout by polling district to the parliamentary (P) and local (L) elections in each time period.

Turnout Maximum Minimum Mean Standard deviation Moran's I adjacent neighbours z-score ward member z-score constituency member z-score N a b c

Time period 1a

Time period 2b

Time period 3c

P

P

P

L

L

L

82.48 44.72 71.16

68.22 11.71 46.86

81.24 51.92 68.02

57.06 18.43 42.91

67.21 13.83 49.15

54.43 3.82 36.22

7.08

8.72

6.03

6.72

7.27

7.40

0.624 10.75 0.632 9.60 0.341 7.89

0.525 9.07 0.556 8.47 0.244 6.94

0.404 7.18 0.530 8.08 0.256 7.64

0.354 6.27 0.492 7.71 0.142 4.73

0.217 3.91 0.395 6.22 0.121 6.69

0.307 5.86 0.442 6.94 0.080 4.37

114

114

114

114

115

115

1979 parliamentary elections, 1978 local elections. 1992 parliamentary elections, 1990 local elections. 2001 parliamentary elections, 1998 local elections.

The Moran's I indexes for percentage turnout are statistically significant at the 5% level for all elections. In time period 1, there is very little difference in the strength and significance of spatial clustering measured by ward membership or polling-district adjacency although spatial clustering at the constituency level is weaker. However, spatial clustering measured by ward membership is stronger in time periods 2 and 3. It can also be observed that the strength of the spatial clustering of percentage turnout is strongest in elections in time period 1 with this strength declining in time periods 2 and 3, regardless of how spatial clustering is measured. This suggests two things: that differences in percentage turnout are becoming spatially more diverse at the pollingdistrict level and that wards are increasingly important for explaining spatial differences in turnout to all elections. The latter suggests that wards are increasingly becoming important sites of campaigning in parliamentary elections and not just local elections. Figures 3 ^ 5 summarise percentage turnout at parliamentary and local elections in time periods 1 to 3, respectively. Percentage turnout has been classified by quintiles across the borough, thus allowing turnout to be compared on a relative basis for


1259

(a)


(b) Figure 3. Percentage turnout classified into quintiles by Brent polling districts for time period 1: (a) parliamentary election 1979, (b) local election 1978.

each election. Marginal wards at local elections are also highlighted. Figure 3 shows that there is a clear geography in turnout in each election. Turnout is much higher relatively for polling districts in Brent North than in the other two constituencies. However, this geographical pattern varies with election, with the 1978 local election appearing to have a greater spatial clustering of high and low turnouts when compared with the 1979 parliamentary election. In Brent North, for example, the solid black mass at the 1978 local election reflects high turnout in polling districts in the very marginal Fryent and Roe Green wards, abutted by polling districts whose turnout appears to be high whatever the election being contested. Figure 4 shows the relative turnout at each election in time period 2. There are clear differences in the geography of turnout with polling districts within the top quintile being more scattered than in period 1. The 1992 parliamentary election shows quite similar patterns of turnout in Brent North and Brent East but rather different ones in Brent South. In particular, the incidence of turnout in Brent South for the 1990 local election is relatively high compared with the 1992 parliamentary election.

1260


(a)



A number of polling districts in Brent East appear in the top quintile for each type of election. However, although this might reflect the parties' desire to build an organisation in terms of fighting the parliamentary marginal in 1992, the relative impact on turnout was in fact limited to a handful of polling-district areas in some of the more closely contested wards. Figure 5 shows the relative turnout in time period 3. Again there are clear differences in the geography of turnout, with high rates of turnout concentrated in Brent North for the 2001 parliamentary election and Brent South and Brent East for the 1998 local electionsöthe latter marking a distinct difference between the 1998 and 1978 contests. It is worth noting that the high turnout concentration in Brent South in the local election of 1998 reflects the only two wards in Brent, Alperton and Barham, won by the Liberal Democrats and the two Labour ^ Conservative marginal wards. Similarly, the polling districts in Brent East which fall into the top or second quintile cover most of the area of the five marginal wards within the constituency.


1261

(a)



The three sets of maps imply that not only do significant differences in relative turnout exist for the two elections but that these differences vary over time. Modelling marginality using multilevel analysis

Three multilevel models were estimated for the parliamentary and local elections in all three time periods. The first model, represented by equation (1), simply calculates the contribution of each level (polling district, ward, and constituency) to the total variance in turnout across all three levels, without taking into account the core vote in particular places. This is the null model; no predictor variables were included besides that representing the intercept. The second model, represented in equation (2), estimates the impact of the rate of European election turnout (that is, the core vote) and ward marginality (in local elections) on the variation in election turnout (the random intercepts model). The final model, equation (3), estimates the interaction of European election turnout at higher levels to examine whether the effects of the core vote on election turnout varies spatially across Brent (the fully random model). The models

1262


were estimated using iterative generalised least squares (IGLS) within the multilevel modelling package M1wiN Version 2.1d (Rasbash et al, 2002). Tijk ajk Xijk vCk Xijk mW jk Xijk ePDijk Xijk , Tijk ajk Xijk b0 T

E ijk

L jk

b1 M vCk Xijk mW jk Xijk ePDijk Xijk ,

(1) (2)

E E Tijk ajk Xijk b0 Tijk b1 MjkL vCk Xijk mW jk Xijk tW jk Tijk ePDijk Xijk , (3)

where i 1, .::, 114 are polling districts (115 in time period 3); j 1, .::, 31 are wards; and k 1, .::, 3 are constituencies. Tijk is turnout at local (or parliamentary) election in polling district i within ward j from constituency k; E Tijk is turnout at the European election in the same polling districts; L is a dummy variable with a value of 1 if the ward is marginal; 0 otherwise Mijk in local elections; Xijk is a column vector which consists entirely of ones; a and b are the fixed parameters to be estimated; vCk , mW jk , and ePDijk are three random parameters to be estimated that represent the variation of turnout between constituencies, between wards within a constituency, and between polling districts within a ward and are assumed to be normally distributed 2 2 with zero mean values and unknown variances sC2 , sW , and sPD , respectively. In equation (3), the random intercept gW jk is assumed to be normally distributed with zero mean value and tW variance. In addition, at ward level the random components are allowed to be correlated with covariance coefficient rW . The first set of models [equation (1)] summarised in table 4, are the simplest since they contain no fixed terms except for the overall intercept (constant). This is the estimated average rate of turnout for the whole of Brent for each election in each time period. The numbers in brackets are the standard errors of the coefficients and can be used to judge significance of the estimated parameters. This works well for the fixed parameters, and is similar to testing significance in ordinary least squares regression. However, for random parameters, Woodhouse et al (1996) recommend using a likelihood ratio statistic. By checking if the likelihood ratios of successive models are significantly different the significance of additional terms may be evaluated. The model allows the variation around the estimated average turnout to be partitioned into variation at the level of the individual polling district, ward, and constituency. There is no significant variation at the constituency level for either pair of elections in any of the three time periods (shown by the large standard errors) indicating that a constituency-level effect does not exist, not even in parliamentary elections. However, the lack of significance may be due to the small number of constituencies in the study (only three). The greatest variation in turnout occurs between polling districts although significant variation in rates of turnout also occurs at ward level, particularly for local elections. The only exception is ward-level variation in parliamentary election turnout in time period 3 which is insignificant, with all significant variation occurring at polling-district level. Due to its insignificance, the constituency level was removed and the models were reestimated but this had very little effect on the estimated parameters. Hence for the sake of consistency, the estimated parameters of all the subsequent models are shown with the constituency included as level 3. A comparison of the proportion of the significant variation in the models (that is, at the polling-district and ward levels) indicates that in time period 1, around 50% of the significant variation in local election turnout occurs at the ward level compared with around only 30% for parliamentary elections. This rises to 55% compared with 15% in time period 2 and then falls to 43% compared with no significant variation in


1263

Table 4. The null model for each time period and for parliamentary (P) and local (L) elections. Numbers in parenthesis are standard errors.

Fixed Constant Random Constituency level Ward level Polling-district level ÿ2 log-likelihood

Time period 1

Time period 2

Time period 3

P

L

P

L

P

L

70.72 (2.72)*

46.48 (2.92)*

67.32 (1.32)*

45.75 (1.20)*

49.19 (1.88)*

36.40 (1.02)*

20.53 (18.01) 12.62 (4.72)* 17.27 (2.68)*

22.17 (20.70) 26.67 (9.28)* 27.62 (4.29)*

3.38 (4.28) 13.73 (5.10)* 18.16 (2.82)*

1.54 (3.56) 22.55 (7.58)* 20.40 (3.17)*

9.12 (8.66) 4.82 (4.32) 37.92 (5.83)*

0.00 (0.00) 23.47 (8.37)* 30.69 (4.65)*

696.05

754.45

698.01

718.46

762.25

761.37

* Significant at the 5% level.

time period 3. Overall, then, around one half of the variation in rates of turnout to local elections occurs at the ward level, whereas this has fallen from 30% to 0 for parliamentary elections. Again this is possibly indicative of the effects of campaigning by ward-based parties in local elections. The second set of models [equation (2)], summarised in table 5, include the influence of the core vote and ward marginality and will allow an assessment of the contextual effects. The differences in the likelihood ratio statistics of the second set of models with the first set of models is statistically significant at the 1% level (Woodhouse et al, 1996), strongly indicating that both variables have an important effect in explaining variation in turnout to local and parliamentary elections. Table 5. The random intercepts model for each time period and for parliamentary (P) and local (L) elections. Numbers in parenthesis are standard errors.

Fixed Constant European election (T E ) Marginality of ward (M L ) Random Constituency level Ward level Polling-district level ÿ2 log-likelihood * Significant at the 5% level.

Time period 1

Time period 2

Time period 3

P

L

P

L

P

L

52.40 (2.01)* 0.67 (0.06)*

23.75 (2.31)* 0.79 (0.08)* 3.95 (1.65)*

42.04 (2.72)* 0.83 (0.08)*

10.49 (2.54)* 0.99 (0.08)* 4.37 (1.13)*

27.59 (2.71)* 0.99 (0.11)*

10.39 (2.32)* 1.18 (0.101)* 4.34 (1.23)*

2.95 (2.96) 4.32 (1.92)* 9.80 (1.52)*

0.00 (0.00) 13.16 (4.65)* 17.81 (2.77)*

0.65 (1.02) 2.20 (1.68) 13.58 (2.10)*

0.00 (0.00) 6.20 (2.18)* 8.18 (1.27)*

4.64 (4.32) 0.00 (0.00) 24.64 (3.30)*

0.00 (0.00) 6.49 (2.84)* 16.08 (2.48)*

618.3

692.1

637.6

603.9

701.2

673.9

1264


The inclusion of the European election and ward-marginality variables has resulted in a decline of the polling-district variance and the ward-level variance although proportionally both levels explain roughly the same amount of the variation in turnout as in the previous models. Hence, even after the core vote and marginality have been taken into account, the ward level still accounts for around 40% of the variation in turnout to local elections (although this has fallen to 30% in time period 3). However, it would seem from the standard error that the ward-level variances for the parliamentary election in time periods 2 and 3 are insignificant and all the significant variation in turnout occurs at the polling-district level. Thus the evidence of spatial clustering in polling-district turnout to parliamentary elections by ward membership in table 3 is a Predicted local turnout (%)

75 65 55 45 35 25 0

(a)

10

20

30

40

50

60

Predicted local turnout (%)

60 50

40

30 20 10

(b)

20

30

40

Predicted local turnout (%)

60 50 40 30 20 10

(c)

0

10

20 30 European turnout (%)

40

50

Figure 6. Predicted ward-level local election turnout against European election turnout in each time period: (a) period 1, (b) period 2, (c) period 3.


1265

result of the wards with the spatial clusters containing a high degree of core voters as opposed to the effect of local campaigning. The significance of the ward-marginality variable in local election turnout across all three time periods is particularly interesting as it lends empirical support to the argument that ward marginality has a positive effect on voter turnout. The random-intercepts models in table 5 assume that, while the relationships between core voter turnout and turnout to local elections in each ward are the same, some wards have uniformly higher turnout rates than others. In order to investigate whether the relationship between core voter turnout and local election turnout varies between wards, the European election variable was included at the ward level [equation (3)] and the models estimated for all three time periods (Duncan et al, 1998). None of the estimated ward-level variance and covariance terms in these fully random models was significant, suggesting that the relationships between core voter and local authority turnout in each ward are the same, as described in table 5. The fully random models emphasised heterogeneity in terms of differences between wards. We also investigated whether there was complex heterogeneity between polling districts. The presence of heterogeneity at the polling-district level would indicate that core voter turnout is variable between polling districts within wards. Moreover, pollingdistrict heterogeneity can have important implications for estimates of ward-level estimates, as what may appear to be ward-level contextual variability may in fact be between-polling district, within-ward heterogeneity (Bullen et al, 1997). Hence the European election variable was included at the polling-district level and models estimated for both elections in all three time periods. However, none of the estimated polling-district-level variance and covariance terms in any of the models was significant and therefore we can conclude that complex between-polling-district heterogeneity in core voter turnout does not exist. In order to investigate in more detail how ward-level local election turnout varies with core voter turnout, predicted local election turnout was plotted against European turnout and grouped at the ward level for each election, based on the random-intercepts models (see figure 6). The graphs show how turnout varies by ward after adjusting for turnout in European elections. The thick line in each graph is the average predicted turnout from the fixed part of each model. High rates of European election turnout result in high ward-level turnout to local elections, although there are some exceptions. Local election turnout between wards becomes more similar through time, indicated by the bunching of regression lines between time periods 1 and 3. Concluding remarks This analysis has presented some initial empirical evidence for the size and potential effects of marginality on turnout and change through time. The analysis has shown that relative turnout at different types of elections is not stable even within tightly constrained time periods. By assuming that turnout to European elections represents the core vote we have shown that there are significant differences in the relative rates of participation between wards and polling districts. For a substantial number of polling districts it is certainly not the case that a relatively high turnout at one type of election automatically means a relatively high turnout at other elections. Indeed, our analysis has shown that quite often the opposite is true, with electors in particular polling districts participating in one type of election but not another. Since the time between each election is quite small, this disparity cannot be explained by the changing social characteristics of the polling district alone. Multilevel modelling allowed us to tease apart the effects of the core vote from the effects of wards on turnout. It would appear that wards do affect turnout in local elections,

1266


although the strength of the effect has diminished slightly through time. Parliamentary elections do not appear to be influenced by ward context with all the variability in turnout occurring at the polling-district level. The analysis has also shown that constituencies have no significant impact on rates of turnout, even in parliamentary elections, although this result is probably due to the small number of constituencies in our sample and may not hold true for bigger samples and for marginal constituencies (see Pattie and Johnston, 2005). In local elections, ward marginality has a significant effect on voter turnout beyond what we would expect given core voter turnout. This may reflect an increase in the concentration of party campaigning in particular wards and could provide an explanation to the observed trend that the differences in the relative rates of turnout are increasing whilst geographically they are also becoming more spatially concentrated. However, since marginality and local campaigning are two potentially distinct effects, care is needed in assigning the ward-level effects to the latter in the absence of empirical data in the models. The fact that ward-level context, such as marginality, has a significant explanation in the variation in turnout in local elections could have important policy implications for increasing levels of turnout in all elections. For instance, one way to increase turnout to parliamentary elections may be to target campaigning within constituencies at particular wards, similar to local elections. There is anecdotal evidence that this occurred to some extent in the 2005 general election, with the Liberal Democrats targeting wards with a high proportion of students in key constituencies. To our knowledge this research is unique in British electoral studies in that it is the first to use polling-district-level data to examine turnout at different types of elections. It would be very difficult to replicate the analysis using a smaller number of larger spatial units, such as wards, since there would not be sufficient variation. The data presented here suggest that electoral campaigning (or its absence) may be the catalyst for increasing electoral awareness and voter turnout. Acquiring details of campaign activity at the ward/polling-district level for different types of election would shed light on the effects of party campaigns and provide greater support for some of the findings made in our paper. A further line of enquiry may be to investigate more directly the influence of population characteristics on differential voter turnout, particularly at ward level. Although previous research has suggested that factors that influence turnout at general, local, and European elections appear to be substantially similar they may contribute in part to some of the ward-level effects on turnout shown in the models. Finally, our data are restricted in their geography. Brent is an outer-London borough and as such represents only one type of residential pattern. Similar data to those examined here are available from a range of electoral registration officers in other parts of the country, including local authorities in suburban and rural areas. Assembling more complex datasets in this manner would greatly extend the scope for analysing multilevel electoral turnout in Britain. Acknowledgements. The authors would like to thank staff in the electoral registration department of the London Borough of Brent, particularly Sean O'Sullivan, for compiling and supplying the polling-district data used. References Anselin L, 1995, ``Local indicators of spatial association ö LISA'' Geographical Analysis 27 93 ^ 116 Blais A, 2000 To Vote or Not to Vote: The Merits and Limits of Rational Choice Theory (University of Pittsburgh Press, Pittsburgh, PA) Blais A, Carty R K, 1990, ``Does proportional representation foster voter turnout?'' European Journal of Political Research 18 March, 167 ^ 181 Blais A, Gidengil E, Nevitte N, 2004, ``Where does turnout decline come from?'' European Journal of Political Research 43 221 ^ 236


1267

Blondel J, Svennsson P, Sinnott R, 1998 People and Parliament in the European Union: Democracy, Participation and Legitimacy (Oxford University Press, Oxford) Bullen N, Jones K, Duncan C, 1997, ``Modelling complexity: analyzing between-individual and between-place variationöa multilevel tutorial'' Environment and Planning A 29 585 ^ 609 Butler D, Westlake M (Eds), 2000 British Politics and European Elections 1999 (Macmillan, Basingstoke, Hants) Butler D,Westlake M (Eds), 2005 British Politics and European Elections 2004 (Palgrave Macmillan, Basingstoke, Hants) Clarke H, Sanders D, Stewart M, Whiteley P, 2004 Political Choice in Britain (Oxford University Press, Oxford) Cliff A D, Ord J K, 1981 Spatial Processes: Models and Applications (Pion, London) Curtice J, 1989, ``The 1989 European electionöprotest or green tide'' Electoral Studies 8 217 ^ 230 Curtice J, 2005, ``Turnout: electors stay home öagain'', in Britain Votes 2005 Eds P Norris, C Wlezien (Oxford University Press, Oxford) pp 120 ^ 129 Curtice J, Seyd B, Thomson K, 2001 Devolution to the Centre: Lessons from London's First Mayoral Election WP 90, Centre for Research into Elections and Social Trends, Department of Sociology, University of Oxford, and National Centre for Social Research, London Denver D, Hands G, 1997, ``Challengers, incumbents and the impact of constituency campaigning in Britain'' Electoral Studies 16 175 ^ 194 Duncan C, Jones K, Moon G, 1998, ``Context, composition and heterogeneity: using multilevel models in health research'' Social Science and Medicine 46 97 ^ 117 Eurobarometer 32 1989, Directorate-General Information, Communication, Culture, Commission of the European Communities, Brussels Franklin M, 1996, `Èlectoral participation'', in Comparing Democracies, Elections and Voting in Global Perspective Eds L Leduc, R G Niemi, P Norris (Sage, London) pp 216 ^ 235 Franklin M, 2004 Voter Turnout and the Dynamics of Electoral Competition in Established Democracies Since 1945 (Cambridge University Press, Cambridge) Franklin M, van der Eijk C, Oppenhuis E, 1996, ``The institutional context'', in Choosing Europe? The European Election and National Politics in the Face of Union Eds C van der Eijk, M Franklin (University of Michigan Press, Ann Arbor, MI) pp 306 ^ 331 Heath A, Taylor B, 1999, ``New Sources of Abstention?'', in Critical Elections: British Parties and Voters in Long-term Perspective Eds G Evans, P Norris (Sage, London) pp 164 ^ 180 Johnston R, Jones K, Sarke R, Burgess S, Propper C, Bolster A, 2003, `À missing level in the analysis of British voting behaviour: the household as context as shown by analyses of a 1992 ^ 1997 longitudinal survey'', WP 3, ESRC Research Methods Programme, http://www.ccsr.ac.uk/methods/publications/Working%20Paper%203.pdf Mattila M, 2003, ``Why bother? Determinants of turnout in the European elections'' Electoral Studies 22 449 ^ 468 Miller W, 1988 Irrelevant Elections? (Oxford University Press, Oxford) Pattie C, Johnston R, 1998, ``Voter turnout at the British general election of 1992: rational choice, social standing or political efficacy?'' European Journal of Political Research 33 263 ^ 283 Pattie C, Johnston R, 2003, ``Local battles in a national landslide: constituency campaigning at the 2001 British General Election'' Political Geography 22 381 ^ 414 Pattie C J, Johnston R, 2005,`Èlectoral participation and political context: the turnout ^ marginality paradox at the 2001 British General Election'' Environment and Planning A 37 1191 ^ 1206 Rallings C, Thrasher M, 1990, ``Turnout in English local elections öan aggregate analysis with electoral and contextual data'' Electoral Studies 9 79 ^ 90 Rallings C, Thrasher M, 1997, ``The local elections'' Parliamentary Affairs 50 681 ^ 692 Rallings C, Thrasher M, 2003, ``Local electoral participation in Britain'' Parliamentary Affairs 56 700 ^ 715 Rallings C, Thrasher M, Downe J, Ridge M, Lowry M, 2000 Turnout at Local Government Elections: Influences on Levels of Voter Registration and Electoral Participation Department of Environment, Transport and the Regions, London Rasbash J, Brown W, Goldstein H, Yang M, Plewis I, Woodhouse G, Draper D, Langford I, Lewis T, 2002 A User's Guide to MlwiN Version 2.1d Centre for Multilevel Modelling, Institute of Education, University of London Reif K, 1984, ``National electoral cycles and European-elections 1979 and 1984'' Electoral Studies 3 244 ^ 255 Reif K, Schmitt H, 1980, ``Nine national second-order elections: a systematic framework for analysis of European elections results'' European Journal of Political Research 8 3 ^ 44

1268


Schmitt H, Mannheimer R, 1991, `Àbout voting and non-voting in the European elections of June 1989'' European Journal of Political Research 19 31 ^ 54 Whiteley P, Clark H, Sanders D, Stewart M, 2001, ``Turnout'', in Britain Votes 2001 Ed. P Norris (Oxford University Press, Oxford) Woodhouse G, Rashbash J, Goldstein H, 1996, `Ìntroduction to multi-level modelling'', in Multi-level Modelling Applications: A Guide for Users of Mln Institute of Education, University of London, pp 9 ^ 57 Worcester R M, Mortimore R, 2001 Explaining Labour's Second Landslide (Politico's, London)

ß 2007 Pion Ltd and its Licensors

Conditions of use. This article may be downloaded from the E&P website for personal research by members of subscribing organisations. This PDF may not be placed on any website (or other online distribution system) without permission of the publisher.