Office of the Deputy Prime Minister

Turnout at Local Elections

CHAPTER 2

Voter Turnout: A Statistical Analysis

INTRODUCTION

The key concern of the research in Stage 2 of this study was to determine how we could account for different levels of election turnout in different local authority wards. The work builds on the analysis of turnout carried out by the University of Plymouth in 1994 at local authority level. Specifically, we use multiple regression techniques to investigate the relationships between voter participation in local elections at ward level in England and a range of political, structural and socio-economic variables, and from this build a statistical model of turnout. The analysis covers the period between 1973-1999.

In addition, the statistical model developed was used to predict voter turnout for individual wards in order to identify those wards in which actual turnout consistently and significantly deviates from these predictions.

The statistical model is designed specifically to take advantage of the attributes offered by longitudinal (often referred to as panel) data. That is, information which is available over a long period of time and which permits comparison across that time. The raw material for the analysis is the University of Plymouth's database containing information on election results, electorates, candidates, parties and votes for all English local elections at ward level since 1973.

THE DATA

The electoral data including voter turnout, covers 72,818 ward elections in each of four types of local authorities since 1973. The types of local authorities are:

County council elections are excluded from the study. Electoral divisions at county level frequently bear little relationship to district wards and thus could not be systematically used in the longitudinal analysis.

Figure 1 shows how local election percentage turnouts at ward level have evolved by local authority type over the review period.

Figure 1 Mean ward % turnout by year and type of authority in England*

*Excluding distortionary general election years 1979 and 19971.
**The final row of the data table shows the total number of ward elections in each year.

Figure 1 shows that overall participation in local elections across types of authority has been in general decline since 1990. Between 1990-1999 the number of voters turning out to vote at each election represented 36% of the registered electorate. Between 1973-1978 the figure was 43%. The average turnout for the period as a whole was 41.7% in 72,818 ward elections.

Figure 1 also indicates the following:

In addition to turnout and authority type the elections dataset includes structural and political data which are used to develop variables for inclusion at the statistical modelling stage. Table 1 lists the variables that we have used in the statistical analysis and shows their mean values for the period and the correlations2 between them and voter turnout. The left-hand side of the table highlights the positively correlated variables. Conservative control of the local authority, the Conservative share of the vote, and the potential to change the political colour of the council in councils with annual elections are positively correlated with turnout. On the right-hand side are those variables negatively correlated with turnout. The size of the electorate, the Labour share of the vote in a ward, councils that are Labour controlled, and councils with annual elections are all negatively correlated with turnout. Another important negative correlation is between voter turnout at the ward level and both the percentage majority in terms of votes at the last ward election and the percentage majority in terms of seats at the last council election. In other words, the safer the ward or council is for the incumbent party, the lower the turnout tends to be.

The socio-economic variables in Table 1 are drawn from the 1981 and 1991 Censuses. However, because the census data cover just two years of the study period, we need a method of generating data for all election years. We do this by interpolating/extrapolating from the 1981 and 1991 census data to cover the same period as the electoral data. We tested two different approaches to interpolation/extrapolation, each yielding similar results. It is difficult to determine the degree to which the interpolations and extrapolations are an accurate representation of the actual variables, but it must be accepted that they might introduce an element of measurement error for the socio-economic variables.

In terms of the correlations between socio-economic variables and turnout, it can be seen from Table 1 that the proportions of the self-employed, of professionals, of owner-occupiers and of pensioners in a ward are associated with higher levels of electoral participation. Conversely, there is a high negative correlation between voter turnout and the proportions of young people aged between 16 and 29, of council tenants, and of the unemployed.

 

Table 1 Means of variables and correlation with turnout (1973-1999)
Political and Structural variables
Positively correlated variables corr mean Negatively correlated variables corr mean
potential change in 'thirds' councils 0.28 0.46 Labour share of vote -0.43 34.92
Conservative controlled LA 0.23 0.29 ward electorate size (logged) -0.38 8.39
Conservative share of vote 0.21 34.56 Labour controlled LA -0.31 0.39
ratio of vacancies to electorate x 100 0.18 0.04 % majority in seats of winning party at previous LA election -0.27 13.05
Lib-Dem share of vote 0.16 19.76 % majority in votes of winning party at previous ward election -0.27 25.42
other share of vote 0.14 10.76 elections for 1/3 of seats annually -0.26 0.59
no overall control in LA 0.10 0.25 same party control for at least previous two elections -0.11 0.62
independent controlled LA 0.08 2.66 single member wards -0.08 0.64
ratio of candidates to vacancies 0.07 0.03 ratio of other candidates to total candidates -0.06 0.77
No. of parties contesting election 0.01 2.83 Lib-Dem controlled LA -0.04 0.04
Socio-economic variables
Positively correlated variables corr mean Negatively correlated variables corr mean
% distribution and catering workers 0.23 31.10 % aged 16-29 -0.39 18.45
% professionals 0.23 20.43 % without a car -0.24 30.76
% agricultural workers 0.23 2.45 % council tenants -0.17 20.59
% self employed 0.19 12.28 % unemployed -0.11 6.00
% no bathroom 0.16 0.73 % of households with more than 1 person to a room -0.11 2.14
% owner occupiers 0.14 67.16 % persons born in New Commonwealth or Pakistan -0.10 4.40
% pensioners 0.14 18.62 % skilled workers -0.08 22.36
% manufacturing workers 0.12 23.26 % students 0.00 5.18
% manual workers 0.00 37.29      

Having identified the variables that might theoretically explain the variations in turnout we now turn to the task of implementing the statistical model.

THE STATISTICAL MODEL

In the previous sections we described the size and of scope of the elections dataset and developed a range of factors that may influence voter turnout either positively or negatively. The main objective of this section is to explore in detail the empirical relationships between voter turnout and the range of political, structural and socio-economic variables set out in Table 1.

The statistical tool used in the following analysis is multiple regression. This is a statistical technique that can be used to explain the variation in voter turnout that arises from the variation in the different characteristics observed in each ward or local authority. The reason we are extending our analysis in this way is to apply a more rigorous test to the data rather than relying on simple correlations between voter turnout and the political, structural and socio-economic variables.

For example, the correlation statistic in Table 1 showing a strong negative relationship (-0.38) between voter turnout and the size of the electorate in each ward might imply that larger wards tend to have lower turnouts. However, before we accept this hypothesis it is important to consider in tandem other factors that may influence voter turnout. For instance, consider the following relationship:

In this case the negative correlation we observe between electorate size and voter turnout is likely to be biased if not spurious. This is because the correlation coefficient will also be picking up the influence of the proportion of pensioners in wards - and indeed any other variables that are correlated both with electorate size and voter turnout. However, in multiple regression analysis it is possible to separate out the effects of the numbers of pensioners and size of electorate on voter turnout, a separation not possible with simple correlation analysis.

The Results
Through our analysis of the dataset we established five multiple regression models. These regression equations were split into two groups. The first two equations were conducted using the Ordinary Least Squares (OLS) method of regression3. One equation used only
the political and structural variables and produced a model which explained 56% of the variation in ward-by-ward turnout; the second added the socio-economic variables and the proportion of the variation in turnout explained increased to 59%. A weakness of this method of regression is that it does not allow us to include all the factors that may influence voter turnout in the models i.e. some important information may be omitted through lack of a reliable measure. But, omitting variables can potentially bias the estimated results. One way of eliminating a significant part of such bias is to estimate voter turnout using a 'fixed effects' approach. In brief, this approach yields estimated equations that take into account the influence on voter turnout of variables that are fixed over time but vary between wards - the size of the ward in terms of hectares is an example. Our second group of equations was based on this method and it is those which shape our reported findings. The first equation uses exactly the same variables as in the second of the OLS regressions; the second adds
a measure of how far any single party had dominated control of the council; and the third further adds an indicator of whether an election held in a council with annual contests could have lead to a change in party political control. We present the results of these three fixed effects models below.

Political Factors Influencing Voter Turnout - Findings

  • a negative relationship between voter turnout and the % majority of the winning party(in terms of votes) at the previous election in the ward - the size of the effect is relatively small indicating a 10 percentage point increase in the majority at the previous election would result in a 0.5 percentage point decrease in turnout in the current election;

  • a negative relationship between voter turnout and the % majority of the winning party(in terms of seats) at the previous election in the local authority - the size of the effect, at -0.07, is relatively small indicating that a 10 percentage point increase in the majority atthe last election would result in a 0.7 percentage point reduction in voter turnout;

  • a negative relationship between voter turnout and the size of the electorate -this result implies that a 10 per cent increase in the size of the ward population would lead to a 0.7 percentage point reduction in voter turnout;

  • the relationship between voter turnout and the ratio of candidates to vacancies is insignificantly different from zero. The marginal effect of the ratio of vacancies to electorate on turnout, although statistically significant, is extremely small - a tiny fraction of1 percentage point reduction.

  • single member wards have a significant negative impact on voter turnout relative to multi-member wards - in the order of 1.9 percentage points - one of the strongest negative effects;

  • the number of political parties contesting the ward election also has a strong effect on voter turnout, but this time the effect is positive - increasing the number of parties contesting ward elections by one is estimated to increase turnout by 1.67 percentage points;

  • another strong positive influence is the ratio of candidates not affiliated to any of the major parties to affiliated candidates. If this ratio increases by one, turnout is estimated to increase by 1.2 percentage points. In other words, the presence of Independent candidates does seem to boost turnout.

  • Control of a council by either the Labour or Conservative parties appears to have a positive impact on voter turnout compared to councils where there is no overall control. However voter turnout declines as their share of the vote - another indicator of political dominance- increases. By contrast Liberal Democrat control of a council has a negative impact on turnout, but as their share of the vote increases so does voter participation.

Socio-economic Factors Influencing Voter Turnout - Findings

  • The proportion of pensioners and ethnic minorities has a positive impact on voter turnout with each one point increase in a ward's population leading to a two-thirds and one-third of a percentage point increase in turnout respectively.

  • The socio-economic, as opposed to demographic, characteristic having the largest effect on turnout according to this model is overcrowding. A one point increase in the proportion of the population living in over-crowded accommodation appears to lead to about a quarter of a percentage point increase in turnout.

  • Perhaps the most significant negative socio-economic impact on turnout is the proportion of young adults in a ward. The inference here is that if this proportion increases by 0.01, or 1 percentage point, turnout would decline by a tenth of one percentage point.

  • Amongst the other statistically significant socio-economic effects are the proportions of owner-occupiers, manufacturing workers, unemployed, agricultural workers, and students. The former three having small positive effects and the latter two small negative effects. The coefficient on council tenants is positive and statistically significant, while all the other variables (professionals, skilled workers, manual workers, percentage without cars, service workers and the self-employed) are insignificant.

Finally, when we added the measures of political competition described earlier to the equation, there did seem to be a link (albeit tenuous) between voter turnout and the competitiveness of electoral politics. In wards in those LAs which hold elections for one third of seats annually and where there was potential to change control, turnout was on average about 0.7 of a point higher than overall mean turnout. However, political dominance, that is where the same party has been in control through at least the previous two elections (which occurred in almost 50% of cases), does not seem to have a significant impact one way or the other on the level of turnout at ward elections.

The overall impact of the political variables is interesting. Almost every measure shows a positive relationship between electoral competitiveness and turnout. Marginal seats, marginal councils, additional candidates, and the scope for an election to change the political control of the council all lead to higher rates of participation. In other words, electors appear more willing to vote in circumstances where they have a greater chance of influencing the outcome and where local politicians are more likely to encourage them to do so.

Robustness
In order to check the constancy and robustness of our results we have estimated some additional models using the fixed effects approach described earlier. These are:

The key differences between the two time periods are the following:

Comparing 1973-86 with 1987-99

  • The effect of the percentage majority in votes at ward level on voter turnout is relatively constant across time periods. But, the impact of the percentage majority in seat at the council level is significantly greater in the 1987-99 period;

  • The size of the electorate has a smaller negative effect in the latest period - approximately a 10 per cent increase in the electorate pushes turnout down by 0.5 percentage points between 1987-99 compared to 1 percentage point in the earlier period;

  • The effect of the ratio of candidates to vacancies was negative in the period 1973-86 and positive in the latter period - the results from the most recent period indicate that an increase in the ratio from 2 to 3 would increase turnout 0.48 of a percentage point;

  • The impact of the ratio of vacancies to electorate on turnout declines to a negligible value in the most recent period;

  • The size of the regression coefficients for the remaining political and structural variables tends to be smaller in the most recent period;

  • There are some significant changes of sign and size on the socio-economic variables

The key differences between local authority types are the following:

Comparing across local authority type

  • The effect of the percentage majority in votes at the ward level and of majority in terms of seats at the local authority level are relatively constant across type of authority;

  • The negative impact of the size of the electorate is significantly greater in London - approximately a 10 per cent increase in the size of the ward electorate in London Boroughs reduces turnout by 1 percentage point;

  • The effect of the ratio of candidates to vacancies is positive for London Boroughs and shire districts with all out elections but negative for metropolitan districts and shire districts with annual elections;

  • Single member wards have a far greater negative effect in shire districts with all out elections - single member wards lead to a 2.6 percentage point reduction in turnout in these areas relative to multi-member wards;

  • Higher shares of the Labour vote have systematic negative effects on turnout across all types - the converse is true for the Liberal Democrats.

These additional equations suggest that some variables may have a varying level of impact on turnout in the two time periods and/or in different types of local authority. However, we are not able to estimate how far these differences may be real, or how far they may be a product of remaining bias, for example in the use of interpolations and extrapolations based on the 1981 and 1991 censuses

Geography

It is possible that geographic factors may influence the turnout in certain wards. We therefore introduced into our analysis two variables that were only available for 1991:

These variables were added to an estimated equation based solely on 1991 data including all the political, structural and socio-economic variables previously described.

The estimated coefficients indicated a lack of correlation between voter turnout and density, but a strongly negative one between voter turnout and migration. In other words, the more stable the population of a ward, the higher the likely turnout.

Local Authority Electoral Budgets
We also looked at the relationship between local authority electoral budgets per elector and voter turnout. This work aims to be purely indicative at this stage. We have run correlations between the budget/elector ratio and change in turnout in 1999 (or 1998) and 1995 (or 1994) using DETR data and our own survey data. The correlations are:

Correlations between budget/elector ratio and the absolute level of turnout in 1999 (1998) are:

The only significant sign showing any relationship is in London. There is a negative correlation between spend and turnout but a positive one between spend and change in turnout. This may suggest that money has been beneficial in helping to stem the rate of decline in turnout - or at least for this to happen artificially if the money is used to make the electoral register more accurate (and smaller). Elsewhere there is little indication that electoral budgets have any appreciable impact on turnout levels.

Local Authority Initiatives
The data used to assess the impact of local authority practices and initiatives on turnout are primarily drawn from the survey of local authorities carried out by the University of Plymouth in Stage 1 of this research project. We also included variables on participation from the De Montfort University study recently carried out for the DETR (Lowndes et al, 1998). The De Montfort study primarily established the number and range of initiatives in use but not the amount of participation going on. The dependent variable in the regression was the most recent turnout figure at the local authority level. Our analysis has shown that there is very little sign that the wide range of local authority initiatives and practices to enhance registration and participation have yet had a positive effect on voter turnout. Indeed, only one variable, the ratio of postal votes to the size of the electorate, was positively significant. However, carrying out a statistical analysis with just a one-year snapshot from the survey is not an ideal method for testing the effectiveness of local authority initiatives and practices. Their impact needs to be assessed over the longer term.

League Tables
For illustrative purposes and as a means of exploring the practical implications of the results from the estimation work, we calculated league tables of wards and local authorities. The estimated equations attempted to model turnout by including the variables we believed to be influential. The approach used to generate the league tables is to look for wards that are performing well even after having considered the range of influences included in the model. Thus, if after explaining a ward's turnout there remains a significant positive element of turnout that is unexplained then this type of ward would be regarded as a high performer. Conversely, a large negative element of unexplained voter turnout placed the ward in the low rankings. Thus, wards are high/low performers after taking into account their political, structural and socio-economic condition. These results were used to inform our selection of case study sites.

CONCLUSIONS

In order to carry out a statistical analysis of voter turnout we have used a dataset that includes a range of elections data and socio-economic variables for wards in England during the period 1973-1999 inclusive. We have used multiple regression analysis to explore the impact of a range of structural, socio-economic, political, and geographical factors on voter turnout at the ward level. The main findings from the statistical analysis indicate the following:

Our main findings from the multiple regression show that the influences on voter turnout are:

The influence of socio-economic variables on turnout appeared to vary depending on the regression method adopted. In one set of analyses it did appear that, for example, the proportion of pensioners and of ethnic minorities had a strong positive impact on voter turnout. The strongest of the negative effects was that wards with higher proportions of young people had lower turnouts. However, using a different regression technique, much of the impact of socio-economic effects on voter turnout appears to have less influence than previous studies have suggested. This is because our preferred statistical approach indicates that previous estimates of socio-economic effects may have been statistically biased. Moreover, it is also the case that most of the socio-economic observations in the dataset were based on extrapolations or interpolations from 1981 and 1991 Census data. Thus, some caution must be exercised with the reported interpretation of the socio-economic results.

1 Local elections coincided exactly with general elections in these years. Turnout at the local elections was 69.4 per cent and 75.2 per cent in 1979 and 1997 respectively.
2 The correlation is an indicator of the strength of the linear relationship between two variables but does not imply causality.
3 Ordinary Least Squares (OLS) is a statistical method which allows you to plot a line through a set of observations which 'best fits' the trend of those observations.

[Previous] [Contents] [Next]

Published 3 May 2000
Return to Local and Regional Government Index
Return to Homepage
Web site terms