Differential rates of postal vote rejections in England and Wales 2007
Colin Rallings and Michael Thrasher,
LGC Election Centre, University of Plymouth
Under the provisions of the Electoral Administration Act 2006 all those electors opting to vote by post are required to provide two pieces of personal information both when they apply for a postal ballot and when they return it. This record of their signature and date of birth isthen used to verify their ballot paper and so provide additional security against personation or other electoral fraud.
This new system was adopted across England for the 2007 local elections and in Wales for the 2007 National Assembly election. Reports on those two elections suggest that the more stringent security arrangements had little effect on the numbers of people applying for a postal vote. In England more than 3.75 million postal votes were issued -12.8% of all those with a contested election in their ward. Exact comparisons with previous years are impossible because of variations in the electoral cycle, but it can be noted that 13.6% of local electors in 2006 had a postal vote and that 12.8% of electors throughout England had one at the 2005 general election. In Wales, 12.5% of electors opted for a postal vote at the nationwide Assembly contests compared with 12.7% at the 2005 general election and 10.7% at the local elections in 2004. Given that electors were obliged to reapply for a postal vote under the new rules this stability is impressive.
Turnout among postal voters also remained much higher than among those obliged to vote in person. In England three-quarters of postal ballots (74.9%) were returned; in Wales the figure reached 78.2% -in both cases more than twice the level recorded at polling stations. At the 2005 general election turnout among postal voters in England was very similar (76.5%); in Wales it was exactly the same (78.2%).
However, a number of electors who try to vote by post are inevitably found to have completed the documentation incorrectly and their ballot papers never reach the count. It seemed that the proportion of such cases might increase this year following the requirement for electors to provide personal identifiers. In England, the overall proportion of postal votes rejected or otherwise not included in the count was, at 3.2%, little different to that in 2006 and only fractionally higher than the 2.4% recorded at the 2005 general election. In Wales, the proportion of postal votes rejected or otherwise not included in the count was 6.2%: rather higher than the 4.6% at the 2005 general election but little different to the 5.8% at the 2003 Assembly election.
It is the purpose of this paper to subject such rejected postal ballot papers to a statistical analysis to see how far the rate of rejection is related to socio-demographic and other characteristics of the population of the electoral unit. England and Wales are treated as discrete entities throughout the reminder of the report and the section on each can be read indepedently.
England
Data availability
Analysis of the data in England uses returns from each individual ward. However, there are a number of constraints concerning data availability.
i)Census data is only available for wards which either existed or whose new boundaries had been formally agreed by 2002. This particularly affects district councils. District councils also provide the largest number of missing cases in terms of uncontested seats. In all, we have useable electoral and census data for up to 807 of 815 metropolitan borough wards; 841 of 957 unitary authority wards; and 3919 of 4855 district wards.
ii)Several authorities used old forms without space for information on verification to report their postal ballot data.
iii)There is evidence that there was inconsistency in the interpretation of the questions asked of returning officers in Form K and/or a failure to complete it fully. In some cases there were no data at all in response to one or more of questions B15, B16 or B17. The situation was exacerbated by evidence that the data were sometimes compiled by different ward count supervisors. This meant there was inconsistency even within a particular authority’s returns.
iv)In innumerable wards all those ballot papers rejected following verification were recorded as being incomplete (B16) rather than having a mismatch of personal identifiers (B17); in innumerable others there was a similar imbalance in the other direction. Given that it is unlikely that this is a reflection of real differences in the ballot papers received, we make no attempt to analyse inter-ward patterns in reasons for rejection. Instead we concentrate on the total number of postal ballot papers rejected as a proportion of the number returned as our dependent variable.
Rejected postal votes
The proportion of postal votes rejected or otherwise not included in the count was, at 3.2%, a little higher than at previous contests, but the difference was not dramatic –see Table E1. Put another way, about one in 31 postal votes were disallowed compared with about one in 42 at the 2005 general election. Of perhaps more interest is the considerable variation in the proportion so rejected. In the metropolitan boroughs, 8% or more of postal votes were rejected in 41 wards, but 0.5% or less in 21. Among the unitary authorities, comparable figures are 43 and 33 respectively; and in the districts, 98 and 395. Testing possible explanations for such differences will form the key part of this analysis. The raw data can be found in the attached spreadsheet –englishwards.xls.
TableE1. % Ballot papers returned by post but rejected before count 2001-7
2007 Locals2006 Locals2005 Gen2001 Gen
3.23.02.42.0
Variations in postal vote rejection rates
Hypothesis 1. The greater the proportion of electors with a postal ballot, the larger the proportion of those returned that will be rejected. This assumes that the more widespread that postal voting is, the more it will reach into those parts of the electorate less engaged with politics and likely to take less care in the completion of their ballot papers.
There are weak and insignificant (at the 0.01 level) correlations between these two variables for every type of local authority. There is no evidence to support this hypothesis for England in 2007.
Hypothesis 2. The smaller the proportion of electors with a postal ballot who return it, the largerthe proportion of returned postal ballot papers that will be rejected. This assumes that a low turnout among postal voters may be indicative of a ward in which the electorate is less engaged with politics and likely to take less care in the completion of their ballot papers.
There is a significant correlation (at the 0.01 level or better) between these two variables in all types of council, suggesting some support for the hypothesis. However the relationship is very weak (0.1 or less) and unlikely to be of substantive import in explaining observed differences in the rate of rejection of postal ballots.
Hypothesis 3. The more a ward displays indicators of social deprivation, the larger the proportion of returned postal ballot papers that will be rejected. This assumes that those with lower levels of education, higher levels of ill health etc will find it more difficult correctly to comply with the instructions for completing their postal ballot paper.
A range of correlations support this relationship in simple terms for each type of local authority. Among metropolitan borough wards, thereare significant bi-variate correlations between the proportion of unemployed in a ward (0.38), the proportion that live in overcrowded accommodation (0.36) and the proportion who are ‘income deprived’ according to the Department of Communities and Local Government’s Index of Multiple Deprivation (0.33), and the proportion of rejected postal ballots.
In the unitary authorities the correlations tend to be less strong, albeit significant and in the same direction. For example, one of 0.22 for unemployment and one of 0.28 for income deprivation. Among the districts the correlations are weaker still, perhaps reflecting a greater social heterogeneity within wards. However, there is a significant 0.13 correlation between the proportion of those in a constituency of working age with a limiting long-term illness and the proportion of rejected postal ballots, and one of 0.15 between the proportion who are ‘income deprived’ and rejected ballots. All the other bi-variate correlations are either equal to or lower than those.
Hypothesis 4. The more a ward has a high minority ethnic population, the larger the proportion of returned postal ballot papers that will be rejected. This assumes that those from a minority ethnic background may find it more difficult to understand and therefore comply with the instructions for completing their postal ballot paper.
The simple correlations do provide some support for this hypothesis, especially in the metropolitan boroughs and unitary authorities. For example, there is a significant, negativecorrelationof -0.33 between the proportion of self-described ‘whites’ in a ward and the rate of postal ballot rejections; and a significant positive correlation of 0.32 between the latter variable and the proportion of Asians in a ward. In the unitaries, the correlations between the same variables are -0.38 and 0.39 respectively. In the districts the correlations, although significant and in the same direction, are much less strong.
A model of postal vote rejection?
Variables drawn from our four hypotheses were subsequently included in a linear stepwise regression analysis to determine the proportion of the variance in the rate of postal ballot rejection for each type of local authority that they could explain. We also included dummy variablesto control for the possibility that electors from regions which had more experience of postal voting (North East, North West, Yorkshire and the Humber, and East Midlands) might be more adept at completing postal ballot papers. It is also possible that the operation of electoral administrators’ networks in some regions may have lead to a common approach to rejected ballots which will be reflected in the regression equations.
The analysis for the metropolitan boroughs explained a total of 19% of the variance in the rejection of returned postal votes. The proportions unemployed and living in overcrowded conditions each made an independent contribution to the model, and wards in the North West also had a lower rate of rejection all other factors being held constant. The application of a similar set of variables explained 33% of the variance among wards in the unitary authorities. The proportion of Asian residents and the degree of overcrowding each had a significant, independent and positive influence on the level of rejections, with the proportion of residents in professional or managerial jobs loading in the opposite direction. In addition, wards in three regions –East of England, East Midlands, and West Midlands- were significantly associated with higher rates of rejection having controlled for the other variables in the regression equation. For district council wards the model was able to explain just 5% of the variance. The proportion of white residents had a significant negativeinfluence on the rate of rejections, with a ward’s score on the Index of Multiple Deprivation acting in the reverse manner. District wards in four regions –North East, South East, South West, and Yorkshire and The Humber- seemed less likely to be associated with the rejection of postal ballot papers –all other things being equal. .
It is possible that our models might have performed better if we had chosen different variables. However, the correlations across a wider range of socio-demographic indicators do provide confirmatory support for the contention that only a limited relationship seems toexist between a ward’s social character and the degree to which votes cast by its postal electors are rejected. To investigate further what might be going on, we turn to a more forensic analysis of the data from individual local authorities.
Patterns within local authorities
A worksheet within the attached file englishwards.xls sets out the pattern of rejection within English local authorities. For each ward we have calculated the expected rate of rejection in terms of the models outlined above and then subtracted that from the actual rate of rejection. A positive figure means that the actual rate was above the statistical prediction; a negative figure that it was below. We then show for the wards across each local authority the scores for these ‘residuals’ in terms of their mean, maximum, minimum and standard deviation from the mean. The symbol ‘+’ indicates that both the minimum and maximum rates of rejection among wards in an individual authority were above prediction; the symbol ‘-’ that they were both below; and the symbol ‘*’ draws attention to an apparent skew in the balance of the maximum and minimum figures.
Among the metropolitan boroughs there are four cases where the rejection rate in every ward is above that to be expected statistically –Calderdale, Newcastle, Rotherham, and Sefton; and five cases where it is always below –Knowsley, Salford, Stockport, Wigan and Wirral. In several other instances there is evidence that the balance of rejection is heavily skewed. For example, in Coventry the minimum residual score is just -0.08 whereas the maximum is 5.88. The data for unitary authorities and district councils show some similar apparent imbalances in the rejection of postal vote returns.
Table E2 looks at the information about residuals in a different way. It shows those local authorities where more than 20% of constituent wards had rates of rejection either three or more percentage points or two or more standard deviations beyond the predicted level. Several of the metropolitan boroughs already mentioned appear again, together with a number of unitaries and districts. In Congleton, for example, every ward had a rejection rate higher than expectation and in 14 out of 16 cases it was more than both 3% and two standard deviations above that predicted.
Table E2. Local authorities with 20% or more of wards in residual categories
AuthorityN. of wards3%+ above3%- below 2 st.dev.+ above2 st.dev.- below
Coventry1884
Knowsley166
Manchester327
Newcastle2510
Rochdale205
Salford204
Sefton211918
Bracknell Forest184
Derby176
Herefordshire3914712
Telford and Wrekin336
Wokingham1844
Barrow In Furness1384
Basingstoke & Deane1955
Blaby1443
Breckland276
Bridgnorth154
Bromsgrove217
Caradon226
Congleton161514
Derbyshire Dales175
Forest Heath1033
Hertsmere1343
Horsham2266
Kennet277
Mansfield1975
Oadby & Wigston1033
Restormel19119
Shrewsbury & Atcham1374
South Shropshire2155
Teesdale72
Uttlesford27117
Discussion
It is not possible to provide precise explanations for these patterns without the detailed inspection of individual postal vote returns, but they are suggestive. All other things being equal, it would appear that elections staff in Sefton were much more likely to reject postal ballots than were their equivalents in Salford. It is worth investigating the practices in those two authorities, and in the other places flagged in the tables, to see whether there are systematic differences in the instructions provided for postal electors, the role of political parties and leaders/community groups locally in guiding their supporters in how to complete the forms, and the interpretation of what is and what is not a correctly completed form within each council’s elections office. There may also be a case for seeing whether individual regional Association of Electoral Administrators’ (AEA) branches did issue separate guidance to their members.
These matters are important because they might imply that electors in some parts of England had their vote recorded whilst those in other local authorities or wards had theirs’ rejected, despite filling in the forms in exactly the same way. Such an outcome amounts to a failure of representative democracy.
To improve the situation it is likely that the forms which postal electors are obliged to complete and the instructions they are given need to be reviewed for clarity and accessibility. Such a conclusion is supported by even the limited evidence that those from poorer and/or minority ethnic backgrounds are more likely to have their ballot papers rejected. There is also a need for electoral administrators to be trained to a common standard in interpreting what is and is not an acceptable return. Our experience in recording Form Ks submitted by returning officers also suggests that consideration needs to be given to streamlining the information required and to spelling out more precisely what data are to be included in what box.
Wales
Data availability
Analysis of the data in Wales is constrained by a number of factors.
i)As is legal practice in the UK, no election returns are published at a level of aggregation below that of the base unit of election. For the Welsh Assembly elections, this base is the 40 NAW constituencies.
ii)Although new constituency boundaries were implemented for this year’s Welsh Assembly elections, ONS has yet to make census data available based on this geography. We have therefore used census data for the old constituencies in Wales in the 36 out of 40 cases where boundary changes affected less than 10% of the electorate. This inevitably limits the precision with which our findings may be interpreted.
iii)Constituency returning officers were asked to supply postal vote information for both the constituency and regional level votes. In practice the differences between the two cases are so small that we use only the constituency vote data in this analysis.
iv)As in England, there is evidence that there was inconsistency in the interpretation of the questions asked in Form K and/or a failure to complete it fully. In some cases there were no data at all in response to one or more of questions B15, B16 or B17.
v)In five constituencies more than three times as many electors had their papers rejected following verification for being incomplete (B16) rather than for a mismatch of personal identifiers (B17); in 13 constituencies there was a similar imbalance in the other direction. Given that it is unlikely that this is a reflection of real differences in the ballot papers received, we make no attempt to analyse inter-constituency patterns in reasons for rejection. Instead we concentrate on the total number of postal ballot papers rejected as a proportion of the number returned as our dependent variable.
Rejected postal votes
The proportion of postal votes rejected or otherwise not included in the count was, at 6.2%, a little higher than at previous contests, but the difference was not dramatic –see Table W1. Put another way, about one in 16 postal votes were disallowed compared with about one in 21 at the 2005 general election. Of perhaps more interest is the considerable variation between constituencies in the proportion so rejected. More than a fifth of postal vote returns were rejected in Gower (22.1%), but less than one per cent in Newport West (0.96%) and Monmouth (0.57%). Aside from Gower, more than 10% of postal votes were rejected in four other constituencies; and there were two seats where only between 1% and 2% of votes were rejected. Testing possible explanations for such differences will form the key part of this analysis. The raw data can be found in the attached spreadsheet –walesconstits.xls.