Dr Christopher Prosser*, Professor Edward Fieldhouse*, Dr Jonathan Mellon*, and Dr Jessica Smith† - Written evidence (ERA0007)

*British Election Study, University of Manchester
†University of Southampton
 

Summary

1. This evidence relates to the experience of eligible voters of the change to Individual Electoral Registration (IER); how this has affected transitions on and off the electoral register; and the potential impact that IER has had on the quality of the electoral registers for the secondary purposes of drawing parliamentary constituency boundaries and measuring voter turnout. It draws on research undertaken between May 2014 and June 2017 as part of the British Election Study (BES) partnership with the Electoral Commission into the impact of IER; and on subsequent research, which uses register, census and Annual Population Survey (APS) data to impact of the changes on drawing constituency boundaries.
2. The BES research on the switch to IER shows that, in broad terms, the impact of IER on registration levels depend on when we are talking about. During the transition period (up to May 2016) we find clear evidence that overall levels of registration decreased and that people in demographic groups that are less likely to be registered in general (i.e. young people and privately renters) were more likely to drop off the register. Comparing the 2015 and 2017 elections, however, we find no evidence that IER affected the level of registration at the time of the elections, nor that is disproportionately affected groups that are less likely to be registered in general. How can we reconcile these seemingly contradictory findings? In short, IER seems to have increased the volatility of electoral registration – people were more likely to drop on and off the register when IER came into force.
3. The result of this volatility is that accuracy and completeness vary during the annual registration cycle and peak around electoral events. Accuracy and completeness also vary geographically. This has implications for future apportionment undertaken by the Boundary Commission, which requires reliable and accurate registration figures to ensure compliance with the 2011 Boundary Act (“Parliamentary Voting System and Constituencies Act 2011” 2011). Parliamentary Constituency boundaries are drawn with the aim of equalizing the number of eligible voters across constituencies. Because IER has not substantially changed the completeness of electoral registers, boundaries drawn using the electoral register still fail to achieve the normative goal of equalizing the number of eligible persons across constituencies. While IER has improved register accuracy, it is still sufficiently inaccurate (containing duplicate or incorrect entries) that it fails to meet the legal standard of equality of registered persons, because accuracy varies enough across regions to give areas with less accurate registers more representation than areas with accurate registers. We therefore conclude that – despite IER – the current register entries method for drawing parliamentary constituencies meets neither the normative or legal aims of the legislation.

Impact of IER on registration levels and movement on/off the register

4. To examine registration levels during the transition to IER, we make use of the BES Internet Panel survey (BESIP). The panel study is designed to follow the same survey respondents over time over a number of survey ‘waves’ and so is particularly useful to study within-person change. We draw on three waves, fielded in May 2014 (after the 2014 local election, when the household registration system was still in place), May 2015 (after the 2015 General Election, during the transition period), and April to May 20105 (before the 2016 local elections – six months after the finalisation of IER). To compare registration levels at the 2015 and 2017 elections, we use BES face-to-face post-election survey data. These data provide a representative ‘snap-shot’ of the population and each dataset includes a new set of survey respondents. For each of the three BESIP waves and the two face-to-face surveys, respondents who gave permission were matched to the electoral register (collected at the time of the survey by the Electoral Commission) using their name and address.
5. One of the most common concerns raised about IER was that it would lead to a reduction in overall levels of registration in the population. The BESIP data suggests that concern was at least partially warranted – between May 2015 and May 2016, we find a six point drop in overall registration levels following the December 2015 removal of names from the register at the end of the IER transition period. However, in the run-up to the 2016 EU referendum there was a flurry of late registration, and the Electoral Commission’s data suggests that overall registration levels at the referendum were similar to those at the 2015 General Election. Likewise, the BES face-to-face data suggests that compared to 2015, there was a small increase (2.5 points) in overall levels of registration at the 2017 election.
6. As well as the impact of IER on overall levels of registration, many people raised concerns about the potential for IER to disproportionately suppress the registration levels of certain groups who are already under-represented on the electoral register, such as young people and students; electors from some BME communities; people in private rented accommodation; and some disabled people, especially those in residential care. The BES data allows us to address the questions of whether IER disproportionately affected young people, students, and private renters. We cannot address the impact of IER on BME people and those with disabilities because we simply do not have enough of these types of people in our data.
7. Across all time points, older people are more likely to be registered than younger people. In May 2016, following the introduction of IER, there was a decrease in registration levels across all age groups. However, the decrease was largest amongst younger people, leading to an increased registration gap between the older and younger age groups. This change can be clearly seen by looking at the youngest and oldest age groups. Between May 2014 and May 2016 there was a 6.8 percentage point decrease in registration rates amongst 18-25 year olds but only a 2.3 point decrease for those aged 66 and over. By May 2016, when those registered under the old system had been removed from the register but before the late surge in registration for the EU Referendum, the registration gap between the youngest and oldest age group was 18.2 percentage points.
8. This is clear evidence that the introduction of IER increased age-related registration inequalities during the transition period. If we examine the relationship between age and registration at the 2015 and 2017 elections, however, we find a similar relationship between age and registration in both years. First, older people are more likely to be registered than younger people. Second, the level of registration within age groups is similar in both years, and if anything, slightly higher amongst those aged 26-55. This patterns suggests that the increase in age related registration inequality during the transition period was counteracted by late registration in the run-up to the 2016 EU referendum and the 2017 General Election.
9. Discussions about the potential impact of IER on electoral registrations amongst the student populations dominated much of the media and political discussion of the new system. For example, in January 2015 the Labour leader Ed Miliband said there were ‘one million’ missing from the electoral register and many of them were students. The government responded by sending a letter to University Vice Chancellors asking for support in ensuring more students were registered under the new system, suggesting targeted campaigns at key points such as during enrolment. The government also gave £530,000 to organisations to encourage student registration.
10.         At first glance, these fears appear to be justified. Compared to the rest of the population, students have lower levels of electoral registration before, during, and after the transition to IER. However, this is not because there is something about being a student that makes people less likely to register, it is because students tend to be young. Before and during the transition to IER, if we take other factors (such as age and housing tenure) into account using a regression model, whether someone is a student does not predict whether they are registered or not. In May 2016, after the transition to IER, being a student does predict whether someone is registered or not – however, contrary to the popular narrative, students were more likely to be registered than nonstudents (allowing for age). Similarly, if we examine the 2017 face-to-face data we find that amongst those aged between 18 and 25, only 67% of non-students were registered compared to 78% of students. The fact that students are more likely to be registered partly reflects that they tend to have higher levels of political engagement, although the regression analysis shows this difference persists even when this is controlled for. An additional factor could be efforts within universities and wider campaigns to drive up student registration, given the concern around the potential effect of IER on student registration rates there. Even though the net registration advantage among young students did not appear to grow after IER, these efforts may at least have had the desired effect of mitigating any drop in student registration, or the original concern may have been unwarranted, though there is no way of knowing which of these is the case for sure.
11.         In all our data, private renters have consistently lower levels of electoral registration. During the transition period, this gap appears to have been exacerbated by the introduction of IER. The registration gap between homeowners and private renters increased from 10.5 percentage points in May 2014 to 23.5 points in May 2016. However, comparing registration at the 2015 and 2017 elections shows there was an increase in registration rates for private renters by just under 10 percentage points compared to 2015. We should note though that despite this increase, home owners are still much more likely to be registered to vote.
12.         The BES data suggests a consistent pattern: during the transition to IER people – particularly young people and private renters – were likely to drop off the electoral register. However, comparing the 2015 and 2017 elections we find few differences in registration, and where there are differences, they suggest that registration levels were higher in 2017. This pattern suggests that people dropping off the register at one point were replaced by people joining the register later. We can examine this pattern more closely using the BESIP data, which gives us the ability to track the same people across multiple time points.
13.         Those who remain registered across the transition period tended to be older, retired and own their own home, suggesting the characteristics of those with stable levels of registration are similar to those who are more likely to be registered in general. Similarly, we find that those are registered at one time period, but drop off the register later, share demographic characteristics with people who are less likely to be registered in general – the young, private renters, and the unemployed. However, those who are younger and rent their home are also more likely to join the register if they had been previously unregistered. In other words, just as the same people tend to remain registered, the same people tend to drop on and off the register – that is, there is a ‘churn’ effect amongst certain demographics of voters, rather than constant levels of under-registration.
14.         We cannot say for sure whether the pattern we observe of people dropping off the register between elections but registering in-time for elections is a feature of the 2014-2017 period, or is something we are likely to continue to observe. If this pattern persists it is both good and bad news for British democracy. On the one hand, it is good news that IER has not negatively impacted registration levels at the time of elections – electoral participation does not seem to have been affected. On the other hand however, electoral registers serve important roles between elections as well – such as being used to redraw constituency boundaries. If these registers are inaccurate – as the December 2015 registers used in the current boundary review are – then decisions that arise from them such as constituency boundaries will be inaccurate as well.

Implications of changes for drawing Parliamentary boundaries

15.         The Boundary Commission final recommendations (2018 review) were presented to Government on 5 September 2018. The electorate numbers used in the review were taken from the electoral registers maintained by local electoral registration officers and the data they provide to the Office of National Statistics. These were published on 24 February 2016 based on the December 2015 canvass, which took place after the completion of the IER transition. At this point registration was 45.5 million, approximately one million lower than in June of the preceding year and 1.5 million lower than in June 2017. This has serious implications for the drawing of constituency boundaries.
16.         There are both normative and legal reasons why the registers used for drawing constituency boundaries should be accurate and complete. Normatively, the aim of creating constituencies is to equalize the number of eligible people across constituencies, ensuring all eligible persons counts equally, regardless of whether they have successfully registered on enumeration day. However, this is not the legal standard set out in the 2011 Boundary Act. The act states, ‘the electorate of any constituency shall be— (a) no less than 95% of the United Kingdom electoral quota, and (b) no more than 105% of that quota’ which is defined by the total number of registered electors. Thus, we have both a legal and a normative standard against which to evaluate the data used for apportionment
17.         There are two major problems that mean these data may not always be fit for purpose: completeness and accuracy. Completeness refers to the percentage of eligible persons correctly registered to vote. The UK Electoral Commission estimated completeness at 85% in 2011 and 82% in 2015. The second class of problems are register accuracy. Inaccurate entries are defined by the Electoral Commission as: ‘entries which have become redundant (for example, due to home movement), which are ineligible and have been included unintentionally, or which are fraudulent.’ (Electoral Commission, Office for National Statistics, and National Records of Scotland 2014). The electoral commission reported that 9% of register entries were inaccurate in 2015 compared to 14.5% in 2011. Additionally there are duplicate entries, which represent electors who are legitimately registered in more than one place.
18.         We use the Electoral Commission’s 2015 accuracy survey to estimate how much register accuracy varies across regions and compare the proportion of register entries in each region to the proportion of accurate register entries (i.e. unique registered persons) in each region. This allows us to assess how much each region is over or under represented on the electoral register due to register inaccuracy (when compared against the legal standard of registered persons). Since regions are much larger and diverse than individual constituencies, the regional variation in inaccuracy will be a strict lower bound on the level of constituency variation.
19.         The table shows that there is substantial variation in accuracy. If accuracy were taken into account and constituencies were apportioned on the basis of the number of registered persons (not simply register entries), London would receive 3.4 fewer seats. This gap is already nearly at the legal level even before we consider constituency variation within regions (of which there will be a significant amount) and other reasons for differences in constituency size (of which there are many). The North West, by contrast, would receive 2 additional seats. Since these differences are lower bounds on the level of constituency variation, it is guaranteed that many seats will deviate by more than the legally allowed 5% from the legal standard of variation in registered persons. This variation is only that induced by register inaccuracy. If there is variation in constituency size due to other factors such as compactness and representation of distinct geographic areas, this will push up the variation in constituency size even further beyond the 5% threshold.

20.         To compare constituency boundaries to the normative standard of eligible persons, we analyse 2011 census data on passport holding (to determine eligibility) at the constituency level. We supplement this analysis with local authority level analysis of 2016 eligibility based on the 2016 Annual Population Survey and registration figures for the 2016 EU referendum. We find that, on average, there are 3.6 percentage points fewer register entries than there are eligible persons. There is considerable variation (a standard deviation of 3.6) between constituencies. The proportion of eligible voters differs from the proportion of register entries by at least five percentage points in 162 constituencies. The number of register entries actually exceeds the number of eligible persons in 56 constituencies, indicating that inaccuracies in the register are sufficient to outweigh incompleteness in these cases.
21.         Whilst the constituency level analysis is based on data that precedes the move to IER, we make local authority district based estimates of variation in completeness using the Annual Population Survey (APS). Overall, the updated analysis suggests that little has changed in the magnitude of the differences in the distribution of eligible persons and register entries. The standard deviation of the percentage difference across constituencies was 3.6 in 2011 and is 5 across local authorities in 2016. The APS suggests that 115 local authorities across the UK have more than a 5 point deviation between their share of eligible persons and their share of register entries, showing that the size of these discrepancies is still well outside the level considered acceptable.
22.         We conclude that the use of register entries fails as a proxy for either the normative or the legal standards, with the deviations likely exceeding the 5% legal threshold when compared with either standard. Moreover, because of increased fluctuation in the register size and the movement of people on and off the register, the use of data from the annual canvass (especially that following the introduction of IER on 2015) exacerbates the problem, leading to under-representation of certain groups and geographic areas. The introduction of IER has improved the accuracy (but not completeness) of the data on which boundaries are determined, but does not solve the fundamental problem. The implication is that the 2018 UK boundary review is unlikely to meet its objective of constituency size parity if we define parity by either a legal or normative standard.

Implications of changes for turnout estimation

23.         The same issue of accuracy and completeness that affect the use of the electoral register for drawing constituency boundaries also affect the measurement of turnout in elections. While turnout in elections is often colloquially talked about in terms of the percentage of eligible voters who participated in an election, the official definition used is the percentage of registered voters who participated in an election. This means that the denominator for official turnout is affected by register accuracy.
24.         Because many entries on the electoral register are inaccurate or duplicated, the denominator of turnout is overestimated. That is, the true number of correctly registered persons in Britain is substantially smaller than the number of entries on the electoral register. Turnout is therefore substantially underestimated in Britain.
25.         We use the same data from the Electoral Commission’s accuracy survey to estimate how substantially turnout is underestimated. We find that turnout in the UK (as defined by the percentage of registered people who voted) is underestimated by up to 11.5 percentage points. The increased accuracy of the electoral registers under IER has modestly reduced this underestimate. However, the change in the quality of the electoral registers also means that 0.4 percentage points of the 2.4 percentage point increase in turnout between 2015 and 2017 can be attributed to the changing quality of the electoral registers.

9 September 2019

1