May’s local elections are the last electoral test before the next UK general election. As such, they are an opportunity for pollsters to refine ways of asking voters about their electoral behaviour, and ways of modelling that behaviour. In particular, they are the second set of local elections following the introduction of new voter ID requirements. In this blog post, we’ll describe some experimental statistics which estimate the proportion of registered voters who lack ID in over 600 London wards.

We describe these as experimental statistics for two reasons. First, they’re estimates of a relatively rare outcome (being registered and lacking qualifying ID) based on a sample size which is small relative to the number of London wards. Second, the estimates use some MRP techniques which we’ve not used before. Although they’re experimental, we think there’s value in them, and we think that experimentation is good. So let’s crack on describing what we did, and what the results look like.

What we did new in the survey

In the survey, we used questions modelled on polling conducted for the Electoral Commission. We asked respondents:

“As far as you know, is your name on the electoral register?”

We also asked all respondents:

“Do you possess any of the following forms of photo identification? Only select those forms of photo identification where you think that you are still recognisable from the photo (it does not matter whether it’s expired or not)”

We asked first whether they had a driving licence or passport; respondents who said that they had neither form of identification were then given a further (much more extensive) list of qualifying documents.

The reason we ask the ID question of all respondents is that we want as much information as possible regarding the correlates of having qualifying ID. If we only ask the question of those who are registered, we lose some information.

What we did in the MRP

We did two new things: we used a bivariate model, and we used a structured prior.

Let’s start with the “bivariate” character of the model. We’re trying to predict two outcomes: whether each respondent is registered, and whether each respondent has qualifying ID. These two models are linked: we include, in our model of whether the respondent has qualifying ID, whether or not the respondent is registered.

The alternative to this is simply modelling a compound outcome: whether or not the respondent was registered and lacked qualifying ID. But asking separate questions and modelling things separately generally improves accuracy.

We know this because we’ve been doing other experiments with MRP for voter intention. In past elections, modelling turnout and vote choice separately improved accuracy, compared to modelling a single categorical outcome which includes non-voters as a category on par with voting for a party.

The “structured priors” are useful because when dealing with small geographies like London wards, we generally have to rely only on demographic variables to use in prediction and post-stratification. We can’t use political variables that we use in modelling outcomes at Westminster constituencies.

Because we can only use demographic variables, we want to get the most out of them. For some variables, that means exploiting natural relationships between categories – or “structure”. In our application, we used a structured prior on age, so that the association between “being 18 to 19” and “being registered” is closer to the association between “being 20 to 24” and “being registered” than it is to the association between “being 60 to 64” and “being registered”. This might sound obvious, but it’s a handy trick which was introduced a couple of years ago and which we’re now incorporating into some work.

What the results look like

We’ve included a spreadsheet with our estimates of the proportion who are registered in each London ward at the foot of the post. We’ve included ward administrative codes so interested readers can combine this data with other sources of information about wards (provided, of course, they acknowledge the source of their data).

In terms of the headline figures, around five percent of registered electors lack qualifying ID. That figure is comparable to the national figure from the most recent Electoral Commission public opinion tracker (treating don’t knows as lacking ID), and one percentage point higher than the cross-break for London respondents. That figure, however, is expressed as a percentage of all UK respondents, rather than those on the electoral register. Our research therefore suggests that the voter ID requirement affects more people in London than previous polling has indicated.

In terms of the patterns across wards, the strongest area-level predictors of the proportion registered without qualifying ID are the proportion with no qualifications (correlation of r = 0.6), and the proportion with no passport (r = 0.49). There are also weak to moderate associations with the percentage of residents who are Black or Asian (r = 0.3 and r = 0.21 respectively). As might be expected, there’s a negative association with turnout in the previous Mayoral election (r = -0.38).

In practical terms, this means that polling station staff in Barking and Dagenham and Tower Hamlets are going to face more difficulties with the voter ID requirements than staff in Westminster or Richmond. Within these areas, however, there’s a big spread. Bethnal Green ward, for all that it’s located within Tower Hamlets, is in a very different situation to Bow West, site of the Roman Road market and just north of Mile End.

Implications for the future

We’ll continue asking questions about electoral registration and the possession of voting ID in our surveys, and may begin to incorporate this into our estimates of national voting intention. We’ve not at the moment assessed the partisan impact of these requirements, but this is something that our ward-level estimates will make possible following the publication of London results.

Get the data

Survation conducted MRP analysis of 2,230 adults in London on their voter registration status and their possession of different types of photo identification between 21st March and 20th April. Ward level estimates are available here.

________________________________________

If you are interested in commissioning MRP or to learn more about Survation’s research capabilities, please contact John Gibb on 020 3818 9661, email researchteam@survation.com or visit our services page.

For press enquiries, please call 0203 818 9661 or email media@survation.com

Survation. is an MRS company partner, a member of the British Polling Council and abides by their rules. To find out more about Survation’s services, and how you can conduct a telephone or online poll for your research needs, please visit our services page.