New telephone poll shows Remain ahead on 46%, and Leave on 43% (DK/PNTS = 11%). Remain 52%, Leave 48% excluding DK/PNTS’s.
Final BMG/Herald Result – Remain 53.3%, Leave 46.7%. Remain lead by 7pts after imputed DK/PNTSs.
Remain lead dips to 5pts among those who say they will definitely vote or who have already voted (Definitely/Already have voted = Remain 52.7%, Leave 47.3%).
Online Poll shows large shift to Leave
BMG’s Online EU Referendum tracker shows big shift to Leave, now on 51%(+6), and Remain down 2pts on 41% (DK/PNTS = 9%).
Labour support critical for remain.
Labour supporters could easily be deciding factor, even staying at home may hand referendum to Leave. Labour are more than two-to-one in favour of remaining in the EU, but if a significant minority end up staying at home, it could be the tipping point that secures victory for the Leave camp.
A class divide.
There is a strong class divide, with those in higher occupational classes (AB or C1) far more likely to vote to remain (AB = 58% Remain, C1 57% Remain excluding DKs), whereas those from the lowest social grades are backing brexit with much greater conviction (DE = 57% Leave). C2 on a knife-edge.
Job security a factor?
Those aged 65+ and those retired favour brexit (58% and 56% for Leave respectively), whereas younger people and those both in work and looking for work strongly in favour of remaining (18-24 64% for Remain).
Messages on potential house price collapse and rising interest rates cutting through?
Home owners and private renters back staying in the EU, whereas council and housing association tenants in favour of Brexit.
BMG Research backs phone over online for EU Referendum polling
After examining both online and telephone results and paradata, BMG backs telephone approach as a more accurate reflection of current public opinion on the UK’s EU Referendum.
Why impute voting intention for Undecideds and Refusals?
It is our view that using predicted voting intentions to impute voting intentions for undecideds and refusals is preferable and more accurate than existing methods of excluding undecideds and refusals.
One of the key lessons taken from the polling industry’s failure to call the 2015 General Election was that key indicators (i.e. leadership and economic approval), were consistently pointing towards a different, and what turned out to be, a more accurate reflection of the state of the parties. In the case of the EU referendum it is our view that simply excluding the undecideds and/or refusers is in itself a judgement about how they will vote (i.e. that they will do so in the same proportion to those who have already responded). Consequently we feel that it is more objective to allocate missing information based on prior estimates derived from their views, in this case their sentiment towards the EU.
To this end, BMG has created an EU sentiment index based on respondents’ views to a series of statements about Britain’s relationship with the EU. The sentiment index is designed to build an understanding of the relationship between key statements and voting intention and thereafter infer the likely voting intentions of undecideds and refusals.
The statements are asked before all voting intention questions. This has the added benefit of asking respondents to think about the major arguments in the campaign prior to asking the voting intention question. We believe it also sparks respondents to consider the arguments as they would on polling day, with the intention of getting them closer to what could be termed ‘the ballot box mindset’.
From this series of questions each respondent is assigned a score based on their response. The higher the score, the more positive the sentiment towards the EU, and vice versa. In order to use the score to impute values for undecideds, the predictive power of the index was modelled in a multivariate regression along with other key variables such as age and social grade (both of which appear to be important determinants of voting intention). See model results below.
While the results show the overwhelming predictive power of the index, they also show that demographics provide a comparatively weak basis for support of either leave or remain by comparison. Though it may well be the case that those who are still undecided and those who refuse to respond could be demographically different from those who do, our analysis of the key demographic information doesn’t support this assertion. The results suggest that the undecideds and refusals break by two-to-one in favour of Remain, slightly strengthening Remain’s lead.
Calling representative samples…and keep calling.
One of the key findings from the polling inquiry after the General Election was that samples for both online and phone modes were unrepresentative. Indeed there has been an open debate about the representativeness of samples for polling in the run up to this referendum that we’re interested in putting to the test.
In order to address this failing, BMG has matched its telephone data, drawn randomly and including mobiles, to external data sources to check the representativeness of the samples in advance. If the sample is significantly skewed across one or two key variables, the sample is redrawn till it is representative.
We also like to spend a little longer in field in order to minimise non-response error. Our latest poll took 6 days to complete for just over 1,000 responses. This is, in part, due to our calling schedule. Some respondents are harder to reach than others, and some are more difficult to convince to take part in a survey. So, much like a face-to-face survey, if there’s no response our interviewers will call the sample of respondents up to eight times and never at the same time of the day. It is for this reason that the majority of our survey responses are collected on the second or later attempt.
Though we recognise the potential for interviewer-effects on surveys, it is our view that calling randomly selected, and therefore more ‘typical’ people, trumps self-selected panellists, especially where the key determinants of an outcome are unknown. Handling objections to participation is crucial, and we feel interviewers are better at responding to different people’s motivations, than an email. We also do not believe you can weight yourself out of this problem, leading us to conclude that random selection is better.
Finally, it is interesting to note that the call data (charted above) from our latest telephone survey implies that, to some extent, pursuing respondents, whether through re-calling or person-to-person interaction on the phone, may be crucial in encouraging ‘harder-to-reach’ respondents to participate. Our results suggest that after one dial, the raw data gives Remain around a 1 point lead, whereas after the second dial the Remain lead is more than five and a half points, before settling at around four points after three calls or more. This suggests that conducting surveys too quickly or skimming through telephone data may underestimate Remain voters.
More details and a full breakdown of the results can be found below:
Fieldwork dates and methodology can be found below:
For a more detailed breakdown of results from this poll, or any other results from our polling series, please visit our website or get in touch by email or phone.
0121 333 6006
Dr Michael Turner – Research Director – BMG Research
Dr Kevin Cunningham – Consultant