“I don’t believe the polls. They got them wrong at the last election. They got them wrong in the referendum. They got them wrong with Trump’s election in America – so I don’t believe the polls”. John McDonnell, Labour MP (and of course Labour being behind the conservatives by 21 points in YouGov poll at the time had no influence on this view).
John’s viewpoint reflects the general cynicism towards the polls at present, none of which predicted the considerable Conservative lead in the 2015 General Election, or that the UK would choose to “Brexit”, or the US election of the now President, Donald Trump.
When Theresa May announced the snap 2017 General Election, I’m sure there were some researchers working in the polls arena who had their own ‘Brenda’ moment (‘Brenda’ being a member of the public interviewed by the BBC in response to the general election announcement): “You’re joking. Not another one? Oh for God’s sake. I can’t stand this. There’s too much politics going on at the moment.” While others were jumping the walls with excitement.
Polling for the 2017 General Election will without doubt be a challenge for the polling companies, none of whom will want to see a repeat of the inaccuracies that have been recorded over recent years.
The apparent poor accuracy of the polls at the 2015 General Election prompted an investigation by the British Polling Council (BPC). The key conclusion in their investigative report was that the primary issue was unrepresentative samples. In the case of the 2015 general election, the BPC deemed it was the over-representation of Labour supporters and the under-representation of Conservative supporters, that led to the failure of the polling companies to effectively predict the Conservative landslide. Polling company YouGov also conducted its own investigation into what went wrong. It also stated sample representativeness as being one of the key issues (in particular the over-representation of politically engaged young people, who were more likely to vote Labour).
And then there was Brexit. The turnout for the Brexit referendum was 3 million people greater than the 2015 General Election. This turnout presented an issue for the pollsters; unlike the General Election they had less of a definitive view on who would be voting, which again raised the representative sample challenge.
We then had President Trump, referred to by some as the ‘King of Social Media’. The polls had Clinton winning by a significant majority, there was no way Trump could become president…but while the polls failed to predict this, analysts using social data could see this coming and they predicted Trump as the next American President. Social media has often been viewed by political pundits as nothing more than an awareness tool, but the outcome of the US election demonstrated its effectiveness as a voter engagement channel. Can effective social data analysis in fact provide the representative view missing from the polls? As the pollsters make their 2017 General Election predictions, closer attention is likely to be paid to social data this time round. What can social data analysis offer that traditional polling approaches cannot?
Poor response rates are a challenge for pollsters, especially with hard-to-reach groups; response rates have been in decline for some time and there is a general reluctance to take part. In light of cynicism about the results of the opinion polls, many do not appreciate the value of their participation. All of this will make participation and securing a representative sample challenging for the polling companies. Meanwhile, those reluctant to engage in a survey are freely sharing their views and engaging with political social media content, daily.
The first opinion poll since the announcement was released by YouGov on 19 April, it placed the Conservatives significantly ahead of Labour with a 24% lead. And shortly followed the sentiment amongst some: “well it’s a full blown conclusion”; “there’s no way Labour can overcome that”, but hang-on didn’t we see similar rhetoric when the polls placed Clinton ahead not so long ago, of course there was no chance of Trump winning then either…
Social data from Tuesday 18th April, the day the election was called indicates it might not be quite the landslide people are assuming. Social data recorded on the 18th April indicates Theresa May has a greater share of voice than Jeremy Corbyn, however the Labour party overall recorded more share than Conservatives (and one thing we can take from Trump’s campaign is that any social engagement can have a positive effect).
On the day of the general election announcement, one of the top 10 links being shared on social (in relation to the election) was join.labour.org.uk. No link to a Conservative website sat in that tier. And those hard-to-reach audiences were engaging too with the Gov.UK register-to-vote link being shared amongst those in the 18-24 year old bracket.
However, social data as with the opinion polls, needs to be representative. Without this there is a danger of naïve sampling and analysis and falling victim to the echo chamber. Social data needs to capture the big conversation, from multiple sources (Twitter, Facebook, forums, news reports etc.) and multiple demographics in order to gain a holistic view.
So in the era of declining survey response rates and the growth in social media communications, where the public are freely expressing their views every minute of everyday, social data should be observed more closely than it has been previously. Perhaps it can even help the pollsters to truly relay the word on the street.
MMRI, have developed an in-house social listening and analysis platform (Klood Radar) to serve the needs of market research professionals. The Klood Radar collects the millions of social conversations being had in real time and helps you to make sense of them, allowing you to identify and analyse social commentary, applying sentiment and topics.
To discuss the use of social data to understand public opinion in the build-up to the general election please contact us at www.mmri.klood.com.