Economists in Glasgow believe they can predict future election results using Google search data after they accurately forecast the vote in last year’s Scottish independence referendum.
Analysing live data on Google searches for the name Alex Salmond combined with commercial polling figures allowed the researchers to predict the yes vote would hit 45% – five days before that actual result in the referendum on 18 September.
Prof Ronald MacDonald, an economist and currency expert at Glasgow University, said they were able to track the yes vote rise with several complex methods routinely used by financial markets to analyse Google Trends big data to predict stock market movements and commodity prices.
In a study likely to catch the eye of political strategists who already mine internet and social media data during campaigning, MacDonald said the same methodology could be used to forecast the result in elections.
They were now considering a further test for May’s general election, he added. “Potentially, it’s a very useful tool but its complementary to the main pollsters. It would be feasible to do it on the day [of an election],” he said.
MacDonald, who backed the pro-UK group Academics Together during the referendum, added that an election was a more complex political event than a referendum, so the tracking would need to be more complex too. A forecast also assumed there was no cataclysmic event in the last days of campaigning.
He added that a single question referendum dominated by two campaigns and a few key personalities was easier to measure than numerous parties and party leaders competing on different agendas, in different parts of the country.
MacDonald said he and researcher Xuxin Mao tracked Google Trends data on searches for Salmond’s name and “SNP”, and factored in opinion polling results every week until August 2014, and then switched to live, daily tracking of searches for nearly six weeks until 16 September.
By 13 September, the model predicted yes would hit 45% with a very high degree of confidence, he said, with a range between 44.8% and 45.2%. In the event, the yes vote hit 44.7%.
They mapped a spike in likely yes support, pushing the vote up by 4.3% to 36.2%, when George Osborne announced the UK government would veto a sterling currency union after independence. It also detected a rise in yes backing after a YouGov poll in the final stages of the campaign gave yes a two-point lead over no, which energised grassroots yes campaigners.
However, that tracking did not link an increase in pro-independence votes with the anti-cuts, pro-NHS campaign run by Yes Scotland in the final weeks, which no campaigners believe cost the pro-UK camp up to five points on polling day.
MacDonald said their modelling was not able to map incremental changes in voting patterns due to one particular topic: it was more sensitive to specific events.
But he added that the famous “vow” from all three UK party leaders promising more devolution in the Daily Record had no clear impact on the final vote – despite assertions by Salmond it had a direct effect on the result. Nor did Salmond’s victory over no campaign leader Alistair Darling in the second TV debate.
MacDonald said in fact, had the referendum campaign continued for several more weeks, the steady momentum by then building for yes could have seen the pro-independence campaign win.
It emerged soon after the vote that Salmond and Yes Scotland believed they would win on 18 September by 54% to 46% because they had used cruder social media data, including Twitter activity. In the event, that prediction from their Canadian pollsters First Contact was significantly wrong.
guardian.co.uk © Guardian News & Media Limited 2010
Published via the Guardian News Feed plugin for WordPress.
Google search data 'could be used to predict election results'
Posts
0 comments:
Post a Comment