Yet one new organization, Qriously, did apparently foresee what would happen.
Qriously used a different technique to the established market research organizations. It used mobile. Harnessing the new programmatic ad serving technologies of the digital advertising world, it redeployed that technology to serve survey questions instead of ads to smartphone owners.
Critically, the day before the Referendum, seven out of eight research agencies incorrectly predicted that “Remain” (in the EU) would account for 51%-55% of the total vote. Qriously forecast just 44%. See their analysis here.
Qriously recruited respondents via quick-paced questions placed in app on mobile devices. This can provide an innovative way of soliciting a spontaneous response with, arguably, a more representative sample.
In contrast, the main polling organizations used either conventional voice telephone interviews or standard online surveys. According to Tom Clark, a highly respected journalist of the UK’s The Guardian, response rates from conventional telephone polls have been falling, in part due to telephone marketers, while online panels tend to be self-selecting, particularly so for political polling.
In 2012, Nate Silver became the preeminent U.S. election forecaster after he correctly forecast that not only would Barak Obama win the presidential election, he also correctly predicted the individual state vote for each of the 50 states.
When Silver reviewed the mainstream British polls leading up to the UK’s referendum, he ran a podcast entitled “British Referendum Polls Are A Mess.”
In defense of the seven research agencies that were wide of the mark, forecasting a one-off Referendum has significant hurdles:
Innovation is the market force behind technology transfer. It implies competitors will quickly acquire any burgeoning technology unless patents are in place to stem the flow. The existing polling organizations are almost certainly reviewing the potential of programmatic as a way to create new panels or ‘rivers’ of potential respondents. As such, two issues arise: