Voting sentiment not revealed in stats

Last week in the run up to the Scottish Independence referendum, I did some digging around to see if search trends could give us any indication of whether ‘big data’ could tell us how the vote would swing.
The first, Can ‘big data’ predict indyref result looked at the number of links coming into each campaigns official website. The research showed a massively higher number coming into the Yes website, yesscotland.net, compared to the No campaign site, bettertogether.net.
And the second, Google reveals dramatic shift in Yes/No trend, demonstrated that search trends had sung in favour of a Yes vote.
Mike McGrail, of Velocity Digital in Edinburgh, also did good research on the impact of social media in the final days of the referendum, both in his twitter stream, his blog and TV analysis before and after the event.
All this had come at a time when the opinion polls had shown that Yes was gaining ground, but the polls only showed a marginal lead. The search data showed a massive lead, and with the company behind it already having made successful predictions before, I thought it may show that there is data out there that is more reliable than the polls.
It wasn’t.
The polls were roughly correct and the No campaign took 55% of the vote.

So what did the exercise prove?

The analysis of big data in this scenario is hopelessly difficult and needs a lot more work. Measuring links and searches clearly doesn’t show the sentiment of the voters unless they changed their mind in the polling station. It’s not just a case of saying more searches equals more votes. It’s just not that simple.
This is not to castigate big data as s source, but it does need much more work to be able to correctly analyse what these statistics can tell us.
If this was a business exercise, it would have been successful research. If we could have shown a client that more people were coming to a website or searching for a term, it would be our job to ‘convert’ those customers when on the site. In a referendum or election, the marketing is deliberately disconnected from the actual voting process. You won’t see pop-ups on these websites pleading ‘Vote Yes now’, or indeed ‘Vote No now’!
If the Yes and No websites allowed voters to vote right from the website, it could well have been a different story. It’s the marketer’s job to lead buyers to a website and make a compelling case to buy products and services. One of the problems here is the disconnect between the marketing and the voting.
Another is that we cannot tell whether the links and searches are actually supporters. In both articles, I suggested that people could easily be visiting the Yes and No websites regardless of their voting intention.
The research, if anything, tells us we need to look deeper into the dynamics of links and searches and correlate them with other data to make it meaningful, but it will be many elections before we get a reliable forecast.