Facebook is running out of growth options via advertising, primarily due to space constraints. It’s tough threading the needle of ensuring engaging content, and its corresponding eyeballs, with monetization of said engaging content. So it’s no surprise Facebook wants to start selling TV ads. Ones that actually appear on TV screens. And if you combine the same 1:1 audience-targeting ability Facebook owns via its audience network, that’s pretty powerful living room targeting.
So how did Nate Silver, of fivethirtyeight .com, the celebrity statistician who predicted the Cubs winning the world series and the 2008 and 2012 elections, get this one so wrong? There are many irregular, exogenous variables this year, such as the difficulty in accurate polling. But one huge, bigly difference may be Silver’s methodology. First thought: he relies heavily on historical voting and party-affiliation behavior. OK, ok, you absolutely need reliable, historical data to forecast outcomes. But, there’s something that sits in the middle of all that data – consumer perceptions, because perceptions change before behavior. Perhaps no one really accounted for the changing perceptions of whole swaths of consumers, like non-college educated white males, who are disenfranchised, angry and out of adaptations. In hindsight the most important variable was perceptions of the economy. That should have been a perception tracked, married to “intent to vote.” The second potential methodological difference is the impact of word of mouth, or how consumers interact with one another. Word-of-mouth exchanges can amplify perceptions in one way or another. This campaign was especially voiced via social media. Silver explains volatility of his modeling late in the campaign, with a sports analogy: An NFL team that kicks a field goal with two minutes to play in the first quarter becomes a 59% favorite to win; but one that does so with two minutes left in the fourth quarter, becomes an 83% favorite. When Silver talks about late-stage volatility, I wonder… Is the volatility in the model? Or does the model pick up the momentum of word of mouth? And then there’s his position that good models ought to avoid changing its rules midstream. With the relentless, ubiquitous news cycle, misinformation, lawsuits, accusations – how can rules not be changed midstream? In any case, a new model with algorithms around rapidly changing perceptions and how those are amplified via word of mouth, will be important in understanding our changing electorate landscape.