Despite not having a stake in the US elections - aside from the indirect ways everyone on earth does - I find them fascinating. Sometimes terrifying - I may not have stakes but I have opinions - but always fascinating. For one thing, they’re the time of year that market research gets the most attention, thanks to the crescendo of polls leading up to election day.
Not all the attention is welcome. In a close election, some firms and pundits are going to be wrong. It’s fascinating when a pollster’s numbers break out of the data peloton and make a bid for glory - as Gallup’s seem to have done this year, moving to a decisive Romney lead while other polls hover a nervous point or two around parity. If Romney wins, Gallup win bigger. If not - well, mistakes are generally forgotten.
But of course the big polling story this year is Nate Silver of Five Thirty Eight, former baseball statistician turned political modeller, New York Times columnist, and the most famous data scientist in the world. Silver - whose mainly accurate predictions in 2008 won him his fame - seems to be loathed, cordially or otherwise, by a bunch of people, including some pollsters and a lot of political pundits. The feeling is perhaps mutual: in the last week or so there have been a series of spats (columns on one side, terse tweets on the other) between Silver and the political commentariat, over issues such as whether the race is a “toss-up” or whether Obama is far more likely to win.
I have very little ability to comment on the rightness of Silver as a modeller, and we’ll all know how good his model is in a few days anyway. What’s interesting from a less technical researcher’s perspective is the tense dance between data and storytelling the Silver wars spotlight. What do researchers learn?
Experts don’t necessarily like data: Two groups of people have non-partisan grounds for beef with Nate Silver: pollsters and pundits. What they have in common is expertise. Pollsters and modellers have been in the game for years and never received a hundredth of the attention Silver gets: even so their concerns tend to be methodological - the details of the model, the assumptions it makes, are wrong. Not so pundits, who rely as much on judgement as on data - their experience of election cycles, the feel of campaigns, and so on. Mark Coddington goes into this distinction here - in a piece RTed by Silver. These are familiar concerns for researchers who often find themselves in a position where they’re fighting - and generally losing against - instinct, the weight of experience, etc. An expert view isn’t often an adaptive view, whereas the nature of data demands some level of that.
People DO like data (when they agree with it): The fact that a data nerd whose posts are naked without a table or two has become the New York Times’ top writer speaks to the appeal of data in election years, but also in general. Experts may be suspicious of it, but people like to know how things stand, and numbers get there faster. Of course, data isn’t often being used to change minds here either - its emotive power lies in confirmation, or sometimes reassurance. People who want Obama to win like Nate Silver, because he says Obama probably will. The acceptance of his model is mostly post-rationalisation. And exactly the same goes for those who read Silver and want Romney to win.
(There’s a side point about emotion and probability here, too. 80% sounds like a reassuringly - or absurdly - high number, because it’s being anchored by an awful lot of poll %ages in the 40s and 50s, which obviously are describing totally different things. On the other hand, if your kid was in hospital waiting for a vital operation with a 20% failure rate, 80% would suddenly seem horribly dangerous odds.)
Narrative and description are frenemies: The 538 model has been roughly constant throughout the cycle: Obama has between a 3 in 5 or 4 in 5 chance of being re-elected. What Silver is offering with the model isn’t a story, it’s a description of a situation. Events impinge on the details of the situation, but the wider situation remains intact - an incumbent president, a bad but gradually improving economy, and all the other stuff Silver talks about. If Romney does pull off a big win this description and its role in the model will seem irrelevant or wrong. (If Romney pulls off a narrow one it will still LOOK wrong but won’t actually be wrong.)
But what the election cycle demands is a narrative, not a description - last-minute escapes, reversals of fortune, re-reversals, scandals and so on. The job of the press is to weave events into a story. This ought to be something we think about more in research, where “storytelling” is a very big deal. When we ask ourselves to be ‘storytellers’, are we actually asking ourselves to paint vivid, compelling pictures - to describe, rather than to narrate. Or do we actually want a narrative, and the drama that goes - or ought to go - with it?
Storytelling is inescapable: We may not have a choice. Where Nate Silver is slightly disingenuous, for me, isn’t in his grasp of the fundamentals etc but in the way he does inevitably tell a story. Five Thirty Eight isn’t just a model, it’s a constantly updated model and he blogs every day about how and why the model has changed. These blogs are enormously interesting but they can’t help but touch on narrative or incline to it, and the up-and-down graphs on the site also speak to that storyline. Silver is a model for any researcher stuck on a tracker - his great gift is finding something to say every single day (most of us wrestle with months!). But his blog also tells us something about storytelling.
Story is the inescapable child of chronology - we simply can’t help but impose narrative on things. In twenty years time Obama’s re-election or defeat will be framed invariably as a thing of stories and a part of wider ones. Voices asking researchers to do more “storytelling” might as well be asking them to do more breathing: we should really be asking ourselves to refine and think about the stories we are already leaking.