This article is co-authored by Stephan Shakespeare and Tiffany Washburn. Tiffany was one of the six-person data modelling group in Obama’s analytics team – part of what has become known as ‘The Cave’. She now works for YouGov.
Both Labour and the Conservatives are claiming to embrace the modern data-driven campaigning techniques of Barak Obama, and have spent large sums making symbolic (but not necessarily practical) hires of senior representatives from his 2012 American election-winning team.
But even in the spin itself, we can see that the most important lessons of the Obama campaign are not being applied at all. Rachel Sylvester told the story in The Times (£) two weeks ago: “The 2015 election will be fought simultaneously on a macro-level and a micro-scale as never before,” and quoted Jim Messina, Obama’s campaign manager, who is now helping the Tories: “Data is absolutely everything.” Using “intensive polling”, Sylvester continues: “The Conservatives have developed a sophisticated ‘segmentation’ strategy… dividing the electorate into eight distinct groups that will then be wooed with targeted campaigns.”
Um, eight? Micro? “Data is absolutely everything”?
While ‘micro-targeting’ and ‘big data analytics’ have become very fashionable terms in politics lately, the segmentation that Sylvester described is something quite different, closer to the gut-based labels from previous decades than Obama-style modeling. Sylvester reported on the “Anxious Aspirationals”, the “In Play Centre”, the “Steady Conservatives”, the “Disaffected Tories”, the “Young Urbanites”, and the ‘Urban Strugglers”. This may be really insightful, really useful stuff, and no doubt based on excellent polling – but you should file it next to ‘Worcester Woman’, ‘White Van Man’, and ‘Soccer Moms’ and not with micro-campaigning, because it’s the opposite.
To understand the most useful lessons of the Obama campaigns, we must grasp the rather basic difference between macro and micro. Obama’s conventional ‘air war’ (the macro stuff, messaging dropped through media) was nothing special – good enough, but no new lessons there for the UK. It was the micro part that we need to emulate.
Using statistical models, Obama’s analytics team scored each and every US voter on multiple dimensions (likelihood of voting on election day, likelihood of supporting Obama, likelihood of donating to a campaign if contacted by various methods, likelihood of “persuadability”- even the likelihood that they lacked a driver’s license). These scores were combined to optimise resources: how do you best use your human effort and media spend to get the votes where you most need them? Relying on traditional broad labels concentrates resources on writing scripts to talk with over 40 million voters, while doing little to determine which of those voters are the highest-value targets for contact. No matter how big the coffers, campaigns are largely about prioritisation, and that’s where the dirty work of data modeling comes in.
Replicating an Obama-style analytics approach in the UK by 2015 is a daunting, maybe impossible, task. The Obama campaign took advantage of a massive store of data that existed before the campaign even started, with the compliments of the Democratic National Committee (DNC). For years, the DNC had been collecting US voter files – which contain name, address, birthdate, and accurate turnout histories for every registered voter in the country. The DNC was also the storehouse for data collected by national, state, and local Democratic campaigns over many election cycles, including records on who had been contacted by each Democratic candidate, how that person had reacted, whether they had agreed to volunteer, and whether they’d ever donated money to a Democratic or Republican campaign. Combined with carefully curated census, consumer, geographic, and aggregate election data (and also, we might add with some pride, YouGov panel data modeling) the Democratic party was able to create a massive storehouse of individual-level information that proved an enormous advantage.
This huge initial investment – plus an intense internal polling operation – was what made all the statistics and analytics possible. Without such a detailed database in place, campaigns don’t have the flexibility to create micro-models of the kind that drove resource allocation in the Obama campaign. Mosaic, for all its benefits, pales in comparison to the micro-data a campaign would need to make probability models for each voter on all the dimensions of interest.
In a political campaign, it’s hardly an exaggeration to say that prioritisation is everything. This is what matters about the Obama campaign, this is the key, new lesson: exactly how you make your money and your energy do the most work. Even with massive cash – the Obama campaign proper spent about $10.37 per vote, to Romney’s $7.11 – the outcome was never a foregone conclusion, and they needed to squeeze out the support where it most mattered. If they’d relied on conventional techniques, Oboma might not have won.
Only vast amounts of micro-data can help you optimise your resources with this degree of nuance. In the past, for example, fundraising calls and letters were sent to voters using a segmentation strategy that involved cross-tabbing by previous contribution, last contribution date, and a handful of other specific groups. But it didn’t actually predict how much money you’d expect to get, on average, from contacting that person. Since calls by paid fundraising firms and mailers cost money, knowing which people’s ‘Expected Value’ is less than the cost of the contact is extremely important. Moving away from segmentation when determining fundraising contacts allowed the campaign to spend more wisely, taking into account dozens of variables with a model that was updated regularly based on the actual results of fundraising campaigns. This was a continuous live-experiment process.
In the 2012 election, 23 per cent of voters were classified by Pew as “swing voters”- undecided, leaning toward a candidate, or open to changing their minds. Of these, a smaller percentage lived in “swing states”- those geographic regions that were really up for grabs. And of those undecided voters in swing states, an even smaller number were actually likely to vote in the election (as the voters most likely to be undecided are also the least likely to vote, on average). We can even go one step further – of this small target group, only a few will actually respond to contact from the campaign, whether in the form of TV commercials, mailers, or phone calls. Undecided voters are often undecided specifically because they don’t pay much attention to what’s happening in the campaign. The moral of the maths is this: if you just go around spending $10 per vote evenly on every voter, you won’t get very far. What you want to do is spend $0 on people who are definitely voting for you, $0 on people who never vote for anyone, $0 on people who will definitely vote for the other guy, and $0 on people who wouldn’t pay attention to a campaign ad if it slapped them in the face. And for the tiny percent of people who live in geographic areas that matter, are likely to vote, unsure whether they’d vote for you, and persuadable if contacted in the correct way – for these, you want to spend $100. And then you want to go out of your way to make sure what you’re saying is your best message for those people.
The upshot is, to confuse micro with macro campaigning is to miss the whole point of Obama’s campaign success. Messina is not at all about micro-analysis. He is a brilliant leader and strategist, and no one expects him to be writing code or implementing multilevel regression and post-stratification models on the side, or even to be overseeing those who do. Great macro campaign consultants like Messina and Lynton Crosby are worth their weight in gold, but don’t confuse them with the quite different power of data-driven marketing.
In any case, to have a truly modern campaign, you need a technical infrastructure (the data and the data-scientists) and a delivery infrastructure (local contact networks to deliver the messages personally), and neither of the main parties actually has that. So Labour are also playing at make-believe Obama-style, while sticking to the methods they know best. There is only one UK party that can do micro, and even they employ the traditional – one might even say ‘artisanal’ – model: the LibDems, who have spent years knocking on the same doors and know their targets inside out, and who have honed the art of getting the most out of sparse resources. It is they, and not the Tories or Labour, who might do better than expected.