Saturday, 24 March 2018

The best 21st century political campaigns will owe more to Saul Alinsky and Readers Digest than Cambridge Analytica

The Cambridge Analytica story has now transmogrified into a good old-fashioned story about how a political campaign some folk didn't like might have broken some rules (I suspect it probably hasn't but, as ever, I'm ready for surprise). The more interesting part of the story, the bit about the use of sophisticated database marketing in politics, has got itself sidelined while the screams of defeated and unreconciled Remain supporters - or in the USA disappointed Clinton enthusiasts and anti-Trump mavens - echo round the halls.

I'm not going to go down the rabbit hole of the latest sensationalised allegations about the referendum campaign with its deliberate conflation of different actions at different times, use of gossip and repetition of allegations already investigated. Mostly because it's not all that important - the UK probably needs to rework its rules on political marketing at elections to reflect modern practices but that's about it really.

Much more important is the question as to whether the ideas, methods and approaches being pushed by the likes of Cambridge Analytica are effective and, if they are, whether we should be concerned about how they will affect the conduct of our public political debate. For a starter, here's Jamie Bartlett from Demos writing in The Spectator:
The shift towards big data elections has profound consequences for the whole of modern politics. If every voter is reduced to a data point who receives not real messages from politicians, but machine–generated adverts finely tuned towards personality and mood, then elections become little more than a software war. And the more politics is a question of smart analysis and nudges rather than argument, the more power shifts away from those with good ideas and toward those with good money or good data skills.
Bartlett has a point but I suspect he is overegging it a bit. The Spectator article describes a pretty standard data-driven direct marketing campaign little different from those we were designing for clients 25 years ago:
Cambridge had a database of around 5,000 data points on 200 million Americans and combined it with the Republican Party’s own voter data to build dozens of these highly focused universes and model how ‘persuadable’ its members were. (For example, analysts discovered during the race that a preference for cars made in the US was a solid indication of a potential Trump voter). Creative types then designed specialised ads for these universes, based on the specific things they were thought to care about. Every-thing was tested, retested, redesigned. They sent out thousands of versions of fund-raising emails or Facebook ads, working out what performed best. They tried donate pages with red buttons, green buttons, yellow buttons. They even tested which unflattering picture of Hillary worked best.
It sounds scary when you set it out like this but the reality is that, although we should absolutely be bothered about data protection and security, these methods really don't make the scale of difference that companies like Cambridge Analytica are wont to claim:
Does it work? Yes – but it’s not a silver bullet. We’d reckon on uplift in response of around 2X or maybe 3X compared to a random selection. Great until you realise that the response to random was around 0.2% - all that clever technology means that, instead of getting ignored by 998 out of 1000 people, you only get ignored by 994.
And those numbers I cited are response figures - people actually responding, doing something we've asked them to do - the reality with message advertising such as that used in political campaigns is that the level of inattention is several orders of magnitude greater (one of the depressing things for advertisers is the realisation that nearly all their carefully constructued 'reach' results merely in people completely ignoring the advertising). This is essentially what political scientists Kalla & Broockman found:
Significant theories of democratic accountability hinge on how political campaigns affect Americans’ candidate choices. We argue that the best estimate of the effects of campaign contact and advertising on Americans’ candidates choices in general elections is zero.
All that noise. All those messages. Billions spent on advertising. And there's no evidence that it makes a blind bit of difference to the outcome (as an ad man I'm not entirely convinced but as a political campaigner it's very unusual for the actual campaign to change much - the 2017 general election is an outlier in this). I note that Bartlett comments that Trump's victory was based on very small margins in three or four key states - not a new phenomenon as fans of hanging chads will remember. But we can't isolate the impact of different campaign elements - TV coverage, advertising and the ground campaign including 'get out the vote' activity.

When we look at the UK, the debate is about the referendum (of course, Vote Leave is very clear that it didn't use Cambridge Analytica or their style of psychometrics-based targeting) where perhaps the communications rules are different. That being said, the outcome was determined largely by unprecedentedly high turnout among older, working class voters - to be blunt, these are those least likely to be influenced by social media activity because they're not using social media (about half of 50-60 year olds and a third of over 65s).

It seems to me that, with the end of the mass media age, political campaigning will change. It also seems to me that very few political marketers have begun to grasp the significance of this change - it's not merely Jamie Bartlett's worry about what might be called the 'agora' but more that social media democratises political debate in a way not liked or expected by the dominant media forms. But first we need to put to bed the idea that algorithmic targeting is the marketing equivalent of a ninth level magic user spell.

Here's Kris-Stella Trump (no relation to The Donald) Program Director of the Anxieties of Democracy program at the Social Science Research Council:
The “Big 5” personality traits (which Cambridge Analytica claimed to use in its work) only predict about 5 percent of the variation in individuals’ political orientations.’s possible to predict personality from online data. But a recent meta-analysis shows that even if you have access to someone’s digital footprint, you can only learn so much about their Big 5 traits. Even if your model does well at first, it will probably be out of date soon, as the things people “like” on Facebook change.

You can improve online advertisements by targeting them using personality data. But the effects tend to be small. In this successful study, researchers targeted ads, based on personality, to more than 1.5 million people; the result was about 100 additional purchases of beauty products than had they advertised without targeting.

Once you know that personality prediction probably didn’t add much value to Cambridge Analytica’s approach, then what it did starts to look a lot like the microtargeting also used by other campaigns, and which the Obama 2008 campaign in particular was famous for.
Dr Trump also makes the killing observation that Cambridge Analytica "does not seem capable of pulling off the large-scale and complex personality-based profiling operation that it claims to have mastered."

So we have a methodology that might add 5% to targeting systems that have been around for nigh on 40 years applied to an environment that works on social interaction, word-of-mouth and endorsement. The result isn't a magic wand but a system that probably won't make a jot of difference to outcomes from social media marketing. Indeed, the successful approaches to social media marketing owe more to Alinsky's theories of community organising than they do to database marketing. This focus on organisation is one reason why Momentum and Corbyn's team have been successful despite the limitations of his political offer.

The manner in which political debate is changing means that the old conduit (or rather the mass media age's conduit) for political messaging is less powerful. It's undoubtedly true that TV coverage probably had more to do with Trump's election than data marketing or social media but it no longer has the field to itself. What we get (and both Trump and Corbyn got this) is a return to the public meeting, the rally and the "impromptu" street event. It's not, as Jamie Bartlett suggests, that political debate and engagement is atomised via the Internet but rather that it will be more like the debate of times before mass media but overlain with a social media element. As a result the skill set for the campaigner will be about organisation and activism rather than slick communications and advertising.

So I guess the advice to would-be political marketers is to step away from the box of tricks that the likes of Cambridge Analytica are flogging and read Rules for Radicals, The Solid Gold Mailbox and David Ogilvy's advice on selling (it's very sexist as it was written in 1935). Then go and do a lot of listening to what people talk about (clue - it's not politics or, at least, obviously politics).


1 comment:

Anonymous said...

I suspect a great many educated, decent and provincial Tories, like yourself (and me), are about to get a nasty surprise at the extent to which the Conservative establishment has gone completely rogue. I'm quite a bit younger than yourself, and after a tech and finance career, have a strong grasp of what happened. The evidence as presented to Parliament by Wylie today looks damning to say the least. Like Wylie, and as a Brexiter, and a Tory, I believe far more in the sanctity of democratic process and fair play than I do in the outcome of the referendum. This cabal, with strong links to the Tories and establishment, are trashing Brand UK - and I'm pretty pissed off by that - because Brand UK has been good for my business to date.