Blinded by Science: Why Advertising Risks Learning the Wrong Lessons from the Politics

Originally written in March 2018

“I understood early that Facebook was how Donald Trump was going to win.”[1]

“(We) built models to exploit what we knew about (voters) and target their inner demons.”[2]

“(We) put almost all our money into digital… when things are digital you can be more empirical.”[3]

In the wake of the Cambridge Analytica debacle, the power – and perceived danger – of digital micro-targeting to influence politics has never been as widely discussed. The problem is, of course, that all of this ignores one salient fact about precise digital advertising in politics: it doesn’t work. This should not be a surprise to advertising practitioners; the main source of surprise should be that politicians and their parties continue to attribute magical capabilities to digital advertising that exist literally nowhere in the world. But in refusing to denounce this myth ahead of the real lessons of great political campaigning, we risk undermining the effectiveness of our work.

We can be confident that micro-targeting online doesn’t change electoral outcomes because last year a meta-analysis of campaigns demonstrated the average effect of advertising on the outcome was literally zero[4]; their “best guess” for the effect of online was also zero, whilst other[5] studies[6] have also shown online advertising in politics to have zero effect on outcomes. This is consistent with what we know about digital advertising elsewhere: a recent Ebiquity report in the UK shows online advertising to be, on average, the least effective media channel in terms of short- or long-term outcomes[7].

Cambridge Analytica have suggested that new technology can fundamentally change the effectiveness of these tactics, but this ignores the inherent issues with advertising, and digital advertising in particular. Advertising works slowly: it generally accounts for less than 10% of an advertiser’s short-term sales; in many mature categories, existing preferences and behaviour creates “long-term equilibrium”[8] that counter-acts advertising; meta-analysis shows that advertising works more effectively at reinforcing memories and defending positions than it does creating them[9]. Imagining that personalisation can mitigate these flaws is optimistic: it ignores that individual advertising effect is unpredictable and messages take years to establish[10]; that people don’t pay much attention to advertising[11], especially those which are smaller in size[12]; that motivation is fluid and diverse. In reality, targeted advertising actually often reduces total effectiveness[13].

Furthermore, the championing of Facebook’s influence ignores its limitations. It is used by lots of people, but those people only use it for around 35 minutes a day[14], browsing content quickly. This speed of consumption means that Facebook themselves recommend creating video between 2-8 seconds long[15]; we know from TV advertising that longer lengths of advertising duration have a higher impact.[16] It is also consumed heavily alongside other media channels, with most consumption passive rather than active[17] and less likely to lead to sales. And for all its scale, in America it has 100 million fewer users than TV[18]; it’s comparatively poor at reaching the older audiences who support Trump. It is a highly valuable platform that allows brands to identify and reach relevant customers precisely, but because of those limitations it is mainly useful as a secondary channel.  

Consultants might argue that the uniqueness of politics demands personalised digital advertising, but this isn’t true either. Political polarisation[19] creates the same “long-term equilibrium” as major product categories; choice is heavily influenced by incumbent popularity, with – unsurprisingly – unpopular incumbents proving vulnerable[20]; personality, whilst connected to politics, is not deterministic or singular[21]. Sophisticated personalisation works less well than simple rules of thumb: the political scientist Eitan Hersh has noted that because consumer data is often noisy and inaccurate, “almost all of the predictive power” in voter models comes from a combination of previous voting behaviour and basic demographics[22].

This observation chimes with basic media planning logic. Perhaps these tactics help find the “persuadable minority”, those who decide during the campaign, yet targeting these groups only makes sense if either if broadcast media can’t reach them, or if the campaign has so little cash on hand it must choose between continuous mass reach or over-emphasising this strategically vital minority. Simply put, if the campaign is spending lots of money on broadcast TV, the persuadable minority will see the message frequently. What value a tenth, fifteenth or twentieth ad exposure on top of a heavy TV campaign?

And, of course, this analysis doesn’t even acknowledge that personalised advertising is used by both winners and losers. Hillary Clinton almost certainly spent millions on Facebook advertising[23] deploying, most likely, just as sophisticated techniques; per Facebook she achieved largely achieved lower “cost per thousands”[24] (the net cost of reaching a thousand people in your chosen audience), meaning her budget bought her more exposure. Equally, pick the other side of the Atlantic: in 2015, targeted advertising was the secret of the Conservative Party’s success[25]. In 2017, both parties used those tactics[26] but only one, Jeremy Corbyn’s Labour, gained a huge amount of share. Why would it work for one party and not another, or vary in its influence so profoundly within a two-year period?

In the case of Donald Trump the combination of political polarisation and eight-years of Democratic rule, plus Hillary Clinton’s personal unpopularity[27], meant that literally any candidate standing for the Republicans in 2016 would have been favoured to beat the Democratic candidate in 2016[28]. Trump underperformed his expected vote share[29] and scraped through by improbably tiny margins. He should have won by more; vote shares were reduced by incompetence not boosted by prowess.

Yet there was one piece of evidence from the US last year that gave a clue as to how campaigns really win – by influencing the cultural agenda. Last year American economists demonstrated the influence of Fox News on the 2008 election[30]: they showed its ideological position maximised persuasive power (rather than viewership) and, over time, caused 28 percent of the Democrats who watched to vote Republican. Without that, the vote for John McCain would have been 6 points lower, below 40% (the effect was actually higher in 2000, when 58% of Democratic viewers switched). By influencing the news agenda, Fox relentlessly shaped the cultural conversation in their preferred candidates’ favour in a way campaign advertising simply cannot.

Using communications to shape cultural conversation helps, over time, introduce topics that favour you and frame debate in your favour. Vote Leave’s core insight wasn’t data-driven advertising but notoriety, using their unaccountability to promote misleading numbers and invented problems, forcing Remain to fight on their turf, boosting salience of key topics[31]. These messages were universally known, with 48% of the population believing the £350m claim[32] and 45% believing Turkey was to be fast-tracked into EU membership[33]. Trump’s messaging was similarly lacking in nuance; how many iterations of “build the wall!” or “lock her up!” were really required?

And the messages were spread by mass channels, not targeted ones. The news media – a form of advertising that politicians don’t have to pay for – was dominated by debates over those issues; they featured on a leaflet that went, identically, to 40m households and on posters on busy streets. That this could be attributed to personalised digital advertising is literally unbelievable: the Leave campaign was reliant on the older and less affluent[34], who consume less digital media than the average person, and turning out non-voters, which advertising doesn’t effect[35].

Advertising theory supports this cultural root of success. It shows “fame” – a shared sense of cultural presence, an understanding that you know your friends know that this campaign is a topic of conversation – is the highest state a campaign can achieve[36]. Famous campaigns are more effective, and shift every metric an advertiser could want, yet, as the single message or idea that is on the tip of everyone’s tongue, they are the antithesis of personalised advertising; they exist as culture, a state almost impossible to achieve if advertising happens in private. And famous campaigns are, by definition, long-term commitments - building the familiarity a campaign requires takes time[37], often multiple years, using associations and references that already exist[38] – and demand broad reach, communicating with all potential customers rather than a subset[39], to shape how people decide and what they expect. They help build a base, and short-term media mops up demand already created.

At its very best, fuelled by the right data, tailored advertising can complement this, reinforcing the appeal and associations of a brand and a story that is already understood or effective, topping up recognition of the features or benefits most salient to you personally, but it cannot do it alone. Both Trump and Brexit were primarily products of longer-term cultural trends cultivated by their proponents, which allowed them – like great brands – to build a base over time and rely on short-term activation to only contribute a tiny amount. Even inside the campaigns themselves, core themes – the corruption of Hillary Clinton and her misuse of emails in the US, rampant EU funding on the other – were consistently amplified in a way that made simple paid advertising redundant. And as David A Graham, writing in The Atlantic, has pointed out[40], you sense that, in a moment of self-doubt, Cambridge Analytica themselves might admit as much – or why else would you combine your all-powerful, Trump-electing, state of the art psychographic ad targeting with a pitch to send Ukrainian prostitutes to engulf a political rival in scandal?

Ultimately, success in political campaigning is not too dissimilar to success in advertising. Brands and politicians need to think first about how they influence culture at its most universal, using their platforms and resources to craft campaigns that create fame, dictate debate or shift how people make decisions. They need to aim five years ahead, not five weeks. Advertising can help achieve this, but only if it’s done at an appropriate scale, reflecting the assets, themes and messages that are most motivating, directing potential customers to the right products when they enter the market. And, above all, precision targeting should be a tool and nothing more.


[1] Brad Parscale, Donald Trump’s Digital Campaign Director, quoted on CBS https://www.cbsnews.com/news/facebook-embeds-russia-and-the-trump-campaigns-secret-weapon/

[2] Christopher Wylie, Cambridge Analytica data scientist, quoted in The Guardian https://www.theguardian.com/news/2018/mar/17/cambridge-analytica-facebook-influence-us-election

[3] Dominic Cummings, Vote Leave, in The Spectator https://blogs.spectator.co.uk/2017/01/dominic-cummings-brexit-referendum-won/

[4] Kalla and Brockman, The Minimal Persuasive Effects of Campaign Contact in General Elections: Evidence from 49 Field Experiments, American Political Science Review, September 2017

[5] Coppock and Broockman, The Effectiveness of Online Ads: A Field Experiment, September 2015

[6] Broockman and Green, Do Online Advertisements Increase Political Candidates’ Name Recognition or Favorability? Evidence from Randomized Field Experiments, Political Behavior, June 2014

[7] Profit Ability: The Business Case for Advertising, Ebiquity and Gain Theory, February 2018

[8] Dekimpe and Hanssens, Empirical Generalisations about Market Evolution and Stationarity, Marketing Science 14 (1995)

[9] Graham, Scriven and Bennett, Brand Loyalty - Plus ça change...? Using the NBD-Dirichlet parameters to interpret long-term purchase incidence and brand choice, The Ehrenberg Centre for Research in Marketing

[10] Binet and Field, Media in Focus: Marketing Effectiveness in the Digital Era, IPA 2017

[11] Byron Sharp notes only 16% of TV advertisements are attributed to the correct brand post-exposure in How Brands Grow, OUP 2010, page 7

[12] Nelson-Field, Visibility: The Attribute That Really Matters, Think TV Australia

[13] For more on the importance of broad reach, see Binet & Field, The Long and Short of It, IPA 2012

[14] Mediakix calculation http://www.adweek.com/digital/mediakix-time-spent-social-media-infographic/

[15] Kat Hahn, Head of Creative Shop, Facebook Northern Europe, speaking at Social Media Week London 2017, https://socialmediaweek.org/blog/2017/09/not-short-video-not-option/

[16] Data2Decisions benchmarking data shows impact of TV advertising grows with duration; 20-30 second advertisements are usually the best combination of effect and cost

[17] Nelson-Field, Visibility: The Attribute That Really Matters, Think TV Australia

[18] Pew suggests Facebook has around 68% of the US adult population (https://www.recode.net/2018/3/1/17063208/facebook-us-growth-pew-research-users), which if c.80% of the US population is eligible to sign-up for Facebook (i.e. over 13), that would make Facebook’s user base around 200m. Nielsen has the US TV universe at c.300m http://www.nielsen.com/us/en/insights/news/2017/screen-time-dollar-signs.html  

[19] Polarization has steadily increased in the US during recent history http://www.people-press.org/2014/06/12/political-polarization-in-the-american-public/

[20] Kalla and Brockman, The Minimal Persuasive Effects of Campaign Contact in General Elections: Evidence from 49 Field Experiments, American Political Science Review, September 2017

[21] Forthcoming research in Personality and Individual Differences suggests that personality traits explain no more than 9% of the difference between political tribes https://www.sciencedirect.com/science/article/pii/S0191886918301466 

[22] Hersh, Hacking the Electorate, Cambridge University Press 2015, p.152

[23] Campaign records suggest Clinton spent $16m on digital ads in the last month of the race https://www.theguardian.com/us-news/2016/dec/09/trump-and-clintons-final-campaign-spending-revealed

[24] Campaign cost per thousands released by Facebook https://twitter.com/boztank/status/968577962223136768

[25] https://www.buzzfeed.com/jimwaterson/how-the-tories-spent-ps12-million-on-facebook-adverts-in-run?utm_term=.eloLBmXXO#.jtw5xJpp2

[26] https://www.theguardian.com/media/2017/jun/07/tories-and-labour-use-social-media-to-target-ads-at-specific-seats

[27] https://fivethirtyeight.com/features/americans-distaste-for-both-trump-and-clinton-is-record-breaking/

[28] https://www.vox.com/a/trump-tax

[29] According to the Vox calculation above, a generic Republican should have received close to 51% of the popular vote, whilst Trump finished with 46%

[30] Martin and Yurukoglu, Bias in Cable News: Persuasion and Polarization, American Economic Review, September 2017

[31] As the campaign ended, more voters mentioned key voting issues being immigration and pressure on housing https://www.ipsos.com/ipsos-mori/en-uk/immigration-now-top-issue-voters-eu-referendum

[32] Ipsos MORI, https://www.independent.co.uk/news/uk/politics/nearly-half-of-britons-believe-vote-leaves-false-350-million-a-week-to-the-eu-claim-a7085016.html

[33] Ipsos MORI, https://www.ipsos.com/ipsos-mori/en-uk/immigration-now-top-issue-voters-eu-referendum

[34] Swales, Understanding the Leave Vote, NatCen, 2016

[35] Krasno and Green, Do Televised Presidential Ads Increase Voter Turnout? Evidence from a Natural Experiment, The Journal of Politics, January 2008

[36] Binet and Field, Media in Focus: Marketing Effectiveness in the Digital Era, IPA 2017

[37] Binet and Field, The Long and Short of It, IPA 2012

[38] Romaniuk and Sharp, How Brands Grow Part 2, OUP 2016, p.90

[39] Ibid, p.109

[40] https://www.theatlantic.com/politics/archive/2018/03/cambridge-analyticas-self-own/556016/

Charlie Ebdy