Mass political advertising campaigns on social media have little impact

Do mass political advertising campaigns via social media really influence voting? Though there have been plenty of claims that they do, experimental evidence has been limited. Now, though, a team led by Minali Aggarwal at Yale University reports their analysis of the impacts of a huge US$8.9 million pro-Biden, anti-Trump digital campaign conducted in the lead-up to the 2020 US Presidential election. Their findings, reported in Nature Human Behaviour, suggest that the impacts on voting were minimal.

The team worked with Acronym, a left-leaning organisation which paid for and ran the persuasion campaign, to design the study. The campaign itself lasted for eight months, and targeted two million voters in five 'battleground' states with promoted news stories and political video adverts via various social media outlets, including Facebook. (The news stories all reflected badly on the Republican party, or well on the Democrats, and were promoted via social media posts that contained links to these articles.) The targeted voters were considered to be politically moderate and to have a lower than average level of political awareness.

A control group of voters in these same states was not targeted with the campaign. (They did still receive whichever ads the algorithms of Facebook and the other advertising platforms decided to show them.)

On average, a participant in the 'treatment' group received (or was meant to receive – a note on this later) 754 advert impressions over the eight months between March 2020 and election day. The team adds that even if these participants didn't click on an ad or watch a video, they were exposed to the anti-Trump or pro-Biden headlines, photographs and accompanying text. "In comparison with most field experimental investigations of the effects of political advertisements, this intervention represents a large dose of pro-Biden, anti-Trump information," the team reports.

And yet, they found that it increased voting among people who had been identified as leaning towards Biden by only 0.4 percentage points, while it decreased voting among Trump-leaners by just 0.3 percentage points. This effect was "much smaller than pundits and media commentators often assume", the team writes.

"Given our findings…the popular narrative that Russia's US$150,000 Facebook expenditure in 2016 could have caused enough differential turnout to affect the outcome of the election is implausible," they add.

As the team also notes, however, there were a few problems with their study. One of the main problems was that not all of the voters assigned to the treatment group could be identified on social media, and served the adverts. In fact, it seems that a "large fraction" of the treatment group didn't get targeted. This fraction could have been as high as 40% or more for campaign content delivered via Facebook. Also, a "small fraction" of the control group was "probably treated", the team reports. "One potential concern is that even though the treatment group was exposed to more political advertising than the control group, the control group was nevertheless exposed to some."

Still, these difficulties may also apply to other similar campaigns. So, the results do suggest that mass, sustained digital political persuasion campaigns don't have major impacts on voting in national elections (even among politically moderate voters) — and so are unlikely to alter the results — while also being very costly to run.