A Facebook Experiment

Solid social science on the opinion pages (needless to say news reporters consider interest in randomized controlled experiments to be opinion.

Christian Caryl explains how it is possible to determine the effect of the Russian influence campaign on the 2016 presidential election.

Sinan Aral, a professor at the Massachusetts Institute of Technology [skip] says he and his colleagues want to study the Russian influence campaign in precisely this geographical context. The MIT scholars have developed a robust methodology for assessing how social media campaigns influence the behavior of their targets — and now they want to bring it to bear on the Russian meddling in 2016 [skip]

“For example, Facebook and Twitter constantly test new variations on their feed ranking algorithms, which cause people to be exposed to varying levels of different types of content,” they write. “One underpublicized A/B test run by Facebook during the 2012 U.S. presidential election caused users to be exposed to more ‘hard news’ from established sources, with effects on political knowledge, preferences, and voter turnout.” Given access to adequate data, the researchers claim they can estimate the impact of the Russian influence campaign in Michigan, Wisconsin, Pennsylvania and Florida “with 95% to 99% confidence.”

Facebook performed the necessary experiment, because they perform experiments all the time aiming to maximize user engagement. These are genuine randomized controlled experiments (because Facebook’s profits are on the line). The names of people exposed to more or less hard news can be compared with the lists of people who actually voted (which are public) only if facebook is forced to cease to protect their privacy (which really means to protect Facebook from proof that they let the Russians trick Americans into electing Trump).

I don’t think the researchers will ever get access to “adequate data” but I do think it is worth fighting for such access.

update: this might not have been clear in the original post. The idea is to use which of the 2 algorithms was used as an instrument for exposure to Russian propaganda. So it is engagement with known russian propaganda regressed on the algorithm used to check the association of (engagement with known Russian propaganda)(demographic characteristics typical of Democrats) and turnout (in particular of African .

To be crude (and explicit) the idea is that African Americans with the hard news algorithm interacted less with Russian propaganda and were more likely to vote. Or not (good experiments are ones where one doesn’t know the result before analysing the data).