The world now knows, thanks to Facebook’s disclosures, that Russians bought a lot of ads in an effort to influence the outcome of the 2016 presidential election. But that still leaves the more important question: Did they succeed?
I think Facebook could do a lot more to find an answer.
To measure the impact of Russia’s meddling, one needs to know whether people who saw the ads acted differently than those who didn’t. The best approach would be a randomized experiment, in which researchers would expose a randomly selected group of people to the ads, then compare their behavior to a control group that wasn’t exposed. Unfortunately, that’s not possible, because the election has already passed.
The second-best approach is what economists call a natural experiment. In this case, it would involve finding some event or circumstance that arbitrarily prevented some people from seeing the ads, but had no correlated effect on their voting behavior. It could be a difference in the way ads are served to mobile devices and PCs, or a difference in the particular times of day people used Facebook, or just the luck of the draw. With enough data, one could then use them as a control group to tease out the impact of the ads on those who did see them.
I’m pretty sure Facebook has enough data. It knows who clicked on the ads, and probably also knows who saw them (“ad impressions” in the lingo). As a business matter, Facebook would keep such records to convince customers — including Russian trolls — that they’re reaching the desired audience. This is why politicians such as Mark Warner can call for users to be notified if they were exposed.
Facebook can also get access to records of who voted, as can anyone with the necessary patience and resources. And now that it requires users to provide their real names (except in Germany), it can relatively easily figure out which of them voted.
The next step would be to identify “similar” users — that is, users whose traits suggest a similar propensity to see a Russian ad. They’d have the same “Russian ad propensity scores,” or something along those lines. Then find the random thing that caused some of them to nonetheless miss the ads, and compare the voter turnout of the exposed and unexposed groups.
How likely is it that Facebook could find such a natural experiment? My guess is pretty likely, given how many Americans use the service. The company is not exactly starved for data, and the more data you have, the more likely it is that such opportunities will pop up.
A further, more involved study could also look at how users voted. Facebook has tons of data on people’s political bents, which would be highly predictive of their choices in the 2016 election. But even barring that, a turnout analysis alone could be super interesting — as Facebook’s famous “I Voted” button experiment demonstrated.
Cathy O’Neil is a mathematician who has worked as a professor, hedge-fund analyst and data scientist. She founded ORCAA, an algorithmic auditing company, and is the author of “Weapons of Math Destruction.”
This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.