Unbiasable

Part 4 · The biased machine

Chapter 19

Microtargeting and the Cambridge Analytica harvest

About 270,000 people took a Facebook personality quiz. The app behind it harvested files on 50 to 65 million, Cambridge Analytica scored voters on psychology, and the ads it aimed left no public record. The receipts are government documents.

What is microtargeting?

Microtargeting is political persuasion aimed at individuals instead of audiences: assemble a data profile of each voter, predict what will move them, and show each one a message the rest of the country never sees. The founding numbers are in a July 2019 Federal Trade Commission complaint. A Facebook personality app that roughly 250,000 to 270,000 users interacted with harvested profile data on 50 to 65 million people, and Cambridge Analytica matched the files to U.S. voter records and scored them for personality. A front page can be compared, argued with, and held up next to a rival's. An ad built for one person appears once, persuades or fails, and vanishes.

Primary source 01 / 11

This action seeks to hold Respondent responsible for its deceptive acts and practices to harvest personal information from Facebook users for political and commercial targeted advertising purposes.

Federal Trade Commission In the Matter of Cambridge Analytica, LLC (complaint), 2019 · Docket No. 9383, paragraph 1

A quiz for 270,000, a file on 65 million

The app was the GSRApp, known publicly as thisisyourdigitallife. It paid survey takers a few dollars to answer questions and log in with Facebook, then used a permission the platform granted developers at the time: access not only to the user who consented but to that user's friends, who never saw the app at all.

Primary source 02 / 11

the GSRApp harvested Facebook user profile data from approximately 250,000–270,000 Facebook users who directly interacted with the app, as well as 50–65 million of the “friends” in those users’ social networks.

Federal Trade Commission In the Matter of Cambridge Analytica, LLC (complaint), 2019 · Docket No. 9383, paragraph 1

For each person the harvest took the Facebook User ID, gender, birthdate, location, and the record of public pages they had liked; from the app's direct users it took the friends list too. The likes were the valuable part. Researchers at the University of Cambridge had shown that an algorithm reading them could score a person on the OCEAN psychometric scale: openness, conscientiousness, extraversion, agreeableness, and neuroticism. The FTC's complaint records the claim that made the data worth buying:

Primary source 03 / 11

their algorithm, which was more accurate for individuals who had more public Facebook page “likes,” could potentially predict an individual’s personality better than the person’s co-workers, friends, family, and even spouse.

Federal Trade Commission describing the University of Cambridge research that Cambridge Analytica set out to commercialize In the Matter of Cambridge Analytica, LLC (complaint), 2019 · Docket No. 9383, paragraph 8

The purpose was never in doubt. The complaint states that Cambridge Analytica intended to offer voter profiling, microtargeting, and other marketing services to U.S. campaigns, and that of the 50 to 65 million harvested friends, at least 30 million were identifiable U.S. consumers whose files could be matched to voter registration records.

At the point where the survey asked for Facebook access, the screen made a promise:

Primary source 04 / 11

We want you to know that we will NOT download your name or any other identifiable information—we are interested in your demographics and likes.

The GSRApp's authorization screen shown to every survey taker, as reproduced in the FTC's complaint In the Matter of Cambridge Analytica, LLC (complaint), 2019 · Docket No. 9383, paragraph 23

It was false:

Primary source 05 / 11

Contrary to this representation, the GSRApp collected the Facebook User ID of those users who authorized it. A Facebook User ID is a persistent, unique identifier that connects individuals to their Facebook profiles.

Federal Trade Commission In the Matter of Cambridge Analytica, LLC (complaint), 2019 · Docket No. 9383, paragraph 24

And the complaint records why the sentence was written at all:

Primary source 06 / 11

Cambridge Analytica, Nix, and Kogan included this representation after finding that half of the survey participants initially refused to grant the GSRApp permission to collect their Facebook profile data.

Federal Trade Commission In the Matter of Cambridge Analytica, LLC (complaint), 2019 · Docket No. 9383, paragraph 24

Half the participants said no, so the operators wrote a privacy promise that was not true. Consent, in this system, was a conversion problem to be engineered around.

The ad only its target sees

A British parliamentary inquiry that heard 73 witnesses across 23 oral evidence sessions treated the same machinery as an election problem. The House of Commons committee's final report, published in February 2019, carried the UK elections regulator's description of what microtargeting removes from public view:

Primary source 07 / 11

Only the voter, the campaigner and the platform know who has been targeted with which messages. Only the company and the campaigner know why a voter was targeted and how much was spent on a particular campaign

The Electoral Commission the UK elections regulator, as quoted in the committee's report Disinformation and ‘fake news’: Final Report, 2019 · House of Commons DCMS Committee, HC 1791, paragraph 203

Every earlier chapter of this encyclopedia describes bias you can inspect: a front page, a broadcast, a feed that at least renders on a screen someone else can compare. Microtargeting withdraws the message from that shared record. It is the paid twin of the filter bubble: the feed narrows what reaches you, and the targeted ad decides what is aimed at you. Propaganda at least had to publish; the whole country saw the same poster and could answer it. The committee's verdict on the rules governing this was blunt:

Primary source 08 / 11

Electoral law is not fit for purpose and needs to be changed to reflect changes in campaigning techniques, and the move from physical leaflets and billboards to online, microtargeted political campaigning.

House of Commons Digital, Culture, Media and Sport Committee Disinformation and ‘fake news’: Final Report, 2019 · HC 1791, paragraph 211

The verdicts on the record

The committee's summary judgment on the company holding the data became the report's most quoted line:

Primary source 09 / 11

Companies like Facebook should not be allowed to behave like ‘digital gangsters’ in the online world, considering themselves to be ahead of and beyond the law.

House of Commons Digital, Culture, Media and Sport Committee Disinformation and ‘fake news’: Final Report, 2019 · HC 1791, paragraph 139

The data itself proved hard to kill. After press reports in December 2015, Facebook demanded deletion, and the complaint records what that was worth:

Primary source 10 / 11

While Kogan and SCL Elections certified to Facebook that they had deleted the data obtained through the GSRApp, individuals or other entities still possess this data and/or data models based on this data.

Federal Trade Commission In the Matter of Cambridge Analytica, LLC (complaint), 2019 · Docket No. 9383, paragraph 30

The FTC charged three counts of deception and attached the order it believed should issue. The remedy reached past the data to the models trained on it:

Primary source 11 / 11

Delete or destroy all Covered Information collected from consumers through GSRApp, and any information or work product, including any algorithms ... that originated, in whole or in part, from this Covered Information.

Federal Trade Commission In the Matter of Cambridge Analytica, LLC (order), 2019 · Docket No. 9383, order provision IV

By then Cambridge Analytica was already in bankruptcy; it filed in May 2018. What microtargeting adds to the story of media bias is a removal: the shared text. When every voter can be shown a different front page, there is no front page to compare, and comparison is the method this entire encyclopedia teaches. The next chapter is the law that decides who, if anyone, answers in court for what the feed shows.

Frequently asked

What is political microtargeting?

Advertising aimed at narrow sets of individual voters selected by data profiles rather than by geography or broadcast audience. A campaign predicts from your data which message is most likely to move you and shows it only to people like you, so different voters see different appeals and none of them see the whole campaign.

What did Cambridge Analytica actually do?

According to the FTC's 2019 complaint, it worked with researcher Aleksandr Kogan, whose GSRApp personality survey harvested Facebook profile data from roughly 250,000 to 270,000 app users and 50 to 65 million of their friends. The data was scored on the OCEAN personality scale, matched to U.S. voter records, and used to sell voter profiling and microtargeting services to U.S. campaigns.

Was the Cambridge Analytica data collection legal?

The FTC charged the company with deceptive practices: the app's consent screen told users it would not collect identifiable information while it collected their Facebook User ID, a persistent unique identifier. The UK's parliamentary inquiry separately concluded that electoral law was not fit for purpose for online, microtargeted campaigning. The company filed for bankruptcy in May 2018.

Why does microtargeting matter for media bias?

Because it removes the shared page. As the UK's Electoral Commission put it, only the voter, the campaigner, and the platform know who was targeted with which messages. A biased front page can be compared against a rival's and answered in public; a microtargeted ad reaches its audience and disappears, leaving no record to inspect.

The primary sources

The documents this chapter quotes. Read them yourself.

The field guide

Now watch it happen in real time.

Every mechanism in this guide runs in tomorrow’s news. The brief lays the same stories across ten worldviews so you can catch the framing yourself. Free.

Free to start. No spin. No yelling.

$5/month, cancel anytime. Secure on Stripe, and nothing is ever paywalled.

One-time, secure on Stripe. No account.