Episode 915: How To Meddle In An Election : Planet Money In which someone runs a science experiment on an actual election, on actual voters, to test the persuasive power of ethically sketchy methods.
NPR logo

Episode 915: How To Meddle In An Election

  • Download
  • <iframe src="https://www.npr.org/player/embed/726536757/726852480" width="100%" height="290" frameborder="0" scrolling="no" title="NPR embedded audio player">
  • Transcript
Episode 915: How To Meddle In An Election

Episode 915: How To Meddle In An Election

  • Download
  • <iframe src="https://www.npr.org/player/embed/726536757/726852480" width="100%" height="290" frameborder="0" scrolling="no" title="NPR embedded audio player">
  • Transcript

AILSA CHANG, HOST:

David Goldstein lived politics for decades behind the scenes, working on this campaign or that campaign, always for the Democrats.

ALEX GOLDMARK, HOST:

His specialty was the digital stuff - choosing which strategy to take online or using data to analyze which voters to focus on.

CHANG: And after years of feeling like he had a handle on how campaigns work, 2016 happened. Two outcomes most people predicted would never come true came true. First, there was Brexit in Europe, and then Donald Trump in America.

GOLDMARK: And when David tries to make sense of it, he realizes that some of the same people were behind both campaigns.

DAVID GOLDSTEIN: It is so rare and so hard in politics to pull an upset. They are few and far between. Maybe in your entire 30-, 40-year political career, you will be on the winning side of a single upset. They pulled off two upsets in back to back years.

GOLDMARK: Specifically, there was one company that worked on both campaigns - Cambridge Analytica.

CHANG: And as we heard about a lot since then, they got caught using personal data without permission, some of it from Facebook. They were accused of breaking laws.

GOLDMARK: But they also claimed to have pulled off new legal tactics for finding especially personalized, even intimate ways to persuade voters using online data.

GOLDSTEIN: There is a lot that was simply innovative, that we hadn't thought of doing before. A big thing was, you know, this kind of setting up your own personal, like, online ecosystem of content meant to persuade people to a certain action, kind of like drawing a fly into a web.

GOLDMARK: They were claiming they could create profiles of people based on their online data and then use that against them to change their minds, maybe without them even knowing it was happening, maybe using misinformation. And it just felt like playing dirty.

CHANG: And he's thinking, this is not how democracy should work.

GOLDSTEIN: I'm not somebody who scares easily, but I am legitimately scared.

GOLDMARK: He's looking at this like an arms race, with the Democrats way behind, as he sees it. And he wants to sound an alarm.

CHANG: He sets up a bunch of meetings at places like Google, Facebook, Twitter, other big online companies.

GOLDSTEIN: And my hope was - is that there could be some kind of, like, cross-industry investigative body that could work with the government and regulatory and private industry. And I was shot down very fast.

GOLDMARK: I imagine the people that he talks to at these big tech companies are, like, obviously, you can target people and persuade them; that is the point of our ad platforms - to get the most persuasive ad in front of every person, exactly in the way that persuades them the most. That is why we get paid so much.

GOLDSTEIN: So (laughter) I was told that. Nicely, though, the same person who shot me down said at the same time, if you want to see something happen on this, you have to do this yourself.

GOLDMARK: You have to do it, too.

GOLDSTEIN: Yes.

GOLDMARK: You have to become a Cambridge Analytica yourself.

GOLDSTEIN: Well, that - yes. Without the illegal part.

(SOUNDBITE OF ADRIAN QUESADA AND SKINNY WILLIAMS' "THE SPAGHETTI WESTERNER")

GOLDMARK: Hello and welcome to PLANET MONEY. I'm Alex Goldmark.

CHANG: And I'm Ailsa Chang. To prove that there is a threat to democracy, David decides to mess with democracy and maybe help his own party along the way.

GOLDMARK: When he says he wants to prove there's a threat, he mean proof, like with a science experiment, on actual voters in an actual election, with data, statistics, controls - all of it.

CHANG: Today on the show, how to meddle in an election, step by step.

(SOUNDBITE OF ADRIAN QUESADA AND SKINNY WILLIAMS' "THE SPAGHETTI WESTERNER")

CHANG: So at this point, David's living in New York City, and his main job - it's not in politics. He's now working at a big corporate ad agency. This science experiment thing he's hatching, it's like a side project, a nights and weekends thing. And the first thing he needs to get it going is some money.

GOLDMARK: And so he calls up some old contacts he has, and most of the donors that he goes to, they're like, are you kidding me? You want me to fund a plan for you to go rogue and be a Cambridge Analytica of the left with potentially ethically sketchy methods on real voters in a real election, but you don't know which one yet? And David is like, yes, exactly; that is exactly what I want. Eventually, he scares enough donors and raises $85,000. And he's in business.

CHANG: It takes David about two weeks to get his whole team together.

GOLDMARK: He's going to be the brains. He comes up with the strategy, picks the messaging and designs what people will see when they click his ads.

CHANG: He needs someone to wrangle the data for him.

GOLDMARK: And he needs people to design the ads he'll be running.

CHANG: He also checks the whole plan over with an election lawyer because he wants to stay legal.

GOLDSTEIN: So we had kind of formed our team, right? And we had all resolved, OK, we're going to do this. But the thing was, it was, like, summer 2017, you know (laughter).

CHANG: The next big elections are more than a year away, and he needs a big election because he's running a science experiment on voters. He needs lots of voters.

GOLDMARK: He needs a way to make two groups of voters, ideally, demographically identical.

CHANG: Voting histories identical.

GOLDMARK: Turnout history, partisan makeup - all the things that he can measure - as identical as possible.

CHANG: And he needs to be able to break those voters in half, into one control group and one test group.

GOLDMARK: Half get his campaign and half don't. And then he's going to measure it.

CHANG: Like a medical drug trial, but for politics.

GOLDSTEIN: It was very tough to figure out exactly how we would cut up the potential electorates in such a way that we could reliably say we had an impact in the test condition because this did not happen in the control.

CHANG: And Alabama comes through. The senator there, Jeff Sessions, he becomes the attorney general. So there's an open seat.

GOLDSTEIN: So what had happened is the Alabama Senate election was called. It didn't seem winnable for the Democrats at first.

CHANG: This is that election, by the way, that ended up with Roy Moore, the Republican candidate, getting accused of being a pedophile, against Doug Jones, the Democrat.

GOLDMARK: But David is assessing this race way before those are the final candidates. And he knows it's been about 25 years since a Democrat won the Senate seat in Alabama.

GOLDSTEIN: But what I point out to my team and what occurred to me was we don't have to win. It doesn't matter if we win. We just have to show an effect. I was like, if he loses by 14 points, but we go back to our group that we hit and he only loses in that group by eight points, we can show that we lessened the loss, if our hypothesis is right.

GOLDMARK: And now to set up the experiment - he takes three state senate districts, matches those against three others.

CHANG: Three test districts, three control districts.

GOLDSTEIN: And three of those, you give them the ads, and the other three, you don't give them the ads, but you measure them like you did, so you can compare. Yeah.

CHANG: And with digital ads, it's easy to pick who will see your ads and who won't.

GOLDSTEIN: Yeah. And in fact, I think what's actually been surprising to a lot of people is this element of what we call negative targeting. So if I don't want someone to see my ads, I can put their name into the system, and they will never see my advertising.

GOLDMARK: He said he could even feed in the names of all the campaign staff, the journalists covering the race, the regulators, too, and make sure that they never see his campaign firsthand - if he wanted to.

CHANG: But the people he does want to target, he can bombard them with whatever ads he can afford to pay for.

GOLDSTEIN: One of the wonderful things - or perhaps awful things - about this new age of digital advertising is the number of times that we can touch you, right?

GOLDMARK: That sounds creepy.

GOLDSTEIN: I know (laughter). And you know, it's referred to as, you know, essentially, a touch. Like, how often can I get my content, what I want you to see, in front of you?

GOLDMARK: With the old ways, a candidate would flood local TV ads and be lucky to get it in front of a voter they wanted to reach maybe twice a day.

GOLDSTEIN: With the amount of time that people now spend on their smartphones, on their laptops and simply online - you know, four to five hours a day - I now have hundreds of opportunities in one single day to hit you and possibly even thousands.

CHANG: He can show you one ad and then another and another, testing out different messages or designs on different websites you might check, trying to figure out what catches your eyes and when.

GOLDSTEIN: Going after the same pool of voters and hitting them over and over and over and over again until they figured out what would make them click.

GOLDMARK: And once the automated ad system has that information, it learns. The next time the system tries to target you, it goes right to the stuff it knows you like - always testing, always learning, very rapidly.

GOLDSTEIN: And that's why this is so devastatingly effective. Because you can cut into people who are typically very sophisticated viewers, people with high educations, you know, people who are very wealthy. But because you're continually exposing them until you find out what flips their trigger, and something always does, you are getting access into that person's mind that you never would have previously.

GOLDMARK: And sometimes looking into that person's mind is relatively harmless. I mean, it's just annoying when the Internet wants you to buy that pair of sunglasses; it gets dangerous when it's trying to change the way you vote.

CHANG: Especially when the person behind all of this doesn't want you to notice they're even trying.

GOLDMARK: And this isn't just Facebook, to be clear; he is trying to reach people all over the web, regular websites, too. His system, it's like any online ad system.

CHANG: And for all this to work, David needs to create a ton of different messages to test out, and those messages need to lure voters.

GOLDSTEIN: The best thing to do, for me, when - sorry, let me think about how to phrase this - on it - (laughter) I'll sound nefarious; that's fine. So basically, what you are going to click on in the future is what you are clicking on now, right? I mean, that - we're creatures of habit.

GOLDMARK: That makes sense.

GOLDSTEIN: So I assumed if I went to right-wing websites and took a look at what the advertisers were doing on those, that would help direct me to what the brief should be for my own content producers and ad-makers.

GOLDMARK: Because you had to actually get inside the mind of the Republicans.

GOLDSTEIN: Exactly.

GOLDMARK: Which was not normally what you do.

GOLDSTEIN: Right, exactly.

GOLDMARK: He does a kind of obsessive study on the look and the styles, the themes, the phrasing even.

GOLDSTEIN: Very big on words having to do with things like - you know, anything that spoke to, like, the flag, the U.S., patriotism - honor was a very important concept that came out of my deep dive. So when I was putting my ads together for them, I was trying to tap into those themes.

GOLDMARK: All right, Ailsa, I want to show you something.

CHANG: Betrayal of Alabama's honor, and there's a dude in a cowboy hat. Alabama's great shame.

GOLDMARK: It looks like a 1990s website level of, like, animation.

CHANG: Right, like you did it at home, not with Paint; it's like a step above. But, like, with stencils that you're borrowing from some program (laughter).

GOLDMARK: He says that's what works.

CHANG: Like amateurish?

GOLDMARK: Enough that I was actually curious why he did that.

That was on purpose?

GOLDSTEIN: (Laughter) Yes, absolutely on purpose.

GOLDMARK: That wasn't because you just had a small budget?

GOLDSTEIN: No (laughter). No, I mean, look - one of the nice things about working in New York is you have access to the best town in the world. And yes, it was tough for me to tell these very, very talented art guys, this is too nice. Make it simpler, you know, more blocky.

GOLDMARK: But those are the ads that fit in with the other ads on the websites that his target voters are going to. So that's what he gets, and now he's almost ready to go.

CHANG: And then the election changes. Roy Moore becomes the candidate on the Republican side, and he is now probably the worst candidate imaginable for a general election because this is the time where he's getting barraged by all these allegations about how he's a pedophile. And there are now prominent Republicans saying Roy Moore is unfit for office.

GOLDMARK: Suddenly, David's science experiment is happening in a nail-biter of a close election, and maybe his science experiment can help his side win.

(SOUNDBITE OF MUSIC)

GOLDMARK: So David has his click-harvesting machine, the ecosystem drawing flies into a web.

CHANG: But what does a click get him? I mean, where do you go when you click?

GOLDSTEIN: What happens when they get the click is they go to a website that I've set up just for them.

CHANG: In this case, different websites for different groups of voters.

GOLDMARK: David's plan for Democrats is just to boost voter turnout. So when a Democrat in Alabama clicks on an ad, they get sent to a website with videos and articles designed to boost their mood, to uplift them and, in the process, make them feel like everything will come down to their one vote.

CHANG: With Republicans, David's plan is a little savvier, a little sneakier. He wants them to vote less. His plan for moderate Republicans is to make them want to write in candidates other than Roy Moore by making them think that that's what everyone else is doing.

GOLDMARK: So he sends the moderate Republicans to a website. It looks like a news site. It's filled with true stories about how other Republicans are planning to write in other candidates.

CHANG: His plan for conservative Republicans is to make them stay home.

GOLDSTEIN: On the conservative website that was just for the conservative Alabamians, I filled it with content that was not written by me; it was written by other conservatives, out somewhere, anywhere on the web, who are writing about how terrible Roy Moore was.

GOLDMARK: So not fake news. This was...

GOLDSTEIN: Not fake news. I never once created article, tweet - nothing, ever.

CHANG: So what this might look like is a Breitbart or a Drudge-style news site.

GOLDSTEIN: But except it has pieces linking to the local newspaper or maybe something a conservative writer posted on his blog.

CHANG: And it just happens that all these posts on the site paint Roy Moore in this really bad light.

GOLDMARK: David is using the site to create a fake view of the world, with real stories. They're out of context, but they are still technically real. And the point is to dissuade voting.

CHANG: So the sites are all ready. His research tells him he'll have the most impact closer to the election. And about a week and a half before election day, everything goes live.

GOLDSTEIN: Essentially, what happens is the ads start to get pushed out, right? And they start to appear in front of people in Alabama, either on their smartphones or on their laptops.

GOLDMARK: And he waits. It's an automated system, anyway. About this point, it's easy to say, oh, come on, this is overplaying the power of ads. Who really clicks on them? Or I'm immune to ads. I lived through Upworthy. But that is not the point. It doesn't really matter if you do it; it only matters if someone does it, a sliver of the people he's targeting.

GOLDSTEIN: You might be immune, right? And the guy next to you might be immune, and the guy next to that person might be immune. But if I only need to change 3% of people in order to affect a given result, then I can go 97 people down and not have an effect, but as long as I have an effect on one, two, three, and only on this one specific action, that all the rest of the time they're immune, but to me they're not, then I can literally change the world.

GOLDMARK: I talked to digital campaign folks, Republicans and Democrats, professors who study this, regular old ad buyers for noncampaign products, and when I told them about David, they all kind of went, yep, that makes sense. Half of them told me other ways that he could have done it that are even more sophisticated. But that, yeah, in principle, this is what campaigning is becoming.

CHANG: And when David checks his early data, the grand science experiment seems to be working.

GOLDSTEIN: I always say, the biggest shock was the most engaged folks were the conservative Republicans.

CHANG: They did the most clicking out of any of the groups. They spent the most time in his ecosystem of websites.

GOLDMARK: To measure the vote impact, when he does eventually get the voter rolls and he compares his control districts with his experimental treatment group to see which groups turned out to vote more, it looks like he has had a real impact.

GOLDSTEIN: The Democrats in our experimental group turned out at a 4% higher level than the Democrats in the control group.

GOLDMARK: Four percent.

GOLDSTEIN: Four percent.

GOLDMARK: So you think that your work caused 4% higher turnout...

GOLDSTEIN: Yes.

GOLDMARK: ...In the places you tried it?

GOLDSTEIN: Yes.

GOLDMARK: That's a lot.

GOLDSTEIN: It is. It is. Most...

GOLDMARK: That is more than the margin of victory in a lot of elections.

GOLDSTEIN: Loads.

GOLDMARK: Yeah.

GOLDSTEIN: (Laughter).

GOLDMARK: It's enough to flip an election.

GOLDSTEIN: Absolutely.

GOLDMARK: And the Republicans, did it work on them?

GOLDSTEIN: It did. The experimental condition was 2.5% lower for the moderate Republicans.

GOLDMARK: OK, so you believe you caused a 2.5% lower turnout for moderate Republicans that you...

GOLDSTEIN: Exactly.

GOLDMARK: ...That you targeted?

GOLDSTEIN: Exactly.

GOLDMARK: OK.

GOLDSTEIN: And then for the conservative Republicans, it was 4.4%...

GOLDMARK: OK.

GOLDSTEIN: ...Of a drop-off.

GOLDMARK: Which is also a lot to take out of someone's base.

GOLDSTEIN: Exactly.

CHANG: Roy Moore loses the election to Doug Jones, the Democrat, by just 1.5 percentage points; that's just a little over 20,000 votes statewide.

GOLDMARK: So I asked David what did he think that his science experiment had caused?

GOLDSTEIN: I would say we could reasonably lay claim to about 8- to 9,000 of those votes.

GOLDMARK: So that sounds like a lot of votes.

GOLDSTEIN: (Laughter).

GOLDMARK: No, I'm serious. Like, it's...

GOLDSTEIN: (Laughter) Yeah, I know.

GOLDMARK: It just - it's a big claim.

GOLDSTEIN: It is. It is. But I'm fine sharing the - you know, I've shared the data with whoever wants to see the data. I mean, it's...

GOLDMARK: No, I would like to take a look at the data.

GOLDSTEIN: Of course (laughter).

GOLDMARK: I looked at the data. And I went and talked to professors who study this and other campaign strategists on the Republican side, and I talked through David's claims with them and his numbers. And it's pretty much impossible to verify his data. They said there's too many variables or too little information. Like, maybe Doug Jones held a bunch of big rallies in one of the test districts, or maybe a big influential pastor of a church in one of the test districts made a big show of writing in his vote against Roy Moore. It could be all kinds of things.

And even by his own numbers and his own calculations, the impact that David had on Democratic turnout is a bit closer to, like, 2% change when you factor in a few more variables. In fact, most of the experts that I talked to were skeptical of how big his numbers were, given how little money he spent. But one of the campaigners I talked to also told me that they've heard of other tests and other studies from other campaign groups that are not ever going to be made public, and in a lot of ways, this isn't so far out of line with them. What's different here is the person who ran this campaign was just doing it pretty much on his own, from an office in a city many states away, making it look easy to build an algorithmic system that could sway thousands of votes for less than $100,000.

What do you do after you meddle in an election?

GOLDSTEIN: (Laughter) Sorry, it's an intense term, but I get it. So you start showing it to other people who have to worry about the next election themselves.

CHANG: After his experiment, David showed his results to Democratic campaign folks, and he said, hire me, and I will show you how to do what I did, or at least learn how to fight against what I did, because anyone can do it or even worse.

GOLDMARK: Is it just inevitable that that's going to keep happening, as long as we have online ads?

GOLDSTEIN: I mean, not necessarily. I think it was far easier to sway people with a TV ad in the early days of TV. There was a way to do it with radio ads. And I think right now we're in a bit of a free and easy period, I think, when it comes to digital online advertising. But, you know, I do have a belief that as we grow more sophisticated, it should make it more difficult for folks like me to have that impact.

GOLDMARK: After I talked with him, I was thinking, this is just an arms race. And I wondered if that ever led him to just want a truce, to go back to the old way. So I asked him if he ever brought that up with anyone. He wrote me an email back, and he said he hadn't. He said this makes it cheaper for campaigns to reach voters, easier to get a message across. So campaigns won't give that up. Also, he wrote, I don't know how to say, I'll stop if you do, in Russian.

(SOUNDBITE OF MARC FERRARI, NATE DONMOYER AND ROSHMOND DORAN PATTEN'S "DYSFUNCTION")

CHANG: Have you subscribe to our newsletter? Do it. It's one PLANET MONEY-style story designed to be read on your phone, like how would a wealth tax work, or are doctors a cartel? Subscribe at npr.org/planetmoneynewsletter. That's npr.org/planetmoneynewsletter.

GOLDMARK: We want to thank Eitan Hersh at Tufts, Dan Kreiss at UNC, Mark Stephenson at Red Oak Strategic, Michael Beach at Cross Screen Partners, Tom Bonier at TargetSmart and everybody else who talked to us for today's episode.

CHANG: And if you liked today's episode, we're going to keep covering how data and technology are changing business and society, and let us know what you want us to cover. Get in touch at planetmoney@npr.org.

GOLDMARK: If you want more PLANET MONEY on campaigns, try episode No. 845, called REDMAP, about how gerrymandering got so out of control.

CHANG: This show was produced by Darian Woods. Bryant Urstadt edits our show.

GOLDMARK: And if you like PLANET MONEY, the best way to show your support is to share it with a friend; help us grow. I'm Alex Goldmark.

CHANG: And I'm Ailsa Chang. This is NPR. Thanks for listening.

Copyright © 2019 NPR. All rights reserved. Visit our website terms of use and permissions pages at www.npr.org for further information.

NPR transcripts are created on a rush deadline by Verb8tm, Inc., an NPR contractor, and produced using a proprietary transcription process developed with NPR. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.