ProPublica

Journalism in the Public Interest

Cancel

Message Machine Starts Providing Answers

Political campaigns use massive databases to target their constituents. ProPublica’s Message Machine is beginning to find some answers on how they do it.

.

January 3rd, 2013: This article has been corrected.

On Aug. 7, Mel Roseman, a retired school teacher in Los Angeles, received an email from the Democratic Senatorial Campaign Committee asking him to give a political donation of five dollars. Fifteen minutes later, he received another — almost identical — email, sent to the same address. But the second email asked him for $25.

It wasn’t the first time he had received near-duplicate messages. When he looked through his emails he found that pairs of messages from the DSCC had been arriving within an hour of each another since 2010. Digging further, he saw that there were other subtle differences. For instance, one version began with the salutation “Melvin J” and the other was addressed to both him and his wife, “Mel and Gladys Roseman.” In 2010, Mel had donated to the DSCC twice, on separate occasions, and had filled out the name field in the donation form with a different name each time. Somehow, the software at the DSCC failed to make the connection, and what’s more had decided that “Melvin J” was likely to be the more generous donor than “Mel and Gladys Roseman.”

Because Mr. Roseman volunteers for a non-profit internet service provider, he told us, “I receive more than one hundred messages a day, but I didn’t pay any attention to” the messages from the DSCC. At least, until he signed up for ProPublica’s Message Machine.

In this campaign season, ProPublica’s Message Machine project has collected more than 30,000 political emails, and identified more than 2,000 distinct mailings with more more than 4,000 variations. For each email blast, the Machine creates a computer model using an artificial intelligence technique called a Decision Tree. These computer models provide hints about the ways that political campaigns microtarget their constituents to optimize donations or some other political activity. Today, you can view the results of these models.

As of this writing, the Machine’s most confident about its models for the Obama for America campaign, though it creates a model for every email from every campaign. Each email in Message machine has an icon denoting the strength of its model.

The amount of data in the Message Machine represents a very small sample of the enormous databases assembled by the campaigns and parties. The model for each email represents our best attempt to understand how the campaigns’ decide which constituent receives which version. The models will get better when they are trained with more data, so if you receive emails from campaigns, forward them to emails@messagemachine.propublica.org.

For the more mathematically inclined, we’ve got a full explanation of our methodology.

Across all campaigns in the Machine, the most significant difference between emails in a single message blast is perhaps unsurprising: campaigns change the amount of money they ask for based on how much you’ve donated in the past. In Obama for America’s case, many emails contain a “quick donate” box that asks for more money when a particular constituent saves his credit card information. Because only previous donors see this box, we use it as a signal that the recipient has given money in the past.

While coverage earlier this election cycle noted the prevalence of the $3 donation request in Obama for America’s emails — Politico called it a “magic number” — this small request is just for starters. Some Obama supporters who had previously donated are often asked for $500.

Along these lines, Obama for America sent emails to supporters from the 2008 election who donated more than $2,000 that cycle, asking for them to match their 2008 support this election cycle. But supporters who donated any amount in 2008 have caught the campaign’s eye: like this message, asking 2008 supporters to “chip in” sent on July 29.

For its get-out-the-vote operations, Obama for America also targets supporters based on their physical location. Voters in swing states can receive slightly different messages than those who live in safer states. In many email blasts from Obama for America, the campaign urges voters from non-swing states to travel to competitive neighboring states.

Despite many attempts to get them to help us understand their techniques, or even to respond to our work on the record, the campaigns remain tight-lipped about the methods they use to microtarget their supporters, and the amount of data they have on voters because, they tell us, they do not want to give up any advantage to their opponents. Adam Fletcher, a press officer for Obama for America, when contacted about our previous story on the campaign’s digital efforts, declined to participate “in stories about our technology/digital strategy for the simple fact that we don’t want the other side to get a heads-up.”

But there are hints at the ways campaigns craft and segment emails. In an undated article for the industry newsletter Winning Campaigns, Ken Strasma, Obama’s head of targeting in 2008, wrote about some of the techniques used in microtargeting. He outlined a list of computer algorithms traditionally used in private sector marketing, “there are a great many statistical techniques that can be used in microtargeting including regression analysis, segmentation techniques like CHAID and CART, and newer techniques,” and continued, “Phones, mail, door-knocking, even cable, radio and broadcast television can be targeted to the most appropriate target universes, saving the campaign money and delivering the campaign’s messages to the most receptive audiences.”

The Obama, and Romney campaigns did not respond to requests about exactly how much they spend on their email operations. An analysis of Obama for America’s emails indicate they send most messages from Blue State Digital’s servers. According to FEC filings, the campaign has paid Blue State Digital more than two and a half million dollars for that service, since 2011. A similar analysis of the Romney campaign messages show that they send their emails from servers run by Targeted Victory, a firm founded by Zac Moffat, who is now Romney’s digital director. This election season the campaign has paid Targeted Victory more than thirteen million dollars, which includes some ads bought by Targeted Victory on behalf of the campaign.

The DSCC did not respond to a request for comments about this story.

The Message Machine will continue to collect emails through the rest of the political season and beyond. Our code will get better at finding targeted emails as more data comes in. If you get campaign emails and would like to discover how the campaigns are targeting you, please forward your political emails to emails@messagemachine.propublica.org.

Correction: This article originally stated an email sent by Obama for America on May 29th, with the subject line "Good Morning" was targeted based on age. Amelia Showalter, OFAs director of digital analytics in 2012, informed us that that email was sent to random segments, and that "fundraising emails were virtually never targeted based on age." That paragraph has been removed.

While this is of a certain esoteric interest, it strikes me as a significant amount of computational and personnel resources being devoted to an effort that is only of peripheral interest to a limited audience…most notably, Republicans who would be interested in obfuscating or misleading such donation-solicitation campaigns.

A better purpose for such resources, IMHO, might be determining how many banks are paying off how many college and university student aid financial advisors to ensure that students and parents are burdened with gigantic student debt mountains when they leave school to find out they can’t get a job.

Or why is it that the Republicans seem to be - according to data input to you from “volunteers” and reported without analysis or explanation - getting three times as much air timer for their advertising dollar as the Obama and other Democratic campaigns?  That is a significant anomaly.

And so on.

clarence swinney

Oct. 22, 2012, 10:21 a.m.

Try blaming Bush Cheney Neo-con Cohorts

That group incresed spending 90% (Obama will 8.6%)—Debt 112% (double in 8)—1400B Deficit—

Lowest job creation since Hoover 31,000 per month to Carter 218,000—Clinton 237,000

Let Wall Street run wild Gamblimg and selling toxic mortgages. Great Recession. Paul Kriugman had been writing Bush since 2002 in NYT articles. He posted them in a book.

Awesome warnings on Housing inflation and Casino Derivative Of America one on one gambling

instead of investing in businesses

Two uncalled for wars.

Imperialism and Debt is what ruined Rome-Spain-Holland-England

Bush tried hard to ruin America. Facts are undeniable.

 

Blame obama. Not for economy. He inherited Hell On Earth.

The single biggest story of the century is very likely to be this sort of “micropropaganda,”  Right now, it’s just fussing with “magic numbers” and the occasional greeting or contest, but give it a few campaigns and see that your neighbor’s view of a candidate is entirely unlike yours due to the isolation of your advertising.

After all, why bother mentioning military deployment to the Middle East to an Arab-American?  It’ll only upset the poor guy and maybe cause him to vote for someone else, whereas you wouldn’t want to preach peace to a war-hawk.  Floor wax and dessert topping…

(And fear not, Steve, I guarantee the companies running both campaigns have already done this analysis and more.  Unlike ProPublica, they’re not exclusively reliant on readers who donate to political campaigns and don’t consider the e-mail spam.)

Eventually, someone’s also going to explain the money factor, which you’d think be fertile grounds for research and journalism.  It’s bad enough that both big campaigns are on target for spending a billion dollars apiece for a job that pays orders of magnitude less while preaching fiscal responsibility.  The very idea that spending more money gets more votes—true or not—should offend every American.  But what’s worse is using “new media” like e-mail for the purposes of raising money to spend on old media, again to buy your votes.

Don’t open a conversation with voters.  That’d be stupid…

I figure that the electric and gas utilities, the banks, the credit card companies, the big retailers…all those people who collect far more data on the American people than Obama and the Democrats can, and who further have far more experience and far more expertise at mining that data - and can throw far, far more money at the problem - probably just give the data to Romney and the Republicans for free - after all, who would they want to win this election? 

It isn’t like Corporate America thinks “long term” about either their corporations or America herself…they all only think about what is going into their pockets right now, and somebody else can worry about tomorrow.  All it takes is the phrase “tax cuts” and Corporate America drops to their knees like…well, like right now.

Besides, giving such data to the Republicans would be just another form of “dark money”. 

So like I said, this effort to analyze what Obama is doing is…a distraction, and may in fact be a disservice to democracy given the likelihood that far darker deeds are taking place “behind the curtains”...back there where your bills come from.