Wednesday, December 16, 2015

On the uses of Big Data in campaigns


Apparently the Ted Cruz for President campaign is making an expensive effort to use the massive trove of information that data analytics companies can collect about all of us these days to build his Iowa campaign. The Washington Post describes a door knocking encounter guided by his data geeks (a company owned by one of his backers):

URBANDALE, Iowa — As Cecil Stinemetz walked up to a gray clapboard house in suburban Des Moines last week wearing his “Cruz 2016” cap, a program on his iPhone was determining what kind of person would answer the door.

Would she be a “relaxed leader”? A “temperamental conservative”? Maybe even a “true believer”?

Nope. It turned out that Birdie Harms, a 64-year-old grandmother, part-time real estate agent and longtime Republican, was, by the Ted Cruz campaign’s calculations, a “stoic traditionalist” — a conservative whose top concerns included President Obama’s use of executive orders on immigration.

Which meant that Stinemetz was instructed to talk to her in a tone that was “confident and warm and straight to the point” and ask about her concerns regarding the Obama administration’s positions on immigration, guns and other topics.

All very high tech and creepy, except that, if you think about campaigning, it really is not.

As Paul Waldman points out, campaigns have been making their best guesses about how to approach voters for decades, as long as they've had fairly accurate lists of who voters might be and where they are. For most voters, some combination of knowing age, ethnicity, neighborhood, voting history and Party affiliation will tell you all you need to know to predict a lot about their voting behavior.

In particular, if the voter didn't vote last time or the time before, she's probably not going to vote this time if she is even actually registered.

So let's deconstruct the interaction WaPo describes here: it is immediately obvious that using the data trove in this interaction at the door requires subtlety and sophistication from Mr. Stinemetz. Whatever his own enthusiasms or fears which have made him a Cruz supporter, he has to be willing put those aside to follow the prompts given by his iPhone. We have to assume that the Cruz campaign has invested a lot of training effort in getting him up to speed on the various approaches to voters they've defined on the basis of their analytics. He is almost certainly paid for his door knocking efforts; campaigns can't achieve the level of messaging discipline their analytics suggest they need with a rabble of short term volunteers.

And, crucially, Stinemetz needs to be a person who is able to do what is asked of him: to assume a different tone and persona at each door in order to utilize what the campaign knows about the target. Hey -- if he's that good at the fundamental skill of retail politics, why isn't he running for something? Maybe not President, but how about dog catcher?

All around, this makes a nice tableau, but it isn't how campaigns work when they get to scale. As Waldman points out without using these terms, increased data is certainly good for identifying who is probably already a supporter based on a demographic profile. Therefore, in an election in which turnout is likely to be low but the campaign has the money to pinpoint likely supporters, the data can be usefully used to target people who must be encouraged to vote. Big data is good for turnout.

What data is less (verging on no) good for is persuasion. If voters need to be given a reason to make one choice rather than another, data helps much less. Most of us ignore social media ads. We are surrounded by TV and other media, but again, being a 21st Century citizen means shutting this stuff out. And we self-sort pretty effectively based on cues from candidates and our accustomed biases. The most accurate and expensive data analytics don't much change that.

Actually, the Wapo description of Cruz' door knocking project does catch what their data is good for: this stuff can help find valuable needles within the voter haystack: plausible campaign workers. The campaign still has to invest massively in training and organization, but it might find some more bodies.

Stinemetz went through his call list one evening and was pleasantly surprised at how quickly people signed up.

“I got three precinct captains to sign up just now,” Stinemetz said, after dialing just a handful of potential volunteers. “It’s like they were just waiting for us to call.”

Now the campaign has to be able to carry the costs of using them effectively ...

No comments: