Welcome to our Day-to-Data series where we will be interviewing data-driven folks from a variety of industries. With the elections just around the corner, this month we’re focusing on the presidential campaign and how specific campaigns are using data.
We had the pleasure of interviewing Grace Turke-Martinez, a Junior Analyst at AFL-CIO. She is passionate about developing innovative, data-driven solutions for progressive causes. Please note, the opinions expressed in this article are the author’s own and do not reflect the view of the AFL-CIO or Umbel.
This year’s campaign cycle has been pretty wild. Can you give us an example of a specific campaign where you felt data has been used correctly and a campaign that has room for improvement?
One failure that stands out is Ted Cruz’s attempt at social pressure mail. Social pressure mail is highly effective, but always controversial. The Cruz campaign was heavy handed and misleading with their “voter violation” rhetoric, which generated a lot of anger and backlash. Examples of data being used effectively are harder to think of since good use cases largely happen behind the scenes and the major campaigns will keep their advances close to their chest until after November.
How have you seen data’s role in politics evolve in the last 10-15 years? Thoughts on the next 10-15 years?
I’ve only been working in political data for the past two years, so I can’t speak much to the past. There have been rapid advances in technology, a growing interest in data science, and increasing education at the K-12 level in computer science and programming. I hope these developments will bring new talent into the field and increasingly professionalize our work.
In the current election cycle, are there any data practices/changes/evolutions that have stood out to you?
I am carefully watching the adoption of online survey panels and other ways to use low-cost digital tools to effectively and accurately conduct surveys. Many people are wrestling with the question of how to get members of the general public to take surveys online, people we can match to our voter file and who are representative of the public at large.
It is difficult to match someone’s online persona to their offline self, and figuring out how to conduct surveys online while minimizing bias in who responds is a major challenge. It’s a question lots of people are asking and I expect to see interesting developments attempting to address the question. Costs are rising as individuals become less and less likely to complete a survey or even answer the phone, and the bias among respondents, who tend to be older and whiter than the general population, makes it difficult to extrapolate what the public at large really thinks.
What data points are most important for campaigns to capture about their supporters/voters, and how do they capture it? In other words, what data are being collected by my politicians about me?
Campaigns make their best efforts to track every interaction they have with constituents – did we knock on your door or talk to you on the phone? And, are you favorable on our issues or candidate (we call these “IDs”)? We also track the number of attempted contacts compared to actual contacts, which helps us evaluate the efficacy or our programs and informs decision making about where it makes sense to spend scarce resources in the future.
We make our best attempt to match data collected in the field to the voter file. Who you vote for is private, but whether or not you turn out to vote is public information. Your voting history (or lack thereof) and your party registration (if available) is very informative to campaigns. It’s all about maximizing votes, which is done in three ways: registering new voters, persuading voters, and turning supporters out to vote (“registration, persuasion, and GOTV” for short). Campaigns use individual-level data to help us do this more effectively. We look for individuals who are likely to support our causes, and try to get them to turn out to vote. We also look for individuals who are likely to vote, and try to persuade them to support our issues.
There has been a lot of discussion around algorithms and AI reinforcing social injustices such as sexism, racism, etc. (e.g. the racist Twitter bot – more here). Have you seen any examples of this in your work, and do you know of any methods that are being used to address it?
I haven’t seen this in my own sphere, but when working with data, there is a responsibility to use that information ethically and thoughtfully.
We use predictive models to identify and target likely supporters or people we think we may be able to persuade to support our issue or candidate. This helps us utilize our scarce resources (whether they be dollars to spend on mail or volunteers going out to knock on doors) more wisely. This also means there are segments of the population we aren’t even attempting to engage, and other progressive campaigns probably aren’t trying to engage, and I assume a similar phenomenon happens on the right.
I worry that this lack of engagement will exasperate the already extreme polarization in our country. In my social circle, I’m surrounded by people who largely think like me, and when I log into Facebook, I see posts from people with similarly progressive views.
How are we going to find compromises, pass new legislation, and move forward together as a country with the level of gridlock and vitriol in political discourse? This really concerns me.
As I mentioned above, campaigns focus on individuals who are likely to vote. People who are very unlikely to vote are largely excluded from attempts at voter contact. It concerns me that their interests may not be represented in the political system, and it’s why it is so important for Americans to turn out to vote and vote the full ballot.
If candidates know how I feel specifically about a campaign issue because of, say, my posts on social media, is there a risk of their campaign sending me a targeted ad stating that they support my belief, whether or not they actually do?
Most data vendors offer commercial variables that are derived from a person’s online shopping habits and other consumer behavior. So, in some sense, a person’s online activities are tracked for the purpose of targeting, although not with that level of granularity – and hopefully, that data is not used in a malicious way.
We use predictive models to target likely supporters. These models improve our targeting. Our knowledge of where an individual will stand on a particular issue is more like an informed guess than a definitive answer. Theoretically, if a campaign were to send ads (digitally or in the mail) to individuals they think are pro-life claiming a pro-life stance and ads to individuals they think are pro-choice claiming a pro-choice stance, I think someone would figure this out and it would generate a huge backlash.
Outside groups might not be as concerned about backlash and have been willing to resort to some pretty disgusting tactics in the past (thinking of the “Swift Boat Veterans” who attempted to discredit John Kerry’s military service or the racist smear campaign in South Carolina against John McCain in the 2000 Republican primary claiming he had fathered a black child out of wedlock). There are people who will resort to the most shameful tactics if they think it will help them win, so I’d never say never.
About Grace: Grace was born and raised in Pittsburgh, Pennsylvania. She studied international politics at Georgetown University and went to graduate school for social science at the University of Chicago, where she became interested in using quantitative methods to analyze social questions. She came into the progressive data and analytics space in the 2014 electoral cycle through an apprenticeship at the AFL-CIO, where she was trained and mentored by more experienced folks on the analytics team (a part of the political department). After her apprenticeship, she spent a year with 270 Strategies, a consulting startup that specializes in grassroots organizing, on their data & analytics team. She returned to the AFL-CIO earlier this year as a member of the analytics team.