Cambridge Analytica has been in the news recently because of its astute use of big data to guide surprisingly successful political races, from Brexit to Donald Trump.
But to some like Matt Oczkowski, head of product at Cambridge Analytica, these upset victories were no surprise at all.
I caught up with Oczkowski at the Global Audience Based Buying Conference & Consultancy and asked him the following questions:
Charlene Weisler: I am fascinated by to how your company used data not only to predict Brexit, but also the election of Donald Trump. Can you give me a quick overview of your approach?
Matt Oczkowski : A lot of the issues in most public polling, and in public perception of the Trump campaign in particular, were that most people had a fundamental misunderstanding of who the voting electorate was going to be this year. It was a turnout and sampling problem.
So the reason we were hired by the Trump campaign was to quantify the Trump effect, which is how Donald Trump is different from the generic Republican. Most data science and data modeling in politics is done “R vs D”: Republicans versus Democrats, like a Mitt Romney versus Barack Obama. The binary question is solved.
But as we both know, Hillary Clinton and Donald Trump are far different from any type of generic Republican or Democrat. So where we had a lot of success was understanding the trends that unfolded on election night in three major demographic groups: a major increase in older rural voters, a massive decrease in the African American vote this year, and a slight increase in the Hispanic vote. Once you start to understand these three trends, you extrapolate who is going to show up on election day.
Weisler: So you had a sense before election day of who was going to show up and who wasn’t?
Oczkowski: Correct. We had the benefit of having access to all of Trump’s donors and all of Trump’s event attendees. When you start to collect this first-party data, it starts to give you an inside peek at what a Trump-specific supporter is like. I can compare them to previous Republican donors. I can look at their voter records and see if they showed up to vote before.
And that, early on, allowed us to start building a picture what a Trump supporter looks like in our heads. But what was a really big indication was when we started to get absentee ballot and early voter returns about a month or so out from the election. Those three trends that I just mentioned were very apparent in the early vote returns.
Wisconsin was probably the biggest surprise in the election this year, and we saw these trends unfolding [there]. Four weeks prior to the election allowed us to change a lot of our tactics in terms of candidate travel, media spend, focus on issues and speeches.
Weisler: So within a four-week period, it was possible to change the course of the election?
Oczkowski: Absolutely. We knew going into it that we had to build a really dynamic data program that could keep up with the candidate — because depending on the week, Trump could tweet one thing and it would change the entire view of the electorate on that week.
So our data program had to be very elastic — which means any time a stone was thrown into the water, we had to be able to track the ripples and see [where] different parts of the electorate were actually moving in a particular direction.
So we undertook a big re-weighting exercise, four to five weeks out from the election, where we resampled all of our polling, reran all of our modeling. It showed a generous approximate three-point bumps across the entire Rust Belt.
The state we were most nervous about going into election day was Florida. We knew that if Trump won Florida, he won the election. We felt really good about Pennsylvania and Ohio, so we put a lot of emphasis on Florida.
The reason why Florida was so on the bubble for us was because of the significant Hispanic population voting in our persuasion universe. Once we started to see the returns from the rural counties and Florida around 8:30 central time, we knew the election was over.
Weisler: Were there any data points that you found particularly or surprisingly valuable?
Oczkowski: Yes. If you were to look at a generic Republican in the database, with the issues they cared about, you would see issue #1 would likely be Jobs and the Economy , #2 would be Security, #3 would be Government size/Taxes.
For an isolated Trump-specific supporter, the three issues were #1 Law and Order, #2 Immigration and #3 Trade.
And when you start to understand that profile, an isolated Trump supporter looks a lot like a Bernie Sanders supporter. It is Blue Dog Democrats. It is people who have been disenfranchised by the political system who feel that the government hasn’t done anything for them in the past and who came out to vote this year. They haven’t been out to a voting booth in several elections (which is a very difficult thing to quantify).
I don’t blame most of the press organizations [for missing these points] because they don’t have access to a lot of this information and data, but I think they were foolish to assume that 2016 would play out like 2012. It was a very different electorate.
Weisler: Are you continuing to look at the data to see how his presidency will play out and whether there are issues he should avoid or embrace?
Oczkowski: I still don’t have a full understanding of the election yet.
There are about four battleground states where I have gotten back data thus far from secretaries of state. There is still a large collection process going on. And as that data comes in we continue to
learn more and more about what happened.
We are working with a number of clients now, taking this program and furthering it to keep this understanding of the American electorate alive for a number of different purposes.
Weisler: So if you were going to give some advice to some Democratic candidates based on the data, what would it be?
Oczkowski: There are two things I like to say. One, it’s throw out the playbook. The Clinton campaign fell into the trap of running the Obama 2012 campaign. Hillary Clinton is not anything close to Barack Obama the candidate. So you need to build a program that is unique towards her.
And the other fatal mistake they made was not focusing on rural areas.
Weisler: I was surprised that of the three top issues for Trump voters, none were the economy.
Oczkowski: The fourth issue, which would be surprising to you, was wages. People felt that they deserved to be paid more, which is not necessarily a Republican issue, raising the minimum wage of any kind.
Economics was a driver, [but] a lot of Trump supporters followed his lead. His first major issue was Immigration this election. Law and order became an issue because of what was happening in the news and the press. Often times we find that the news media drives the discussion in the presidential election.
Weisler: Is the Trump universe large enough to carry the midterm elections?
Oczkowski: It’s a combination of two things. It's Trump winning his people and bringing new folks into the Republican party...
One of our struggles early on was winning back the traditional Republicans that had never seen a style and tenor of candidate like Donald Trump before. So if he can win these two factions, absolutely.
And midterm election maps are much more favorable to Republicans than they are for Democrats this year because of turnout. The question is, will Trump’s people come out again when he is not on the ballot? I think that’s what we are trying to understand here.