Behind the scenes of election night projections

John Lapinski, director of elections at NBC and the Robert A. Fox Leadership Professor of Political Science at Penn, discusses projecting elections and what to expect from the midterms.

Poring over mountains of data on Tuesday will be John Lapinski, director of elections at NBC—who happens to double as the Robert A. Fox Leadership Professor of Political Science at Penn, as well as the faculty director of the Robert A. Fox Leadership Program at Penn and Penn’s Program on Opinion, Research, and Election Studies (PORES). 

John Lapinski surrounded by Penn undergraduate workers with NBC's Elections Unit
Penn Political Science Professor John Lapinski, third from right, poses with Penn undergraduate students working at the NBC Elections Unit in November 2016. (Photo: Alex Schein)

And it’s not his first go-around as a designated numbers-cruncher on the most important day of the year for political scientists, data scientists, and statisticians. 

Lapinski began working with NBC’s Elections Unit—which encompasses NBC, MSNBC, CNBC, Telemundo, and digital properties—during the 2000 election, as he was finishing his Ph.D. at Columbia, before taking on a more significant role shortly after. He rose through the ranks in 2008 before becoming director after the 2012 election. He works to not only call elections, but is involved with analyzing data that instructs editorial content, a “privilege,” he says, that affords him the invaluable insights, as an academic, that come with knowing exactly how news is made. 

Here, ahead of Tuesday’s midterm elections, Lapinski discusses his role at NBC, the undergraduate students at PORES afforded the opportunity to be involved with the Decision Desk on election night, and what knowledge to keep in mind as results stream in. 


What is your job with the NBC Decision Desk, and what does that involve?

I am director of elections at NBC News, and also director of their data analytics lab. That means I run the Decision Desk at NBC, and we’re in charge of doing all the election night analysis for the entire company. I am the head of this, where I project all national races as well as gubernatorial contests. So, for example, when we projected President Trump to have won in 2016, that’s my team that does that. We also conduct all the exit polls, interview more than 100,000 voters on Election Day, and my team does all the analysis. All of the data-driven stories that come out of the company, we produce all the data and are analyzing that for the umbrella of media shops. We’re the center of everything, calling hundreds of races. For the Senate, governors’ races, House races. There’s a lot of stressful jobs in the company, but the entire reputation of NBC is on the line that we do a good job. No one wants to have a 2000 election again, where the networks projected the wrong winner. That led to Congressional inquiries and had huge implications for the country. 

I am also director of the Fox Leadership Program. The Fox program has been incredibly generous in creating a number of Fox-PORES undergraduate fellows. We have created an opportunity where I’ve brought a lot of Penn undergraduate students to work with me, and a few other Penn faculty members, on election night at NBC’s Decision Desk. We have dozens of people working at NBC on the Decision Desk, which includes a number of my distinguished colleagues at Penn, as well as our students.  We are in charge of aggregating the vote for the country, which we do in a collaborative effort with the National Election Pool. NBC, ABC, CBS, and CNN are the members who make up this pool. The logistics behind doing this effort on election night is kind of crazy. It takes an entire team of technical people of all types to make this happen.  

In terms of making the sausage, how does that data come in?

We work with a vendor. And we send out more than 3,000 vote reporters to counties throughout the country, and take secretaries of state who have election feeds and either collect it manually or electronically if the state or county provides it. We collect it all in real time, analyze it, and turn that into projections and other analyses.

How long from the time you get data to making a projection?

Completely depends, right? Where there’s huge spreads in races, we can, with our models and data, call races very quickly with the data we collect. Sometimes, we are able to call races at poll closing because we have exit polls in states, and sometimes we have split poll closes—think: Florida with two time zones—where we get the vote before the latest poll close time for the state. We never project a race until all the latest poll closing time for a state. In some instances, we’ve interviewed people doing in-person early voting, and we also send out [people to] do 85,000 in-person intercept interviews on Election Day. Depending upon the race, we use a combination of exit poll data and vote data at different levels of granularity, including state, congressional district, county, and precinct-level vote data, to make statistical models which we use for projection purposes. Some races will not be called on Election Day, because a lot of states have significant amounts of votes not counted on Election Day. This is a particularly big issue in the Western states. For example, people don’t understand how much of the vote is not counted on Election Day. If you’re in California, maybe 35 to 40 percent of the vote is not counted on election night. And people also don’t realize the election started a long time ago—we already have over 35 million people who have voted.

Can you tell from exit poll data where a race is going to go, in some instances?

An exit poll is still a poll and, therefore, it gives us an estimate of the vote breakouts in a state. It gives us guidance, but it’s still a poll—a high-quality, in-person exit poll of real voters, but in tight races you need to use a lot of actual data to see exactly where a race is heading. 

What gives us strength in our modeling efforts is that we have all these different types of data. We have the exit poll, which is the only thing that actually allows us to tell the full story of election night, because it tells us how different demographic groups like women voted or African Americans voted. It also tells us what they are thinking about politics and the election. It really is a rich source of data. It can help us answer questions like whether people are splitting their votes across parties or whether they are voting for candidates not from their party. For example, in Tennessee, if the Democrat is going to win, that Democrat—who happens to be a very popular former governor—will need to win some support from people who identify as Republicans because Tennessee is a Republican state. The exit poll helps us see if that is happening early on in the night.  

And Penn students contribute?

We train our students to understand and analyze this data. I have two undergraduate students who are Fox-PORES fellows who will be working with us on election night at NBC, Hannah Kanter and Sarah Lentz. We take students as Fox-PORES fellows and teach them the skills to be real contributors with us on election night at NBC News. It is the ultimate experiential learning opportunity for them. We have had Penn students with us on the Decision Desk at NBC for multiple election cycles. We plan on doing this well into the future. The truth of the matter is the Decision Desk is the effort of many people. We’ve brought in a lot of students, professors, and people in different circles to help in the effort; it’s not just a ‘me’ effort—I could never go it alone. I have to build the right team. And NBC has been very open to working with the people I know or who teach at Penn, and it’s been an invaluable opportunity for our students to see how the sausage is made on election night and be part of the whole thing.

Has there been anything unusual about this election cycle?

It has been an unusual cycle. Just think about it, the enthusiasm—potentially on both sides—is crazy. We most likely will have unprecedented turnout for a midterm election. We don’t know for sure, but this election is looking to be somewhere between a normal midterm and a presidential year where you have a lot more people vote. Aside from the high enthusiasm and interest in this election, so much has happened over the past few months that could be impactful. Just over the past month or so, we have had the Kavanaugh confirmation process, the continuation of the Mueller investigation, and a scary amount of extremist violence, among other things. We’ve seen so much going on that I think in this election cycle there’s a lot more uncertainty. It looks like the Democrats are the favorites for the House and the Republicans are the favorites for the Senate, but I really feel like there’s more variability in this election in that we could observe a number of different outcomes in this election and it would not surprise me.

Any new techniques this year, in terms of reading data, or anything new in your toolkit?

We spent a lot of time—and a lot of my Penn colleagues have worked with me—to improve upon the exit polls as well as our data collection efforts. We are perfectionists, and always know we can improve upon what we do. The work is constant. There’s a lot of statistical work to make that even better than it has been in the past. It’s always an ongoing process. We’ve done a lot of work on creating new statistical models that analyze the data, and we’re working on getting more granular data, to get data that goes all the way down to the precinct level. Because the more granular we have data, the better for us to do a good job in making our projections, and also for analyzing data.

What should people bear in mind as they’re watching results come in on Tuesday?

I think one of the things we learned from 2016 is people often have strong priors, or strong beliefs, on what they think is going to happen, and they should set those aside and actually see how the night is evolving. I think fairly early on we’ll see—in the House, for example, there are a lot of bellwether House districts that we’re going to see which way they break, and to be honest, I don’t think we know right now. I think there’s a difference between predicting elections and projecting elections: I project elections at NBC News. We call winners on election night and in the days after the election, in the case of extremely close races. There’s a lot of people out there predicting elections, and I think people often get overly fixated on those numbers. This is what happened in 2016.   

But once we get to election night, you can start seeing real data and see what’s actually happening fairly early on. For those that tune in, and of course they should only tune in to NBC or MSNBC, they will have a much better sense of the night. I think we’ll have a much better sense in the 8 p.m. or 9 p.m. hour on the direction of what’s going on in the House. I don’t know when we’ll project it or call control, but I feel we’ll have a very good sense of the night as we see what is happening in early-poll-close states like Virginia. One thing I am feeling fairly certain about is that we’re going to have a ton of close elections, especially in the House races as well as gubernatorial contests.   

This election feels, of all the elections I’ve done in the past, it just feels like there’s more close elections than I’ve ever seen before.