Lenski on Exit Polls in the Coming Elections
In an exclusive interview, Joe Lenski, co-founder and Executive Vice President of Edison Media Research discusses the November midterm exit polls. He reflects on conducting his first election day survey following the death of his former partner, exit poll pioneer Warren Mitofsky. He also reveals the steps that he and his colleagues have taken to avoid problems associated with the 2004 poll. Lenski is an expert in the operation and organization of survey research and has been involved in every major exit poll conducted in the last decade for the television networks. Under his supervision, in partnership with Mitofsky International, Edison Media Research currently conducts all exit polls and election projections for the six major news organizations — ABC, CBS, CNN, Fox, NBC and the Associated Press.
Interviewer: Andrew Kohut, President, Pew Research Center
It’s hard for me, and just about everyone to imagine an exit poll without Warren Mitofsky [creator of the exit poll at CBS in 1967, who died on Sept. 1, 2006]. But it has to be especially hard for you, who worked with him so much. I wonder if you could start off by telling us about an incident or a moment that stands out in your memory as the quintessential Warren Mitofsky experience at doing exit polls.
Well, there were several good stories told in the memorial service on September 5th. I thought the one Murray Edelman told showed Warren’s sense of the importance of things. The 1992 election was the first presidential election in which just one consortium was doing these polls for all the networks and the Associated Press. A little after 10:30 pm, I believe, it was time to call Ohio for Clinton, which would give Clinton the 270 electoral votes needed to win. All the people in the decision room were looking at Warren; Warren told everyone to be quiet. He turned around and pointed to the TV screens because something momentous was going to happen. Then Warren entered the call for Clinton in Ohio and we just stopped and watched the TV screens as ABC, CBS, CNN, NBC popped up with the next president of the United States, Bill Clinton. And that sense of the importance of what we do and how many people share that information — Warren had a good sense for that.
I was thinking about a comparable moment perhaps in this election when you might be forecasting or declaring that the House or the Senate or both had changed hands. Does it give you the willies a little bit to be in that situation?
I think we’re going to have a really good team of people together on Election Day. A lot of people with experience, from before, and new people we’re bringing in. But what I’m going to really miss is that final debate. When Warren and I looked at each other and before we finally pulled the trigger, we looked at each other and said, ok is there anything else we are neglecting to look at here before we make that call? And sometimes it would be me pushing for the call and Warren holding back and sometimes it would be the other way around with Warren pushing for the call and me holding back. But it was that final debate when we could argue with each other and then come up with the right decision. That, I’m going to miss.
Are you going to be bringing in someone else to play the role of methodologist? A statistician ?
Yes, we’re going to have political people, we’ll have statisticians; we’ll have people who are familiar with the computer systems that we’re using — people who have been involved with the survey process, and the survey weighting. Since 2000, the desire for the consortium to make calls quickly has lessened. Frankly, the real desire, since 2000, is to make sure the data is correct first… absolutely correct. And that the actual calls, if they’re 10 or 15 minutes later than they’ve been in the past I don’t think anyone is complaining. I think our main task is to make sure the data is being delivered and is accurate to the best of our abilities. Each of the members at the networks and the Associated Press have their own experienced teams that have been looking at this data for years, and if we are a little slower then they are, they’re not going to mind that.
What’s your biggest challenge going to be in this particular midterm election?
Well, there are a lot of interesting things here…one is the relatively small number of races that is going to determine control of both the House and the Senate. The Senate is basically down to 6 or 8 seats that will determine whether the Democrats get the 51 seats they need for control or whether the Republicans hold on with 50-plus with Dick Cheney as the deciding vote. So even national trends aren’t going to be as important as state-by-state trends particularly in those states that are going to decide the Senate. Similarly in the House, you’ve seen in most of the national polls that the Democrats have a sizable lead in what’s called the “generic house vote,” but you can win the popular vote for the House and still not win 218 congressional districts, depending on how those votes are proportioned out among districts. As we learned in 2000, you can win the popular vote but not win the Electoral College. So, I think what people — including those of us on the decision teams as well as lay people looking at the results — have to be careful to remember that just because an exit poll shows more people voting for Democratic House candidates than Republican House candidates, that doesn’t mean that, when you look district by district, the Democrats get or don’t get 218 seats. That’s going to be one of the trickiest things on election night.
In the interview that we did, right after the Mexican presidential election, I asked Warren [Mitofsky] why he thought he did so well in calling Mexico. He said it was because they were more rigorous in the way they had the interviewers sample people coming out of the voting booths. Faulty sampling was a key to the bias that was apparent in the exit poll in 2004. What are you doing to address the issue of more rigorous sampling of the individual voter?
Yes, the evidence did support Warren’s contention in that regard. At the least, it showed that when the interviewing rate was less — when you were interviewing 1 out of every 10 people rather than 1 out of 2 people — we tended to have a higher survey bias in 2004. Failure of the interviewer to strictly adhere to the sample rate is probably the major cause of that. Since then we have reevaluated the entire hiring and training process of our exit poll interviewers. In conjunction with the NEP members we have gone through and put together new recruiting guidelines, new training guidelines, a new training video, a new training call, a kind of quiz to make sure that after the training the interviewers retain the most important pieces of knowledge with interviewing rates being one of those most important pieces of knowledge. So, I think we have done a lot of revamping of the training to make sure that the interviewing rate or the sampling rate doesn’t get lost in the training procedure for the individual interviewers and that they know — with a lot of emphasis — that it’s one of the most important factors in gathering correct and accurate exit poll data.
How many different interviewers do you think you will be using this election day?
In 2006, we’ll be using around 1,000, give or take.
1000 interviewers…that’s quite a lot…
And in addition to that, there will be several hundred back up interviewers in case something happens, so that if interviewers get sick or have an emergency, they can be replaced. We’ll also have several dozen supervisors who will be roaming around and making sure that people are in the right place and not having problems, etc. So not only are there the interviewers, but there are also the back up interviewers and the supervisors, and they need to be trained as well.
How will they transmit the data?
All the information is read over the phone to multiple input centers we have set up around the country
But it’s being read manually, it’s not being transmitted electronically
No, It’s read manually. Also, after the election, we’ll get back all the paper questionnaires to compare them and make sure that all was done accurately. But on Election Day itself, all that information is transferred over the phone. We also have several hundred phone operators who need to be trained.
Are you doing special samplings in congressional districts that you think are competitive?
We are only doing statewide samples. So the only House races that we will be covering are the handful of states that have a single congressional district.
Warren spoke, and others have spoken, about the efforts that are being made to reduce the chances of a leak…
Well, I’ll tell you a brief description of what we will be doing with the networks and the Associated Press. We’re going to put in place systems in which no one, even at the networks, can view any of this data before 5 p.m. on Election Day. This will be very similar to procedures used in England and Mexico and other places to strictly control the dissemination of data, and the number of people who get to look at it before 5 o’clock. We’ll have one or two rooms in which people will,in essence, be in a bubble, quarantined. They’ll have to give up their cell phones, their pagers, any internet activity, anything of that sort and stay in that room until 5 o’ clock in order to view the data so we know there is no possibility of communication with the outside world.
I’m going to ask you another question about a potential problem that occurred to me…Are you printing the names of the networks on the questionnaires, as you did in ’04 and every other election?
Is there any fall out with respect to subscribers this year? I know all of the networks and AP are participating, but what about newspaper subscribers?
Again, I wasn’t part of VNS in 2002, the last off-year election. The subscriber rate is less than in a presidential year, but that’s to be expected. So we have several dozen clients this year and we seem to be on pace to have the number of subscribers we were expecting.
I know what you did in ‘04, but maybe you could describe it to us for ’06: What are you doing to accommodate changes in voting procedure, such as early voting and relaxed rules?
Well, as we did in 2004, in all the major states with significant percentages of absentee or early voting, we will be doing telephone surveys a week before the election to speak directly to people who have already voted, or already have a ballot and plan to vote before Election Day. Then we incorporate those surveys with the Election Day interviews done at the polling place, and merge those together in proportion to the expected percentage that will vote earlier.
Will there be Spanish language versions of the exit poll available?
Yes, in precincts that have been identified with significant Hispanic populations.
What about Asian languages such as Korean?
The only language other than English will be the Spanish language. Once you get to Asian, there are so many different languages you have to consider and in the polling places that we’ve sampled, you never really know, until you start doing the research on the sample, what community was in those locations. Ideally you would, but in terms of the production of an exit poll, you can’t really do that.
A few final questions. How did you meet Warren?
It was when I walked out of college and went to work at CBS in 1987. Warren didn’t interview me, but Murray Edelman, his deputy did, and I was hired to be a statistical associate in the CBS news election survey unit…and that’s how I met Warren, and that’s how I got to work both on the exit polling and on the CBS News/New York Times polls.
And when Edison Research is not doing exit polls, what’s its specialty?
We do a lot of research for media properties, a lot of research for radio, a lot of research on music, both in the United States and around the world. We also have taken our exit polling methodology and started using it more and more in commercial settings, whether it’s baseball stadiums, concert halls, theaters, shopping malls, school buses, subway stations, basically anywhere there is a flow of traffic and people being exposed to advertising and other stimulus at those locations.
How many people work at Edison?
That’s an interesting question, because it really varies depending on the type of year, whether it’s an election year or not, or who you consider a part of Edison. We have people who work for Edison here in New Jersey who are full time employees, we also have people who work for Mitofsky International in New York City who are part of the election team. We also have programmers in Yorktown Heights, N.Y., who are working as part of this team, though they are not on the Edison or Mitofsky payroll. And then you have the people who work for one day around the country. So, my technical full-time payroll is around 25 people, but on Election Day this year, we will have more than 3,000 people working for us on that single day.
That’s quite an operation. So now you find yourself stepping into the shoes of someone who was the father of exit polling and to you, Joe, I wish the best of luck.
We have a good team that Warren put in place and a good system that he designed and I think a lot of lessons that he taught us. We’re going to do our best to do him proud on November 7th.