Articles and Analysis


Joe Lenski on the Iowa Entrance Poll

Topics: 2008 , Exit Polls , The 2008 Race

Last week, I had the chance to briefly interview Joe Lenski, the executive vice-president of Edison Research, who will conduct exit polling in 2008 on behalf of the network consortium known as the National Election Pool. The following is a transcript of the interview.

Q. I want start by asking about the entrance poll that your company will be providing for the National Election Pool in the Iowa caucuses January third. What is an entrance poll and how does it differ from the exit polls that we've heard so much about?

Lenski: An entrance poll is similar to an exit poll in that we have interviewers that go to a random sample of precincts throughout the state and ask attendees to fill out these questionnaires. The only difference is we're asking caucus attendees to fill out these questionnaires as they are entering the caucus as opposed to a normal primary election or general election where our interviewers approach the voters and ask them to fill out the questionnaires after they have voted.

Q. What do you provide the networks with?

Lenski: We provide the networks with an estimate which updates very rapidly during an entrance poll because we're doing all of the interviews in a 60 to 90 minute period of time, as opposed to an exit poll where you're interviewing throughout the election day and in most states that's somewhere between 12 and 15 hours of voting. We only get the chance, in a caucus situation, to interview people as they're entering and that's usually 60 to 90 minutes before the caucus begins. So, these results get updated very quickly throughout the evening as interviews must be completed before the caucus begins. What we provide for an entrance poll is similar to what we provide our clients and subscribers for an exit poll on an election night and that is an estimate of how the voting is going, demographic detail of who's voting for whom, what voting groups are there, what their sizes are and what factors seem to be driving the vote that evening.

Q. Do you do anything in the questionnaire or in the analysis to try to anticipate the reallocation that occurs among Democrats when a candidate does not meet the 15% threshold within any given caucus location?

Lenski: Yes, there will be a question on the Democratic Iowa entrance poll asking "If your first choice does not gather enough supporters, who will be your second choice?" We can use that and we did use that in 2004. We did see that indeed a majority of the Kucinich voters chose Edwards as their second choice, matching the deal that the Edwards and Kucinich campaigns came up with the morning of the Iowa caucus in 2004 to have Kucinich voters switch to Edwards in precincts where the Kucinich group wasn't viable.

Q. Do you do that analysis as the aggregate for the whole state, or do you try to reallocate within each precinct?

Lenski: No, we only have sample sizes to do it for the whole state. Remember there are 1,781 distinct precinct caucuses. Actually, as an aside, some of the blogs say 3,562 -- that's combined Democrat and Republican caucus sites. Each party has 1,781 precinct caucus sites, and the dynamics within each one of those is very different and just because a candidate is receiving 15% statewide doesn't mean they are meeting that viability threshold on the Democratic side in each precinct. In fact even though John Kerry won in 2004, he didn't even reach viability in about ten percent of the precinct caucus locations. So, even the leading candidate may not reach viability in every single precinct.

Q. How many precincts are you sampling for each party?

Lenski: For each party we will be covering 40 precincts with the entrance poll. And at a larger sample, we will also have reporters there recording the initial preference so we have a statewide estimate of initial preference as well.

Q. I assume this a question you get a lot, but I've seen some commentary about it in the blogosphere recently. Why is it okay to sample only 40 precincts out of about 1,000?

Lenski: It is about the same sample size we use in most statewide exit polls. We will be interviewing over 1,000 actual caucus voters on each side, Democratic and Republican, and that's much larger than any of the pre-election likely caucus samples in the pre-election polls. We'll have over 1,000 actual Democratic caucus goers and over 1,000 actual Republican caucus goers in our entrance poll samples.

Q. I assume that in past years you've done the exercise where you've gone back and looked at the actual results from those sample precincts and asked the questions "What would you have done differently? Did the actual results predict the statewide results to assess whether that sample is appropriate?"

Lenski: That's easier to do on the Republican side because it is a straight straw poll and you get an official result: one person one vote. On the Democratic side it's a little more difficult because the final official result is the state delegate equivalence number that the Iowa state Democratic Party calculates on election night. So, it's not really an apples-to-apples comparison on the Democratic side. But we do look within our sample precincts at what responses we get on the entrance poll and what the initial preference breakdown was in those particular precincts because we have our vote reporters stay for the initial preference and record the initial preference breakdown in those sample precincts.

Q. And that's part of what gets reported?

Lenski: That's used in our estimate on election night for our statewide estimate of initial preference on the Democratic side.

Q. Last question, I've written a bit about this issue of whether there's a non-response correction possible with an entrance poll analogous to what you do for the exit poll? Could you explain that?

Lenski: In all exit polling, we do a non-response adjustment in which our interviewers record the gender, the race and the approximate age of those who refuse to fill out the questionnaire. We do find generally that younger people are a little more likely to fill out questionnaires than older people so this non-response adjustment corrects for that tendency. In past entrance polls -- and I've been involved in every Iowa entrance poll since 1988. I believe the first one was conducted by CBS in 1984 -- we have not done a non-response adjustment mainly because of the crunch of time. It's an added task for the interviewers to do, and in the crunch of time we try to get as many questionnaires completed as possible. But this year, pre-caucus telephone surveys show a difference in how, at least on the Democratic side, people are planning to vote by age. Younger people are more likely to support Obama, older Democrats more likely to support Clinton. We have decided this time to add a non-response adjustment to the entrance poll as well.

Q. So is that going to require additional people looking over the shoulder of the interviewer, or are they just going to have a little more to do?

Lenski: They're just going to have a little more to do. We do have two interviewers at each precinct for the entrance polls, again because of the crush of the number of interviews to be conducted in a short period of time. For our typical exit poll, there's only one interviewer all day. But for the entrance polls we always have two, so this will just be another task for the two of them to divide.

[Editor's Note: Andrew Kohut of the Pew Research Center did a longer interview with Lenski recently that went into more detail about the exit polls planned for the February 5 primaries and the rest of the year].


Mark Lindeman:

And they're still doing all this by phone (see Kohut). Amazing.

One thing to flag: when Lenski answers why it is "okay" to sample only 40 precincts, the question is, okay for what? If the networks had to "call" Iowa before the caucuses even voted, then 40 precincts probably wouldn't be nearly enough. But to tell a decent story about, say, how heavily young voters turned out and whom they supported, it should be OK. (For that particular purpose, it's really important for them to tally non-responses by age!)



Wouldn't filling out an entrance questionnaire take a certain amount of time, meaning that only the people who show up early are included? I'm not sure if that'll make a difference, but first-time caucus goers might be worried about being late.


Mark Lindeman:

ryan, that's an interesting question. Yes, it seems possible that first-time caucus-goers might be less willing to participate in an entrance poll.

If so, then to the extent that first-time caucus-goers are young voters, the non-response correction may pretty much fix it. Given the total sample size, it will be really hard to tell.


all the entrance/exit polls do is try to manipulate voters into thinking their candidate is too far behind to catch up. The major flaw i see in this system is that we don't know which precincts they are polling. Will they intentionally avoid highly populated evangelical precincts so it appears as though romney is running away with it. I really believe this will happen because romney is apart of the establishment and huckabee is an outsider.


all the entrance/exit polls do is try to manipulate voters into thinking their candidate is too far behind to catch up. The major flaw i see in this system is that we don't know which precincts they are polling. Will they intentionally avoid highly populated evangelical precincts so it appears as though romney is running away with it. I really believe this will happen because romney is apart of the establishment and huckabee is an outsider.


Randy L:

Answering Mark L, it's "okay" statistically. And an increase of sample size -- even 5 times the sample size -- would result in very little decrease in the potential error range.


Post a comment

Please be patient while your comment posts - sometimes it takes a minute or two. To check your comment, please wait 60 seconds and click your browser's refresh button. Note that comments with three or more hyperlinks will be held for approval.