While taking marketing classes at the University of Hawaii, Rebecca Ward recalls being “fascinated with the idea of trying to understand human behavior and describe it, which is what polling is.”
Thirty-two years into her venture as a pollster now located in downtown Honolulu, her Ward Research is known as a leader in market research for private sector companies and nonprofits, conducting polls for Hawaii’s Top 250 corporations and in political races.
The company’s political polls are taken for the news media, including the Star-Advertiser, which she said “precludes us from working with major candidates or parties.” Those political polls, while the most well-known among the public, comprise less than 5 percent of the company’s work, which mainly involves research for private companies and organizations.
Ward emphasizes that polls produce public opinions at the time they are taken, not predictions of what will happen at polling booths or with consumer activity.
“If we really thought we could predict the outcome,” she said, “likely I wouldn’t be in polling. I’d be in the stock market.”
QUESTION: Is it important who does the polling, who is paying for it?
ANSWER: It can make a difference. It doesn’t have to make a difference, but it can. Candidates can structure the questions in a poll such that it would lead people in a certain direction. So where we start right off the bat is asking people if the election were held today, who would you vote for. Candidates may start with other questions that could tend to lead people a certain direction. For example: “From your perspective, is the country on the right track or the wrong track?” And so if someone is saying wrong track, it starts to build some thought processes. So assuming that they’re working with a reputable pollster, and assuming that all the scientific part of the sampling and all is in line, there are still some things that can change the outcome of a poll. So candidates, of course, only release polls if it’s advantageous to them, so you have to wonder how it became advantageous.
Q: What’s the process for determining who is included in a political poll?
A: Polls can only either work with a voters’ sample that’s purchased from companies that maintain, buy the information from the government and will resell it for campaign purposes, so we can work with the voters sample or we can work with a random sample from the population. But what’s important is that we screen first for registered voters and, as it gets closer to election time, for likely voters. So the question that we use is how likely are you to vote in the November elections — very, somewhat, not very, not at all — and we only take those who say they’re very likely to do so.
Q: Those who are chosen in a telephone poll, is their party affiliation asked?
A: The way we ask and the way we think is appropriate in Hawaii is to ask what party they usually find themselves voting with. In Hawaii we have open primaries, and we don’t have to register with a party.
Party identification is not quite so strong in Hawaii as it is in a lot of other places where you have to register. We’ll get a certain proportion telling us that they’re independent, that they don’t necessarily find themselves voting with either party most often. But generally, when you look at how the vote tends to fall, generally in the high 50s or 60 percent in a general election will vote for the Democrat candidate in many of the elections, so we’re looking for about 60 percent to say that they usually vote Democrat. However, we don’t use that as a screening criteria. We don’t predetermine that there should be X number of Democrats and X number of people who usually vote Democrat or who usually vote Republican in the poll. We control for age and ethnicity here in Hawaii, and let the party that they usually vote for fall out of that.
Q: How many people need to be interviewed for a poll to be valid?
A: Generally, the industry standard is 400, and that limits sampling error to plus or minus about 5 percent. We will include more people in a poll when we have to start looking at subsamples. So, for example, we might have 750 in a poll statewide, but that’s because we want to be able to look separately at Congressional Districts 1 and 2, so that splits the sample.
Q: How is the margin of error determined?
A: Margin of error is based on laws of probability and statistical formulas, so that with any population greater than, say, about 3,000 people, the sampling error is calculated the same way, whether it’s a million people in the population or a hundred thousand or a hundred million. … One thing people don’t always understand about margin of error, it’s plus or minus. So if there are two candidates in the race, the one on top could actually be minus, and the one on bottom could actually be plus, so it’s a bigger range than people really understand.
Q: In Hawaii political polls, how does ethnicity come into play?
A: We’ve talked about it a lot over time. It’s probably less a factor than it used to be, with younger people coming into the voter population, but at the same time it still plays a role, in that in the Democratic Party there’s a stronger representation of Japanese, Filipino, and there’s always been a tendency for candidates of one ethnicity to hope to draw those from their same background.
Q: Are people of some ethnicities more reluctant to participate than others?
A: They are. If we took whoever answered the phone and we didn’t control for ethnicity, we would have a sample that’s predominantly Caucasian.
Q: Why is that, do you think?
A: Because we’re more willing to speak. We (Caucasians) are often more willing to share opinions, and a little bit more forthcoming than are some Asian populations.
Q: Has that changed over the years?
A: I don’t know that that’s changed. No.
Q: One thing that’s been recent is cellphones. How do you handle that?
A: We purchase cellphone samples from sampling houses on the mainland, and I assume they buy the numbers from the cellphone companies. There are companies that sell samples, they sell cell samples, a list of phone numbers. … They purchase them and resell. What we don’t know is any name associated with it. We just get a list of phone numbers. Just recently they’ve improved those lists. It used to be that we would just buy area code 808, and if we needed a survey of people on Maui only, we had to screen through a lot of people. But now they’ve added billing centers, so we can buy Kahului billing center for a cellphone sample, and buy cellphone numbers on Maui.
Q: When did you start doing that?
A: Cellphone sampling probably became more commonplace I’d say two years ago, something like that.
Q: Are there a certain percentage of your calls to cellphones?
A: It’s difficult to know exactly what proportion of households in Hawaii or anywhere have a land line; that’s data that the phone company holds onto fairly closely. Estimates are somewhere around 35 percent without a land line. It’s very difficult to sample and find cellphone-only households; it’s a costly proposition. So generally what we do is we’ll split samples, 50-50 land line and cell, because cell samples also produce more younger people than we get on the land line. Having said that, reaching likely voters, the cell sample is important but because the likely voters skew older, we don’t need to rely on the cell sample quite as much.
Q: Have there been other changes over recent years in the way you operate?
A: Other challenges, the proliferation of answering machines and caller ID. That certainly impacts our completion rates. And I’ve have to say that in the past six months or so, with all of the political polling going on and all the calling that is coming in from the mainland, it’s more difficult to find cooperative people than it used to be.
Q: There are some media organizations nationally that will decide that some polls are not reliable and others are. What are those criteria?
A: These days, the use of a mixed mode, land line and cell. These organizations are more and more circumspect about land line only calling. They generally want to see all the wording for all the questions, and I think that’s something that people should expect when they’re reading poll data, and I think that’s something that reporting organizations should require, and that’s that transparency. Because the questions that are asked prior to a question that the campaign is releasing can influence the later data in the survey.
Q: Your polls showed Mufi Hannemann holding a 65 percent lead to Tulsi Gabbard’s 20 percent in February and 43 percent to her 33 percent less than a month before the August primary in the U.S. House 2nd District race, but Gabbard won with 55 percent over his 34 percent. Were you surprised?
A: I was shocked by the margin, and I trust you’ll also point out how accurate we were on the others, on the mayor and the Senate race. It’s appropriate here to point out that polls are snapshots in time.
So one month out, do I believe that reflection was accurate? I do. From the very same poll, we had data that was on the money for the mayor’s race and the Democratic primary for Senate, so these all came from the same poll. I believe that nothing happened to change the outcome of the mayor’s race and the Senate primary, but a lot happened to change the outcome of Congressional
District 2. I really believe that the electorate in CD2 had to feel comfortable with Tulsi Gabbard, and I think over time that was building. Now when people asked me around the time of our poll in July, I said I think Tulsi could win an election in November, meaning that I thought it was going to take her months to continue with that momentum. Clearly, she had a tipping point for someone to move ahead that much. And we saw money coming into her campaign in the last couple months. I really believe that what happened was that people started to feel that she really was a viable alternative to Mufi, believing earlier that they only had one choice, and as they started to hear more and see more, that they felt comfortable and she had a tipping point and just took off. … In political polling, we’re not trying to predict an outcome. We really are reflecting thinking at the time, and so when I talk about us being accurate, I believe we’re always accurate at the time of the poll. In a lot of races, nothing changes. In the U.S. Senate primary, we showed that data exactly the same in February and again in July, and that’s what happened in August at the polls. So nothing happened to change those races. I think there was a tremendous amount that changed that race in CD2.
