������Ƶ

Prior to the August primary we wrote about how we produce our polls. We discussed some unique challenges to conducting surveys in Hawaii, and we dissected a 2012 poll that was notably off the mark. As part of Civil Beat’s commitment to transparency, it is important to review our most recent polling results now that the concrete election results are known.

Before we get into this year’s polling, it’s a good time for a reminder about one common misconception about election polls: They are a great tool for predicting the outcomes, but they are not intended to be a prediction of the exact results.

Polls are a snapshot in time of public opinion and voter intentions, both of which are moving targets. Most elections have some last minute twists and turns, and every campaign has strengths and weaknesses in the critical final days that often materially affect the outcome.

This is particularly true in lower-turnout elections, such as primaries, where small changes in circumstances can have outsized effects on the results. And it’s fair to say that the recent primary election had a number of unpredictable elements in the closing days!

What we got right: We were pleased to correctly project the winners in all four races we polled. In the races for the Democratic nominations for lieutenant governor and for the First Congressional District seat, our poll results bucked conventional wisdom, and were correct.

Where we struggled: We overestimated the margin by which Brian Schatz would win the Democratic nomination for Senate. Although ours was and did , the narrowness of his victory was outside our margin of error. We did project that Colleen Hanabusa would outperform her polling numbers (as she has done consistently in the past), but not that she would bring the race within 1 percentage point.

And in the governor’s race, while we were the first to detect David Ige’s surprising lead, which we confirmed in two subsequent surveys, our snapshot in time did not capture the magnitude of his surge. We had reason to believe that his win would exceed 20 percentage points (more on that later) but his 35 percent margin shocked just about everyone.

Race by Race

Now, let’s look at all four races in more detail:

In the lieutenant governor’s race, we showed incumbent Shan Tsutsui holding a 45 percent to 35 percent lead over his better-known challenger, state Sen. Clayton Hee, with 20 percent undecided. Undecided voters typically break proportionately to the percentages of those expressing a preference (particularly in a low-turnout, down-ballot matchup, where many who say they are undecided do not end up voting at all). Given this, our poll projected a 13 percent win for Tsutsui. The actual result, Tsutsui by 16.6 percent, was within our poll’s margin of error.

In the First Congressional District primary, we found Mark Takai surging and drawing away from Donna Mercado Kim. He led 30 percent to 23 percent with 33 percent split between all of the other candidates and 14 percent undecided. We noted that support for trailing candidates almost always decreases as voters like to gravitate toward the winner. Predictably, Takai benefited the most and ended up with a 15 percent victory.

This race is an example of the difference between a snapshot in time and a prediction of results. Takai’s win did exceed our margin of error. At the same time, the result mirrored what the poll suggested: Takai held a lead that was projected to increase at the expense of the trailing candidates.

In the race for U. S. Senate, there were from several different outlets in the final weeks, ranging from one poll showing that Schatz led by 15 percentage points, to another poll that found Hanabusa leading by 8 points. In our polls of the Senate primary, we generally found a modest but consistent lead for Schatz, ranging from a tie in February to his 8 percent lead that we found last month.

We did identify a potential vulnerability for Schatz in our crosstabs. There were strong correlations between Hanabusa and Ige supporters, and between Schatz and Abercrombie supporters. For instance, Hanabusa was leading among Ige supporters by a margin of more than 20 percent, and Schatz was doing even better among Abercrombie supporters. Ige’s lopsided victory over Abercrombie in the governor’s race may have played a role in increasing Hanabusa’s vote total while suppressing Schatz’s.

Some have suggested that we had . As we noted previously, our voter-list provider offers a solid estimate of the percentage of Caucasian voters who participate in Hawaii’s primaries (just under 40 percent in 2010 and 2012, and the Caucasian percentage has continued to increase during the last decade).

When the report of who voted in this month’s primary is made available later this year, we will do an analysis of the demographic breakdowns to see if that holds an answer. It is possible that Caucasian voting in the Democratic primary was suppressed to some degree. We know from previously unreleased crosstabs from our final poll that Americans of Japanese Ancestry (AJAs) participated in early voting to a greater degree than did Caucasians. It would not be surprising to find that some of the Caucasian voters who intended to vote on election day stayed home, whether due to the significant weather issues, the steady stream of reports that Gov. Abercrombie was almost certainly going to lose, or a combination of the two.

Finally, in the governor’s race, our polling was the first to identify the looming political upset, and we were able to track David Ige’s evolution from a little known candidate who was known mostly as “Not Abercrombie” to a candidate who was able to double his name recognition and favorability in just a few short months. By our final poll, Ige’s double-digit lead had us confident that he was about to win, especially because of unreleased results among early voters.

In Hawaii a majority of people now vote by mail, and 25 percent of the sample in our final poll consisted of people who had already cast their ballots. Among that group, Ige had an eye-popping 25 percentage-point advantage (and even that turned out to be lower than his actual 30-plus point margin among early voters).

The question of whether to report early voting crosstabs is a gray area and a tough judgment call for any news organization. We have decided not to do it, because it is too similar to releasing exit poll results before voting has concluded — .

Regardless, what these results showed is that we were relatively close to the mark regarding Ige’s strength among early voters, and far from the mark regarding Abercrombie’s competitiveness among Saturday (Aug. 9) voters. That suggests a campaign in free fall.

Our conclusion is that our snapshot in time somewhat underestimated Ige’s strength (finding a 25 point lead among early voters that was actually over 30 points), while Ige’s continued surge and Abercrombie’s struggles accounted for the rest of Ige’s margin.

One final note: while early voters made up more than 50 percent of the electorate, our survey was comprised of 25 percent of voters who had already voted. That is because the survey was conducted about two weeks prior to the election and many early voters had yet to return their ballots.