Tulsa mayor's race: A tale of two polls

The first thing to say about any poll in the Tulsa mayor's race is that it doesn't matter. It shouldn't affect how you think about your own vote. Because there is a runoff in November, you can vote for your favorite of the top three candidates without worrying that you'll inadvertently help your least-favorite candidate win. Read my earlier entry on the Tulsa non-partisan election and runoff process for an in-depth understanding of why this is.

SoonerPoll.com, the firm whose pre-Vision2 poll incorrectly showed both propositions winning, has published a poll showing Kathy Taylor and Dewey Bartlett in a near tie and Bill Christiansen trailing in advance of Tuesday's mayor of Tulsa primary. The "top-line" weighted numbers from SoonerPoll.com have Taylor at 36.4%, Bartlett at 34.6%, Christiansen at 20.1%, Branch at 0.1%, Kirkpatrick at 0.0%, and 8.7% don't know/refused. The percentages add up to 99.9%, but the raw numbers for their sample of 400 add up to 401.

Meanwhile, the Bill Christiansen campaign has posted the results of their internal poll, which shows Taylor at 34%, Christiansen at 31%, Bartlett at 24%, and 11% undecided. The sample size was 468, and a voter was called only if they are high-propensity voters (4 or 5 times out of the last five elections).

Back before the Vision2 vote, I published a critique of SoonerPoll.com's methodology, particularly the age of the poll, their use of random-digit dialing and their weighting of subsamples.

As with Vision2, some of the SoonerPoll.com sample is over a week old and predates several debates and forums. They are weighting even more subsamples in this poll than they did for Vision2.

This poll has additional problems. (To their credit, SoonerPoll.com has published the questions they asked.)

First, note how they screen for likely voters. They simply ask:

On June 11, Tulsa residents will vote in a nonpartisan primary election for Tulsa mayor. Do you plan to vote in this election?

Is the person on the other end of the phone even registered to vote? Does he know where to vote? Was she even aware that there was an election before she picked up the phone? Younger voters tend to overestimate their likelihood of voting; older voters tend to underestimate. I suspect that their sample weighting by age is intended to correct for that.

The second problem: In the "ballot test" question, they don't simply read the names as they will appear on the ballot. They add a description for each candidate.

If the election were held today and you were standing in the voting booth right now and had to make a choice, for whom would you vote? [READ & ROTATE] 1. Dewey Bartlett, Jr., the current mayor 2. Jerry Branch, a pipefitter/welder 3. Bill Christiansen, former city councilor 4. Lawrence Kirkpatrick, a church volunteer 5. Kathy Taylor, the former mayor 6. DK/Refused [DNR] [SKIP Q5]

In the voting booth, it's going to be up to the voter to remember which candidate is which. Here's 2013 City of Tulsa sample ballot. Just names, no party affiliation, no description. I think voters would be well-served if candidates could add a few words of description or abbreviations for endorsing organizations, but right now that's not allowed.

(Oddly, the ballot header says "INDEPENDENT MUNICIPAL OFFICERS." Independent of whom? Shouldn't it simply say NON-PARTISAN, as it does for judicial and school offices?)

The methodology summary says, "Results were weighted by gender, age and party." The poll has six different age brackets, two sexes (sex is a biological term, gender is a grammatical term), and three party affiliations. That implies 6 x 2 x 3 = 36 subsamples. The best case is that each subsample would have 11 respondents, each with a margin of error of 29.55%.

And this is puzzling: They say the results were weighted, and yet magically every number or responses is a whole number. If you're weighting a result, you're multiplying the actual number of responses, which would be a whole number, by some weighting factor, which probably won't be, you're more likely than not to get a fractional number. My suspicion is that the percentages accurately represent (to one decimal place) the weighted result, which was then multiplied by 400 and rounded to get the weighted number of responses. On several questions, the number of responses adds up to 401, due no doubt to cumulative rounding errors.

Let me repeat what I said in reaction to SoonerPoll.com's Vision2 poll:

It may be that all these flaws cancel each other out, and I don't mean to cast blame on SoonerPoll.com, which is no doubt doing its best to gauge public sentiment in an increasingly difficult environment. We'll find out on Tuesday.

Why does it matter? Poll results can be used to create a bandwagon effect, particularly when an issue isn't strongly partisan. Without any strong sense of what to do, some voters will go along with whichever side they see in the majority.

The antidote to a poll-driven bandwagon is to create your own bandwagon -- put signs in your yard, post to Facebook and Twitter, email and phone your friends, and let them know which candidates you're supporting.