What I Learned About Polling From a Georgia House Survey

by Nathan L. Gonzales March 27, 2017 · 10:54 AM EDT

I was initially skeptical of a recent poll in the special election for Georgia’s 6th District, not because it utilized Interactive Voice Response, or IVR, technology or because it was conducted by a GOP-friendly firm or because a Democratic candidate was leading in a Republican-leaning district. But it only gave respondents the option to choose from less than half of the candidates, proving the limits of automated polling, or so I thought.

The March 15-16 automated survey conducted by Clout Research for zpolitics showed Democrat Jon Ossoff leading with 41 percent followed by two Republicans: former Georgia Secretary of State Karen Handel and wealthy businessman Bob Gray, who had 16 percent each. Former state Sen. Judson Hill and three other Republicans combined for nearly 17 percent while former Democratic state Sen. Ron Slotin received 3 percent.

The problem is that the survey of 625 respondents included just eight of the 18 candidates running to replace former GOP Rep. Tom Price, who left the House to become President Donald Trump’s secretary of Health and Human Services. All candidates run together in the April primary, and the top two advance to a June 20 runoff if no candidate receives a majority in the initial balloting.

There are plenty of pitfalls when it comes to polling including cost, low response rates, etc., but presenting respondents with an accurate reflection of the ballot isn’t supposed to be one of them.

“We feel strongly that replicating the ballot, in form and function is critical to getting an accurate read on the outcome,” said Republican pollster Adam Probolsky of Probolsky Research. “That includes the ballot order (or rotational scheme if there is one), any other information such as party (and how that is presented), notation of incumbency or other information.”

“It takes effort to dial in these details for each client, but without that next-level of inquiry about a ballot for a specific race, the data can be skewed minimally or worse,” Probolsky added.

In the not-too-distant past, IVR pollsters had to be creative with how to handle elections with more than 10 candidates, since respondents were limited to pressing one of the digits on their phone.

“When the Republican primary for president had 17 at least somewhat serious candidates, we just set it up as two questions,” explained Tom Jensen of Public Policy Polling, a Democratic firm which helped mainstream IVR polling. “We listed off all the candidates, then let people press 1 to 8 for the first 8, or press 9 if they wanted one of the other candidates, and if they pressed 9 then we read them the options for the rest of the candidates. Doing it that way worked out decently well but it was still more cumbersome than if we could have squeezed it into one question.”

“If it’s easy to narrow it down to 9 or less serious candidates, we do take that approach,” Jensen continued. “When that comes up, it’s usually on a poll that we’re doing for a private client, and they generally tend to have a pretty good sense of who the serious candidates are and who’s going to end up with 1 percent or so.”

And that’s what happened in this case in Georgia.

Technology was not a limitation in the 6th District poll, according to Clout Research partner Fritz Wenzel. The client, zpolitics, chose which candidates should be included, something Wenzel called a “wise choice.”

But that decision could still have an effect on the results. Even though the 10 candidates omitted from the poll aren’t likely to win, each percentage point they garner makes it more difficult for another candidate to reach a majority in the initial balloting. Even if the three omitted Democrats receive 1 percent each, that would likely put Ossoff below 40 percent and further out of reach of a majority.

Through diving into this survey, I was surprised to learn that automated polling can now accommodate large candidate fields.

“[IVR] technology is getting so good these days, there are very few limitations,” said Wenzel.

The firm can increase the amount of time given to register a response to allow for more digits to be entered, when previously the program would only accept the first digit, or it can allow for respondents to speak the name of their candidate of choice and the program will translate that response into a numeric value to be tabulated on the back end.

Ossoff recently told The Associated Press that “the goal is to win outright” in April. While that scenario still seems unlikely in a district with a Democratic performance close to 38 percent, Ossoff has a path to shock the world by continuing to consolidate the Democratic base and get 10 percent of GOP voters, as he does in the Clout poll according to Wenzel.