In the hotly contested Kansas Republican primary elections this year, polls generated a lot of interest. In two Kansas Congressional districts, independent polls did a good job of predicting the vote for all candidates except the two winners, and a candidate’s own poll may have been undermined by large voter turnout.
In a KWCH/SurveyUSA poll of the Kansas first Congressional district, the poll accurately (within the margin of sampling error) predicted the outcomes for all candidates except for victor Tim Huelskamp. The survey predicted 24 percent of the vote for him, and the actual vote was 35 percent. This poll had three candidates tied, so it didn’t predict a winner.
The same group also polled the fourth Congressional district. For three candidates — Jim Anderson, Wink Hartman, and Jean Schodorf, the poll predicted the exact percentage that the candidates actually received. The exception was winner Mike Pompeo. The poll predicted he would win and receive 31 percent of the vote. He did win, and his actual vote total was 39 percent.
An election eve poll by political consulting firm Singularis had mixed results in the fourth district, but is notable in that it predicted eventual winner Pompeo’s vote total closely. The poll indicated 37 percent of the vote, and the actual was 39 percent.
In the fourth district, Schodorf released four polls that her campaign commissioned. Each poll showed her support increasing, until in the third poll, she took the lead. In the fourth poll her lead increased.
When comparing this poll to actual election results, we find that Schodorf’s poll overstated her actual performance by six percentage points. The performance of Anderson and Hartman were understated by six and seven points. For winner Pompeo, the final Schodorf poll understated his performance by 13 percentage points. (These polls did not include candidate Paij Rutschman.)
In a conversation before the election with Schodorf’s pollster, he indicated several reasons why the numbers in her surveys were different than the KWCH/SurveyUSA poll numbers.
One difference between the polls was the source of the voters called by the pollsters. The KWCH/SurveyUSA polls started with a list of households. To determine likely voters, the pollster would ask respondents if they were going to vote. Schodorf’s polls used voter lists as a source, calling only on voters who had a history of voting in August primary elections.
Because many people look at voting as a positive civic duty, it is thought that people will overstate their actual tendency to vote, and this is a reason why polls might decide to use voter history as a selection device, especially in primary elections where turnout is generally low. It is standard practice of campaigns to use voter lists in their voter contact efforts.
But this year voter turnout was high. The Wichita Eagle reported voter turnout in Sedgwick County — home to about 71 percent of the population in the fourth district — was 25 percent. That’s higher than the 19 percent turnout predicted statewide, and higher than in most primary elections.
Considering Republican voters, the Sedgwick County election office reports there are 104,558 registered Republicans, and 49,967 Republican ballots were cast. That indicates a turnout of almost 48 percent, considering Sedgwick County only.
By calling only those with a history of primary voting, many people who voted in this election would not have been sampled by polls based on voter history.
The Schodorf polls were conducted by live operators, while the KWCH/SurveyUSA polls were automated response. This can lead to a difference in the types of people that respond to the poll.
In the Republican Senate primary between Jerry Moran and Todd Tiahrt, the final KWCH/SurveyUSA poll had Moran ahead by 49 to 39 percent, with eight percent undecided. The actual totals were Moran winning with 50 percent to Tiahrt’s 45 percent, so that poll understated Tiahrt’s total by six percentage points while correctly choosing the winner.