Calbuzz seldom puts much stock in polls pushed by campaigns, for one simple reason: they don’t tell you the results when they’re not good for their candidate. On the other hand, when a campaign is willing to give us the whole survey and we can see if it makes sense, we’re more likely to bite.
So when the Ro Khanna campaign touted a new poll showing their guy even at 38% with U.S. Rep. Mike Honda in Silicon Valley’s 17th CD, we were willing to take a look, if they gave us the entire thing, from San Francisco pollster David Binder. Which they did.
According to Binder’s poll, the Khanna-Honda race is – against the odds — up for grabs, with Khanna pulling 33% plus 5% of leaners and Honda drawing 34% and 4% of leaners. That leaves about one-fourth of the voters – 24% — undecided and up for grabs with less than three weeks to go.
Some Very Strange Results: But no sooner had the Khanna poll hit the internets when the Honda campaign suddenly discovered, by golly, they also had a poll in hand, claiming to show that Honda is leading Khanna 42-27%, with 31% undecided. But Honda’s campaign didn’t release the whole survey, wouldn’t tell us how voters were selected and contains some inexplicable weird results.
For example, 21% of the respondents in the Honda survey, done by David Mermin of Lake Research Partners, volunteered to pollsters that they had already voted. But that’s not possible: ballots had just been mailed out to voters in the district and even Mermin acknowledged “people are lying – there’s a lot of over-reporting.” The Honda poll also includes 38% Asian-Americans (compared to 28% in the Khanna survey). That’s not impossible, but it sure seems unlikely.
Even stranger is this: If the Honda survey is correct, not only would Khanna have made no progress since the primary, when he got 28% of the vote, but Honda would have lost 6 percentage points, or 12.5% of the 48% he got in the primary. And this is supposed to show strength by the incumbent? Hmmm.
According to Binder’s poll for Khanna, his candidate’s favorable-to-unfavorable rating is now 57-14% while Honda’s is 57-28%. So Honda is better known, but his negatives are considerably higher.
The questions on voter preference and favorability were, according to the Khanna survey we’ve been given, asked before any “push” questions that campaigns use to test various themes. That makes them “clean” questions, not affected by information imparted to respondents by later questions.
Testing An Attack Line: The one “push” question in the brief survey of 400 likely voters, was as follows:
Press reports have suggested that that Honda’s Congressional office illegally coordinated with campaign staff by inviting people to an official State Department event for the purposes of securing campaign donations, and that his staff violated ethics rules by running his personal errands on official time. If these allegations were proven true, would you be very concerned, somewhat concerned, not too concerned, or not concerned at all?
The results: very concerned 32%; somewhat concerned 35%; just a little 8%, not at all 20%.? (1% said it’s not true and 4% preferred not to answer.)
The survey then asked: If these ethical issues about Congressman Honda are proven true, would this make you more likely to vote for Khanna, more likely to vote for Honda, or does this make no difference to you?
As far as we can tell, the campaign tested no other attack lines against Honda and did not attempt to measure support after this one, except to see if it has some effect – which is obviously does.
The Honda campaign may have tested other communication points, but the campaign did not reveal them.
As far as their voter models, there’s not much difference: both assume 46% Democrats; Khanna has 23% Republicans, 27% Decline-to-states and 4% minor party voters; Honda has 21% Republicans, 30% no party preference and 3% other.
The Khanna survey was conducted Oct. 8-9 among 400 likely voters with a margin of error of +/- 4.9%. The Honda poll was conducted Oct. 7-12 (just as absentee ballots were going out in the mail) among 500 likely voters with a margin of error of 4.4%.