Reporters Hyperventilate Over, Then Discount, Poll — After a Call to Eric Davis
When word spread Wednesday afternoon that Castleton State College's new-ish polling center would release fresh results on the bitterly contested Democratic primary race for attorney general, reporters drooled.
At least, I did.
After all, we live in one of the least-polled states in the union, leaving political reporters to simply guess what real people are thinking or, worse yet, to dust off the rolodex and query retired Middlebury College professor Eric Davis — Vermont's Pundit Laureate, as Green Mountain Daily's John Walters endearingly calls him — who will readily provide the latest conventional wisdom.
So when Castleton released the first publicly available poll of the AG's race since May — back when the contest was still in its infancy — reporters surely went straight to Castleton Polling Institute director Rich Clark to seek his insight into what his poll tells — and doesn't tell — us about the state of the race. Right?
Uhhh, not exactly. Their speed-dials seemingly frozen on Eric Davis' number, at least three news outlets went straight to the Oracle of Middlebury to see what he thought about a poll he didn't conduct.
In a story titled, "The leader is ... uncertain: Poll results doubtful," the Vermont Press Bureau's Peter Hirschfeld brings in the Pundit Laureate in the fifth graph, before quoting Clark himself saying much the same:
“I have serious doubts about the validity of this poll,” said Eric Davis, professor emeritus of political science at Middlebury College.
Terri Hallenbeck over at the Burlington Free Press goes so far as to lead her story with Davis' take-down:
A poll released Wednesday suggests incumbent Bill Sorrell leads challenger T.J. Donovan in the Democratic primary race for attorney general, but one political scientist argued the poll has little validity.
VTDigger's Taylor Dobbs, meanwhile, doesn't even bother bringing Clark into the story, instead devoting most of his piece to Davis' critique:
“I have very, very serious doubts about the validity of a poll that says it has that many people who are likely to vote in a primary,” Davis said. “I think this Castleton survey over-reports likely voting in the primary by three, perhaps as many as four times.”
Does Davis have a point? Does a survey that relies upon respondents to self-identify whether they'll vote in a snoozer of a late August primary they probably won't actually vote in paint an incomplete portrait? Totes. As I argued back in May — and as Clark himself readily admits in a summary of his results —over-reporting intent to vote is a chronic problem in public opinion polls and is exacerbated when turnout is especially low.
Would a poll of those who've actually voted in the past couple of Democratic primaries be more accurate? Sure. Would a larger sample size have reduced the poll's 7 percent margin-of-error? Mos' def. Would a pair of questions gauging the candidates' name recognition be informative? No doubt.
But do these failings render the Castleton poll invalid? Hardly. Polls rarely pin down precisely where an uncertain primary electorate stands, nor are they terribly good at predicting the outcome of races that have little in the way of precedent. Keep in mind that next Tuesday's election is only the second since the Legislature voted to move primary day three weeks earlier — into the summer — and the first such election without a blockbuster gubernatorial primary at the top of the ballot.
What this poll does provide is a second, time-stamped look at the preferences of a particular slice of the electorate — dudes who claim they're going to vote in the primary — three months after a similar group was surveyed by the same pollster. By establishing a baseline in May of where Vermonters stood on the AG race before it had truly begun — and relying upon a similar number of respondents — Castleton's latest poll tells us two important things:
- That neither candidate has appreciably pushed the needle one way or the other. Sorrell was the favorite of 49 percent in May, compared with 44 percent now. Donovan, meanwhile, drew 23.2 percent in May, compared with 24 percent now.
- More importantly, the percentage of people who have no friggin' idea who they'll support has actually increased as the campaign has progressed: from 25 percent in May to 31 percent now.
Am I headed to Intrade to bet my last paycheck on Clark's results mirroring next Tuesday's outcome? No way. But in a poll-starved state like this, I say more information — in the appropriate context — is always better than less.
Without it, dear reader, you're left listening to blowhards like me — informed by guesswork, campaign spin and occasional interaction with the outside world. Or, even worse, drivel like this:
"If the primary had been held three weeks ago when the early voting period began, I believe Donovan would have won. Sorrell has made it a tighter race over the last couple of weeks."
And this:
"I would say at this point, I would give Donovan about a 55 to 60 percent probability of winning the primary. But at the same time, I wouldn't be surprised if Bill Sorrell were able to pull out a narrow victory at the very end."
Both of those predictions came last weekend from Eric Davis himself, speaking with VPR's Peter Biello Saturday morning.
Leaving aside for the moment the Pundit Laureate's characteristic equivocation, what I would like to know is precisely what methodology the Oracle of Middlebury used to determine a great shift in the mood of the electorate these past three weeks. More importantly, just how is he able to pin down Donovan's probability of winning to between 55 and 60 percent? That is mighty precise.
The answers of 223 randomly-sampled, self-identified Democratic primary voters may not tell us everything we want to know about what's going to happen next Tuesday. But they tell us a lot more than the guesstimates of a sample of one: a retired professor reading his newspapers in Middlebury, waiting for the next reporter to call.
Photo of Rich Clark provided by Castleton State College.
