The Brown-Warren race and the ghosts of a 2010 poll

With today’s Boston Globe poll reporting that Democratic challenger Elizabeth Warren leads Sen. Scott Brown by five points, get ready for Brown’s defenders to dredge up an infamous Globe poll from two years ago — the one that showed Attorney General Martha Coakley leading Brown, a Republican, by 15 points.

Globe-bashers like Howie Carr love to point to that earlier poll as a sign of the paper’s liberal bias — and I’ll predict right now that that will be the subject of Carr’s next column in the Boston Herald.

In fact, Globe polls are not Globe polls — they are conducted by the University of New Hampshire Survey Center, a respected, independent polling operation.

So what went wrong in 2010? My theory: Nothing. The story about that earlier poll is protected behind a paywall (I’m a subscriber, so I’ve reread the whole thing). But as you can see from this excerpt, the poll was conducted between Jan. 2 and 6, and the election to fill the U.S. Senate vacancy created by Ted Kennedy’s death was held on Jan. 19.

Thus it’s likely that the poll was accurate when it was conducted. People were just getting back to their normal routines coming out of the holidays. The race broke very late for Brown. By the time the story was published, on Jan. 10, the race was already trending away from Coakley, and within days, other polls were reflecting that.

What does that mean for Brown now?

First, the margin of error in the new poll, which shows Warren with a 43 percent to 38 percent lead, is 4.4 percent. In other words, if the election were held tomorrow, Brown could beat Warren by several points without calling the validity of the poll into question. The race is still essentially tied.

Second, this is not a low-turnout special election, and as the Globe story notes, Brown faces some harsh realities. By wide margins, people like Brown and like the job he’s doing — but they are increasingly leaning toward Warren because of the enormous enthusiasm among Massachusetts voters for President Obama. I suspect you would not be able to get Brown to utter the words “Mitt Romney” these days even if he were being waterboarded.

Third and most important: It’s still early. No, it’s not as early as it was during the pre-Labor Day period, when you could argue that most people weren’t paying attention. But it’s early enough for things to change dramatically if Warren stumbles badly. That’s why I think Brown is making a mistake by putting a torch to his nice-guy image with his continued attacks on Warren’s claim that she’s part-Native American.

David Bernstein of The Phoenix offers some further analysis of the Globe poll. And Nate “The Great” Silver of the New York Times takes a deep look at conservative claims of liberal bias in polling — and buries the assertion in an avalanche of well-marshaled data.

Illustration (cc) by DonkeyHotey and republished under a Creative Commons license. Some rights reserved.

About these ads

22 thoughts on “The Brown-Warren race and the ghosts of a 2010 poll

    1. Mike Rice

      I agree with you, really creepy. Warren looks like she could take down a Sierra redwood in nothing flat and Brown looks like Dan Aykroyd.

  1. Tony Schinella

    Good piece here Dan and I would tend to agree with much of your analysis.
    While I’m not a fan of the Globe for a variety of reasons, at the time that 15% poll was taken, it was probably accurate for all the reasons you state. Everyone knows that Brown surged at the end and tracking polls showed that. In fact, the surge story – and Coakley attempting to hold back that wave – were positive stories for Brown in the last days of the campaign. There were also a lot of idle hands not doing much in January and many of them quietly helped Brown raise last minute cash for ads – that one day money bomb which raised a ton of $$$ – which helped elevate his presence.
    Poll results are a snapshot in time and have so many factors around the presentation of data that it is completely impossible to surmise much from them at all. It’s fun for junkies but there are more positives than negatives around polling. I’m glad though that some people are bringing up the issue of skewed and weighed polling as a method of creating positive news around a candidate’s chances. The Colbert Report can laugh about it all it wants to. But the methodology used and the people polled play a large role in the outcome of the results. For years, I’ve advocated media outlets not spend a dime on polling and instead, put money and resources into actually covering the race, what all the candidates on the ballot do and say, and then allow the voters to choose the best candidate. If the media took the horserace aspect of campaigns out of the storyline, the voters might actually have to do a little bit of studying and find out which candidate they like best – instead of basing their votes on which candidate will win, based often on manipulated data collection.

  2. L.K. Collins

    Good luck on that dream, Tony. Without the polls they journalists might have to do some thoughtful analysis. When thoughtful analysis is confronted by looming deadline, looming deadline always wins…with the journalist making a New Years-type resolution to do better next time.

    Rinse and repeat as often as necessary.

  3. Martin Callaghan

    Good analysis Dan. While you are correct that “By the time the story was published, on Jan. 10, the race was already trending away from Coakley” we should not and cannot forget that Coakley was also trending away from the race.

  4. Deb Nam-Krane

    I am privy to nothing, but my guess is that in the two weeks right before the race Warren ads will spend a lot of time linking Brown and Romney. Possibly sooner if Romney has another 47% moment. I think they’re saving it but they previewed it in the debate where she twice mentioned that she believed Obama should be president/commander-in-chief. It looked like he wasn’t breathing when she did that.

  5. Steve Stein

    “First, the margin of error in the new poll, which shows Warren with a 43 percent to 38 percent lead, is 4.4 percent. In other words, if the election were held tomorrow, Brown could beat Warren by several points without calling the validity of the poll into question. The race is still essentially tied.”

    I really have to take issue with that, Dan. You have to do the math (which, in this case, is rather complicated). What those poll numbers really mean is that if the election were held tomorrow, Warren would have more than an 85% chance of victory. That doesn’t sound “essentially tied” to me.

    1. Dan Kennedy Post author

      @Steve: You’re better at math than I am, but if Brown beat Warren 42.4 percent to 38.6 percent, that would be at the outer limits of the margin of error. I understand what you’re saying, because the MOE tilts in her direction. But I think in lay terms, a race that is within the MOE can be considered more or less tied. And we’re not even getting into assumptions about turnout and who’s a likely voter, always fraught with uncertainty.

      1. Steve Stein

        It’s hard to explain this without diagrams, but “margin of error” describes a bell curve, such that 95% of the area under the curve falls within the given percentage, plus or minus the MOE. A 42.4% showing for Brown (at the edge of the MOE) would be a 1 in 40 event (2.5% – the other 2.5% is on the other end). Similarly for Warren. So Brown beating Warren 42.4 (or more) to 38.6 (or less) would be a 1 in 1600 chance. Not impossible, but very very slim.

      2. Dan Kennedy Post author

        @Steve: Not to be contentious, because you know what you’re talking about. But you’re assuming these things are precise and not just an approximation.

      3. Steve Stein

        Kevin Drum had an article about this here (with an admittedly unfortunate “blast-from-the-past” quality) to describe the convoluted math. (Literally “convoluted”! The math behind it is known as a “convolution integral”.)

      4. Steve Stein

        Au contraire, Dan. This is ALL about how these things are NOT precise. The mathematics behind polling are precise though, and this one poll says Brown would have about a one-in-six chance of winning. That’s definitely not zero. And other polls are better for Brown. But *this* poll doesn’t show the race “essentially tied”.

  6. Steve Stein

    Then again, you can look at this another way. IF the race was really tied, and you took 12 polls with this sample size (a little more than 500 I think), you would DEFINITELY expect one result like this. (And a corresponding one showing Brown up 5, and more in the middle.)

    Math is hard. Which is why we have Nate Silver.

    1. Dan Kennedy Post author

      @Steve: Brilliant as Silver is, he is attributing a level of precision to these things that doesn’t really exist. Remember, he made his mark with baseball statistics, a much more honorable (and less malleable) calling.

      1. Mike Benedict

        You misunderstand what Silver is saying with his polls. What he doesn’t say is, Candidate X will win. What he does say is, given the population that was surveyed, here’s where they stand. Many journalists and less ethical forecasters tend to extrapolate from what the data are actually designed to reveal without also noting the problems of the survey. Silver is equally rigorous in pointing out the flaws.

      2. Steve Stein (@SteveZStein)

        I think election polling is much more accurate predicting elections than baseball stats are at predicting a player’s next year performance. Or a team’s. How many predicted the Sox’s finish this year?

  7. Mike Benedict

    I should add, as far as I know, Silver doesn’t actually conduct any polling. He assesses others’ polls and tries to meld all of them into a single data point, a kind of crowd-sourcing of the analysts, if you will.

Comments are closed.