Marquette Law poll has Walker ahead

Although honestly? When you run a poll showing a large spread and then have to declare that spread within the margin of error, you really shouldn’t have bothered to poll. From the release:

The poll finds Walker ahead of Milwaukee Mayor Tom Barrett by a 50 percent to 44 percent margin. Walker leads the only announced Democratic candidate, former Dane County Executive Kathleen Falk, with a 49 percent to 42 percent margin. Walker leads former Congressman David Obey by 49 percent to 43 percent. Janesville Democratic State Senator Tim Cullen receives 40 percent to Walker’s 50 percent. While Walker consistently leads his Democratic opponents, the size of the lead is within the poll’s margin of error for all but Cullen.

That would make the MOE over 7 points! Oh good grief.


  1. “The margin of error for the size of the lead
    in the governor’s race is 7.0-7.2 points.”

    Yep that’s pretty hefty. Likely budgeting considerations.

    And if you’re going to do a Marquette poll, why farm it out to PA?

  2. jimspice, I’m afraid it is unfortunately before 5pm and I am unable to follow your logic. 🙂

  3. What’s significant is that Walker is at or very close to 50% against everyone. Also significant is that he has the leads even though Obama is up by8 points in Wisconsin. That means that independents, including Democrat-leaning independents, are sticking with Walker in number similar to his election last year. I wonder if the recall organizers wish they had timed the election to coincide with the general election instead of the summer. If Obama is winning Wisconsin handily come November, there might be more Democrats at the polls vs. the summer. Granted, participation is pretty high in Wisconsin, anyway, but if the race comes down to a handful of votes, that might be the difference.

    As it stands, the race could be a precursor to the general election. If Walker survives, it could boost the GOP overall in November. If Walker is toppled, it would be a major motivator for Democrats.

  4. Just saying that confidence intervals shrink as the number of respondents (the “N” in stats speak) goes up. Ideally, every poll would have a huge sample resulting in tiny margins of error, but that’s easier said than done; time and budget constraints are always in play.

  5. And generally about 1,000 respondents will meet that goal. Why make the effort if you aren’t going to do it correctly?

  6. People might say “Hey, Marquette. Isn’t that where (R) yes-man Rick Esenberg works? I’m sure the results are rigged” or somesuch. But it’s not in the best interest of any polling agency to provide sub-optimal results, because the discordance with actual results will become apparent soon enough and the polls reputation is then sunk. Besides, Marquette has Charles Franklin attached to the project, and his reputation is solid.

    There are calculations you can do given the the size of the population in question and the MoE you are shooting for that will supply the minimal number of pollees. However, there are further considerations, such as when only a subset of the whole anwers certain questions, or variations on a single question are given (e.g. Walker or Barrett, Walker or Falk etc.) which also theoretically depress confidence in the accuracy of the results. Too often these considerations are ignored. Polling is just like a hard science experiment; the method of the experiment can actually influence the results (like the act of observation in the classic Schroedinger’s Cat example). Each time you poke the data, it becomes less reliable.

  7. Randy in Richmond says:

    Another reason this poll is shakey is that polls of ‘registered voters’, as this one is, are traditionally much less reliable than polls of ‘likely voters’ and tend to favor Democrats because usually they have a higher rate of registration.

  8. jimspice, yes, yes, trust me. I made excellent grades in all of my stat classes. And if my memory serves me correctly, polling in anticipating of some dropping out to the interior questions of about 1,000 to respond gets you there, as I mentioned earlier.

  9. I wouldn’t be surprised if they rushed it in order to get it out this week. Plus, since the result was within the margin of error (albeit at the edge of it, in all 3 cases), it keeps it “interesting.” If a number of polls came out showing Walker with a significant lead outside the MoE, that could deflate the recall initiative.

  10. Good catch; I also calculated the MoE at about 7 percent, and that immediately told me that this polling attempt was poor. Of course, that MoE is tied to the substandard sample size of only 700 — and others looking at this have pointed out problematic weighting within that number that further negates the usefulness of this poll.

    Then I looked at the survey instrument, and no wonder that Franklin got only 700 respondents from more than 4,000 or something (I saw that sample size on some other analysis, as well). The number of questions would take at least half an hour and probably a lot longer to answer! And a lot of the crucial questions come near the end, with no explanation of how much dropoff must have occurred, how the weighting (Repubs vs. Dems and “leanings” on both sides) was affected, etc.

    I know that Franklin was tied to a conservative organization, with some problems there, but his overall reputation is so solid that I was glad to see him take on this project. Now, from what we’ve seen of the first effort, I’m disappointed and will cast a very critical eye on future efforts. I hope that some other polling outfit tests the state soon, so we can have some basis for comparison with Franklin’s poll to get a better sense of the public sensibilities.

  11. Of all the pollsters, Democratic-leaning Public Policy Polling was the most accurate in last year’s recall votes and the referenda in Ohio. They are like the Rasmussen of the Left. While they have a political bias, their polling methods are very good. I would look for their polls.