Wednesday, May 30, 2012

MU Law poll massively rigged to favor Walker. Ignore it. (Part 1)

Given the clear momentum that Tom Barrett seems to have grabbed, you just sort of knew there would be a garbage poll coming out to try to show Scott Walker ahead to try to blunt the movement. And Marquette Law Professor Charles Franklin did not disappoint, throwing up a poll today showing Walker up 52-45 among likely voters, and 49.93- 44.35 among all voters. So given that Franklin and Marquette notoriously skewed their last poll to show Walker up 5, I figured I'd go inside the numbers and test this out as well.

Immediately what jumped out at me was this question.

What region of the state are you from?
Milwaukee City 13%
Rest of Milwaukee media market 33%
Madison media market 17%
Green Bay/Appleton media market 18%
Rest of state media markets 19%

That's right, this sample is based on 46% of the state being from the Milwaukee media market. Immediately, this seems like bullshit, but I wanted to test out this theory with the 2010 final results from the Walker-Barrett governor's election, which had a relatively high Milwaukee (and especially 262) turnout. So I took the results from Dodge, Kenosha, Milwaukee, Ozaukee, Racine, Sheboygan, Waukesha and Washington Counties, and compared it to the rest of the state.

Total votes cast in "Milwaukee media market", 2010- 866,576
Total votes cast for state- 2,160,832
Percent of votes cast from Milwaukee media market- 40.1%

So Franklin's oversampling them by 6%, and probably more than that given that other parts of the state (like the Madison market) will probably have higher turnout than we saw in 2010. Now interestingly, Walker only leads 48-45 in this part of the state, but that's still is a whole lot better than the 54-40 he's losing by in the Madison media market (which is equally undersampled at only 17% of all respondents and even 54-40 seems low for Barrett, because Walker might be lucky to hold Barrett to 54% in ANY county in the Madison media market.) This move to emphasize Milwaukee seems sketchy at best, but there are more obvious ways to show how skewed this poll is.

And the best is through ideology questions. As mentioned before, the the 2010 Wisconsin exit poll showed ideology at 21% liberal, 43% moderate and 36% conservative. So how does Franklin measure up? Check out question number 13 on the cross-tabs.

Ideology, Marquette poll
Very conservative 9
Conservative 33 (42% total)
Moderate 32
Liberal 16
Very liberal 7 (23% total)
Don't know (v) 2
Refused (v) 2

Somehow a 42-32 con-to-mod breakdown might go a bit better for Walker than the 36-43 one like he had in 2010, don't you think?

So let's turn these numbers and match the responses to the actual 2010 turnout (which is itself is more favorable to Walker, as the 60-65% predicted turnout on June 5 will probably skew more moderate and liberal). And we get...

Walker 48.62 - Barrett 46.11. Very close race, even with the pro-Milwaukee skew allowed. And do you really think any undecideds would vote for Governor John Doe?

On a realted note, Franklin also makes the same mistake last week's St. Norbert poll made- underpolling 2010 Barrett voters compared to 2010 Walker voters.

(Among those answering) Who did you vote for in 2010?
Walker 50.34%
Barrett 42.36% (actual result, Walker 52.25%, Barrett 46.48%)

So that's 2.2% of a higer gap than the real result. A 2.2% drop in Walker's edge among possible voters would lower Walker's advantage to 3.3%, and would be portrayed as a near toss-up, which wouldn't sound so good if you wanted Walker to stop Barrett's momentum now would it? Strangely, Franklin refuses to give cross-tabs on this question, and you can't see who those people are voting for specifically, but it's probably a good guess that they'd be more likely to favor Walker. And while Franklin includes a lot of interesting questions about attitudes and how people may have done things like discuss politics and yard signs, but he mysteriously dropped a question he had a month ago on whether a person had signed a recall petition. Makes you wonder why. Maybe a low number of voters woud expose this sample as even more GOP-leaning (remember, the equivalent of 42% of the 2010 voter turnout signed a Walker recall petition).

And the last skew I want to bring up goes to another major problem I've had with Franklin's polling all the way through, the Marquette Law School's habit of weighting certain groups if they can't get them on the phone. As Marqutte Law explains in their methodology statement
Post-stratification, or weighting, compensates for patterns of non-response that shift sample characteristics from
known population values. In telephone surveys it is common for potential respondents who are younger to
exhibit higher rates of non-response resulting in these groups being under-represented in the sample. To
compensate for these non-response effects the sample is weighted to bring sample characteristics into line with
the population values.
And they sure lived up to it in this poll, as they talked to virtually nobody that wasn't near Boomer age in this poll.

Unweighted average of Marquette Law poll
18-29 6%
30-44 16%
45-59 34%
60+ 43%

That's right, 77% of respondents were 45 or older. Sure, the Marquette people tried deflating down from that with their weighting, but let's be real about this, it's going to be flawed. You can't get anything close to a representative poll with that lack of age diversity, especially when you realize that more than half of the voters in Wisconsin in 2008 were under 45. Does anyone really believe that Walker leads 47-46 among voters under 30, when that same demographic voted 64-35 for Obama in 2008? That doesn't pass the "Bullshit" test, but that's possible when you have such a small sample size of young voters.

I'm going to make this into two parts now due to length, because in Part 2, I'll explain why Franklin let this faulty data in. And the evidence points to propaganda, and not necessarily shoddy research. But be sure, this poll is basically junk, and that's without any suspicion of the motives behind the poll and the pollster.

1 comment:

  1. Another great post, Jake. No reputable pollster would release a poll so obviously skewed.

    Here's what I found out about the outfit who actually made the calls, the Universal Survey Center, NYC. They are a market research firm, not a public policy polling organization. To wit, they do not belong to the American Assn of Public Opinion Research (AAPOR), which has high standards. USC does have a polling operation. But here is what USC’s own website says about their ‘Public Insights : Vote’ polling:
    “A dedicated team of Universal Survey professionals works directly with you to understand and execute the requirements of your project. This team approach provides you with undivided and personalized attention. Our teams are championed by Account Directors who are your research partners. Their role is to understand your needs and provide you with the right guidance and support. Dedicated Client Service Managers are responsible for the implementation of your project. They ensure you have the critical information required to help manage your project.”

    IOW, if Franklin wanted to poll a respresentative sampling of the electorate, this outfit would deliver it. But he doesn’t. He commissioned a GOP propaganda tool – and that is exactly what they gave him.