Wednesday, July 11, 2012

Marquette Law was more lucky than good on #wirecall polls

Noticed that Marquette Law School and PPP had polls come out in the last 2 days, both of which show Obama up in the high single digits over Romney and the Senate race being close (although they're way off on the GOP primary, PPP has Hovde leading Tommy, while Marquette Law has Tommy up double digits). But I wanted to do a bit of looking back at the Marquette poll that was conducted before the June recall elections, and see if I owe them a mea culpa after my article in May ripping them for having a biased sample. Well that same poll said Walker was up 52-45, and he ended up winning 53-46, so at first glance it looks like Chuck Franklin and co. at Marquette Law knew what they were doing. But let's find out for sure.

Let's first look at my complaint that the sample was too Milwaukee-centric. Franklin's final poll had 46% of respondents from the "Milwaukee media market", which I have defined as Dodge, Kenosha, Milwaukee, Ozaukee, Racine, Sheboygan, Walworth, Washington and Waukesha Counties. Then I checked the final results from the GAB to see how many people in those areas actually made up the vote. I also criticized Franklin for underplaying the Madison media market (which I'll ID as Dane, Rock, Columbia, Sauk, Jefferson, Green, Iowa, Lafayette and Grant Counties), as he only projected them as 16% of the vote.

Milwaukee and Madison media market % of 2012 recall vote
Milwaukee- Marquette poll- 46%, Actual vote- 40.0%
Madison- Marquette poll- 16%, Actual vote- 18.6%

So yes, those numbers were indeed off. What was also undeplayed were the numbers in the rest of the state, as they accounted for 41.4% of the total votes, and Franklin only had them at 38%.

Another area I had issue with Franklin's poll was the ideological ID, which I felt skewed overly conservative (and went along with the pro-Milwaukee area skew), and oversampled "independents", which helped Walker's numbers because many independents are really Tea Baggers who don't admit they're Republican (for example, I heard Bagger caller C.J. on Sly's show insist he was "Libertarian" and not "Republican. Give those of us in the above-ground world a fucking break).

So let's compare these figures with the "revised" New York Times exit polls, which accurately had it 53-46 Walker.

Ideological and Party ID, Marquette poll vs. NYT Exit poll
Marquette poll- Liberal 23, Moderate 32, Conservative 42
NYT Exit poll- Liberal 21, Moderate 44, Conservative 36

Marquette poll- Democrat 33, Republican 28, Independent 36
NYT Exit poll- Democrat 34, Republican 35, Independent 31

Yep, they really oversampled the cons, got a few too many libs, and didn't talk to enough moderates. On the flip side, the Marquette poll traded far too many "Independents" for straight-up Republicans, which also leads me to think that a lot of those polled conservative independents were really GOP in disguise.

A couple of other demos I want to look at was the Franklin theory that Republicans would be more motivated to vote, and 2010 Walker voters would be more likely to show up than 2010 Barrett voters. I thought this was a wrong assumption, but it turned out that not only was Franklin closer to correct, he underestimated this effect.

(Among those answering) Who did you vote for in 2010?
Marquette poll- Walker 50.34%, Barrett 42.36%
NYT Exit poll- Walker 47, Barrett 34, Did not vote 13

Not sure what happened to those people that voted Barrett in 2010 (they also could have voted early, as Dem constituencies encouraged this), but this definitely had a big effect, especially since Barrett won new voters 53-45.

Lastly, I criticized Franklin for underestimating the younger vote, as nearly 2/3 of the respondents 45 or older (and that's only after he weighted the poll, because he apparently couldn't get any Gen Xers and Yers to answer). So let's put that up against the exit poll and see how Franklin and co. did.

age of Marquette poll answerers (vs. NYT exit poll)
18-29 years old 16% (16%)
30-44 years old 18% (22%)
45-59 years old 36% (44% 45-64 NYT)
60 years or older 29% (18% 65+ NYT)

Not that far off, though a bit underestimating of Gen Xers like me.

So in looking at Charles Franklin's "bang-on" final poll for the Wisconsin guv recall election, there were some things he got right, mostly involving age of voters (so maybe there's something to this weighting, even if you don't get the actual people on the phone) and he sort of attached to the enthusiasm gap that I didn't think existed but apparently did in the 920 and 715 area codes.

But in other areas, Franklin was comically wrong. He overestimated the Milwaukee and conservative votes in the election, and overestimated the number of Independents that would vote. So when taking it on the whole, I'd give Franklin a BC. That's the grade you probably deserve when you have the right answer to the essay, but the reasoning and logic behind the conclusion isn't very accurate or coherent.

That won't stop the lemming-like Journal-Sentinel from blindly accepting Franklin's MacIvered numbers as totally accurate, but I'd hope this breakdown of the numbers has shown that they deserve to be treated with at least a grain of salt.

No comments:

Post a Comment