Some deep background on the polls

Reader Frogg passed along this really interesting commentary from a professional pollster who is (a) a die hard Obama supporter and (b) laughing at the public polls:

I was having dinner a night ago with a friend of mine who is a statistician for a well-regarded private polling company. They do some work for Republicans in California, but most of the work they do is for Democrats or Democrat-leaning operations (Unions, etc.). Anyway, her shop was retained to do a few Presidential polls for targetted states on behalf of a union so the union could decide where to spend their ad dollars for the last week. They did Ohio, Florida, North Carolina, New Hampshire, Pennsylvania, Colorado, New Mexico, Nevada and Missouri. After mocking the hell out of the voter id spreads used by Rassmussen, Zogby, etc. (and this is coming from a committed Dem who will be voting for Barry O) she said the results of their polling lead her to believe that McCain will definitely win FL, OH, NC, MO and NV.

She says Obama definitely wins New Mexico. She said that Colorado and New Hampshire were absolute dead heats. She said she thinks there is a 55% chance Obama holds on in Pennsylvania and a 75% chance McCain wins Virginia. She absolutely laughed at the public polls showing Obama leading Virginia-and pointed out that all of those polls rely on Dem turnout being +4 and as much as +7, when in 2006, Republicans actually had the advantage by +3. She also pointed out that the numbers for Obama in SWVA look absolutely awful and that McCain is running 10 points better then Allen did in NoVa.

Anyway, her company's conclusion is that the election will come down to Colorado, New Hampshire and the Republican leaning district in Maine, which in her opinion might very well decide the Presidency (apparently the district in Nebraska that Obama thought he might be able to get is now off the table). She said she has very little doubt that the public polling is part of a "concerted voter suppression effort" by the MSM. She said IBD/TIPP was the only outfit doing public polling that was "worth a bucket of warm piss".

The turnout model scam of 2008 will be the big story of this election. In my last post on polls I noted how many of the polls out there are starting to fall into two camps. The first camp is the traditional approach using historic turnout models where party affiliation is actually quite close as both sides turn out their base equally well. The second model is new and unverified, and assumes there will be a huge difference in party affiliation turnout this year.

The first class of polls show a very tight race, the second class are so heavily weighted towards the dems they assume all the anger with the direction of the country has no impact on Democrats and their media puppets. The DC/NY Political Industrial Complex is mostly liberal in its make up, and it has recently become more and more isolated in its echo chamber. Just look at how it reacted to Sarah Palin. The end result is the polls of polls are showing a classic bimodal result with clusters of polls at one of these two modes: "traditional' and "extended" - just like the two turnout models now being openly produced by Gallup.

I think this pollster has it about right. LJStrata and DJStrata have been out doing the GOTV stuff this week and what they find here in our part of Northern Virginia is there are about equal numbers of Obama and McCain supporters, but the Obama supporters are more open, and the McCain supporters are a bit down because of the tilted polls. The GOP support is quiet, but just as massive as the Obama support.

Sadly, one of the better pollsters has gone to the dark side of the bimodal results. Rasmussen has decided to increase his Democrat edge to new heights, under the naive assumption Dem and GOP supporters are equally engaged in answering pollsters:

Like all polling firms, Rasmussen Reports weights its data to reflect the population at large. Among other targets, Rasmussen Reports weights data by political party affiliation using a dynamic weighting process (see methodology)..... The targets are not set arbitrarily. Rather, they are established based upon survey interviews with a separate sample of adults nationwide completed during the preceding six weeks. A total of 500 nightly interviews are conducted for a total of 21,000 interviews over the six week period.

While it sounds reasonable, the fact is willingness to participate in polls may not be uniform between the left and right. And pollsters are more and more considered part of the problem, part of the Political Industrial Complex which the public blames for the country being on the wrong track by 90+%. There is a `pox upon all your houses' mentality in the country that is starting to lump the media, the pollsters, the consultants in with the politicians from both sides. Trust in the political class is gone.

Rassmussen now gives the dems a questionable 33-40 edge in turnout (Indies at 27). Previously in September he had Rep-Dem-Indie (RDI) at 34-39-28. Each time he increased the dems edge McCain lost ground - duh!

Let's look at the latest Newsweek Poll which has Obama up 12%. One look at the internals gives the answer to the large lead. Supposedly from 1204 interviews they distilled out 1092 Registered Voters, which were further distilled down to 882 Likely Voters. Of these they claim 299 Reps, 380 Dems and 381 Indies.

If we do the math the total sample size for party affiliates is 1060 - which cleverly doesn't map to interviews, registered voters or likely voters. What is also interesting is they have error bars on their party ID? They claim the Rep error bar is +/-7 (which comes out to 2%.) and the indie/dem error bar is +/- 6. Not sure why they are not sure what party people claimed they were associated with was in doubt, but there it is.

Anyway, if we do the math the Newsweek RDI turnout model is 28-36-36. Recall Rasmussen's model was 33-40-27. I mean they are not even close!

One of the problems is a lot of people are still on the fence to some degree and pollsters are trying to push them into a lean for one candidate, but this is very shaky ground to be on. Look at this Minnesota poll which is making people sit up and take notice:

A St. Cloud State University poll shows that Democratic presidential candidate Barack Obama leads Republican opponent John McCain 42 percent to 37 percent in Minnesota.... After earlier surveys showed DFL challenger Al Franken with an edge, the SCSU poll shows Republican U.S. Sen. Norm Coleman leading with 36 percent, compared with 27 percent for Franken. Independence Party candidate Dean Barkley trails with 16 percent.

Not only is McCain down only 5%, there are an enormous number of undecideds given the low numbers for both candidates. I included the surprise Senate seat race for comparison because it shows similar numbers for Coleman and Franken (with the Rep out front), but there is a substantial number voting for a 3rd party candidate in that race. There is no major 3rd party candidate on the national level. In MN and with this poll, the race is wide open because of the enormous number of uncommitted voters. Obama could easily lose this state if this poll is to be believed.

So what do we have? We have polls using dynamic turnout models which link willingness to take a poll to willingness to get out and vote (a very suspect assumption IMHO). Especially since different pollsters are deriving vastly different turnout models as a result of this approach. All these polls share one thing, they tend to show massive Obama leads, while the traditional turnout models show a tight or tightening race.

Much more here

Posted by John Ray. For a daily critique of Leftist activities, see DISSECTING LEFTISM. For a daily survey of Australian politics, see AUSTRALIAN POLITICS Also, don't forget your roundup of Obama news and commentary at OBAMA WATCH (2). Email me (John Ray) here