McCain Moving Up Or Pollsters Playing Around?
By: Mark W Adams

Note to Obama: Don't take the bait and play in the sludge with Exxon John. His only hope is to bring you down because even if he had been running an "honorable" campaign, his best case scenario still loses unless he can steal a significant amount of your support. McCain can't win by building himself up alone. He must bring Obama down to prevent a Democratic tsunami.

When's the last time we had a President smart enough to be president of the Harvard Law Review and not just swear to "preserve, protect and defend the Constitution of the United States," but also teach what the Constitution means. I don't even think the current Resident has even read the thing. Obama winning is a very good thing.

You'll notice some movement in McCain's direction if you've been
following along. don't be alarmed. It's as more about how I'm
allocating close races than anything McCain has done (or anywhere he's
threatened to bomb). Note that Obama's numbers are still rock solid
and he's got a cushion of 14 electoral votes while even in this latest
map, McCain's best so far, he needs 98 more electors with only 82
available in "Toss-Up" states.

Rather than flip a coin or roll a 12-sided dice to decide which of the three polling aggregaters I'd rather rely on between the very different projections Real Clear Politics, Pollster.com and FiveThirtyEight.com present, I did what pollsters do and averaged their results, matched that with the closest fit and ran the numbers to show you how I wanted you to see the world.

I really like the look, the color scheme of Pollster.com, and their interface is superb. You can tell at a glance the state of the race and you can delve into the details with ease just by mousing over what you want to see -- a click takes you deeper inside the numbers and graphs. I also like their trend analysis since it's simple and something a non-statistician can grasp. 538 uses a much more sophisticated method of analyzing and weighing polls to project the outcome, and while it probably is highly accurate you can't tear the conclusions apart to see where they came from like the other sites. But they put a ton of info on one page.

That leaves RCP who's editorial preferences are not my cup of tea, and they're way too stingy when it comes to awarding a state to one or the other candidates (ie. they have Obama +7.2 in Wisconsin, but it only "leans" his way). However they do attempt to do what you're supposed to do with these things, provide a snap-shot of the current state of the race and not necessarily predict the future. That's what all polls really are, at least according to all the pollsters who simply can't resist the temptation to predict rather than merely report, despite their best intentions.

RCP w/147 toss-ups: Obama +65
RCP w/o toss-ups: Obama +70
538.com, no toss-ups: Obama +34.4
Pollster, 85 toss-ups: Obama +115

Both of RCP's electoral vote counts are close to the average of all the aggregaters, so I used their numbers as a tie-breaker between the other two when there's no clear consensus on a particular State. Interestingly enough, RCP's "rolling average" and their simple division of leaners vs. strong states based on whether a candidate has more than a 10 point lead seems arbitrary compared to the "confidence interval" methodology 538 and Pollster use. But since these more professional polling annalists come up with a radically different spread in the electoral count, I'll go with the "less is more" theory that RCP uses when in doubt if it seems reasonable.

It's that or resort to a modified version of the Keith Number since I have questions about all three sites' category selections processes. RCP's rolling average has no scientific value. 538 uses a weighted method that handles the age of a poll in a unique and intuitive manner, but also weighs each polling firm by their reputation and reliability, a completely subjective way to analyze any particular data point offered by professional polling firms.

Pollster is chalk full of professional analysts, and I can't fault their methodology. It's what they're trying to show I have a problem with. I don't necessarily want to know what the "trends" are "telling us" will happen later on. I want to know what it says now, period. I don't particularly trust crystal balls.

I know that explanation of polls can make a lot of people's eyes glaze over, but in a nutshell that's how I decided to make North Dakota and N.Carolina pink and put Arkansas safely in McCain's red zone. The other two sites didn't agree with Pollster.com, where I was getting my original data. Majority wins -- except in Alaska where only Pollster is using the Hays/IBEW poll that has Obama ahead and the Anchorage Press poll of the other state-wide races this week paints a picture of a near total collapse of the corrupt GOP machine there. Some new AK polls will be out in the next week or so for the presidential contest. AK stays yellow despite what the experts say because I think it'll be turning light blue by next month. It's wide open on the last frontier.

For anyone who's interested in Ohio (and aren't we all) RCP and 538 have the race as a toss up although Pollster (and me) rate this as leaning Obama. Here's where trends can eliminate the number-bomb an outlier poll can cause to a spreadsheet. Rassmussen put out a poll on July 21st with McCain 10 points ahead, an 8 point jump from their own numbers from the previous month and 12 points off a PPP poll that came out the same week which had Obama leading by 8. Obama was up by 2 in a Quinnipiac survey in the same time frame.

Obama's numbers stayed steady in the two Ohio Rassmussen polls from June and July, so McCain wasn't gaining any strength due to Buckeye fans falling out of love with him, which is almost required nation-wide. There is no way Maverick Man convinces all the fence sitters to fall his way while Obama holds all of his followers even if McCain rips Cindy's shirt off and rides a Harley over a shark tank.

When RCP averages the last three polls and 538 assigns greater weight to more recent polls even though they came out within a couple of days of each other, an outlier poll swings all the numbers wildly when everything's averaged together. Pollster's graphic representation of the trend lines tell a better story in this, case accounting for the outlier but keeping the big picture in mind.


Again, for those who haven't seen my state-by-state electoral maps (cartograms), the size of each state is drawn to indicate the number of votes each gets in the Electoral College.


Barga said...

the states you have on ths map are actually not how they are currently polling, you should update it

And how the hell can Alasak be toss up?

Barga said...


Mark W Adams said...

Gee, update these maps, maybe on a weekly or semi-weekly basis even. What a great idea.

Thank you for your concern, I'll get right on that. This map is almost a week old.

As for Alaska, I believe two or three updates ago I went into great depth on that anomaly. By this time next week the Anchorage Press should be publishing another poll, and things will change again -- unless they stay the same.