Monday, September 15, 2008

Accuracy of polls

A friend wrote to me:

Happened upon this article while perusing the NY Times election map and poll data today:


I read this article to mean that the polls listed at the NY Times web site are quite responsibly designed. I'd want to know the agendas of people who criticize the design of these polls. (We all know about Truman's "defeat" predicted right up to election day in 1948; I agree that polls can be wrong.)

Today's viewing of this electoral map and polls data was heartening:

The national polls with post-convention (post-Palin) data show a very close election and McCain gaining. But the big Gallup poll bounce that put McCain ahead is losing steam and the red lead is narrowing quickly.
A first glance at the electoral vote map also shows McCain with recent gains. But examination of some state-level data suggests that the map overstates those gains and understates Obama's situation:
Michigan (17 electoral votes) in a recent poll shows Obama gaining in a close race. The trend is acceleratingly blue, but the state is shown as a tossup.

Ohio (20) also shows a trend toward Obama and a decent September lead of 49-45 (error margin 3%). It's also shown as a tossup.

If we add these 37 electoral votes to the 238 already shown as at least leaning to Obama, he is elected. I really can't imagine Michigan voting red in November. The tougher the Republicans care to be about federal loans to the auto industry, the more that will be true. McCain then must win Ohio -- not easy for him this year, because of Republican scandals in Ohio.

I read the September polling data in Minn (10), Wisc (10) and Iowa (7) as all solidly blue. But all three are shown as just leaning to Obama. This suggests that his leaning-to states are pretty reliable. [Obama's lead in Pennsylvania (21 critical electoral votes) is thin, but I think economic issues will carry that very industrial/urban state.] New Hampshire (4) is shown as tossup but a Sept poll has Obama with a decent lead in a traditionally Republican state.

Missouri (11) polls show McCain ahead but not beyond statistical error. Other states in that condition are shown as tossup but MO is shown leaning red. That may actually be a mistake that overstates McCain's standing.
And the latest Wall St news from this weekend cannot help the Republicans. Is expected the financial carnage will linger for some time as losses are finally made public and firms held accountable. Independents should trend blue-ward as they grow aware of the wreckage wrought by Republican greed.

***

Back to my own ramblings:

I surfed over to the Gallup website, but didn't want to wade through a lot of stuff to find how their polls were constructed. But it does leave me with the question: Do the polls constructed by reputable organizations try to balance party representation as some of my sources claim? If not, why do those stories get told? Is it a way to comfort us before the election? Does it not blind us and allow us to avoid doing what we can?

1 comment:

  1. "Ohio (20) also shows a trend toward Obama and a decent September lead of 49-45 (error margin 3%). It's also shown as a tossup."

    That's because of the math. 3% error margin means either-or-both figures could be inaccurate by three percent; in this case, with a 2% error margin it'd be tossup, 3% means it's impossible to call - because Obama could actually be on 46% with McCain on 48% (3% different from the figures, note).

    ReplyDelete