.comment-link {margin-left:.6em;}

Born at the Crest of the Empire

Friday, June 20, 2008

Polling

As everyone talks about the Newsweek poll, Obama 51, McCain 36. I thought I'd mention again the top question of 2008.

How are the polls modeled? What is the mix? How do they project turnout by party, gender, age, demographics?

The Gallup Daily tracking has it Obama +2. Rasmussen Daily has Obama + 4.

2 Comments:

  • The big issue, per fivethirtyeight.com, is the weighting of the samples.

    Do you pretend that the mix of D to R is the same as in 2004, or do you account for the record increase in Democratic self-identification and decline in GOP affiliation? If the latter, you are just guessing at the ratio.

    That's why we see two different polls less than a week apart show different results in states like Florida.
    Quinnipiac has a ratio of 47/43 (D/R) and shows Obama up by 4.
    Rasmussen has a ratio of 39/47 and shows McCain up by 8.
    Of course, Quinnipiac has a sample about three times larger than Rasmussen, so I'd consider them more reliable in this case.

    These national polls you linked to don't mean that much, though. It's the state polls that indicate how the electoral votes will break down. The national polls break their samples into regions rather than states, distorting things even more.
    Other polls, such as Newsweek, use weak criteria for "likely voter" status. Others apply strict standards for "registered voter" status in states where you can register at the polls.

    By Blogger Todd Dugdale , at 1:07 PM  

  • I dunno. Those are the big questions, but it's not just Dem/Repub/Independent.

    You've also got to throw in there the Demographic turnouts. Black, Hispanic, Evangelical, how do the age brackets balance out.

    All of those groups are a big question this year, and their weighting makes a substantial difference.

    (And in most polls, they constitute such a small actual number that their reports are virtually meaningless +/- 7.)

    By Blogger mikevotes, at 1:18 PM  

Post a Comment

<< Home