Two Newsweek Polls

[gendjinn]

I read this story on DailyKos and figured the poll junkies here would likewise be interested. Two Newsweek polls one on June 20th and another on July 11th, the first showed Obama 15 points in the lead the second only 3 points. Cue all sorts of stories about how great McCain and his campaign are....

Thing is when you look at the poll methodologies detailed in the DK article with links to the Newsweek details you find that the biggest change was in the demographics of those polled. Like experiments, poll results are only as good as the methodologies employed and in this instance the methodology was tweaked to make a horse race where there isn't one. It's better for ratings you know.

The world of polls isn't all bad, Pollster.com has a great composite of polls and provides trend line graphs for (almost) each state. Checkout Virginia, Michigan, Indiana for instance. Clicking through the states the trends are universally favourable for Obama and unfavourable for McCain.

In summation: Media polls must be investigated before being trusted and composites can help ameliorate the bias of individual polls.

[cyberianpan]

I would suggest this thread ought be appended to this one

cYp

[setanta]

Really useful link, Gendjin. Thanks.

[FutureTaoiseach]

The problem for assessing polling-accuracy is the weighting respective polling companies give to Democrats, Republicans, men (more likely to vote GOP), women (ditto for Democrats), and ethnic-minorities (usually Democrat except Cuban-Americans).

[hiding behind a poster]

The problem for assessing polling-accuracy is the weighting respective polling companies give to Democrats, Republicans, men (more likely to vote GOP), women (ditto for Democrats), and ethnic-minorities (usually Democrat except Cuban-Americans).

Probably a bigger problem for them this time is that they weight the samples according to historic turnout levels among the different voter groups - and this time round its impossible to know just how many of the 50% who don't vote Obama will bring in. All that can be said with certainty is that he'll bring in more than McCain, because he has a Unique Selling Point that will appeal to far more of them than McCain has. Which makes it a pollster's nightmare.

[gendjinn]

I would suggest this thread ought be appended to this one

cYp

Well this was more about the deception Newsweek embarked on in their two polls and the methodology of polling rather than McCain vs Obama polls.

[gendjinn]

The problem for assessing polling-accuracy is the weighting respective polling companies give to Democrats, Republicans, men (more likely to vote GOP), women (ditto for Democrats), and ethnic-minorities (usually Democrat except Cuban-Americans).

It's all about accurately estimating likely voters and constructing the most accurate turnout model. The models are informed by history and do not deal well with tipping points or shifts in voter demographics that this year looks likely to produce.

The rise of cellphone only voters has also frustrated polling accuracy and this is exacerbated this cycle because the millienial generation is a significant component of the cellphone only voters, this age group has historically been unlikely to vote but this year they will buck that trend.

Then there's the factor that heavily influences actual turnout - how energised your supporters are and that looks likely to depress Republican turnout this election.