Why aren't the Polls Converging on a number?

[kennetha]

Though the weighted and unweighted averages of the polls have been pretty stable, the different polls also remain spread about those averages pretty much as they have been for quite awhile, I think. With early voting running apace and with potential voters intentions probably firming up, it seems to me that you might expect the various sampling methods, to the extent that they are reliable, all to start converging on a model of the electorate. And you would think that once they started to converge on a model of the electorate, they would start to converge in their numbers.

But that doesn't seem to be happening. The polls today seem to be nearly as spread out around the average as the polls over the past several weeks have been. I'm not sure what this lack of convergence means, exactly. Maybe it means that the models of the electorate that the polls used aren't sensitive enough to change as votes actually get cast and intentions start to firm up. But that would be pretty astounding. Wouldn't it? I'm no stats rat, so I'd appreciate any enlightenment.

I suspect that we may basically be flying blind in a way with these polls and don't have any real basis for judging whether it's a 3 point race, a 5 point race, or an 11 point race.

[onenote]

1. because they generally don't

the things that cause them to diverge don't magically go away the closer it gets to election day.

[kennetha]

5. But am I wrong to think

that as more people (a) actually cast votes and (b) firm up their intentions about whether to vote and who to vote for, then the models should be able to pick that up? You could think that as they pick that information up, the polls would all start to zero in on a number. Not that there wouldn't be some divergence, but that the divergence would be reduced.

[Drunken Irishman]

2. Rarely do polls ever really agree.

But the common theme is Obama generally leads from 3-10 points. That's a pretty good thing for Obama.

[CreekDog]

3. they are samples

they don't converage, but they do create an average.

[NRaleighLiberal]

4. Some other factors - lots of early voting. Cell phones (young or first time voters).

The polling is trying to catch up with the times, which are changing!

Fear not!

[Frumious B]

6. A lot of the disparity is probably due to different modeling of likely voters.

You can see it best illustrated in Gallup's two different sets of likely voter figures.

[scheming daemons]

7. The daily trackers are all converging on a number around 5 or 6

[grantcart]

8. because of the youth and 'relapsed' voter quotient

Each poll is giving this group a different weight

[kennetha]

10. I get that

But do they not adjust their turnout models in light of actual voting behavior (i.e. early voting) and firmed up intentions as the election gets closer? You would think a sensitive sampling technique would be able to pick up that group X is voting in higher numbers than previous samples anticipated or that Group X is firming up its intention to vote or that Group X is firming up its intention to vote for candidate Y.

I would think that as more information became available good sampling techniques would all adjust their models in light of that information and so would tend to converge.

But, hey, the last time I took a stats course I was a 20 year old college student and that was a LONG time ago,

[ShadowLiberal]

9. Actually some daily tracking polls like Gallup have been consistent for over a month

Don't forget there's a margin of error in all polls.

In the Gallup daily tracking poll both Obama and McCain have shifted 2% one way or the other (meaning they've both been in the same 4% of the poll for over a month, though of course their 4% is at a different part in the poll).

[Schulzz]

11. I think national polls are less likely to have the same results than state polls.

National polls are polling a much smaller percentage of people, compared to state polls. For example, a poll of 600 voters in Ohio is probably more reliable, than a poll of 1000 voters nationwide.