We need to run like we're 10 points down. Don't think for a moment that this could be a landslide.

This sound a lot like Schrodinger's cat.

Washington State just had it's first primary in ages where voters were required to select a partisan ballot.  So voters had a choice to vote for Dem. candidates only, Rep. candidates only, or soley in non-partisan races.

600,502 people chose Democratic ballots, and 460,891 people took Republican ballots.  Democrats took 56.5% of 1,061,393 ballots cast.

Obviously which races were competitive across the state affects this.  For instance in Jefferson County there was a contested Democratic primary for County Prosecutor that would essentially be decided in the primary with no Republican in the race.

But across the State it looks like at the very least the Democrats are more energized.

PortDork

This post doesn't make sense at all. These polls are allegedly measuring the same reality. Some are more accurately measuring than others. There aren't two realities. Clearly the issue comes down to sampling and criteria for "likely voters." Just using these polling numbers as an argument to fire up the activistists might be nifty but it's not pertinent to the statistical argument of polling.

...I struggle to buy the argument that both the close and the big-margin likely voter polls are 'right'.

How can this not be an artifact? (Different definition of likely or something like.)

If all the likely voter polls are genuinely polling equivalent samples, a pattern like you found - the two clusters wide apart - surely wouldn't occur?

Are there precedents? If not, why would the phenomenon occur this year?

Time to hit the hay...

Obviously, an exit poll taken tomorrow would be very accurate, assuming there were any polling places to exit from. Just as obviously, only one of these polls is right--as far as tomorrow's nonexistent election is concerned.

I do think the makeup of the electorate will tell the story, and therefore the pollsters with the most accurate likely voter screens will predict the results most accurately.

Pollsters only intrepret elections. The point, however, is to change them.

There are two things, not one, that need to change.

First, as you say, activists have to crank up the intensity. If you work 16 hours a day, and the GOP crazies work 17, you will surely lose. And make no mistake, they really know how to work.

Second, the era of bad feeling that starts with the Rahm and Howard show and runs all the way down to pissants like me tangling assholes with Hillary-baiters--all this divisive shit has to stop.

It's okay to hate Hillary and even Bill Clinton. It really is. But PLEASE shut the fuck up for a while.

For my part I'm going to go on the right-wing sites and claim I'm a loyal Republican but Nancy Reagan is a cunt and her husband shit his pants daily for the last two years he was in office. That being the case, why bother to vote?

It's okay to hate Hillary and even Bill Clinton. It really is. But PLEASE shut the fuck up for a while.

At least you said please.

Now to the subject:

- In my NY-24 race, both Arcuri(D) and Meier(R) report polls claiming huge advantages for themselves.  While the seat is generally reported by the pollsters and pundits coast to coast as the single Republican seat in NY likely to become Democratic, I would bet on the troglodyte who is sadly much worse than the retiring Republican.  If I was into gambling, I would rather put my money on Eric Massa even without the spread. Eric is given little chance by most.

The only Democrat that anyone here can ever remember being elected as a representative from this area was elected during the "Truman landslide."  The area voted overwhelmingly against Dewey, mainly because Dewey was from downstate.

Dewey's anti-coattail effect may well be true of Hillary Clinton in this district though 9/11 has palpably lessened the hostility to NYC.

Why the heck wouldn't the polls show all this?  What is wrong with the CI (Confidence Interval) provided with polling data?

People are not digits.  They are not even clones.  Formulations utilized to reduce "error" are mostly ad hoc rather than deeply analytical but all are doomed to fail to some degree.  The one thing that polls can do is change the dynamic of the race by creating a bandwagon effect.  Does even the lowest grade moron not understand why politicians rush favorable polls out to the public?

You trying to elect Republicans here, stevehigh? :-)  The best way to do so around here is to enforce a ban against criticism of Hillary Clinton in my opinion.

Freedom is a very good thing, stevehigh.

Best,  Terry

Ben Tripp has a sobering article at Counterpunch:

http://www.counterpunch.org/tripp0916200 6.html

I can foresee the Republicans keeping control.  There are still only a few slam dunk Democratic gains emerging (TX 22, AZ 8, IA 1, that's pretty much it).  And I don't see how the Democrats are dealing with any of the three factors Tripp brings up.

Chris, I wouldn't get too caught um in the Schroedinger's cat analogies.  It really is far more likely that the discrepancies are due to the "likely voter" models being used.  The fact that the numbers are more similar for the more objective "registered voter" sampling would tend to confirm the simpler explanation.

Sorry, it's not equally probable. I deal in odds every day and have for two decades. I get annoyed when people who have no background in the matter simply guess.

You didn't specify what surpassing the Republican gains of '94 means, but if you're talking in terms of +40 or +50 seats or anything close to that, the odds would literally be hundreds if not thousands to one. The odds of a small Democratic gain, let's say 5 seats or less, would be significantly lower, perhaps 3/1 or slightly less than that. You could only get big odds if you wanted to wager Republicans would gain seats, and huge odds if you started declaring Republicans would gain 10 seats or more.

The base number right now would be roughly +12. The odds would would move slowly in either direction as you move away from the base, particularly upwards. But once you move 10 or more seats away and particularly 20 or 30, the odds change exponentially. I'm talking about true odds. Bookmakers are notorious for ripoff prices especially at high numbers. Teams that should be thousands to one are offered at 200 or 300 to 1 because they oddsmakers know the suckers won't know true odds and jump on the low ripoff numbers. When I worked in sportsbooks and later consulting oddsmakers we used to joke all the time that the Excel math put the true odds at let's say 5000 to 1 but no sense putting up more than 300 to 1 because you'll get plenty of square money at that price. The Chicago Cubs, in particular, are notorious for drawing ridiculous sums of future bet money no matter how absurdly low their odds are set.

If I see odds on a bracket of House or Senate pickup, like Democrats +5 to +9 in the House, for example, I'll post it here. I've seen that previously but the offshore outfit that offered it has changed owners and has not dealt political odds so far this year.

I am worried about dem turnout.  We need turnout because they will get theirs out come hell or highwater...

3 things on this. (1) there are a lot dependencies and "priors" to account for when interpreting poll results. i'd urge anyone who's serious about this to read up on Bayesian statistics, quantum uncertainty has nothing at all to do with this, even in a metaphorical sense. i remember a good number of cycles where polls have favoted Dems, only for them to fall short by a critical few points. we have all kinds of information available to us on top of polling information.

(2) i agree with RickD that likely voter models explain this. however, Chris is absolutely right that the partisan differences are suspect. the media have been totally fucking naive about interpreting poll results as somehow being empirically objective. if i had to guess, i'd say that partisan pollsters either selectively choose factors or massage parameters and screening questions in order to inject expected bias. because the media are so naive, the release of biased polling data itself is part of the persuasion war. a campaign can gain momentum by releasing, or playing up, a poll that shows them with gains.

(3) confidence intervals for polls are probably underestimated. chi-square makes assumptions that are violated when using likely voter models. i'm going out on a limb here because i'm not a pollster, but i do know statistics, and i can't see how MOE is the same when you start making assumptions about who's likely to vote.

(4) by overestimating gains this year, Dems run the risk of unrealistic expectations in the media that will significantly influence post-election spin. confidence is good, but i think that confidence is much better spent on talking about Dem leadership than talking about Dem returns. say Dems gain 9 seats in the house. not bad, but next-day headlines may read "Dems defeated" and that will seriously influence voter perception, which bears on 2008.

here's a post-election spin i'd like to see, no matter how many seats we gain: "Dems finally competitive on national security issue"

I bet the Group Republican is based more on prior turnout than Group Democrat, while Group Democrat is based more on intensity and stated likelihood.  As steveolson said, Undecideds are a big factor as well.

As far as Jagakid's analysis, he clearly hasn't been reading about the concept of a "wave" election, as Charlie Cook has researched and written about.

Why I believe it's closer to the 0-3% differential:

1) Likely voters are now being increasingly factored into these polls

2. Democrat registration is higher than Republicans (a recently increasing trend, and one of the few bright sopts for Dems)
   
3. Republicans have more than proven that they are FAR superior at getting out the base
   

Slightly off topic, but check out the interview Two Political Junkies just did with Jason Altmire, the Dem candidate in PA-04. Altmire explains the genesis
of the recent Anzalone Research poll he had done showing him within MOE of Hart.

http://2politicaljunkies.blogspot.com/20 06/09/jason-altmire-interivew.html

Interesting also are Altmire's anecdotes of voters
saying if he's running against Missy Hart, he's got their votes. That tells me the LV models that measure enthusiasm rather than past voting record
of the pollees will be the ones who are right this year. The past voting models probably favor Republicans while the enthusiasm models favor Dems.

It might be interesting to go back and look at 1992 and 1994, both true change elections, to see
which pollster's LV model came closest to the actual result. That might give a good clue to who's got it right now also in, IMHO, a change year.

There are the districts and states that are likely to swing, and those that are not. Polling is imprecise and the generic ballot's weakness is that we don't know where those respondents were polled, in the America that will decide this election or the America that won't.