FL: Rasmussen: McCain 49, Obama 44

[Alcon]

Right now, I would say that, without a major McCain collapse, both OH and FL are lost to Obama.  CO and NV are probably leaning McCain.  PA, VA, tossups, probably a slight lean to McCain.  MI, NH, WI, unknown.  NM, possibly Obama.

PA is leaning McCain already?  Or are you still maintaining that "the polls in PA are wrong" thing?

[Alcon]

Right now, I would say that, without a major McCain collapse, both OH and FL are lost to Obama.  CO and NV are probably leaning McCain.  PA, VA, tossups, probably a slight lean to McCain.  MI, NH, WI, unknown.  NM, possibly Obama.

PA is leaning McCain already?  Or are you still maintaining that "the polls in PA are wrong" thing?

I said "probably a slight lean to McCain."  I'm looking at the trend.  PA has tightened and I suspect a Bradley Effect (small, but there).

Right now, I would say that, without a major McCain collapse, both OH and FL are lost to Obama.  CO and NV are probably leaning McCain.  PA, VA, tossups, probably a slight lean to McCain.  MI, NH, WI, unknown.  NM, possibly Obama.

Hahahaha.

Didn't you say that about McCain getting closer in PA? 

Dude, you can run a mathematical trend via the national polls.  FiveThirtyEight does it.  It comes to Obama +2.  Without weighting for pollster quality (a "pure" weight) it's close to Obama +2.5.  Slight lean McCain?  Based on one Rasmussen poll showing a tie, two days ago, with tracking polls tightening since then?

Edit: Date corrected, rest of post stands.

[Alcon]

OH, VA, NV, and CO are tossups. PA is definite lean Obama, as is MI, WI, NH and NM. FL is strong lean McCain.

Dream on.  OH is solidly McCain.  PA has slipped from solid Obama to toss up.  VA, back to tossup (with perhaps a slight lean McCain).  MI, WI, NH, well, I'll wait for poll data, though I'd expect one to be Obama's.

This is one of the peaks of McCain's performance thus far.  His best poll is +7.  The trend-adjusted average currently stands at +3.

In PA, on the other hand, Obama's best poll was +12.  At one point, his trend-adjusted average was +7.

Yet Ohio is "safe McCain," and Pennsylvania has been a toss-up for a while?

Where is this coming from?

[Alcon]

Dude, you can run a mathematical trend via the national polls.  FiveThirtyEight does it.  It comes to Obama +2.  Without weighting for pollster quality (a "pure" weight) it's close to Obama +2.5.  Slight lean McCain?  Based on one Rasmussen poll showing a tie, two days ago, with tracking polls tightening since then?

Here is what I'm looking at:

2008-09-14   Rasmussen   Tie   47%   47%   2%   4%   5%   500 L   0
2008-09-09   Quinnipiac University O +3   45%   48%   1%   6%   3%   1,001 L   0
2008-09-07   Strategic Vision O +2   45%   47%   3%   5%   3%   1,200 L   0
2008-09-07   Rasmussen O +2   45%   47%   2%   3%   5%   500 L   0
2008-08-26   CNN/Time   O +5    43%   48%   1%   5%   4%   669 R   0

The Time/CNN was before the convention.  Since then, there has been a drop.  I believe with two exceptions prior to the DNC, no poll had Obama winning PA by less than 5.  Since then, well, you can see the results.  Closing numbers.

I think it's a trend.  Today's Rasmussen isn't part of the bounce.

I also (like the Obama campaign), think those undecideds will break for McCain.

It's close and subject to chance, but if the election were held tomorrow, I'd give the state to McCain.

If you plot those relative to national poll averages, and then track accordingly, it comes out to Obama +2.  You're basically saying, "I'm pretty sure if you did the math, the trend would show McCain up."  It doesn't.  Is it possible that McCain has surged in PA more than elsewhere, and we under-estimate this trend?  Yes, if the two recent PA polls were off too.

Is it proven by evidence?  No.  And I think you are (as I think you frequently do) allowing your own cognitive bias to color your analysis.

[Alcon]

Why is it more likely that the PA trend continues McCain-ward, when the national trend seems to have leveled or started trending Obama?

[Alcon]

Why is it more likely that the PA trend continues McCain-ward, when the national trend seems to have leveled or started trending Obama?

I'm not saying it's more likely, but I am saying it is, to an extent independent[/i] of the national polls. 

But if your prediction is that PA is more likely to have an edge to McCain than Obama, that's exactly what you are saying.

Unless your assertion is that the polls aren't picking up on some phenomenon (secret cache of voters, heavy late breaking in undecideds, racial anxiety) that you think likely will make that two-point swing.  But isn't that breaking a self-attributed rule of yours?

[Alcon]

I didn't say that.  But is Pennsylvania more likely to move against the national numbers than with?  That's required by your assertion, since McCain hasn't led in a poll there, and a trend calculation with the national numbers comes out to Obama +2.  You've explained why McCain could be ahead fine; you haven't addressed why he probably isn't.

But either way, you have to accept that you are using more than raw, or even trend-adjusted/reweighted polling data -- you are making the claim about an unseen political force.  That's risky, but valid.  But I thought one of J. J.'s rules applied to claiming the polls were not picking up on a certain phenomenon, in general?  No?

[Alcon]

J. J.,

You spoke of a trend.  If your claim isn't that the poll numbers are off, and that you thought McCain is probably leading, then you'd have to be assuming an opposite trend from PA versus the national numbers.  And you'd have no proof of that.

Now, I'm seeing that your claim almost entirely rests on the Bradley effect.  I assumed by, unseen, you meant unseen by polls.  By whom, exactly, would these voters be "unseen"?  J. J.?

[Alcon]

Right.  If you take the Rasmussen poll alone, assume that it must be right, then McCain is likely ahead.  But if you average out polls, even with the other that came out today, it shows that Obama is likely ahead.  That includes mathematical adjustments for sample size.  You would have to heavily over-weight Rasmussen to get McCain leading ahead in a trend-adjusted poll line.  In other words, you have to claim something beyond the polls, here.

PA and the United States:  They may be mostly independent of each other, but not purely so.  I mean, Pennsylvania is a component part of the United States.  They're not totally independent, and I do imagine they're more likely to correlate than to inversely correlate.  I'm arguing likely here, not definite.  I think it's more likely that PA moves with the nation than directly against, even if the latter is possible.

Again, "unseen" by whom?  Polls, or what?  There are a great number of claimed effects that are not unseen--the cell phone factor, for instance, which is even quantifiably provable.  It's not that I deny the possibility of a Bradley Effect (even if I don't believe in adjusting polls a few points for it), but I want to know how J. J.'s Second Law of Elections defines "unseen."

(What's J. J.'s first law, again?)

[Alcon]

People without landlines are more likely to be young and urban.  Not tons of them vote, but plenty do.  They're increasing rapidly.  I'm one of them--I vote.  It's going to be a problem eventually.  I'm not convinced it is now, but by your rules, you can't just dismiss it.

In any case, Obama's trend-adjusted polling margin in PA is >0.5, so...?

[Alcon]

A gap between PV locally and nationally has nothing to do with local movements correlating with national trends...I didn't claim a perfect correlation, just one.  Besides, you just proved my point.  The national margin moved right, the PA margin also moved right.  That was my point...that event being more likely than the inverse.

So, your rule only applies to situations where people are being polled, but you claim they're giving inaccurate answers?  It doesn't apply to claims that people aren't polled at all?  Um, why?  That doesn't really make sense.  All things being equal, if a group has x property and isn't polled, that's going to affect the poll results more than if they have x property, are polled, and lie at <50% net rates.