pageok
pageok
pageok
Presidential Race Tightening in Two Tracking Polls.

The two tracking polls that I follow most closely are showing a slight tightening of the Presidential race (though some other equally recent polls are showing a huge lead for Obama).

The Rasmussen Poll of likely voters shows Obama up by 4% (50 to 46%). And the Gallup Poll of likely voters (using their traditional likely voter model) shows Obama up by only 2% (49 to 47%).

josh bornstein (mail) (www):
I find this interesting, and not that surprising. Most people have been expecting the polls to tighten a bit. I think the real crucial stage will be the polls that come out over the weekend (ie, including the 2-4 days after the 3rd debate).

If the numbers continue to close, then McCain has somehow turned a corner, in spite of his debate performance and the continued bad economy.

On the other hand, if these later polls reflect the 'snapshot' right after the debate, then I would expect the polls to stay where they are now (around 5-6, on average), or even inch up a point.

Of course, this is not dispositive, since the electoral math still does not look promising for McCain. But if he can get the conversation moved to "look, McCain is making one of his patented comebacks," then who knows . . . .
10.16.2008 4:15pm
jukeboxgrad (mail):
If you think this is a significant shift that is generally overlooked, you can go make some money over at Intrade.

McCain is currently trading at about 15. That's a drop of roughly 50% in roughly a week. He hasn't been that low since before he won the nomination.
10.16.2008 4:17pm
J. Aldridge:
Only polls that matter are the one of each state both need to win.
10.16.2008 4:18pm
Ben P:
Could be something, could just as easily be nothing.

In my opinion this is at the initial glance no different than the "Obama up by +14 in Poll" that was ricocheting around the internet yesterday.

Even "careful" polling will occasionally hit a sampling bias. In that case the +14 was still inside the margin of error of the +/- 5% polls that show him up by 8-9 points.

He's not up by 14%, polls reliably show him up by 5% to 8%.

The true measure is how continued polls show him over the next week.
10.16.2008 4:22pm
Obvious (mail):
In the latest Real Politics summary of polls, there are also signs the race is tightening. Last week, Obama had a lead over McCain that was uncountably infinite. This has narrowed now, with the former community organizer having only a countably infinite lead over the former prisoner of war
10.16.2008 4:25pm
David Larsomn (mail):
In other news, the latest polls show that 84 per cent of Americans want the Federal government to be their mommy AND their daddy, that 77 per cent believe that they would personally be better off under a communist government, that 81 per cent believe that "if I can't have it nobody should be allowed to", and that 86 per cent had never heard of "Chairman Mao", "Joseph Stalin", or "George Orwell."
10.16.2008 4:34pm
A.S.:
I am curious why Prof. Lingren prefers to follow those two specific polls, as opposed to the RCP average?

Polls are quirky. Some have biases and, even without biases, some are just better than others. And of course polls have flukes. I find that the best way to limit the problems with using just one or two polls is to follow the RCP average. With the number of polls RCP uses to calculate the average, the biases of one poll or another are likely cancelled out, and the problems associated with bad polls and fluky results are minimized.

RCP has Obama with about a 7-point advantage. With that advantage, it doesn't pay to start looking at the state by state polls - he's going to win no matter what. If McCain get the Obama advantage down to about 3 points, I'll start to look at the state-by-state polls again.
10.16.2008 4:45pm
Lior:
Excuse me, but why should we care about national polls at all? Until and unless there is national voting for president, national opinion polls of "likely voters" are completely irrelevant.

If you want to predict the winner of the election, use electoral college models with state-by-state polls. If you want to get a sense of the national view then "likely voters" are not necessarily the people you want to poll. This introduces additional unnecessary biases.
10.16.2008 4:45pm
musefree (www):
James Lindgren -- It seems much nicer to follow a weighted aggregate of all the tracking polls rather than just two. Less volatility and more accuracy. I cannot recommend Five Thirty Eight highly enough.
10.16.2008 4:45pm
Kelly (mail):
As I understand it, the "traditional likely voter model" does not count most newly-registered voters - assuming that they are not likely to actually go to the polls. I think the chances of that being an accurate assumption this year are very, very low.
10.16.2008 4:53pm
Uh_Clem (mail):
Last week, Obama had a lead over McCain that was uncountably infinite. This has narrowed now, with the former community organizer having only a countably infinite lead over the former prisoner of war

But what I want to know is whether the lead passed through an intermediate state in between the two, or whether it jumped right from uncountable to countable.
10.16.2008 4:54pm
Arkady:

But what I want to know is whether the lead passed through an intermediate state in between the two, or whether it jumped right from uncountable to countable.


It went from aleph-1 through aleph-dull to aleph-null.
10.16.2008 4:59pm
commontheme (mail):
musefree - your post assumes that Mr. Lindgren is interested in accuracy, not merely keepin' hope alive for McCain.
10.16.2008 5:08pm
JosephSlater (mail):
RCP poll average as of now is O +6.8;
Pollster average is O +7.9;
InTrade has Obama at around 86%;
Electoral college math looks really tough for McCain even if the race tightens a couple of points;

And all this reflects polling done before the third debate which almost every semi-scientific poll indicated Obama won.
10.16.2008 5:16pm
Uh_Clem (mail):
I am curious why Prof. Lingren prefers to follow those two specific polls, as opposed to the RCP average?

There's nothing to be curious about, it's just plain old cherry-picking.
10.16.2008 5:21pm
EricPWJohnson (mail):
Interesting that the generic ballot shows the dems losing a whopping 9 pts in a week as well
10.16.2008 5:24pm
Anon21:
I think there is some rationale for following the Gallup and Rasmussen trackers specifically--they have by far the largest sample sizes of any of the trackers. However, the nature of tracking polls is such that it's safer to assume that any apparent movement is simply statistical noise until it's sustained over a period of 5-7 days.
10.16.2008 5:29pm
ifoughtthelaw (mail) (www):
I'm really hoping McCain wins the popular vote and Obama wins the electoral vote.
10.16.2008 5:36pm
darelf:
I think it's important to stress right before the election that historically polls have overestimated the Democrat candidate's support by between 6 and 10 points. If McCain is trailing by 5 in the polls that probably means he is winning.

Another outrageously wrong poll point is the so-called "youth vote". They just never seem to get around to actually materializing into actual votes. Historically speaking.
10.16.2008 5:39pm
musefree (www):
@Anon 21

Yes, if you specifically insist you must follow only two polls, it makes sense to use the Gallup and Rasmussen ones. But when there are many other polls available, it makes sense to take a weighted average, as Five Thirty Eight does. Pollster on the other hand takes a non-weighted average, while RCP takes a non-weighted cherry-picked average.

@ifoughtthelaw

Less than 5% chance of that happening. Still that's probably likelier than McCain winning.
10.16.2008 5:41pm
musefree (www):
@ dareself

Please do the research before posting such falsehoods. I have gone through the (average) polling predictions of the last 6 elections and the average error has been less than 2% and it has gone in both directions.
10.16.2008 5:42pm
JosephSlater (mail):
More good news from Obama in state polls today:

(CNN) — John McCain continues to lose ground in three battleground states, new CNN polls of polls out Thursday suggest.

According to the new surveys, McCain is still slipping in Pennsylvania, Florida, and even Colorado — a state that hasn't voted for a Democratic presidential candidate since 1992.

In the new Pennsylvania poll of polls, McCain now trails Barack Obama by a 13 point margin, 53-40 percent. That deficit is up from 12 points in a poll of polls released earlier this week.

In Florida, the state that handily voted for President Bush in 2004, McCain trails Obama by 4 points in the latest CNN poll of polls, 49-45 percent. That's up from a 3 point gap there over the weekend.

And in Colorado, a longtime Republican stronghold where McCain has held the advantage for months, the new CNN poll of polls shows Obama on top by 6 percentage points, 50-44 percent. Earlier this week, Obama held a 5-point lead over McCain.


If Obama gets Florida OR Colorado it's over (or any one of VA, OH, NC, NV, IN, or MO).

And dareself -- no, that's entirely wrong.
10.16.2008 5:48pm
Uh_Clem (mail):
Only polls that matter are the one of each state both need to win.

There is no such state, i.e. a state that both need to win. The states to watch are:

Virginia
North Carolina
Florida
Ohio
Indiana
Missouri
Colorado

McCain needs to win all seven. Obama only needs to take one. Good luck. You'll need it.
10.16.2008 5:48pm
Jiffy:
musefree:

According to 538, the chance of McCain winning is slightly higher (4.9%) than the chance of Obama losing the popular vote and winning the election (3.2%). But it's close!

I'm very impressed with 538, by the way, but not enough to take Obama on Intrade, which shows a lower percentage chance of an Obama win (about 86% vs. 95%).
10.16.2008 5:52pm
Kazinski:
Uh-Clem:
There's nothing to be curious about, it's just plain old cherry-picking.

I happen to like the RCP average myself, but any serious stathead will tell you that the RCP average is as likely to provide a "sum of all errors" as it is to give you a correct answer. Lindgren happens to be a serious stathead, so that is likely the answer. You are more likely to get a correct answer with a consistent methodology, like either weighting for party affilliation or not, or counting likely voters or not, but don't mix the two.
10.16.2008 5:52pm
musefree (www):
@ Jiffy

Only 3 weeks or so back, Intrade had a 50 on Obama (at a time 538 had close to 80). That's when I put in some money :)

(Actually not on Intrade, which did not accept my Credit card, but on another betting site, which gave good odds)
10.16.2008 5:54pm
musefree (www):
@ Kazinski

I know a little bit about statistics, and that's plain wrong. Perhaps you are confusing errors with absolute errors. When errors can go in either direction, taking a large number of similarly accurate polls and averaging them gives you a better estimate (a simple consequence of the law of large numbers). In practice the polls are not equally acurate, so you take a weighted average based on past accuracy.
10.16.2008 5:58pm
JosephSlater (mail):
Uh_Clem:

All true, but you should add Nevada, where Obama has a small lead. Nevada's 5 EVs plus the 264 that Obama has in the bank = 269, and given the likely composition of state House delegations, a tie would go to Obama.

In short, McCain needs all eight current "battleground" states, and he's losing in almost all of them currently.
10.16.2008 6:02pm
JosephSlater (mail):
I should add, if we're listing close states McCain has to win, see also West Virginia (5 EVs, recent polling, M up only 2 points).

And FWIW, recent polls have Obama ahead in North Dakota of all places (although with only 3 EVs, it's not likely to matter much).
10.16.2008 6:13pm
Andrew Janssen (mail):

In short, McCain needs all eight current "battleground" states, and he's losing in almost all of them currently.


And making the probably false assumption that McCain has a 50/50 chance of winning any one of those eight battlegrounds, his chance of winning all eight is only .39% (1/256).
10.16.2008 6:13pm
A.S.:
538.com's weighting is through some made-up formula that Nate decided to use. That seems to me to be highly subject to manipulation and doesn't strike me as likely to produce any more accurate results than simple averages like pollster.com or RCP. The difference between pollster.com and RCP, as far as I can tell, is that pollster.com includes the Daily Kos poll and the ARG poll. I don't think that including the Daily Kos poll makes the result more accurate. So, to me, that leaves RCP as the best average.
10.16.2008 6:14pm
Syd Henderson (mail):
Except 538 was showing Obama going ahead in Missouri and North Carolina before he actually did so, so they have some predictive power. Indiana's bouncing from party to party and he now has it pink. You can also look at all the polls he's using and make up your own mind. I think he's a little bit low on the popular vote and about right on the electoral vote.
10.16.2008 6:28pm
Syd Henderson (mail):
The GOP's pulling their funding out of the Louisiana Senate race. This is going to be the second election in a row the GOP failed to capture a Senate seat held by a Democrat.
10.16.2008 6:31pm
MS (mail):
A.S.,

Nate's weighting is "based on that pollster's historical track record, the poll's sample size, and the recentness of the poll. More reliable polls are weighted more heavily in our averages."
10.16.2008 6:34pm
musefree (www):
"538.com's weighting is through some made-up formula that Nate decided to use"

Hate to sound like a 538 agent, but the formula is not made-up. It all there in the FAQ. I went through it, and think it is plausible.

As for RCP, it omits several other polls. There was a discussion in 538 on this topic.
10.16.2008 6:34pm
mmaletic (mail):
Regarding the complaint that national polls don't matter: actually, they're a pretty good predictor of what's going to happen in the swing states. Swing states are that because they closely reflect the demographic make-up of the nation as a whole. So looking at a national poll is a fairly good indicator of where the swing states will be, I'd assume.
10.16.2008 6:37pm
LM (mail):
Lior:

Excuse me, but why should we care about national polls at all? Until and unless there is national voting for president, national opinion polls of "likely voters" are completely irrelevant.

When reliable national polls have a candidate leading by a couple of points over the margin of error, it's implausible that that candidate would trail in in a state by state electoral college poll. In this election, even more so given Obama's electoral college advantages.
10.16.2008 6:55pm
A.S.:
Hate to sound like a 538 agent, but the formula is not made-up. It all there in the FAQ. I went through it, and think it is plausible.

Sure it's plausible. But there are lots of things he could do that are plausible. His weighting by pollster quality seems helpful (although I'm not that sure about his calculation of quality), but the "recentness" fudge factor seems a bit arbitrary. I prefer just to look at the polls out in the last few days, like RCP and pollster do.
10.16.2008 7:12pm
A.S.:
That all said, I think there is a 99.999% chance that Obama wins, and have thought continually since the spring that his chance of winning is 99%+. (Too bad I am not a betting man...)
10.16.2008 7:15pm
musefree (www):
"Sure it's plausible. But there are lots of things he could do that are plausible. His weighting by pollster quality seems helpful (although I'm not that sure about his calculation of quality), but the "recentness" fudge factor seems a bit arbitrary. I prefer just to look at the polls out in the last few days, like RCP and pollster do."

Plausible was a bad choice of word by me.

But anyway, even forgetting the recentness fudge factor (which I don't find arbitrary) there is another important point that makes 538 great. It is the trendline correction. Let's say that states A and B have similar demographics and are polled on October 5. Over the next week, Obama gains several points nationally. These gains are also reflected when state B is polled on Oct 12. Then it is natural to expect that the next time State A is polled, it will show much higher than before. If you are just averaging the last few polls, and state A is not polled as often, you miss out on this information.

It is adjustments like these (and other small touches, like house-effects, incorporation of demographics in the model, an eye at history, etc) that give 538 its predictive power.
10.16.2008 7:32pm
Asher (mail):
That likely traditional voter model is useless. Their expanded likely voter model, which factors in increased turnout from minorities and young people, shows a 6-point lead. The traditional discounts the stated intent of minorities who say they'll vote because in the past, they haven't turned out at high rates. Of course, in the past, a black candidate wasn't on the ballot.
10.16.2008 8:41pm
Aleks:
Re: I think it's important to stress right before the election that historically polls have overestimated the Democrat candidate's support by between 6 and 10 points.

Please substantiate this bizarre claim. At no point in autumn of either 2000 or 2004 did the polls show such a thing in favor or either Gore or Kerry. Gore actually did slightly better than the polls predicted, Kerry slightly worse-- emphasis on slightly in both cases. The polls have generally been very accurate predictors of victory during the month or so before the election, though not necessarily of the exact margin.
10.16.2008 8:56pm
James Lindgren (mail):
538 finds the new Gallup Likely Voter model (about 66% turnout) more plausible than the traditional Likely Voter Model (about 60% turnout). While the 2% result in today's latter model may be doubtful in itself, the traditional Likely Voter Model, which considers past voting behavior, is likely to be a better model overall on most days (even of not today).

In 2004, the Gallup traditional LV model was very close to the actual outcome, while yet a third model, the registered voter model, had Kerry up by 2%.
10.16.2008 10:28pm
Dionysius (mail):
The numbers in Virginia and Pennsylvania are suspect, I think. The polls were taken just as the first shocks of the coming hard times were felt and people's anxiety got in the way of their ordinary prejudices. The numbers were good for Obama even in southwest Pennsylvania! Wait a while. People will get over the shock and return to their residual prejudices. Maybe Obama will win in Pennsylvania and Virginia, but it will be by a squeak, if at all.

As for Missouri, from which state this note comes, I can't see this state voting for a black man for president.
10.16.2008 10:54pm
Hoosier:
My poll of 5 and 10 year old boys in Indiana who live in my house is currently at 50-50. The race for our family room is too close to call.

Candidates are campaigning hard for the vote of the undecided 2 year old girl. (My guess is that Sarah Palin will soon make a campaign swing through here dressed as "a princess." That would almost certainly tip the vote to McCain/Palin, according to Barrone.)
10.16.2008 11:12pm
Asher (mail):
But Prof. Lindgren, 2004 was 2004. Kerry didn't have much of a turnout machine. Obama's got the best one we've ever seen. I just would argue that this is a 7-point race, judging by most polls, and there's nothing that McCain can really do to change that. He's clearly decided to not make a big new policy splash. He's not a brilliant campaigner anymore. This isn't 1948, and Obama isn't sitting on his lead by any means. Only a terrorist attack, God forbid, can save McCain.
10.16.2008 11:17pm
KeyComments (mail):
Uh_Clem, in your 3:54pm post, you posed the CONTINUUM HYPOTHESIS!!!!!!!!!!!!!
10.16.2008 11:20pm
CB55 (mail):
In American political jargon, an October surprise is a news event with the potential to influence the outcome of an election, particularly one for the presidency. It is so called because the presidential elections are held in early November, and therefore events that take place in October have greater potential to swing votes. "Historically, news outlets avoid investigative pieces critical of candidates within days of an election to avoid appearing partisan." [1] Particularly since the 1980 election, the term has been pre-emptively used to discredit late-campaign news by one side or the other.

The term usually applies to the acts of a sitting president, especially in military or foreign policy matters. But it can also apply to news stories unfavorable to the incumbent administration. - Wiki

False flag operations are covert operations conducted by governments, corporations, or other organizations, which are designed to appear as though they are being carried out by other entities. The name is derived from the military concept of flying false colors; that is, flying the flag of a country other than one's own. False flag operations are not limited to war and counter-insurgency operations, and have been used in peace-time; for example, during Italy's strategy of tension. -Wiki
10.16.2008 11:28pm
KeyComments (mail):
Andrew Janssen,
you calculate a 0.39% chance of McCain winning in your scenario, but you forget that those joint probabilities are almost certainly not independent so you cannot just multiply them together. If McCain can win any one of them, his chances of winning any or all of the others rise potentially significantly.
10.16.2008 11:31pm
Cold Warrior:
I second the compliments for fivethirtyeight.com

Those of you who are serious baseball fans are probably aware of some of the fine work Nate Silver's done at baseballprospectus.com

Silver has taken the same rigorous methodology he applied to baseball pennant races and applied it to the electoral college map. And he's pretty clear about his methodology; that transparency is admirable.

Under Silver's system, McCain now appears to have somewhere around a 5-6% chance of winning the electoral vote. Slim? Yeah, sure. But to continue the baseball analogy, I think McCain's chances of winning are roughly equivalent to this situation (I stole this from insidethebook.com, a great baseball website started to promote an even better book):

Your team down 1 run, batting in the bottom of the 9th, 1 out, runner on first base.

If you've watched as much baseball as me, you're not about to head for the exits in this scenario ... not even close.
10.16.2008 11:35pm
Asher (mail):
If you've watched as much baseball as me, you're not about to head for the exits in this scenario ... not even close.

Right, because:

1) Baseball fans don't quit on their team when they still have a 5% chance of winning the game, and

2) You've probably watched thousands of baseball games. So you've actually seen many games where your team's won in that situation, and think it can happen again. In a way, all your experience leads you to overestimate the odds, because of course you don't think about, when your team's in that situation, all the hundreds of times when they were in the same spot in the past and they lost. Presidential elections, on the other hand, you've probably only seen about 5-15 of.
10.16.2008 11:58pm
Not Danny Darwin:

Your team down 1 run, batting in the bottom of the 9th, 1 out, runner on first base.

If you've watched as much baseball as me, you're not about to head for the exits in this scenario ... not even close.


This election is more like Obama up 3 runs in the ninth with Mariano Rivera on the mound. 1 out, 1 runner on first base.
10.17.2008 12:06am
Cold Warrior:
"Not Danny Darwin"

I just wanted to say I like the moniker. At least you didn't go for the Donnie Moore reference.

But on a serious note, my "win probability" scenario is based on real-world baseball history. And that's what a 5% chance of winning looks like in the 9th inning. If you're a McCain supporter, you ought to be very, very pessimistic ... but not yet resigned to losing.
10.17.2008 12:19am
A. Zarkov (mail):
Indeed a weighted average of polls is better than an unweighted average in the sense that it will have a smaller error. Each poll is weighted by its precision, which is the reciprocal of the error variance. You can calculate the error variance from the reported maximum margin of error (MOE) reported by the poll by squaring it. The MOE assumes puts the percentage at 50% and then applies a formula (see Wikipedia) which is accurate enough for political polls in the US where the electorate breaks almost evenly.
10.17.2008 12:30am
Cold Warrior:
By the way, I overestimated McCain's chances in my baseball analogy. If fivethirtyeight.com is right, McCain's roughly 6% chance of winning the electoral vote is akin to:

Bottom of the 9th, down 2 runs, 2 outs, runner on third.

With, of course, a completely average pitcher facing a completely average hitter, as based on the real-world results in baseball from 1957-2006 (excluding, for whatever database quirk, 1999).

I just find this to be an interesting way of looking at things. He basically needs to hit a 2-out bottom of the 9th HR to tie it, and then needs to win the thing in extra innings.

http://www.tangotiger.net/wins.html
10.17.2008 12:40am
dave h:
Well, the game last night should be heartening to McCain then. Sox were down to 0.6% chance to win at one point.
10.17.2008 10:16am
Angus:
To keep things in perspective: sites like fivethirtyeight.com do not predict what will happen on Nov. 4. They predict what would happen if the election were held today. Changes in the polls can and have influenced the 538 win projections.

There does appear to be a tightening of the polls over the last 24-48 hours. Part of that was expected given that polls do tend to narrow as election day approaches, except in that rare case of an electoral blowout. Anyone who thought this would be a popular vote blowout in this divided climate was nuts.

The question now is how far the swing to McCain will go. Will he remain behind, pull even, or move ahead? Early signs are that Obama will retain a slight edge as trackers pick up the full post-debate effect (most recent one day results for Zogby were Obama +4, for R2000/Dailykos Obama +6)
10.17.2008 10:44am
Jiffy:

sites like fivethirtyeight.com do not predict what will happen on Nov. 4. They predict what would happen if the election were held today.

Actually, 538 does purport to predict what will happen on election day (although the prediction can and does change as more recent poll results come in). In contrast Princeton Election Consortium uses its own poll aggregation methodology to estimate what would happen if the election were held today.
10.17.2008 12:52pm