New AP Poll - McCain & Obama EVEN

[Sir_Galahad]

According to today's AP Poll, McCain has pulled into a virtual tie with Obama. If you follow such things, this is a good sign for McCain.

I just keep hoping that Biden keep talking.

[TheConfessor]

According to today's AP Poll, McCain has pulled into a virtual tie with Obama. If you follow such things, this is a good sign.

Wouldn't that depend on which candidate you prefer?

[Weyoun]

While I would like McCain to win, honestly, I just don't see this is plausible.

[Sir_Galahad]

While I would like McCain to win, honestly, I just don't see this is plausible.

Probably because you have been listening and/or reading too much of the liberal media who is trying to
convince you that it is all over and you should STAY HOME and not even bother to vote.

[Bob Juch]

Polls are snapshots of highly fluid campaigns. In this case, there is a margin of error of plus or minus 3.5 percentage points; that means Obama could be ahead by as many as 8 points or down by as many as 6. There are many reasons why polls differ, including methods of estimating likely voters and the wording of questions.

Charles Franklin, a University of Wisconsin political science professor and polling authority, said variation between polls occurs, in part, because pollsters interview random samples of people.

"If they all agree, somebody would be doing something terribly wrong," he said of polls. But he also said that surveys generally fall within a few points of each other, adding, "When you get much beyond that, there's something to explain."

The AP-GfK survey included interviews with a large sample of adults including 800 deemed likely to vote. Among all 1,101 adults interviewed, the survey showed Obama ahead 47 percent to 37 percent. He was up by five points among registered voters.

[gsabc]

While this is interesting regarding the poll:
"A significant number of the interviews were conducted by dialing a randomly selected sample of cell phone numbers, and thus this poll had a chance to reach voters who were excluded from some other polls."

I find this contradictory:
Opening line, 2nd paragraph of story: "The poll, which found Obama at 44 percent and McCain at 43 percent, ..."
But then, further down: "The AP-GfK survey included interviews with a large sample of adults including 800 deemed likely to vote. Among all 1,101 adults interviewed, the survey showed Obama ahead 47 percent to 37 percent. He was up by five points among registered voters."
So which is it? Were there that many more Obama backers in the "unlikely to vote" group? And who did the deeming?

While over in the "it ain't over till it's over" aisle:
"Some 24 percent of likely voters were deemed still persuadable, meaning they were either undecided or said they might switch candidates. Those up-for-grabs voters came about equally from the three categories: undecideds, McCain supporters and Obama backers. "

Gonna be an interesting election night.

[Weyoun]

While this is interesting regarding the poll:
"A significant number of the interviews were conducted by dialing a randomly selected sample of cell phone numbers, and thus this poll had a chance to reach voters who were excluded from some other polls."

I find this contradictory:
Opening line, 2nd paragraph of story: "The poll, which found Obama at 44 percent and McCain at 43 percent, ..."
But then, further down: "The AP-GfK survey included interviews with a large sample of adults including 800 deemed likely to vote. Among all 1,101 adults interviewed, the survey showed Obama ahead 47 percent to 37 percent. He was up by five points among registered voters."
So which is it? Were there that many more Obama backers in the "unlikely to vote" group? And who did the deeming?

While over in the "it ain't over till it's over" aisle:
"Some 24 percent of likely voters were deemed still persuadable, meaning they were either undecided or said they might switch candidates. Those up-for-grabs voters came about equally from the three categories: undecideds, McCain supporters and Obama backers. "

Gonna be an interesting election night.

Traditionally, yes, those "registered but not likely to vote" are more likely to be Democrat. Which is fine by me.

[15QuestionsAway]

Although the frontline result in this poll shows Obama +1 (44-43), this is only after AP's likely voter model is applied. The raw results show Obama +10 (47-37). Have a look at the internals.

I question the likely voter model in this poll. Nate Silver (fivethirtyeight.com) does as well. However, even if it's appropriate, this poll looks like an outlier.

Probably because you have been listening and/or reading too much of the liberal media who is trying to
convince you that it is all over and you should STAY HOME and not even bother to vote.

Two ridiculous assertions in one sentence. The traditional media wants as much of a horse race as possible. More interest in their product. And why wouldn't reports of big Obama leads depress Obama turnout as well?

Obama voters (at least) are highly motivated. Look at the statistical breakdown in early voting. If McCain voters aren't as highly motivated, perhaps it has to do more with the candidate.

[ToLiveIsToFly]

And why wouldn't reports of big Obama leads depress Obama turnout as well?

The historical tendency has always been that a not-close election depresses turnout among discouraged supporters of the trailer more than it does among supporters of the leader who don't feel the need.

[ToLiveIsToFly]

One other thing to note is that when polls list their margins of error, that's a 95% confidence interval. Which means we should expect that for about 1 in 20 polls, the reality is outside the margin of error. And we're getting 20 national tracking polls every two days or so.

[Bob Juch]

One other thing to note is that when polls list their margins of error, that's a 95% confidence interval. Which means we should expect that for about 1 in 20 polls, the reality is outside the margin of error. And we're getting 20 national tracking polls every two days or so.

And the margin of error is 8 points!

[Timsterino]

Do national polls really mean anything anyway? Nope. What really matters is what each individual state does. Candidates do not win elections based on the national popular vote anyway regardless of who is leading.

[ToLiveIsToFly]

Do national polls really mean crap anyway? What really matters is what each individual state does. Candidates do not win elections based on the national popular vote anyway.

True, but the national polls have some relation to the state polls, and they can inform them. We're not at the point where we have a dozen or so state polls coming out every day. If the last poll for a given state is X days old, and the national polls have moved Y points toward one candidate or the other, it's likely that if another state poll were conducted today, it will have moved somewhat toward the same candidate.

It's way above my pay grade to tell you how much based on all the information out there, but the movement of national polls DOES mean something.

[silverscreenselect]

Obama voters (at least) are highly motivated. Look at the statistical breakdown in early voting. If McCain voters aren't as highly motivated, perhaps it has to do more with the candidate.

And the anti-Obama Democrats I have heard from are highly motivated to vote as well, so that may not mean that much. In 2004, fewer than 1% of early voters said they would have changed their votes if they had waited until election day, so what you have here are the most partisan. Those who might change their votes, along with the true undecideds, which I have heard could be 15-20% of the total, won't be voting early, and those will make the decision. Traditionally, those wind up breaking one way or another pretty heavily.

I feel that there are a number of anti-Obama Democrats, blacks and independents who don't want to tell a pollster, especially a young sounding pollster who they plan to vote for, so they either refuse to be polled or lie about who they are backing.

[Bob Juch]

Obama voters (at least) are highly motivated. Look at the statistical breakdown in early voting. If McCain voters aren't as highly motivated, perhaps it has to do more with the candidate.

And the anti-Obama Democrats I have heard from are highly motivated to vote as well, so that may not mean that much. In 2004, fewer than 1% of early voters said they would have changed their votes if they had waited until election day, so what you have here are the most partisan. Those who might change their votes, along with the true undecideds, which I have heard could be 15-20% of the total, won't be voting early, and those will make the decision. Traditionally, those wind up breaking one way or another pretty heavily.

I feel that there are a number of anti-Obama Democrats, blacks and independents who don't want to tell a pollster, especially a young sounding pollster who they plan to vote for, so they either refuse to be polled or lie about who they are backing.

Are you one of those?

[silverscreenselect]

Obama voters (at least) are highly motivated. Look at the statistical breakdown in early voting. If McCain voters aren't as highly motivated, perhaps it has to do more with the candidate.

And the anti-Obama Democrats I have heard from are highly motivated to vote as well, so that may not mean that much. In 2004, fewer than 1% of early voters said they would have changed their votes if they had waited until election day, so what you have here are the most partisan. Those who might change their votes, along with the true undecideds, which I have heard could be 15-20% of the total, won't be voting early, and those will make the decision. Traditionally, those wind up breaking one way or another pretty heavily.

I feel that there are a number of anti-Obama Democrats, blacks and independents who don't want to tell a pollster, especially a young sounding pollster who they plan to vote for, so they either refuse to be polled or lie about who they are backing.

Are you one of those?

If polled, I would have no problems telling anyone who I would vote for. I have no intention of putting bumper stickers on my car or wearing anything that might indicate I'm voting for McCain, because I live in a neighborhood where that is an open invitation to trouble.

[Bob78164]

Do national polls really mean crap anyway? What really matters is what each individual state does. Candidates do not win elections based on the national popular vote anyway.

True, but the national polls have some relation to the state polls, and they can inform them. We're not at the point where we have a dozen or so state polls coming out every day. If the last poll for a given state is X days old, and the national polls have moved Y points toward one candidate or the other, it's likely that if another state poll were conducted today, it will have moved somewhat toward the same candidate.

It's way above my pay grade to tell you how much based on all the information out there, but the movement of national polls DOES mean something.

That's exactly what Nate Silver does at FiveThirtyEight.com. In each state's results box (along the right-hand margin), there's a "Trend-Adjusted" result. That's what he thinks the reported polls for that state would have said (on average) if updated to today.

He then applies a further adjustment based on demographics to come up with a "snapshot" (his best guess of the state result if the election were today), and then still another adjustment to project the race to Election Day itself. --Bob

[silverscreenselect]

Do national polls really mean crap anyway? What really matters is what each individual state does. Candidates do not win elections based on the national popular vote anyway.

True, but the national polls have some relation to the state polls, and they can inform them. We're not at the point where we have a dozen or so state polls coming out every day. If the last poll for a given state is X days old, and the national polls have moved Y points toward one candidate or the other, it's likely that if another state poll were conducted today, it will have moved somewhat toward the same candidate.

It's way above my pay grade to tell you how much based on all the information out there, but the movement of national polls DOES mean something.

That's exactly what Nate Silver does at FiveThirtyEight.com. In each state's results box (along the right-hand margin), there's a "Trend-Adjusted" result. That's what he thinks the reported polls for that state would have said (on average) if updated to today.

He then applies a further adjustment based on demographics to come up with a "snapshot" (his best guess of the state result if the election were today), and then still another adjustment to project the race to Election Day itself. --Bob

What moves voters in one state may not move voters in another. Suppose Obama came out today and proposed a constitutional amendment banning discrimination against gays. That would elevate his standing in some states and hurt it in others and have little effect in others. So even if the national polls moved as a result, it doesn't mean the state polls would necessarily move the same way.

[Estonut]

In this case, there is a margin of error of plus or minus 3.5 percentage points; that means Obama could be ahead by as many as 8 points or down by as many as 6.

I am not a statistician, but this doesn't make sense. If the margin of error is +/- 3.5 points, how does that translate to ahead by up to 8 or down by up to 6? Wouldn't the estimated high and low have to differ by no more than 7?

[silverscreenselect]

In this case, there is a margin of error of plus or minus 3.5 percentage points; that means Obama could be ahead by as many as 8 points or down by as many as 6.

I am not a statistician, but this doesn't make sense. If the margin of error is +/- 3.5 points, how does that translate to ahead by up to 8 or down by up to 6? Wouldn't the estimated high and low have to differ by no more than 7?

It's more complicated than that. Suppose that a poll shows Obama 45, McCain 45 and 10 undecided, with a three point margin of error. What that means is that there is a very high probability, in the neighborhood of 95 to 98%, that Obama's actual total will be somewhere between 48 and 42 percent, and an equally high probability that McCain's total will be somewhere between 48 and 42 percent. So under this scenario, it is within the margin of error for either McCain or Obama to be up by as much as six percent.

Plus, the fact that there is a 95% chance that the results are somewhere in that range means that there is a 5% chance that the results are outside that range. That means that on average, one in 20 polls will be an outlier. Considering that there are anywhere from 10 to 30 state and national polls being released each day, that means that there is a fair chance that one or two of them will be outliers. This can explain how you get an occasional poll that seems way out of line. It's not "fixed," it's merely the operation of the laws of probability, just as you occasionally roll doubles in Monopoly three times in a row and wind up in jail.

And all of these assumptions are only accurate if the polling sample is representative of the populace as a whole and if those being polled are giving accurate answers.

[Estonut]

I am not a statistician, but this doesn't make sense. If the margin of error is +/- 3.5 points, how does that translate to ahead by up to 8 or down by up to 6? Wouldn't the estimated high and low have to differ by no more than 7?

It's more complicated than that. Suppose that a poll shows Obama 45, McCain 45 and 10 undecided, with a three point margin of error. What that means is that there is a very high probability, in the neighborhood of 95 to 98%, that Obama's actual total will be somewhere between 48 and 42 percent, and an equally high probability that McCain's total will be somewhere between 48 and 42 percent. So under this scenario, it is within the margin of error for either McCain or Obama to be up by as much as six percent.

You've illustrated my question exactly. Either McCain or Obama could be up by as much as six percent, which is double the margin of error. That doesn't explain how the window could be 14 points in the previous example where the margin of error is +/- 3.5 points. In other words, if they say candidate X is at 45 with a margin of error of 3.5 points, then he is really between 41.5 and 48.5, with the range being exactly double the margin of error.

[silverscreenselect]

That doesn't explain how the window could be 14 points in the previous example where the margin of error is +/- 3.5 points. In other words, if they say candidate X is at 45 with a margin of error of 3.5 points, then he is really between 41.5 and 48.5, with the range being exactly double the margin of error.

Because you actually have two different findings here. A poll reports both candidates' totals and each total is separately subject to the margin of error. So if Obama is at 45% with a 3.5% margin of error, then his total could be anywhere from 41.5 to 48.5. McCain's total could also be anywhere from 41.5 to 48.5. And it's actually more likely than not that if candidate A's total is higher than predicted, than candidate B, who is drawing from the same pool of voters is lower than predicted. So in order for a candidate to be ahead of another candidate outside the margin of error, his lead would have to be at least double the amount of that margin (in this case, 7 points). Most margins of error in this election have been 2 to 5 points, depending on the size of the sample and other factors, so a candidate would have to be ahead anywhere from 4 to 10 points in that state (or the country) to be outside the margin of error.

[Bob Juch]

In this case, there is a margin of error of plus or minus 3.5 percentage points; that means Obama could be ahead by as many as 8 points or down by as many as 6.

I am not a statistician, but this doesn't make sense. If the margin of error is +/- 3.5 points, how does that translate to ahead by up to 8 or down by up to 6? Wouldn't the estimated high and low have to differ by no more than 7?

Because the poll actually showed Obama ahead by one point, not even.

[Estonut]

D'oh! I ran some examples with different numbers and it is clear now. Your example of 45/45/10 illustrated my misunderstanding perfectly. I now see that if O were at 44 and M at 43, then O's real range would be 47.5 to 40.5, while M's would be 46.5 to 39.5. This shows that if O were really at (his max) 47.5 and M at (his min) 39.5, O could really be up by 8. On the other hand, if O were really at (his min) 40.5 and M at (his max) 46.5,k then O could really be down by 6.

What a bunch of non-helpful crap!

[gsabc]

Just remember the three kinds of lies:

Lies

Damned Lies

Statistics