The Biden Bounce
It’s a dead heat in the race for the White House. The first national poll conducted entirely after Barack Obama publicly named Joe Biden as his running mate suggests that battle for the presidency between the Illinois senator and Republican rival John McCain is all tied up.
In a new CNN/Opinion Research Corporation poll out Sunday night, 47 percent of those questioned are backing Obama with an equal amount supporting the Arizona senator.
“This looks like a step backward for Obama, who had a 51 to 44 percent advantage last month,” says CNN Polling Director Keating Holland.
“Even last week, just before his choice of Joe Biden as his running mate became known, most polls tended to show Obama with a single-digit advantage over McCain,” adds Holland.
So what’s the difference now?
It may be supporters of Hillary Clinton, who still would prefer the Senator from New York as the Democratic Party’s presidential nominee.
Sixty-six percent of Clinton supporters, registered Democrats who want Clinton as the nominee, are now backing Obama. That’s down from 75 percent in the end of June. Twenty-seven percent of them now say they’ll support McCain, up from 16 percent in late June.
That’s all great fun, but I still don’t buy it. Those Clinton voters are going to vote for Obama – most of them, anyway. They’re just feeling a little jilted right now…
Agreed, I don’t see them voting for McCain nor are they likely to stay home or vote for one of the fringe Dems.
Obama will get a nice bounce out of the convention and should be able to sustain it because much of it will come from those, currently, miffed Clinton supporters.
Also agreed.
Lets not forget were dealing with a large group of angry women. Obama’s campaign is now dependant on them and they are the most unpredictable group known to human kind. If they behave at the convention people will probably calm down about it and he could seal the victory. If not the negative momentum will be a huge black eye. Well see in the next few days.
All the more reason for McCain to offer the VP to Hillary.
In fact, he should double down and offer Secretary of State to Bill.
Obama leads McCain among women voters by a whopping 49% — 39% . (I’m sure the percentage is even larger among women democrats, though you gotta pay to get that number.)
If only the election hinged on the votes of those “angry” Clinton supporters! Obama would win in a landslide.
Hey people-despite around the clock media pimping for the last 8 months and massive fundraising advantages the race is tied according to the biased MSM. Doesn’t that tell you something? Do you remember that Obama consistently over polled during the primary/caucus season and that was with just Democrats? In states where he has been spending large sums of money his numbers have gone….down. People are looking at him and are now realizing what many Hillary supporters and Republicans have known for months-He is an empty suit with no accomplishments and totally propped up by an adoring media.
You’re conflating groups, the poll is of registered voters, not female Clinton supporters. I understand you are countering the argument that Obama’s campaign is “dependent” on female voters, but are you also suggesting McCain isn’t helped, and Obama hurt, by defections of Clinton democrats to the McCain campaign?
I much prefer polls of likely voters than of registered voters, although, I don’t put much stock in polls, as an absolute measure, of these things anyway; more valuable as an index.
What I’m saying is that there is zero empirical evidence that “Clinton democrats” are defecting in even relatively insignificant numbers to the McCain campaign. This is just a myth, propagated by a lazy media and McCain operatives ( http://www.youtube.com/watch?v=O8Ta8IZ9sa0 ).
http://www.youtube.com/watch?v=597YG23mAWs
Ok, one is probably insignificant, but it is one.
Jacques,
I’ll take your cherry-picked poll that says nothing, not even what you claimed was your point in your next post, and give you one in return:
From CNN:
Since CNN/ORC don’t make their poll results available (unlike USA Today/Gallup, which is the one I quoted), I can’t comment. But fiverthirtyeight was harshly critical of the “analysis” of the poll results presented in that article you linked to.
In any case, the comment (#3), that I was responding to, was specifically about women, and there, the “angry women” meme is flatly contradicted by the facts.
If you want to broaden the question to be about Democrats overall, Gallup shows 10–13% supporting McCain, and that number hasn’t budged since June.
Maybe you have reason to trust ORC over Gallup. But, in light of the fact that they don’t deign to publish their numbers, and in light of the criticisms linked-to above, I am rather less than inclined to take them seriously.
Im not saying that all Clinton supporters are angry and Im not saying they are defecting toward McCain. Im saying the group of angry Clinton supporters which I am assuming is made up of mostly women (but I could be wrong) is threatening to seriously hurt Obamas chances at winning the election. The election doesnt “hinge” on Clinton supporters, but if the angry ones throw a big enough fit about losing its going to keep a substantial portion of key voters home on election day. A handful of protests/tv ads claiming Obama cheated the primary or is otherwise unfit to be president will boost McCains momentum enough to put him past the dead heat stage and into the lead.
Jacques,
Saying that 538 was “harshly critical” is disingenuous to say the least. He was somewhat critical about two points, and one of them is completely false. Nate somehow infers out of thin air that the point of the released analysis was to show that the Biden pick caused this turn of Clinton supporters, when nothing of the kind is suggested or implied in the CNN analysis.
The other thing he was critical of was them comparing results from a June poll in one part and a July poll in the other. Well, fine. You can easily go back and look at the June poll if you like, instead of the July poll and then the whole thing is a June/August comparison. The one line about July was that in July Obama led by 7, and now it’s tied. Ok, in June he lead by 5. I found this in about 2 minutes from looking at the poll archives in RealClearPolitics, but Nate couldn’t be bothered to do this, apparently. It was more important for him to find flaw in the analysis of others than to do any himself.
This is why you should avoid Nate’s blog posts unless he’s talking about mathematics. His analysis is almost always weak. He criticizes people for not taking any effort to do proper research or for their inconsistency and yet he’s a marvel of inconsistency himself.
As to why I might trust CNN/ORC over Gallup, one reason might be that their “house effect” is much closer to zero. That by itself doesn’t mean a whole lot, but their standard deviation is much closer to zero as well. You can see a good analysis of “house effects” here.
Now, would I prefer CNN/ORC publish more details like what the breakdown of Reps/Dems in a poll is? Sure. Of course, for the question at hand since we’re only looking at Clinton Democrats that question is meaningless, and therefore, so is the criticism.
1) The 95% confidence interval (that’s 2σ, not 1σ), plotted on that graph for CNN/ORC is ±2.5%. The 95% confidence interval for Gallup is ±1.5%. I don’t know why you think the former is smaller than the latter.
2) The CNN/ORC “house effect” is, according to that analysis tipped towards Obama, whereas the Gallup house effect is tipped towards McCain.
3) In any case, if you are looking at trends in the time-series generated by one or another of these polling outfits, then the house effect ought to cancel out.
Huh? Here’s what the CNN article says:
(emphasis mine)
Sounds to me like they’re implying exactly what Nate says they’re implying.
Which would be less useful, for the purpose at hand, than a July/August comparison.
I fail to understand what you are saying.
They obviously asked people their party affiliations and they evidently asked the Democrats who they supported during the primaries. That’s how it is that they reported numbers like
So Nate’s complaint about the omission of the July figures, which would have a direct bearing on the question of “what’s the difference now?”, is valid.
Now, I will agree with you that Nate is a bit over-the-top about CNN/ORP’s secretiveness about their data. They do provide enough information to recover the 95% confidence interval for their sample of Clinton Democrats: it’s ±18.4% (1σ is ±9.2%).
Still, Gallup’s sample size is 2.5 times as large, and they run a poll weekly, rather than monthly. So, if you are trying to discern trends (Clinton Democrats shifting to McCain), their data is bound to be more reliable. And their data shows no such trend.
I wrote:
I must be particularly fumble-fingered with the calculator tonight. The 95% confidence interval for their sample of Clinton Democrats is ±7.5%.
Sorry.
Jacques,
1) I’m referring to the trending shown here and here. As you can see, Gallup is all over the map, sometimes 2-3 points above the standard trend, sometimes 3-4 points below it. That makes looking at a single poll and adjusting it for the “house trend” rather meaningless. However, the CNN/ORC (with many fewer data points, which does limit the effectiveness of my argument) polls follow the standard trend much more reliably.
2) So? Who cares? We’re discussing the reliability of the poll not the direction of it’s bias.
3) Huh? All I’m doing is answering your question as to why I might find the information in the CNN/ORC poll to be more valuable.
Your quote is useless. It doesn’t take the context of the article into account at all. I’ll cherry-pick some quotes too:
Note that that was the sentence right before yours that you conveniently left out.
and
Hmmm…I see a lot of “since June” in there, not “since last week”. I fail to see how any reasonable person could see this piece as an attack on the Biden pick.
Yes, obviously I’m not making myself clear. Nate complains (and rightly so) that the details of the polls are missing. He doesn’t specifically say what kind of details he’d like to see, but typically when you look at information from other pollsters, you get things like “how many Democrats polled”, “how many Republicans”, “how many Independents”. Some even break it down further. Zogby likes to do “how many Bush voters”, “how may Kerry voters”, etc.
These kinds of details are totally missing from the CNN/ORC poll. And not having them is a valid reason to criticize the poll. We have no way of knowing if the “race is tied” because they over-polled Republicans and didn’t account for that. But, our question here relates to one thing and one thing only. How many Clinton supporters are planning to vote for McCain? In such a question, how many Republicans were included is meaningless. Really, the only question that’s important is how many Clinton supporters were included in the poll. I gather that it’s not a terribly large number since the MOE is 7.5. I could probably work up an estimate, but it’s been a while since I did any calculations of MOE. Off the top of my head, I’d say the number is around 200, but I wouldn’t be surprised if I’m way off there.
I disagree totally with your reasoning about the July/August data. The point of CNN’s analysis (and Nate has a point here that it’s hard to tell with CNN whether the polls drive the analysis or the analysis drives the polls) is to compare June and now. If I wanted to show that people are paying less taxes now than they were in 2000, I’d look at now and 2000. Looking at 2004 would be somewhat irrelevant.
This is the first thing you’ve said that I mostly agree with. Let me break it down.
1) Gallup’s sample size is 2.5 times as large. That certainly makes their data more reliable, and by quite a large margin. No disagreement there.
2) Gallup runs a poll weekly, rather than monthly. Certainly that’s better for trend analysis, and if the numbers show that Clinton supporters are moving towards McCain, you might be able to analyze the trend and find a moment in time where that started to happen. Trend analysis showed that Kerry started losing support 2004 after the Swift Boat Veterans started speaking, for example.
3) If you are trying to discern trends (Clinton Democrats shifting to McCain), their data is bound to be more reliable.
This is where you’re wrong. Or at least it appears to me that you are. I’ve yet to see a Gallup poll where that question has been asked, or that data has been included in the matrix. Perhaps I missed it. I wouldn’t be surprised.
Now there is some Gallup data that would appear to support your claim. It certainly appears to show that there is no significant trend of Democrats to McCain, over the lifetime of their polling. (13% now, 13% in June) Certainly you can take specific data points and attempt to make a claim about them. In early July, he was getting 10% of Democrats, and now he’s up to 13%. Of course, the difference is within the MOE, so perhaps there’s no difference. And this is where you’re absolutely correct. More data points are necessary for accurate trend analysis.
But the data does somewhat support the claim that’s been made by news organizations other than CNN, namely that initially the Obama support by Clinton backers was lower, then rose, and now has fallen again (sorry, I don’t have a link). And that may indicate movement of Clinton Democrats to McCain. It’s too early to say, and I really wish that Gallup had actually asked that question (since the beginning. it helps little to start asking it now).
Since the Daily tracking poll has a sample size of ~1000, the MOE is ±3%. So the scatter you see on the first plot is exactly what you expect to see. If their poll data hugged the trendline significantly more closely, you would have strong reason to suspect they were faking their data.
On the second plot, the CNN/ORC poll (with only 9 data points) shows about as much scatter as you’d expect for their sample size.
In both cases, I think the public (and, perhaps, you) would be better served if the data points were plotted with their statistical error bars, a practice routine among us scientists.
No it doesn’t. “House effect” and MOE are entirely separate issues.
More than that. The USA Today/Gallup poll interviews 10,000 people/month. The Gallup Daily Tracker poll interviews 30,000 people/month. CNN/ORC interviews 1000 people/month. When you slice their sample into particular demographics, it gets even smaller. As you say, the Clinton Democrats in their poll constitute only a couple of hundred people/month, compared to an order of magnitude more in the Gallup survey.
Given the inherent graininess of the temporal resolution of the CNN/ORC poll (and the total lack of information on the precise time-period over which the poll was conducted), and the tiny sample of Clinton Democrats therein, you are perfectly justified in comparing the average of the USA Today/Gallup polls taken during the month of June with the one CNN/ORC poll. Statistically-speaking, the former blows the latter out of the water.
To my knowledge, you didn’t miss it. But we do know that Clinton Democrats constitute slightly less than half of all Democrats. If CNN/ORC were correct, and there had been a huge (11%) swing in the support of Clinton Democrats, then we would be surprised, indeed, to find no discernable shift at all in the overall support among Democrats in the USA Today/Gallup poll.
The only explanation would be a bizarre, severe, under-sampling of Clinton Democrats in the Gallup survey. That seems far more unlikely than the natural explanation: the MOE of the CNN/ORC poll among Clinton Democrats is 7.5%, and they just cherry-picked an outlier.
(That’s not entirely fair. The correct way of presenting their data is to say that: in June, 16±7.5% of Clinton Democrats said they supported McCain; in August, 27±7.5% supported McCain. Stated that way, (with just those two data points) the result is entirely compatible with the hypothesis that support for McCain among Clinton Democrats did not budge at all from June to August. )
Hmmm.
I would take issue with the “46%” If you look at the graph, Obama’s support among married women stayed pretty constant, at 42-44% (surely, well within the MOE, which they don’t explicitly state, but which must be at least ±2%). Then, in the first week of August, a blip up to 46% — very likely an outlier, though it may still be within the MOE. Then, in the succeeding two weeks, a slide to 40% and 39%.
That’s still a drop (of 3% or so), but not as dramatic as a 7% drop (per the quote above).
This illustrates exactly the problem I complained about with respect to the CNN/ORC report. It is terribly tempting to pick an outlier, in order to sex-up your findings, by making the shift (in whichever direction) seem more dramatic than it really is.
When Gallup succumbs to that temptation, you can look at their data and say, “Hmmm, looks like there’s a real effect here, but not as dramatic as they said.”
When CNN/ORC does it, you have no recourse to a reality-check.
If I were polling Czar, I would demand that polling firms put the error bars on the graphs that they publish (and state the MOE for every percentage that they cite). And I would forbid them from engaging in the stupid practice of drawing straight line segments connecting the central values. Doing so makes it look like the poll numbers are bouncing around, even when nothing is actually happening (they’re actually constant, to within the MOE).
Jacques,
I don’t believe this has ever happened before.
I agree with every single word in your last post.
It’s called syzygy.
(Also possibly the best word to spell in Scrabble, after quizzical.)
Syzygy … there’s the ticket!
Anyway, I’m always glad to see when we can agree on some things.
If we could never agree on anything, it would be kinda pointless to keep arguing, wouldn’t it?
Syzygy-Quizzical ’08!
Oh, sorry, that’s not what you meant by “there’s the ticket!”
Good point, Jacques. I’ll keep arguing.