Brad DeLong's Semi-Daily Journal: A Weblog

N.Y. Newsday columnist Jimmy Breslin asserts that today's presidential phone polls leave out 170 million cell phones -- possibly distorting the results in favor of Bush because they leave out more of young voters. Maybe there is hope for Kerry.

Denis Drew
ddrew2u@comcast.net

Coincidentally, here is an excerpt from Bush's speech at the UN a couple of days ago,

"Well, I don't give a flying phooey what the polls say. I believe it is time we send a man to Mars. In fact, if we send anyone, it will be my good friend, Bob, who deserves more than Cheney, even. You can find his harrowing yet inspiring journal at http://www.sendbobtomars.com and it's a darn good read, y'all."

Thanks, Dubya. Right back at ya.

It interests me because it tends to confirm a pattern I think I see: that Kerry would probably win the electoral college if, say, one point behind in the popular vote. Bush's support is more concentrated.

Other analyses are welcome.

It interests me because it tends to confirm a pattern I think I see: that Kerry would probably win the electoral college if, say, one point behind in the popular vote. Bush's support is more concentrated.

Other analyses are welcome.

Anybody been following IEM trading? The election markets there currently have Kerry way behind on the winner take all market, but not too far behind on the vote share market.

http://128.255.244.60/quotes/78.html

On the winner take all market, prices suggest Bush likely to win (65% probability), and more likely to win by a larger margin than a smaller one (28.5% chance of winning by less than 2 percentage points; 37.3% chance of winning by more than 2 percentage points). Massive movement in prices in this market since the Rather stuff (check the price graph here: http://128.255.244.60/graphs/graph_Pres04_WTA.cfm ).

On the vote share market, it's looking more like 52.5-47 Bush win. But, the winner take all market is way thicker.

I liked the Bush speech part where he talked about democracy in Iraq. We put in a dictator there and then kill thousands of Iraqis who are against this. I wonder if we will do this here, too.

...or a dishonest voting system...

Notice also that the polls in Colorado are Bush-46%, Kerry-45%, Nader-3%.

Notice ARG's discussion for how to produce a political advt. I'll rewrite them slightly:

1. Tell a powerful story
2. Include a call to action in the story
3. Make basic emotional appeals
4. Use easy arguments
5. Show *and* tell.
6. Use concrete, rather than abstract, language and images.
7. Make the audio and video work together.
8. Hold scenes long enough for impact.
9. Use powerful video--there, add no words.
10. Bind your campaign ads together with signature music.

There's another way to think about it: why go to the trouble of taking lots of data if your model errors are that bad? There's not much point in getting your sampling MOE down to +- 1 point when it's likely that your systematics are off by at least 5 or 6 points (as a glance at the Pollkatz presidential-job-approval aggregation would imply).

All that said, I think it's clear that the trend right now is in Bush's favor, and I can't really think of how Kerry could turn that around. It's turned around before, of course.

Chef Ragout had a correspondence with the president of ARG in which it became clear that the ARG guy didn't know how to calculate the standard error on the difference of shares in a poll. This doesn't give me a lot of confidence in ARG's methodology.

http://ragout.blogspot.com/2004/08/more-margin-of-error.html

Though I wsih it were so, I find these poll results highly suspect. Results of state polls, from organizations that have been accurate in election after election, show John Kerry to be in trouble. The evenness os the race in New Jersey should be believed, and should be a prime worry for Democrats. The problem John Kerry has had attracting women voters seems clear. I am worried!

http://www.nytimes.com/2004/09/22/politics/campaign/22women.html

Kerry in a Struggle for a Democratic Base: Women
By KATHARINE Q. SEELYE

WASHINGTON - It was no accident that John Kerry appeared Tuesday on "Live With Regis and Kelly'' and recalled his days as a young prosecutor in a rape case. Or that he then flew from New York to Jacksonville, Fla., to promote his health care proposals. Or that on Thursday in Davenport, Iowa, he will preside over a forum on national security with an audience solely of women.

These appearances are part of an energetic drive by the Kerry campaign to win back voters that Democrats think are rightfully theirs: women.

In the last few weeks, Kerry campaign officials have been nervously eyeing polls that show an erosion of the senator's support among women, one of the Democratic Party's most reliable constituencies. In a New York Times/CBS News poll conducted last week, women who are registered to vote were more likely to say they would vote for Mr. Bush than for Mr. Kerry, with 48 percent favoring Mr. Bush and 43 percent favoring Mr. Kerry.

In 2000, 54 percent of women voted for Al Gore, the Democratic nominee, while 43 percent voted for Mr. Bush.

Democratic and Republican pollsters say the reason for the change this year is that an issue Mr. Bush had initially pitched as part of an overall message - which candidate would be best able to protect the United States from terrorists - has become particularly compelling for women. Several said that a confluence of two events - a Republican convention that was loaded with provocative scenes of the Sept. 11 tragedy, and a terrorist attack on children in Russia - had helped recast the electoral dynamic among this critical group in a way that created a new challenge for the Kerry camp.

Mark Mellman, a pollster for the Kerry campaign, said that the campaign was not especially disturbed by the reduced support from women. "I don't define it as a problem,'' Mr. Mellman said. "I define it as an opportunity.'' He noted that a group of widows of Sept. 11 victims endorsed Mr. Kerry last week and offered that as evidence that the women "thought he was better able to protect the country.''

Just as Mr. Kerry is trying to win back women, Mr. Bush is seeking to do what Karl Rove, the president's chief political adviser, has vowed to do this year: challenge the Democrats on their own turf; in this case, make a pitch directly to women.

Mr. Bush frequently tells audiences about the newfound freedoms for Afghan women who were liberated when the United States toppled the Taliban. His campaign rallies often feature signs saying "W stands for Women.''

I wasn't being facetious. I think hiding model error behind sample error really is part of why sample sizes are the size they are. It is strange that all polls use similar size samples,no ?

I think the electronic markets numbers are consistent. It was odd when the vote share was as it is now and Bush win was the same as expected vote share. The current numbers are guess Bush ahead by 5.5% standard errof of dif about 7% which actually still sounds large to me. I am tempted to put money on Kerry.

Kerry wins electoral if loses popular 1 % fits the ARG polls exactly (that is totally ignoring standard errors) but with tiny changes in Florida and Penn Kerry would lose electoral.

On ARG methodology, Brad might remember that a quite smart person (not Brad) mamaged to get confused about that briefly 4 years ago. I read somewhere that in 2000 ARG did much better than Gallup (which isn't saying much). People reading this have probably read arguments that Gallup and CBS clearly have screwy samples as there are too many self identified Republicans. CBS asked a retrospective question. With there sample Bush won by a landslide in 2000.

My sense is trend is weakly pro Kerry, but that might be pure wishful thinking.

Robert Waldmann

Hope you are correct.

"I think hiding model error behind sample error really is part of why sample sizes are the size they are."

In some cases, it doesn't. The dKos community caught one poll with Bush ahead by about 11%. The poll also asked who the people voted for in 2000, and Bush won there, too, by about 12%. Problem? We all know Gore won the popular vote. So the sample size and MoE was what you'd expect from a typical poll, but they oversampled Republicans by a whopping 12%. This would've been tougher to do with a large sample size.

In short, it's not really cover-your-ass. Many polls out there really intend shape public opinion, not track it. A high MoE serves as a smokescreen to deliberately hide inaccuracies caused by push-polling or biased sampling. And if one poll shows the race tied and another shows Bush up by 10? People will assume Bush is ahead by 5, even if the latter poll is completely fraudulent.

What, does anyone here believe a poll would value statistical accuracy and methodology over pleasing a client with fat pockets?

The Election still lies some distance away, but a prominant problem with Surverys can be cited: most of polled are older voters, as younger voters are harder to reach by phone. Guess-estimates of myself see Bush losing the military vote, and 7% of the men who voted for him in 2000. The key States will be Michigan and Ohio, along with Florida. I expect Bush will be cut to half of the Pairre States, simply due to the opposition to the Casualties in Iraq. An interesting note may be the South, a New Englander may break one or two States loose this Election, because Southerners are beginning to realize Bush is big Corporate business not concerned with Jobs for working people.

I suggest Kerry has a better record with the Ladies than expected, and could easily win in the Electoral College, with even 53% of the Popular vote. I smell Regime change, though it may only be wishful thinking. lgl

The Wall Street Journal did a piece a few days ago reporting that the pollsters who have been giving Bush a big lead are also reporting a switch in party affiliations to the GOP -- while those who show the race close are discounting any such party switch in advance on the grounds that it's implausible, and adjusting their polls to show the previously existing Republican/Democratic voter balance (and thus a closer race). I'd be more inclined to believe the former group than the latter.

Still, the latest NBC/Wall Street Journal poll (just out) agrees with Zogby and Rasmussen in showing only about a 3 or 4 point lead for Bush, which is also in accord with what I'm projecting from the recent polls of swing states done by the previously very accurate Mason-Dixon.

"I read somewhere that in 2000 ARG did much better than Gallup (which isn't saying much)."

Actually, in 2000, Gallup's final prediction was 48-46-4, Bush-Gore-Nader, which was quite close to the actual result. The table I'm looking at doesn't have an ARG poll from 2000.
http://www.electoral-vote.com/info/polls-2000.html

"People reading this have probably read arguments that Gallup and CBS clearly have screwy samples as there are too many self identified Republicans. CBS asked a retrospective question. With there sample Bush won by a landslide in 2000."

I think it was Bush by 36-28 (with a lot of people having not voted). But that's from memory.

People also tend to misreport how they voted. Jack Kennedy's retrospective numbers were MUCH higher after his assassination than in the actual election.

I see Kerry winning Penn and Florida, which probably means he wins. Seems those are the two major trophies up for grabs.

With respect to ARG and Gallup, what people are remembering reading were comments by Alan Abramowitz that Ruy Texeira passed on, which can be found here:

http://www.emergingdemocraticmajorityweblog.com/donkeyrising/archives/000692.php

He had two points: that the September, 2000 ARG state-by-state turned out to be pretty damn accurate and that it was more accurate than the "wildly gyrating" gallup poll results.

Bruce, with respect to the question of republican counts, Ruy Texeira also discusses this here (because what would i know about this):

http://www.emergingdemocraticmajorityweblog.com/donkeyrising/archives/000693.php

The short of it is that we're supposedly seeing a bigger swing to republicans than we saw in the aftermath of 9/11, when bush had 90% approval ratings.

That said, Bruce, i didn't read the journal article, but i don't believe that any polling organization is arbitrarily rebalancing the sample.

I wish I had a feel for what the women are thinking. The anecdotal evidence of shifts among Republican men away from Bush and of young people voting when they didn't before, and voting for Kerry is so persistent that one has difficulty absorbing the poll shift towards Bush, which has got to be coming from frightened females, if it is coming at all.

I still think Mr. K will win big. If it is the frightened females who are moving the polls, they have a lot of things to be frightened about besides forgetting where they left their duct tape. Medical care and their spouses' employment status to begin with. This campaign has a long way to go.

Knut, the way i look at the election is that i don't really care if bush wins texas and alabama and mississippi 99-1; i don't care if kerry wins california and new york by 50.1 - 49.9. I only care who wins the likes of ohio, michigan, florida, etc.

What is still extraordinarily difficult for me (and i think anyone) to judge is whether the horse-race lead that bush shows is a function of winning texas 99-1 or winning ohio, michigan, and florida, but the ARG poll offers a hint that it may be the former....

I'm not a reporter and I do have an intermediate understanding of statistics and survey sampling. But I'm having trouble understanding your modeling error versus sampling error comparison and explaining it to someone on my staff. Am "discussing" (debating?) your modeling error versus sampling error issue with a somehwat dogmatic Sociology PhD who scoffs at the assertion that "only amateurs worry about sampling error. Professionals worry about modeling error." He says that's because Economists don't understand the statistics behind surveys. I responded that I'd rather ask the author than make that assumption.

Can someone clarify the assertion? (Brad? Robert?)

Thank you.

"I wish I had a feel for what the women are thinking."

Well, you can start by not lumping a whopping 51% of the country's population into an uber-demograph called "WOMEN".

I have never noticed Matt McIrvin to be wrong in the comments (he has corrected me any number of times) and I think he has it right here. The is analogous to meta-analyses in epidemiology. The often aggregate a lot of data to remove statistical noise, but they introduce model errors by combining things that were not done the same way, and in the end they often are not all that reliable (I should note that very careful meta-analysis of data can be useful, but often it is simply a cheap publication, IMHO). Look to the razor of Ockham, in this and all other endeavors. If you cannot get your model error down below 3 or 4 percent, why bother calling all those people?
Also, I have to question the error analysis on this poll-did they just aggregate the state numbers for the national? They took it over at least four weeks (mid-Aug to mid-Sept). Does anyone really think that PA in mid-Sept plus CA in mid-Aug = PA + CA?

Am I too cynical, or does Kerry need to forget any plan for winning that counts on his getting Florida? A Slate article suggested that the fix is probably in there, unless I suppose Kerry has a landslide. I wonder what you more knowledgeable commenters think about that.

But I do think that polls are underestimating the drive that the ABB people will have to get out and vote this year, including we young, cell-phone-bearing folk. My vote has no chance of counting, as I'm in Mississippi, but I'll be casting it from sheer pique.

(Btw, Miss. in April: 60/31 for Bush. In Sept: 51/42 for Bush. That's pretty amazing for the Magnolia State.)

HERE'S TEH LATEST ECONOMIST/YOUGOV POLL:

http://economist.com/displaystory.cfm?story_id=2907805

(Cross posted with Robert Waldman & Matt Yglesias)

Response bias! Add this to your list.

One more source of error ignored by sampling margins is the likelihood of certain types of people will agree to answer questions about the presidential race.

Pollsters generally admit that people are becoming less willing to respond to telephone surveys & those hangups may well be correlated with people's feelings about the race. More republicans respond around the time of the RNC convention, and v.v.

In addition, the spread of cell phones as primary voice line is increasing un-noticed sampling bias. It's beginning with young people, who matter less for voting since turnout is lower, but it may be weighting response further toward the republicans.

I have two anecdotal questions. Does anyone know someone who is changing their vote from 2000? If so, what is the change? I do not know one person who voted for Gore or Nader that is now voting for Bush, and I live in Ohio. I do know people who voted for Bush and Nader who are going to vote for Kerry. One thing about the ARG poll that I found very promising for Kerry was the percentage of undecided voters in battleground states where both canidates are under 50%.

What's troubling in all this is the fact that female voters (in many subsets - I'm sensitive to your point, Dragonchild) find Bush/Cheney appealing. "Homeland Security" won't count for much if you're uninsured, if you or your spouse or both are unemployed or chronically ill, or if you're the subject of sexual discrimination in the workplace. There's a really depressing Associate Press photo in the Chicago Tribune today of an Afghan woman in a bourkah, begging for money in the middle of a Kabul street with cars buzzing all around her. It accompanied a story about the failure of the Administration to make any genuine progress on womens' rights in Afghanistan. So - can anyone clarify this disconnect for me?