Yet, another opinion on polls. This one by John Zogby in an

[RoundRockD]

interview in the Washington Post. Note the last sentence.

Brooklyn, N.Y.: Mr. Zogby -- last week CNN-Gallup and WAPO-ABC released polls showing Bush with a 4-6 point lead, while you (whose polls I put much more stock in) released a poll showing Kerry with a 3 point lead. Presumably, this does not come from the margin of error -- how is your selection technique different from the other two polls?


John Zogby : It really isn’t very different. I use listed telephone numbers instead of random digit dialing. At times I apply a weight for political party identification while at other times I don’t. But again, I think the polls are much closer to each other than they are apart and I would ask this question – after the two weeks that the president has had do people really believe he got a bump in the polls?

[buddy22600]

1. Ok, so he fudges the numbers?

[Caution]

2. Ummm huh?

I'm trying to figure out how you got that from the response of Mr. Zogby.

[Cassandra]

3. I was wondering that, too

[buddy22600]

4. He said sometimes he applies different weight to answers

that is what I am saying. If he wants to show some result all he has to do is weight those answers

[Ediacara]

8. ALL pollsters do that

That's why polls have demographic questions at the end. He applies answers you give coupled with your demographic responses to a statistical model built on demographic data.

All pollsters do this. In fact they HAVE to do it because the raw data has a huge degree of randomness. Some pollsters have better statistical models than others, and Zogby's seem to be better than most.

[izzie]

12. I am not sure what that means Weight? What?

I will say that I once head one of Rush's rants and it was this is the only guy to believe in polls. Guess if it is good enough for Rush it is good enough for the GOP. So I will look at it week or so.

[Caution]

16. Reason for weighting

If you randomly dial listed numbers and ask who they vote for for president and 90% of respondents say "John Kerry" and 9% say "Ralph Nader" and 1% say "Dubya" and then you ask what party they belong to and 75% of the 90% who said Kerry answer "Democrat" and 100% of the 9% who voted for Ralph say "Throw My Vote Aways" then you can safely surmise that the 1% who voted "Dubya" have an IQ of the same value as their percentage of votes.

Seriously though. If you poll 3500 people and 3400 say Kerry but all 3400 are registered Democrats through sheer chance, you have to weight the results or the poll doesnt accurately reflect the data.

[sandnsea]

6. Ummm, right here

"At times I apply a weight for political party identification while at other times I don’t."

WTF does that mean??

[Ediacara]

5. More like WaPo/CNN etc

Fudged theirs to give the appearance that Bush didn't suffer.

[buddy22600]

7. I'm not saying that on his latest poll

but he said he weights answers sometimes and then doesn't do that

[Ediacara]

10. If the responses he gets are skewed versus his statistical model

Like in a state with 50-50 voter registration, and traditionally close races, he happens to call up a sample that's 74% Republicans, he'll alter the responses so that they can better fit the standard statistical model of 50-50.

Or he could scrap that data.

Or he could make more calls.

Or he could alter the model.

It's not something HE ALONE does however. Pollsters have to do it, or chance will give us poll numbers that fluctuate by 10s of percentage points on a daily basis.

[daa]

9. Just remember, Zogby is usually the one on the money nt

[Nicholas_J]

11. Not really

Zogby was amazingly off during the primaries and usually right for Presidential Elections is unproven as Zogby has only been around snce 1999 anyway. He predicted correctly in 2000, but since that, he has not been as accurate, especially during the Democratic Nomination Primaries in which ARG and a few other pollsters more closely nailed down the final results.

Facts are that this election may be the closese in history, with the lowest amount of crossover votes between parties, with more Republicans crossing over to vote for Kerry than Democrats crossing over to vote for Bush (at least a two to on ratio, more likely three to one I am estimating that a mere 2 to 3 percent of Democarts will vote for Bush, while anywhere from 6 to 8 percent of Republicans will vote for Kerry.)

Right now the percentages are close, it is likely that less than ten percent of the people who have already decided that who they will vote for will change their minds. Leaving something like 7 percent average undecided to make up their minds over the next 6 months.

This election will be closer than 2000, and there are 4 states which are too close to call which have 20 percent of the voters for popuar vote. Those are Florida,Ohio,Minnesota, and Pennsylvania. Rght now Florida is too close to call, with Kerry having a tiny lead. He has Minnesota locked up, Bush has a small lead in Ohio right now, but that was only in the one last poll, in a larger number of polls, Kerry was ahead, and in Ohio, it will be jobs and the economy, not Iraq that wins the state, Bush leads sligtly in Pennslyvania.

Gore won the popular vote by 500,000 votes in 2000. I seriously doubt that Bush is going to be able to pull even a small percentage of those who voted for Gore in 2000 from Kerry. On the other side, significant numbers of Republicans are indicating that they will vote for Kerry, not Bush. GIven these facts, I estimate that Kerry will win the popular vote by anywhee from 2 to 4 points. The electoral college strategy may go in BUsh's favor. but I sincerely doubty thatn anyone is going to allow a Florida Scenario to occur again. Certainly not the Democartic PArty, who will be watching closely, not just in Keye states, but in every state.

[TruthWins]

13. Don't mean to be negative

But anyone that thinks all the Gore voters from 2000 are going to vote Kerry is off the mark. Most will of course, but I am absolutely certain that there are people who will go Bush this time that didn't last time. A big part of those are middle-aged white men who like a good war and feel only Republicans can wage it. (And I'm a 34-year-old white man voting for Kerry, so I'm only slamming my peers.) I hope I'm wrong, but I'm not counting on having all the Gore voters on the Kerry side and having to just convince a small number of Bush supporters from 2000 to cross over. And don't forget Nader. Who knows what impact he's going to have. He could pull less than a half percent in some state and have it be enough to swing the whole election, that's how close this could be.

[Nicholas_J]

17. Over the last few months

As it has become more and more obvious that Kerry will be the nominee, the percentage of crossover votes with Democrats voting Republican has dropped to a historic low since the start of both parties. While it was thought that Dean would be the nominee, the polls indicated that as many Democrats would cross over to vote for Bush as did at the highpoint of the Reagan Democrat craze, which was 20 percent Dems stating in polls that they would vote for Bush if Dean was the nominee. Republicans who would vote Dem was ranging between 10 and 12 percent. By the time of New Hampshire, the percentage of Republicans who stated they were going to vote for Dems was 12 percent, but the percent of Dems who stated they would vote for Bush dropped to 10 percent. Right now polls tracking this same data have Republicans who will vote Dem down to 8 percent, but Dems who will vote for Bush has dropped to 2 to 3 percent in where this info is polled. Polls like ARG and a few orthers who do in depth polling like this are all coming up with similar data. Dems are more firmly supporting the Democratic candidate more firmly since any elecion since 1980.Far from appealing to those middle class white males (he does to a vocal group of these) these people are falling away from Bush rather rapidly.

19. Do you have any links for this?

Where are you getting these numbers from? I agree it's ridiculous to assume every person who voted for Gore in 2000 will vote for Kerry this time. Some may feel we're in the middle of a "war" and we shouldn't change presidents, some may have decided they like Bush ok after all, some may not like Kerry, etc. There's no way to know how many Gore voters will vote for Kerry.

[Nicholas_J]

20. Actually

It would take a few dozen links...you can go directly to AmericanResearch.com and Quinniapiac and the other pollsters who do in depth polling and go back to the pools they did before and through the primaries...

realclearpolitic.com has links to pretty much all of the pollters and its just a matter of going back through all of the old ones. The better pollsters break down the data by Democrats and REpublicans and who they say they are voting for.

A few pollsters are not listed on realclearpolitics, like sufolk university who did in depth polling for the New Hampshire primaries with similar info, because they are a local pollstet that does polling in New England. I have been closely following this stuff, and James Carville has done so as well and his analysis is pretty much the one that I have had for the last few months. That this will be one far closer than 2000, and that Bush will get far far fewer votes from Democrats than he did in 2000. The polls have indicated an ever smaller number of democrats who will vote for Bush.

SAme stuff has been turning up in National Polls that do this kind of breakdown, but as I said, there are dozend of links to dozens of polls that show a consistand drop for Bush among Democrats. ARG and Suffolk University showed the comparison of each Dem candidate against Bush indicating that Dean would have the largest number of Dems who would vote for Bush if he was the nominee. Kerry had the lowest rate at the time of the New Hampshire Primaries. As well as the highest percentage of people who voted for him who beleived he could beat Bush. Dean had the lowest of all the major candidates with only 22 percent of the people who voted for him in New Hampsire beleivingthat he could beat Bush if he were the Democratic Party nominee. Out of those who voted for Kerry, 54 percent beleived he had a chance of beating Bush. Clarke and Edwards fell between thise two. If it were matter of one poll, I would gladly give one link. But I am not going to search down the info I have been following in every poll over the last six months. Too much work.

[PA-DEM]

14. Just facts please

Anyone who ever visited Zogby's web page knows he's been around since 1984.

[RoyGBiv]

15. Weighting

"Weighting" is not a conspiracy, it's not fudging, it's not anything similar. It's a standard practice of statistical analysis that is supposed to take into account variations in the poll sampling data caused by factors related to the polling process. There are different weighting procedures, different methods of employing them, and different instances in which weighting may or may not be used.

In short, pretty much everyone who conducts a scientific poll uses weights, and their intent is to reduce the influence of bias, non-responses, and certain categories of targets being unable to respond.

A standard textbook on statistical analysis can explain this much better than I can.

[salonghorn70]

18. Two Points

Rush's love affair with Zogby came in 1996 when Zogby was showing the Clinton-Dole race closer than other polls. That was when Rush thought Zogby was the greatest pollster in the history of civilization. Rush failed to mention Zogby much in 2000. :)

Ok, Statistics Experts. I am trying to understand why there would be a difference between Telephone book numbers and randomly generated numbers.:shrug:

[RoyGBiv]

21. Two Other Points

1) Rush is a waste of oxygen. But then you already knew that. ;-)

2) Random digit dialing finds people with unlisted telephone numbers, or people who've changed their number recently. When Zogby refers to telephone book numbers, he's probably referring to something called a phone matched list from which numbers are randomly chosen. This is basically a list of likely voters for which a phone number has been matched to a name. People who have never voted before, those who move around a lot, and very wealthy individuals tend to be under represented in these polls. On the other hand, random digit sampling can yield skewed results as well because sometimes the pollster doesn't ask "do you plan to vote?" or people can claim they will vote and then not.

So, it's controversial which method is preferable, but this is where weighting can come into play. When a person has an established track record of voting, his or her answer might be given more weight than an answer from someone who has never voted.

I am certainly no expert on this. I've had to use polling data for some research projects, and I somehow ended up with a few friends who use statistics in their jobs. They explain stuff to me, and I have to say "okay" because all the equations make my head hurt just looking at them. One of my friends said today that in his view random digit sampling is better for polls that simply gage overall public opinion. The reason for this is that one can hold an opinion passively. Matched lists are better, again in his view, for voting preference because there are very many people who have opinions, will answer the poll sometimes even lying about whether they plan to vote, and then never actually do anything about it. People who have voted in the past are more likely to vote in the future, and these are the people that comprise a matched list.

Corrections welcomed, but I think that's about right.

[DemocratSinceBirth]

22. With Radomly Generated Numbers You Get People Who Have Unlisted

Phones.....


I wish he would expanded on his reasons why...


Of course you have to weight....


Your poll has to represent the populations you whose views you are trying to measusre...