Mixed poll results in VA Senate.

[longship]

Election Predictions is reporting two new polls in the VA Senate race. Mason-Dixon has the Allen/Webb race tied at 43/43 whereas SurveyUSA has Allen ahead of Webb 50/44. Both cannot be right. What's going on with these polls? Is anybody tracking the internals on these things?

I think polling this year is a bunch of bollocks. Does anybody know what they are doing?

[Singular73]

1. The fact that its tied is a huge victory, in any poll

You can't believe the 50..he hasn't polled over 50 in a long time.

[Lasher]

5. Actually, Rasmussen gave Allen 50% on 9/12/06

Here is their latest poll on the Virginia Senate race:

http://rasmussenreports.com/2006/State%20Polls/September%202006/VirginiaSenate.htm

[Tiggeroshii]

2. Most polls have given Allen a slight lead.

A few have had it tied and one(Zogby), gave Webb a 8 point lead. This would lead me to believe that Allen has a very slight lead in reality, much slightler than the average of what the majority of polls say he has(5 or 6 point lead). I would realistically give him a 3 or 4 point lead. It may or may not get better as time goes by, depneding on what people think about his alleged racism.

[Zynx]

3. Mason-Dixon is more reliable.

Survey USA gets it right by accident sometimes.

[Awsi Dooger]

6. I agree, Mason-Dixon is superior

I have more faith in them than any other polling company. They don't always release the crosstabs but when the newspapers they work with do provide the crosstabs they always make sense.

SurveyUSA will give you the crosstab percentages plus the raw numbers of respondents in each category, but sometimes the numbers don't jive. Like too many Republican-leaning Hispanics in virtually every SurveyUSA poll this year. Zynx also correctly pointed out in another thread that the percentage of blacks in the Virginia senate poll by SurveyUSA is too low at 14% when it should be low 20s. Also, very hard to believe Allen will earn 27% of those blacks as SurveyUSA is indicating.

When you make logical adjustments in that Virginia poll it comes out very close to even, same as Mason-Dixon. The only problem is Webb's favorables aren't very high which prevents him from taking full advantage of Allen's problems, especially with the womens vote.

[Zynx]

7. Interestingly, Mason-Dixon even pegged the white vote too high.

They had it at 81% or so as I recall. In reality, it's now about 71% white, 20-22% black, ~4% Hispanic, ~4-5% Asian.

[Lasher]

4. I have been tracking the Senate races closely for months.

I summarize where we stand according to the latest polls, based on when they were conducted, not released. Election Predictions is doing a good job, but Real Clear Politics is more timely in their poll updates.

http://www.realclearpolitics.com/epolls/2006/senate/va/virginia_senate_race-14.html

If you hit the above link you will see the Mason-Dixon poll that reported a tie, but you will also see the SurveyUSA poll that is more recent by one day. It gave Allen a 50% to 44% advantage, and it is the one that should be cited at this moment in time.

[Zynx]

8. It may be more recent, but its internals were off.

Mason-Dixon is much more reliable than SurveyUSA.

[longship]

9. How can one day mean a 6 point spread difference?

One of the two polls is flawed, or maybe both. As another poster has indicated, the SUSA poll may have flawed internals.

It's my opinion that many polls this season are screwed up because their assumptions are going to be just plain wrong. On top of that, they do not poll cell phones and they do not check their models by doing any door knock polling. They are going to overstate Repuke strength and they are not going to catch the problem until they find out on election day.

[Lasher]

10. You might be right but I just list the latest one regardlessly

I'm not what you would call a sophisticated poll tabulator or anything like that. I'm just a guy with a spreadsheet. Making a reliability judgment on one poll would lead me to do the same for them all. That is beyond the scope of what I am trying to do, which is just to keep a running tally of where we stand on the Senate races, based on the latest polls. As it is, with all the polls coming in so fast now it's sometimes a struggle for me to do even that.

I guess I could just visit a couple of sites where projection summaries are provided, based on more sophistocated methodology than I am able to employ, but this is just something I started doing for myself awhile back to compare against combined forecasts like those.

My favorite of these sites (for Senate races), Real Clear Politics, is currently predicting we will pick up 3 or 4 Senate seats in November.

http://www.realclearpolitics.com/latestpolls/

I also like Election Predictions, partly because they project both Senate and House outcomes. They're currently showing we'll have a net gain of 5 Senate seats (including the Independents) for an even 50/50 split. In the House, they're saying we'll end up with a 219 to 216 Democratic majority.

http://www.electionprediction.us/

[longship]

13. Most prognasticators use a concensus of polls.

They use an average a certain number of the most recent polls. It is most certainly incorrect to blindly accept the latest poll when there are such glaring discrepencies between different recent polls. So, I suggest that your methodology is not robust enough.

Without looking at the internals of the poll, you're not going to be able to make the necessary value judgement as to the polls validity. I've seen too many polls this year making unrealistic assumptions about Republican strength which overstates the number of conservative voters in the sample. This will skew the results.

Being charitable, I think that the pollsters really don't have an idea of how things have changed politically in the past year and are applying flawed assumption that the political demographic has not substantively changed. We're not in 2000 or even 2004. Those models ain't gonna work in 2006.

Some will say that the pollsters are deliberately skewing their results, and there has been recent press about precisely that. However, I take the more tame opinion that the pollsters really are trying to do their jobs correctly. I just think that they are substantially missing the target because it's location has changed so dramatically in the past 12 months.

In spite of all this, the polls overwhelmingly paint a very grim picture for the Repukes. There is really not a single Democratically held House seat in serious jeopardy. The closest Dem Senate seat is NJ-Menendez, in a very Democratic state.

I think the issues this year violate O'Neill's maxim, "all politics are local". This year, "all politics are about Chimp and his rubber stamp Congress." That's how things are going to play out. And nothing that ChimpCo is currently doing is going to change that dynamic. All they have is "stay the course".

[Lasher]

14. I don't claim to be anything more than one who tries to identify...

the latest Senate polls. That's all. I understand the complexities of weighting, taking into account the differences between RV & LV, and all that. When I speak to the issue here I'm not trying to make myself out to be someone who claims an ability to predict the final outcomes based on these factors. In that context it is not incorrect to blindly accept the latest poll when it is exactly that which I seek to convey, and nothing more. I started doing this just for myself but have found that there are times here at DU when I can help others by supplying this simple information.

I have learned a lot about polling from my friends here at DU, and I suspect you might be able to answer a question that has been on my mind for awhile. Will you please visit this thread, where I have so far not received a useful response, and help further my education?

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=132&topic_id=2855850&mesg_id=2855850

I would be grateful for that.

Lasher

[longship]

15. Aha! Yes, the Zogby Interactive polls.

I think Zogby is attempting to put together models to increase the validity of such strictly voluntary polling techniques. It's obvious that this kind of polling would result in highly skewed results. That's why few people will cite them. But Zogby seems to think that there's some valid information there, and they may be correct. The only issue is, how do you make the silk purse out of the sow's ear?

I participate in these, and I imagine many other DUers do, too. But we all realize that they are going to be almost hopelessly skewed.

[Lasher]

16. So I guess you helped answer your own question in your OP

RCP, for one, is tracking internals in that they list Zogby as well as those that are funded by the political parties, but do not use them when they develop their averages. You would think that RCP would explain this somewhere at their website. Maybe I missed a footnote somewhere.

Still, this does not reconcile the difference of 6% you mentioned between Mason-Dixon and SurveyUSA. I've noticed that you and some other folks don't like SurveyUSA, but RCP doesn't exclude their polls.

A lead for Allen ranging from 2% to 4% would be within the 4% MOE cited in each poll, so that's one way to reconcile the 6% difference.

Surely you've noticed, the variance between these two polls is not unique, and not as pronounced as some others.

[Imagevision]

11. 2 weeks ago pollsters found guilty of fraud favoring Lieberman & Bush

It would appear extremely damaging to GOP leading up to elections if Allen showed losing in the polls, Webb most likely is ahead in the true/honest polls.

[paulk]

12. AFAIK, Survey USA does random telephone polling

which doesn't take into account party affiliation, or potential turnout - that doesn't seem to be a very accurate way to predict elections, but they claim to have been the most accurate poller in 2004.