Why were Zogby polls so different from election results in IL 6?

[AtLiberty]

Zogby released the following poll results for IL 6:

Oct. 2, 2006

Duckworth 43% Roskam 38%


Oct. 29, 2006

Duckworth 54% Roskam 40%


ELECTION RESULTS:

Roskam 52% Duckworth 48%


Can anyone explain an 18-point flip in just one week?


Zogby was 10 for 10 in the Senate races, a little "off" in a few Congressional races, but "off by a mile" in IL 6. ?!?!?

I can provide Zogby data from this race if you write me with your e-mail address.

[skipos]

1. If it was one of their "interactive"polls, those tend to be pretty irratic. nt

[OnTheOtherHand]

8. actually, apparently it wasn't, but...

http://pollster.com/polls/?state=IL06&race=house_race

The Reuters/Zogby polls are phone polls. (See also http://elections.us.reuters.com/content/midterms/zogby5.html .)

But this result was way out of line with every other poll in the district. I think it's not only Zogby's interactive polls that can be erratic.

[AtLiberty]

9. So was Zogby's Oct. 2 poll off as well?

Duckworth 43% Roskam 38%

[OnTheOtherHand]

11. why ask me?

Look at the data for yourself, form your own opinion, and argue for it if you like. 43% in early October obviously isn't exactly a lock. (Am I vouching for the vote count? of course not. So far we're talking about polls.)

[Teaser]

2. Because Zogby is smoking metal shavings

nt

[On the Road]

3. If Zogby Uses a 1% Confidence Interval,

It means that four polls of House races were likely to have been outside that interval.

[mirandapriestly]

6. what does that mean?eom

[Febble]

12. I think he means

that if you do enough polls, a predictable proportion will be wrong outside their "MoE". If, for example the 95% Margin of Error is 2 percentage points , that means that 1 in twenty polls will be outside that margin of error. There were a heck of a lot of polls before the election, so certainly some would be outside their MoE.

But I can't help as to specifics, as I don't know what the MoE was for the Zogby polls, what the confidence criterion was, or how many polls he did.

[philb]

40. Do we have any reason to think that "official results" are more accurate than a well designed poll

in recent years? In many cases not; with all of the switching, glitches, machine problems, and other types of
manipulation widely reported. A difference with the polls could likely mean look into possible manipulations.

[slackmaster]

42. If they aren't then the "officials" running the elections need to GO!

:nuke:

[On the Road]

13. It Mean That Polls are Not Exact

If you have two October polls, one showing 43% and one showing 54%, and the vote comes in about halfway between, that's not especially troublesome. Look at the variability of the Bush polls from various sources. That kind of spread is not too unusual.

There are a variety of reasons polls can be off, even just on a random basis. Pollsters often use a 5% confidence interval. That means when you're dealing with over 400 House and Senate races, over 20 races (5% on average) are going to be outside the pollster's margin of error.

[AtLiberty]

4. Zogby used the same polling methods in all races...

Hmm...

[diva77]

5. has Duckworth conceded?

will there be a recount for this one?

[AtLiberty]

7. Duckworth conceded at 11 pm on election night...

Recount highly unlikely.

[AtLiberty]

10. Has anyone heard if Zogby conducted exit polling here?

I read on their website that exit polling was conducted in Illinois. Does anyone know where in Illinois?

[ISUGRADIA]

14. The Zogby poll was what is called an outlier compared

to every other poll of the race from the time period.

[BeFree]

15. Oh?

Here are TIA's polling numbers 


 Average polls                                    
      Early     Latest           NOV.7        Fraud  
Dis    Dem GOP Dem GOP...      Dem GOP      Dem GOP 
IL 6   48 47    54 40          57.6 42.4    52.0 48.0

[AtLiberty]

16. OMG -- you used the "F" word!

Wash your keyboard out with soap this instant!

[nicknameless]

17. Worse than that. (Typo)

Fraud

Dem GOP
48.0 52.0

[BeFree]

22. Good eyes, nick

Looks like theft of some kind in that race. Anyone know what kind of vote capture devices were used? DRE's?

BTW went back and looked at the original TIA numbers, it was a typo from the get-go. Still, I should have seen it when I posted it here. Thanks for not raking me over the coals.....

[ISUGRADIA]

24. Yes, ignore any data (other polls) that support a pre-

determined outcome (fraud). Other polls showed it was close, Republicans pour hundres of housands more $ in the last week, Republican wins. Plus the poll listed was a week old! One of the base rules of statistics and polling is a poll only reflects opinion the day or days on which it was undertook.

No statistician is going to say there is a 100% calculated chance for an outcome based on data a week old. Various events ($$$ in GOP ads, turnout efforts, etc.) impact actual day of the election numbers.

[Name removed]

25. Deleted message

Message removed by moderator. Click here to review the message board rules.

[ISUGRADIA]

27. Please ask questions, there are things that stats can be used for

in elections very effectively. The huge undervote in the undecided FL House race for one example. How the butterfly ballot in 2000 caused a huge discrepancy in Buchanan votes compared to absentees beyond any realistic explaination (Costing Gore 2-3 thousand votes and the presidency from it alone) is another.

But for people to take this seriously, those who have access to number crunchers who have training and know stats, the analysis has to be sound. I questioned the TIA analysis because it fails on the stat 101 level in many areas. I think 100% chance of victory based on polls that can be flawed overshadows many of the concrete facts of election rigging that can be done in a legal or barely legal way (Electonic touchscreen voting - More time need to vote = longer lines = people giving up trying, Undersupplying voting equipment to minority areas, to name two). Analysis is needed but in a scientifically accurate way.

One more example of how shoddy research can be used to discredit election reform. In Mark Crispin Miller's book he tried to paint Minnesota as a possibly stolen state in 2002 for the Senate race because of election poll discrepancies i.e. the machines stole it. Problem was no US Senate ballots were counted by machine. It was took late to configure the scanners bacause of Wellstone's death so all ballots were hand counted. Miller has backed off on the claim he made for MN but some thorough research would have helped him right off.

[BeFree]

28. Excellent!

So you know what type of computers were used in Illinois 6?

[ISUGRADIA]

29. I'm not sure, I think that info might be on the Secretary of State

webpage for the county or on an election reform site.

Any county or city that uses touch screen machines should be watched because, let's just assume for fun no fraud exists, PCs can fuck up. We've all seen it happen. At home we can reboot. Something major happens such as in Florida in this House race because of software problems NO PAPER BACKUP. A manual recount of paper ballots could prove a 16% undervote in the race if they had ballots. A scanner tabulator at least has a piece of paper marked by human hands.

[ISUGRADIA]

30. DuPage County uses tabulator scanners and touchscreen

It looks like they have a touchscreen machine in each precinct because of HAVA requirements. Claim is only a few voters will use the touchscreen "the optical scanning equipment that Saar suspects 90 to 95 percent of DuPage County's voters will use"

http://www.suburbanchicagonews.com/napervillesun/news/124607,6_1_NA05_ELECTION_S1.articleprint

[BeFree]

38. Good info!

Can we be certain that most votes were cast on a paper ballot then scanned?

The next question is: was an audit conducted?

[eridani]

18. These are not EXIT polls

Therefore discrepancies are pretty meaningless.

[Awsi Dooger]

19. The 54-40 didn't make much sense

Illinois-6 is a red leaning district in the western Chicago suburbs. According to my notes it is trending our way but still had a -3 partisan index in 2004, compared to -10 in 2000. That means in a presidential year it voted 10 points more Republican than the national average in 2000, and 3 points in that direction in 2004.

It was an open race (no incumbent) after Hyde retired and in a wave year I certainly thought we could take it, but some reports indicated Roskam was more polished during the campaign than Duckworth. She is the double amputee Iraq war veteran.

The odds were always in the 50/50 range, not heavily in Duckworth's favor. That Zogby poll was indeed an outlier. Every other poll had the candidates within 5 points either way. Remember, Duckworth barely won the primary versus the 2004 Democratic nominee. I forget the specific margin but it was a handful of points.

That was also a robocall district with tens of thousands, if not hundreds of thousands, of annoying phone calls in the final week, designed to harm Duckworth. A late poll, by Zogby or anyone else, might have caught that.

Here's a good link to evaluate the demographics of House districts. Illinois-6 is drifting our way, partially due to slowly increasing numbers of Hispanics and Asians, but notice it has less than 3% blacks, compared to more than 14% statewide.

http://fastfacts.census.gov/servlet/ACSCWSFacts?_event=ChangeGeoContext&geo_id=50000US1706&_geoContext=01000US%7C04000US17&_street=&_county=&_cd=50000US1706&_cityTown=&_state=04000US17&_zip=&_lang=en&_sse=on&ActiveGeoDiv=&_useEV=&pctxt=fph&pgsl=040&_content=&_keyword=&_industry=

[OnTheOtherHand]

20. great post, but I'm laughing a bit at one thing

Can you imagine being one of the interviewers who is trying to conduct a late telephone poll during a barrage of robocalls? I'm glad I don't have that job. But really, anything that leads people to associate a ringing phone with an annoyance is bad news for pollsters. (Not that that factored in with IL-6, because as you noted, there were no late polls in IL-6.)

[Awsi Dooger]

35. Ha, let's hope it was dueling IVRs and robocalling

Great stuff, OTOH. I never thought of that. Maybe there will be anecdotal comments from pollsters who were abused during robocall week.

It might be interesting to check which districts, if any, were polled late and also on the robocall list. Someone has probably done that, but I didn't notice.

Might have been a tough week for pizza delivery drivers also.

[Elmer1007]

21. When did Duckworth come out for Gun control.?

Could her position on Gun Control have affected the vote vs. the earlier polls?

[Awsi Dooger]

34. I wasn't familiar with her gun control stance

That's not a district I followed closely but it did receive national publicity, partially due to Duckworth's vet status. I think McCain made a stupid comment while campaigning for Duckworth's opponent, something like, "many soldiers have lost limbs, you know."

I keep basic data on competitive districts and try to watch C-SPAN debates, plus look at the polls. For some reason the House polling seemed to end early this year, especially compared to senate and gov. One more late round by Majority Watch would have been helpful. I think they'll do that in '08, particularly since their numbers held up this time.

[slackmaster]

23. Human behavior does not always follow rational trajectories

People change their minds on short notice and often with little apparent stimulation.

[BeFree]

26. Computers don't always follow rational trajectories

Computers change the numbers on short notice and often with little APPARENT stimulation.

Now then, if you have proof the computers didn't change numbers, we'd all love to see that proof.

It's like leaving an open gun laying around the house... eventually the trigger's gonna get pulled, eh?

[slackmaster]

31. You are asking me to prove a negative

Computers generally do exactly what they are programmed to do. Only in cases of hardware malfunction do they behave otherwise. Most reported bad behavior by computers is the result of operator error or misunderstanding.

Did anyone get video footage of any touch-screen voting machines changing votes last Tuesday BTW?

BTW BeFree, your analogy of the unsecured firearm is flawed. I'm certainly not in favor of voting systems that are both hackable and unauditable any more than I am of leaving loaded weapons lying around, but many here would take the presence of a gun as evidence that it has already been fired. If it's not warm or smoking, that's carrying things too far IMO.

[BeFree]

33. Of course

When we speak of computers we are talking about the human programmers as well. Same with guns. Guns don't kill people: Computers don't steal votes.

So computers don't follow a rational trajectory, computers follow human beings.

[slackmaster]

36. I take it by your lack of comment there is no video

No smoking gun.

[BeFree]

37. Video?

No video, no matter how instructive, would ever be seen by those whose eyes and minds are closed.

I see no smoking gun, but I do see a few smoking computers. And a whole lot of funny numbers, don't you?

[slackmaster]

39. There are two kinds of "funny" numbers

Ones that are auditble, and ones that are not.

The ones that are auditable can be challenged and, if incorrect, fixed.

Unauditable systems are obviously unacceptable.

No video, no matter how instructive, would ever be seen by those whose eyes and minds are closed.

Translation: Despite all the bluster and outrage and calls for people to take cell phones and video cameras into their polling places to catch maliciously programmed voting machines in the act of thwarting voter intent, there is no video.

The Conspiracy has no clothes.

[slackmaster]

32. Funny how Zogby is both praised and condemned on DU

When a Zogby poll suggests Democrats should have done better than the reported election results, Zogby is praised as an election reform hero.

When Zogby polls report bad news for a Democrat, there are always DUers who will accuse them of whoring for the GOP.

[philb]

41. Is anyone auditing the race for undervotes, absentee irregularities, etc.?

See the thread on 8 undecided races for suggestions on what to audit