Final Zogby poll numbers VS Results of Election Night (A FR Archive thread)

Zogby VS. The Results ^ | Nov. 6, 2002 | Jern

[ambrose]

SurveyUSA's worst polling came from SC.

And best polling was in Oklahoma. I don't think anyone else saw that upset coming.

[KQQL]

POA AKA SCOTT POllS might be BAck:

Check out his last polls:

MN:Norm Coleman defeated Walter Mondale by 2 percentage points to become Minnesota's new Senator. Scott Rasmussen's latest survey, conducted Sunday, had projected a toss-up race 48% to48%.

SD:Scott Rasmussen Public Opinion Research found that the Johnson and Republican John Thune were just one-tenth of one point apart.

[Howlin]

jern, thanks so much for doing this!

[KQQL]

overall SurveyUSA did a good job..
They caught the last wave about races in CO, NC,OK
AZ, MI,CA,TX etc..

Only screw up was SC

[BlackRazor]

I've just completed a cursory analysis of the accuracy of independent pollsters for the 2002 cycle. The only poll numbers considered were an organization's final poll for a given race, and only if it contained polling dates Oct. 21 or later. Only Senate and Governor races were considered. I also lumped together the qualifying results for the various local polling agencies (e.g. Ciruli, Quinnipiac, Field Poll, ARG, etc.) to come up with their composite accuracy. A final note - pollsters were not penalized for races in which the actual margin of victory was greater than 25 points (it's easy to miss by 10 or even 20 points in blowout races, thereby skewing the overall accuracy average).

A polling organization was reviewed only if it conducted 5 or more polls that met the above criteria.

The results, in order of descending accuracy.

1.) Gallup
Average error: 2.8 points
Correctly called 4/5 races (80%)

2.) Mason-Dixon
Average error: 3.4 points
Correctly called 17/18 races (94%)

3.) Rasmussen Research
Average error: 3.5 points
Correctly called 2/5 races (40%)

4.) SurveyUSA
Average error: 4.6 points
Correctly called 26/28 races (93%)

5.) Local Pollsters (Composite)
Average error: 4.9 points
Correctly called 71/79 races (90%)

6.) Zogby
Average error: 5.0 points
Correctly called 12/17 races (71%)

7.) Research 2000
Average error: 5.2 points
Correctly called 3/5 races (60%)

Gallup was the most accurate in the races they polled, but I hesitate giving them the overall prize since they only polled 5 races (that I'm aware of). I'd probably give the year's award to Mason-Dixon for best accuracy among extensive pollsters. Zogby comes in dead last among that small group, both in terms of average error and in ability to correctly call winners. I think it's safe to say that the Zogby mystique (if there really was one) is gone.

Despite alot of the criticism they have taken, I wouldn't be quite so harsh on SurveyUSA in the final analysis. They're clearly not in the upper echelon, but they're not that bad, either -- perhaps slightly better than your average local pollster. Their overall numbers were somewhat skewed due to being so far off in a couple races in SC and IA. Remove those, and their average error was a pretty respectable 3.7 points. They polled MI, OK, CO, MD, OH, AZ and TN as good or better than anyone else. For their cheap price and large number of polls they conduct, I can't really criticize them.

[nutmeg]

Nice work! Thank you for posting this.

[woofie]

Heck he was only of by a few numbers on most of em

[KQQL]

Largent lost this race, when he mouthed off,,,
AR GOP lost because of
Hutch wife problems.....

[KQQL]

O>K and AR ..only southern states where Rats are still loved.....

[KQQL]

Thsi thread should answer your questions..
ZOGBy was wrong, because he spins his polls, and this time
his dislike of W showed up in his polling bias...

Gallup , Mason -Dixon and Survey USa out performed the
ZOg Man...
SurveyUSA worst polls were in SC and IA..other wise they did o.k

[ffrancone]

Here's the reply from KQQL.

[Guillermo]

Zogby was 5 for 9 in picking winners over losers and 3 for 9 in coming close to teh actual results (within 1-2 points on each end).

Like I've been saying all along. Either believe them all, or believe none of them, because there is no way really to discern which one of them are correct prior to the actual election.

I'd prefer it if no polling was done, other than a generic "who will you vote for GOP or Democrat" poll.

[PolishProud]

According to a Fox News election day poll regarding the Allard-Strickland race, newly arrived Liberal voters canceled out native or long time Colorado conservative voters, but senior citizens went 53 to 46 for Allard.

Does anyone know how senior citizens voted nation wide?

[Hurricane]

I had no less than 5 pollsters call prior to the election. I told them all to go pound sand. I refuse to be a part of any misleading poll.

[Always Right]

Out of 9 races, looks like 3 were outside the margin of error. Perhaps a larger margin of error is called for.