Gallup accuracy

[shockingelk]

Here's a histogram of the deviation between Gallup's last poll before the election and the popular vote:



Party
overestimated a Democrat's popular vote 7 times and underestimated it 10 times
overestimated a Republican's popular vote 9 times and underestimated it 8 times
underestimated Democratic candidates' popular votes a net total of 7.5 points
overestimated Republican candidates' popular vote a net total of 1 point

Incumbency:
overestimated an incumbent party's popular vote 7 times and underestimated it 10 times
underestimated an incumbant Democrat's popular vote by a net total of 8.2 points
overestimated an incumbant Republican's popular vote by a net total of 3.1

Wartime (WWII, Korea, Vietnam):
overestimated the incumbent wartime party's popular vote a net total of 3.4 points

I used the data from here: http://www.gallup.com/poll/content/?ci=1258

[yellerpup]

1. Wow. The truth is a beautiful thing. n/t

[yellowcanine]

2. Conclusion - take Clinton out of the mix and Gallup has underestimated

Dem support in every election since 1968, except for 1984. - and their accuracy in general recently resembles the 1940s more than 1950 to 1980.

[shockingelk]

4. I think this years problem is ...

The ABB vote. The questions that they use to determine likely voters eliminate a lot of ABB voters, ie "How closely are you following the campaign?" "When was the last time you voted" ...

[SW FL Dem]

3. Interesting

For the past 20 years, Gallup has been pretty close on the Repub vote but way off on the Dem vote. Hope that trend continues.

[SoCalDemocrat]

5. That's still not accurate and this is why....

What they polled immediately before the election means far, far less, than how they skewed polls the month leading into the election. That is the period of time where poeple are making up their minds on how they're going to vote. On November 1st, it's pretty much decided, people have made up their minds.

Gallup in the 2000 election presented heavily biased and unsupportable numbers in favor of Bush leading into the election. The last day or two they showed more realistic numbers. Their results were not consistent with more credible agencies such as Zogby.

Gallup is doing the same thing in this election, showing Bush with rediculous 2 digit leads that other polling agencies like Zogby are unable to duplicate. They'll put their house in order when pressured or right before the election, but in the interim they are influencing a federal election with their biased publications.

Finally, look at the source. Gallup is reporting on their own bias?

Get their data that is 2 weeks to a month out from election and then compile the bias results. It will be off the chart.

[shockingelk]

6. I think what _you're_ saying is not supportable

I see no sudden change the days immediately before Election Day:

http://www.cnn.com/2000/ALLPOLITICS/stories/11/05/tracking.poll/index.html
http://www.cnn.com/ALLPOLITICS/1996/polls/cnn.usa.gallup/tracking/