Projected New Senate:     48 Democrats     52 Republicans    

Polling data in Excel Battleground states Previous report Next report News Strong Dem (43) Weak Dem (2) Barely Dem (3) Exactly tied (0) Barely GOP (5) Weak GOP (0) Strong GOP (47) No Senate race   Map algorithm explained

Sep. 05	New polls:	MD MN MO MT NM OH PA
	Pickups:	Montana New Jersey Ohio Pennsylvania Rhode Island

Welcome	General FAQ	Polling FAQ	Votemaster FAQ
Donations	All Senate races	Hot House races	Poll Graphs
More election data	Senate info	Political websites	Articles and books
2004 Election	EVP track record	Political humor	For bloggers
Contribute a page

News from the Votemaster

I found the dates on the Gallup polls I mentioned Saturday. They were all Aug. 23-27. Here are the numbers.

The main surprises here are that Sen. Rick Santorum (R-PA) is even further behind state treasurer Bob Casey (D) than he had been and Sen. Jim. Talent (R-MO) is even further ahead of state auditor Claire McCaskill (D) than he had been. I suspect these numbers won't last though.

It is worth pointing out that Gallup was involved in a huge controversy in 2004 about party normalization. If you took a large enough sample, say, 100,000 people, you would find out how many people are Democrats and how many are Republicans from the sample itself. But sampling so many people would be outrageously expensive, so typical state polls sample 500-1000 people and many pollsters correct for party identification. As an example, imagine that a state poll sampled 900 people and found that 300 were Democrats, 300 were Republicans and 300 were independents. Now just suppose that the exit polls from 2004 showed that 40% of the actual voters were Republicans, 35% were Democrats, and 25% were independents. But the pollster believes that 40% of the people who actually vote are Republicans (based on the previous exit polls). What a pollster can do is compute the poll results by weighting the 300 Republicans for with 40%, the 300 Democrats for 35% and the 300 independents for 25%. This method corrects for the lack of an accurate sample by party.

The rub comes in knowing what fraction of the electorate identifies with each party, if any. It was once thought that party identification was very stable, sort of like religion. Recent data call this assumption into question. What Gallup did in 2004 was use an assumed percentage of Republicans in the electorate that other pollsters felt was much too high, thus counting the opinions of the sampled Republicans more heavily than they should have been, consequently skewing the results towards the Republicans. Gallup defended its methodology, of course, but the area is still controversial.

I often point out that while some people are fixated on margins of error, they are small potatoes compared to methodological issues like this and how do you get a truly random sample and what happens if the only people willing to talk to the pollsters are little old ladies with nothing better to do, and how do you sample whose only phone is a cell phone, and so on. For more on polling issues, see the Polling FAQ on the menu above and also the many articles on the Articles and books page above.

The NY Times had a good article yesterday in which widely-cited political gurus Charlie Cook and Stu Rothenberg both think the Democrats are likely to retake the House. Both think the Senate will be more iffy. I concur with them (but of course we are all looking at the same polls). For more on the House races, click on the Hot House races link above.

WWW www.electoral-vote.com