What poll should I consider accurate?

I’m wondering what poll webistes are the best to go to. I usually check CNN, RCP and the one on Yahoo, but they all have differning numbers. While all of them look pretty good for the dems I’m not sure which one is the most accurate. there must be a other polls that are. What polls pad the numbers and which ones are trustworthy?

http://www.fivethirtyeight.com/ is a site which tries to take the bias out.

http://3bluedudes.com/ProjectDatabase.htm is a list of sitea like fivethirtyeight.com so one can see the overall trends.

You should trust a poll that shows it’s polling process and the breakout of the party affiliation of the people polled.

Best of luck with that.

They all provide that, but you have to pay to get access to that info.

In short, you shouldn’t rely on any one poll. There is too much statistical noise. Your best bet is to follow the average of a bunch of polls, like www.realclearpolitics.com or a weighted system like www.fivethirtyeight.com.

Don’t use any single poll, use tracking sites like fivethirtyeight as others have suggested.

The reason 538 is arguably the best is that it actually weighs individual polls based on the pollster’s transparency, methodology, political bias, and accuracy. Pretty standard statistical regression other than that, but it gives you a much better idea. Nate Silver is a democrat, but he tells you clearly the justification for his weightings, and any exclusions or inclusions. The other aggregate poll sites not so much.

RCP isn’t transparent enough for me to trust them. That’s the really big deal about 538; not only are Nate’s methods arguably the best, but he’s open enough about them that we can verify that.

electoral-vote.com seems good as well. He takes an average of all non-partisan polls in the last week or so*.


*Take the most recent nonpartisan poll. Call it P. Find the day in the middle of the days that P covered. Now take all nonpartisan polls whose middle day was within five days of P’s. Average the results of all these in with the results of P. Post.

In addition to following some of the excellent suggestions in this thread, here’s an additional thing to check out…

FWIW I’m making an appointment to listen to the final hour of “Bob Brinker’s Moneytalk” on Sunday, Nov. 2 on ABC Radio.
There’s a tradition that on the Sunday before a major election for Brinker to bring pollster John Zogby on the last hour and go over the rolling averages.

Two years ago Zogby hit the exact numbers for the changes in the House and Senate.

A big difference between the results is how they judge a “Likely Voter” (there is less difference between the “Registered Voter” numbers of different polls than the “Likely Voter” numbers). As discussed at fivethirtyeight.com:


I’m more than a little surprised to see that nobody has recommended fivethirtyeight.com yet. You ought to check it out.


The poll on November 4 should accurately show who will be elected.

Yeah right. You probably said that in 2000, too, and look at what happened. :stuck_out_tongue:

Last time somebody asked this question, someone posted a link to Abulseme.com’s Electoral Projection graph. I thought it was pretty cool although I’m not a statistician or anything. Scroll down and there’s an explanation for the graph and also a link to earlier discussion on the Dope about his methods.

Aside from that, I owe Nate Silver my sanity during this long national nightmare of an election season.


Polls are worthless. Even if they’re accurate RIGHT NOW (which they’re not) they’re virtually guaranteed to differ from election results as some people change their minds, as a campaign makes a successful advertisement, as new info is released on candidate positions and backgrounds, as candidates get caught on camera making a goofy face, etc. etc. Polls don’t measure anything actually useful, except possibly for trend direction. Even then, polls can’t predict how long a trend will be sustained and how powerful the trend will be.

I’m sure John McCain will be pleased to know that it doesn’t matter that he’s 8 points down in the polls, since they’re worthless and all.

The folks paying for the polls would be happy to know too, what a waste of money!

Hehe, yeah exactly.

It amazes me how many people simply don’t understand the validity of statistical sampling within a given confidence interval; that it is actually very accurate, reliable and predictive at measuring political preferences, of course, subject to all the usual caveats about methodology, survey question design, registration and intention controls and data sets.

They just have this idea in their heads that all polling is equally inaccurate, or that because polling tests contemporaneous opinion, that it must be absolutely meaningless to the vote.