The Wild Differences in The Polls, Explained

[JimGinPA]

If you’ve been watching Presidential preference polls over the past week, you might feel a bit whipsawed.

For the second straight day, Gallup’s daily tracking poll Thursday has John McCain and Barack Obama tied. Both candidates dropped a point from yesterday’s tracking poll, down to 44 percent. The margin of error is +/- 2 percentage points.

The Rasmussen daily tracking poll in the same period of time has shown a 3-7 point gap between the two candidates.

Neither of the daily tracking polls square with two polls from that show Barack Obama holding a stunning double-digit lead. A Los Angeles Times/Bloomberg poll showed Obama with a 12 point lead over McCain, and a Newsweek poll had Obama leading by 15 points, if Ralph Nader and Bob Barr were included in the mix.

Why the difference? Were the Newsweek and L.A. Times biased in favor of Obama? Do the Rasmussen and Gallup pollsters favor McCain? Or maybe the public is wildly changing its views, daily.

All unlikely, says Richard Morin, a senior editor at the Pew Research Center. In an interview with CQ Politics, he said the discrepancy is probably a result of the Newsweek and L.A. Times/Bloomberg polls over-representing Democrats.

“When I look at those results, I know something is going on,” said Morin.

“The first place that I look when I see these discrepancies, I look for the percentage of Republicans, Democrats and Independents in the sample. We know that the best predictor of how someone is going to vote is their party ID.

“Both the L.A. Times/Bloomberg and the Newsweek polls have (too) large percentage of Democrats and a (too) small percentage of Republicans.”

While there are indeed more people who identify themselves as Democrats than Republicans in the country, Morin says the other polls, including Gallup, are more in line with the actual disparity than the Bloomberg or Newsweek polls.

“Interestingly enough,” Morin said, “if you do the math and apply the proper percentages to the L.A. Times/Bloomberg and the Newsweek findings, you find that their results change dramatically.”

In fact, Morin says, if the two polls that show Obama winning by a large margin were to modify their findings using the same percentage of Democrats and Republicans as other polls, Obama’s lead would come down to somewhere between a toss-up and a small, single digit lead for Obama.

More at link

http://www.cqpolitics.com/wmspage.cfm?parm1=1&docID=news-000002907401


I think the polls that show it being close are under-sampling Democrats myself.

[liberal N proud]

1. Sampling

Select the samples you know will provide the answers you want.

Somewhat like selecting the intelligence you want to justify a war, this is for setting up another stolen election.

[Bullet1987]

2. You would HAVE to sample more Dems than Repugs

Since more Dems are likely to vote in the GE anyway because we're energized. Wasn't this guy paying attention to the primaries?

[kitfalbo]

3. Heh

Stats.

Combine them all for the most accurate result.

[Overseas]

4. You see, my friends, you just can't trust those darn polls. We need you to remember that.

Remember all those earnest discussions in the corporate media in 2004 about how exit polls could be just so far off the mark that they had the final vote tallies flipped, showing Kerry winning? Even though we trust exit polls to evaluate the soundness of elections around the world, somehow, U.S. polling firms with solid reputations were impugned during the TV talk about the 2004 election. Those discussions took place to drown out all the additional comments from the field about various methods of vote manipulation employed by the Republicans-- distribution of voting machines-- too few in Democratic areas, challenging particular voters to slow voting down to a crawl, not counting provisional ballots, voter roll caging lists, etc.

That's why it looks to me like this year the corporate media is just starting earlier to question polls to make sure to give the public a longer-standing memory that you just can't trust those darn polls. Let those wild swings and discrepancies sink into public consciousness now, so that when the election is stolen again, people will have those memories and say to themselves-- "See, you just can't trust those polls."

I may as well go a step further and wonder which organizations might be deliberately skewing their polling samples to demonstrate that you can't trust polls. Whoops, you see, you're right, we just screw up sometimes.

[phrigndumass]

5. It's always difficult to project turnout, but turnout is never split 50/50 evenly along party lines

A bigger sampling of likely voters is always better. Trust the ones with lower margins of error, because those are the ones with the bigger samples. If a poll reveals more likely voters who are Democrats, then so be it. A poll doesn't need to be weighted by party unless the number of likely voters for one party largely outnumbers the number of likely voters for the other party.

I agree with you, Jim ... Gallup is under-sampling Democrats. K/R

[mrreowwr_kittty]

8. Dem turnout will be massive this year. Repub turnout will likely be average, or less. nt

[Teaser]

6. Morin doesn't know what he's talking about

partisan reweighting doesn't work.

We used in 2004 to "adjust" every bad Kerry poll to a result more favorable to him.

Sometimes we adjusted ourselves into 10 point leads. Leads which never showed up in the actual tabulation.

[mrreowwr_kittty]

7. The thing about re-weighting samples

Is that pollsters ask people what party they "identify" as, not what their actual registration is. That's an important point. You wouldn't believe how many people swear up and down they are registered as a particular party and then turn out to be something else when I look them up on the voter file. That's why we always collect far more signatures than we need for petitions because we know that up to 20% of them may be registered as Republicans. So what's happening is these polls are being re-weighted when they were probably close to being properly weighted with just the first 1000 people they spoke to.