Monday, September 29, 2008

62% of people say so, so it must be so

The major pollsters like Gallup, Rasmussen and Zogby weight their polls. These weights are in an attempt to mirror the general population. So if the population is 40% Democrat, 35% Republican and 25% independent, they'll call people until they get a sample that's 40/35/25. Makes sense.

Except it doesn't. The 40/35/25 is an estimate. It's baed on nothing substantive. Their estimates are based on other polls of self-identified party identification, voter registration numbers and past elections. But it's still at best an educated guess. And every pollster has a different weight, because it is a guess and nobody really knows. Rasmussen, Gallup and Zogby each have their own weights.

I question why any weighting is necessary at all. A poll is supposed to accurately represent the population as a whole. So if you randomly call 1000 numbers, and the population is truly 40/35/25 as you think it is, then you should get a breakdown of 400/350/250 in your sample. If you don't, then it's not a random sample and your methodology is incorrect. Either that or the sample is too small and you have to start calling more numbers.

It's like flipping a coin. Every two flips should yield one heads and one tails. If you flip 2 times the odds of getting 2 heads or 2 tails is high. If you flip 20 times the odds get closer to 10 of each and the odds of getting 20 heads or 20 tails is almost 0. If you flip 1000 times, the odds are close to 100% that you will see 500 heads and 500 tails.

Same with polling. If you call 10 people, you can get 8 Democrats and 2 Republicans or vice versa. If you call 100 people the chances of you getting a true representation of the population as a whole is better, but you could still get something freaky from a small sample. If you call 1000, the chances get exponentially better to find a true representation. And so the need to weight based on some arbitrary breakdown strikes me as unnecessary...if and that's a big if, 1000 is a high enough sample.

As I see it, by purposely weighting the calls you are essentially saying your sample is not random. And if your sample is not random, then the entire validity of the poll - and polling in general - is called into question.

There is one major poll that doesn't do any weighting. That's the BattleGround poll. It is also the one poll out there that shows McCain winning. That poll calls 1000 people at random without any weighting. It is a daily poll and has also been the steadiest poll throughout the campaign basically showing a tie with either of the two up 1 or 2 percentage points.

And that to me says than the numbers themselves. The major polls have seen wild swings during a short period of time. One day Obama is up by 8, the next week McCain is up 7, week after that Obama is up 9 all in the same poll. I find it hard to believe people change their minds that radically that quickly. Sure, one candidate or the other may say or do something to change some people’s minds. But I just don't see 20% of the electorate gyrating so much over a 2-3 week span.

If people are truly that easily swayed by a speech or a commercial, it's a sad commentary on the American voter.
And one more comment about polls. I have heard the valid explanation that youngens are not represented accurately in polls. This is because lots of youngens have cell phones only and are never called. And youngens are heavily Democrat so the polls have to be weighted to take this into account. But what I don't ever hear is the other side of the equation. Evil rich republicans (like me according to Obama) don't answer the phone to take polls. The small business owner who is overwhelmingly Republican who is out working 12 hours a day isn't home to take the call. Nor is the consultant who is away at a client site 3 or 4 nights a week and is never home to take the pollster's call. And any poll conducted over the weeked will way undersample the evil rich republicans on weekend trips. However, the welfare recipient is at home 24/7/365 to take that call. And the teacher as well as all government "workers" who are home by by 4:00 every day has all the time in the world to chat with a pollster.

Best thing to do is ignore all polls.

4 comments:

Anonymous said...

Ed

Check these out. Luntz is a manipulative piece of s*** and has been outed. This is how these polls truly work. He is a regular on Faux but he is really in the tank for Barak America.

youtube.com/watch?v=pFL-LubDF9c

There was another video where they had a 'random' person in the audience, who spoke with authority, describe how Ron Paul was a psycho blah blah blah ... well it turns out that he was another crony for Luntz but I couldn't find that video. Luntz is the worst.

All MSM info is manipalitive in one way or another.

Ed said...

Do you remember the CNN/Youtube GOP debate earlier this year where an "undecided" voter turned out to be a Hillary campaign staffer?

Speaking of Ron Paul...man if only he had decided to run. With all this bailout crap going on, he may actually have gained some traction. He wouldn't have won, but maybe he would have received enough votes to make people pay attention.

Anonymous said...

I can't thank you enough for the info you share on this blog.

Sean said...

anyone know where I can find a "Blame Me, I didn't vote for Ron Paul" bumper sticker?

Ed, I think your use of "Exponentially" in the following sentence "if you call 1000, the chances get exponentially better to find a true representation." is questionable....