FR EXCLUSIVE - Poll Weighting Demystified...and the Bug in Rasmussen Polling

FR | 10-18-08 | Bob J

[Bob J]

For those wanting to know the inside story on how and why polling companys weight their samples, Rasmussen has done a pretty good job of explaining it.

Here you can get an overall explanation of the methodology and rationale;

The Value of Party Weighting for a Tracking Poll

Next they explain why there may be large differences between polls;

Why Polls Sometimes Show Different Results ,p>

Rasmussen establishes the weighting for the polls by conducting a party affiliation poll of 15k every month. Here they show the results since 2004;

Summary of Party Affiliation

Here's the bug. Since 2004 their Party Affiliation numbers have see-sawed somewhat since they are based on political conditions at that time. IMO in 2004 they saw an increase in republican identification leading up to the election due to concerns over the war in Iraq/terror and doubts about Kerry. In 2006 there was a likewise trend toward dems due to electorate frustration about Bush over the WOT and the economy, leading to a dem victory in the midterms.

The party identification numbers in the last four years hav averaged...2004 = +2.4% dem, 2005 = +1.9% dem, 2006 was +4.1% dem and 2007 was +4.8% dem.

However in the first 2 quarters of 2008 there was a jump to +9% dem. Huh?

This jump in dem identification is not only historically without precedent, there is nothing from a national, cultural or financial perspective to explain it. In Jan 2008 we started to see the US winning in Iraq, the economy was basically sound, the DOW at 13000 and there were no pressing cultural issues. Rasmussen weakly explains this by saying the "Pubs were having a hard time".

Baloney. The only plausible explanation for such a dramatic and unprecedented change in party affiliation over such a short period of time is manipulation.

Now I don't know if this was being done by Rasmussen "Party Affilation" polling employees (how do you "weight" this poll, BTW?), Operation Chaos or dem/ACORN fraud, but for Rasmussen to use these obviously manipulated numbers as a basis for his entire polling methodology is at best suspect and at worst professional incompetence.

FReepers have been right all along. When looking at Rasmussens numbers we should skew at least 4% back toward McCain to balance the error in party affiliation.

At worst, McCuda are tied and probably up by 1-2%.

[lonestar67]

I agree. These rallies are bigger than the Bush rallies.

I went to several Bush rallies in 2004 and they were awesome.

The current Palin rallies dwarf it.

She is sweeping the electoral floor with Obama.

[1curiousmind]

Great work. Very helpful indeed.

Thanks for posting.

[JasonC]

Um, they didn't vote *against* Hillary, they voted for her, against Obama.

The dem race tightened by at least 5 points as soon as the republican race was settled, at the same time "turnout" rose. Why? Because people chose a dem ballot as the only one that still made a difference, by the time voting got to their state.

Unprecendented passion to vote *against* Obama is what was going on.

[Bob J]

Notice I didn’t accuse Rasmussen of being biased. After going through their website I’m of the opinion they try to do a good job. Pollsters aren’t taken seriously unless they nail the results pretty close. If they have a system, such as using the party affiliation numbers to weight their samples, they have to go with it.

My beef is why they didn’t question that dramatic jump in the 1st Q of ‘08. It sticks out like a sore eye yet there isn’t one word or explanation with the esception of “pubs are having a bad time”.

Really Rasmussen? How so? How about the 10% approval of the dem majorities in Congress?

[lonestar67]

I think you are forgetting what happened. This has some conservatives rather upset because they now fear that Obama may have been a stronger opponent in the end.

Its true, regardless of how they voted they had to switch their registrations. Its obvious.

I think if you go back and look you will find Rush encouraging people to derail Hillary by voting for Obama. Conservatives initially galvanized themselves against Hillary— not Obama.

[codercpc]

How do they come up with the numbers in states that don't declare party registration?

I live in Wisconsin, and I have never, ever "registered" with a party. The closest I have come is to vote in the Republican primary. This year both the Republican and the Democrat primaries were held on the same day. When I got my ballot I was actually given both party ballots. I filled one out, and then the other was put into a discard box under the watchful eye of the poll watcher. There was no way to assert which party I voted for. How do they then get a weighted scale in Wisconsin.

BTW, by the time we voted here, McCain was the only viable candidate left. I think Huckabee was still technically in the race, but it was a foregone conclusion that McCain had won the primary, so more people in my county voted in the Hillary/Obama (Operation Chaos) race than they did in the Republican primary. I live in one of the most Republican counties in the state. Since the late 1800's we have only voted non Republican once.

[bereanway]

You’re right we hear this every election cycle for various reasons. Youth vote, minority vote, general dissatisfaction. These polls are now hugely skewed and I’m convinced many identifying themselves as dims don’t want to say they’re not voting for 0 (maybe it’s the Bradley effect pre-vote). 0 is not going to hold certain traditional dim constituencies very well and Palin has energized the pubbie base. 11/4 is gonna be fun.

[MediaMole]

Would someone who switched party to vote in a primary tell a polling operation that he was now a democrat? Operation Chaos was an attempt to help the GOP. It was not a shift of voters’ allegiance.

Turnout numbers for the primaries could cause a big shift in Rassmussen’s numbers, but self-identification shouldn’t shift that substantially.

[LegalEagle61]

I agree, very good points. I think PUMAs will help alot and I do think the Bradley Effect will be in play as well. The WSJ had an article that should wake people up too, they talk about the government being in total liberal control, that thought needs to be communicated to the American people, THAT will scare them enough to rethink things, but it needs to get out there better. I can’t and won’t imagine such a scenario, I have to admit, I have been praying to God about this election more than I normally do, it is THAT serious.

[Intolerant in NJ]

This jump in dem identification is not only historically without precedent, there is nothing from a national, cultural or financial perspective to explain it...stupidity and brainwashing of the American people......

[nutmeg]

However in the first 2 quarters of 2008 there was a jump to +9% dem. Huh?

Wasn't that the height of "Obama-mania"? In Connecticut anyway, our 'RAT Secretary of State has been boasting about all the "new young voters" registering by the hundreds of thousands in CT, especially during the first quarter of 2008, hinting of a "big turnout" for the Messiah.

Great analysis, Bob... I'm trying my best to be upbeat in the face of these insane polls... ;-)

[nutmeg]

The WSJ had an article that should wake people up too, they talk about the government being in total liberal control, that thought needs to be communicated to the American people, THAT will scare them enough to rethink things, but it needs to get out there better. I can’t and won’t imagine such a scenario, I have to admit, I have been praying to God about this election more than I normally do, it is THAT serious.

Same here, LegalEagle. Do you know where to find that WSJ article?

I fear three branches of 'RAT control will do decades-worth of damage to this country... McCain/Palin MUST be elected.

[zebrahead]

I agree completely that the poll weighting appears to be biased and it is pathetic. Here’s a quick example. Rasmussen and SurveyUSA have Obama up by an average of 7 points in Missouri. Only SUSA posts their internals and they weight 38% Democrats and 31% Republicans. Looking at the exit polls from the 2006 Senate race in Missouri, Republicans made up 39% and Democrats 37%. A drop in Republican support of 8% is exactly Obama’s margin overall.

[jveritas]

Excellent analysis Bob.

[xzins]

If someone calls you on the phone and asks if you consider yourself to be a dem, rep, or indy, what do you say?

Do you say, “Well, I registered as a dem in operation chaos, but I’m really a rep, however, you better write me down as a dem, since that’s the last thing I had registered?”

Or would you say you consider yourself to be a republican.

[Bob J]

When they do their 15k party identification poll each month, there are three choiices, dem, pub or indy. When they do their actual political polling, they weight the samples to these numbers to offset statistical anomalies.

Here’s an example. Their PA numbers are 40% dem, 35% pub and 25% indy. They poll 1100 people but find they were 400 dems, 400 pubs and 300 indys.

They keep the 400 dem responses, throw out 50 pubs and 50 indys to get the 40-35-25 ratio.

[Bob J]

BTW - They don’t ask party registration, they ask “affiliation”.

[codercpc]

See my post right above yours, in my state we don't even declare parties, in the primary, by the time we voted, McCain had basically won already, so there was nothing else for Republicans to vote for. I live in one of the most Republican counties in the state, yet we had twice as many people voting in the Democrat primary as we did in the Republican, how do they weigh our state polls? I am 100% sure that come November 5th, we will once again give McCain/Palin the vote regardless of what the primary numbers were.

[surfer]

What we need to do is promote some type of pre-election party event to celebrate the Obama win the night before at all the college campuses.

It should run all night and have massive amounts of free booze.

That would do the trick. By the time they decided to get their slacker arses out of bed it would be Wed.

Not that I support voter suppression or anything...lol

[Bob J]

Changing party registration might have been a legally grey area in some states. When called, they might mistake affilation for registration and say how they were registered so as not to invite legal complications.