Bush Takes Back The Lead As 'Primary Effect' Wanes

[The Vorlon]

This one seems a tad out of line on the high side for Bush, I think we should just lump it in with the other 5 polls showing the race even to perhaps a very slight Bush lead...

Add this to the collection of polls - by my count 9 in the last 14 days...

As an aside... the person who wrote the article that goes with this poll OBVIOUSLY knows absolutely nothing about randon error, margins of error, or indeed anything of any kind related to polling.. so bad it is almost funny to read!

enjoy...

http://www.investors.com/editorial/general.asp?v=3/13

The Firm that conducted this poll also ran a presidential poll in the year 2000.  Their Final 2000 poll had Bush leading 48% to 46% for an error of 2% - Of the 10 major polls publically released just prior to the 2000 vote TIPP was the 3rd most accurate

Other polls for your consideration...

6 polls showing (collectively)  a modest Bush lead...

Investors Business Daily - Bush +5
http://www.investors.com/editorial/general.asp?v=3/13

NBC News - Bush +2
http://www.msnbc.msn.com/id/4499877/

National Public Radio - Bush +2
http://www.npr.org/news/specials/polls/mar2004/mar04.pdf

Rassmussen Research - Bush +2
http://www.rasmussenreports.com/Presidential_Tracking_Poll.htm

Associated Press - Bush +1
http://apnews.myway.com/article/20040304/D813Q6E01.html

Fox News - TIED
http://www.foxnews.com/story/0,2933,113412,00.html

Three polls showing a big kerry lead....

ARG - Kerry +6
http://www.americanresearchgroup.com/presballot/

ABC/Washington Post - Kerry +8
http://abcnews.go.com/sections/us/Politics/bush_kerry_poll_040308.html

CNN/UsaToday - Kerry +11
http://www.usatoday.com/news/politicselections/nation/polls/usatodaypolls.htm

No person's liberty is safe while Congress is in session - Thomas Jefferson

[Reaganfan]

It's like everyone has said. Kerry was ahead because his campagin was in full swing. Back in 1984, polls showed Mondale was 20 points ahead of Reagan, in 88' polls had Dukakis over Bush by a large margin, and in 1996 polls showed Dole would beat Clinton by a landslide. Bush is ahead, and may continue to get ahead.

[Filuwaúrdjan]

I have never heard of the outfit that did that poll before...

[WalterMitty]

the cnn poll is laughable.  kerry may be ahead, but not by 11 points.

[Reaganfan]

the cnn poll is laughable.  kerry may be ahead, but not by 11 points.

11 points... LOL

[zachman]

I'd say 2-5 points for Kerry.

[© tweed]

It's like everyone has said. Kerry was ahead because his campagin was in full swing. Back in 1984, polls showed Mondale was 20 points ahead of Reagan, in 88' polls had Dukakis over Bush by a large margin, and in 1996 polls showed Dole would beat Clinton by a landslide. Bush is ahead, and may continue to get ahead.

Well, your facts are wrong.  Reagan only led Mondale by a couple point at some times in 1984, but always held the lead.  And, Dole lost his lead on Clinton around 2/96.  Dukakis did hold a substantial lead on Bush through the summer (including the famous 55-38% poll).

[Fmr. Gov. NickG]

It's not clear from the graphic: is this a poll of registered voters or ADULTS?!

[zachman]

It looks like 150 or so from the sample were unregistered.

[Gustaf]

Most people who write about polls in papers have no idea how they work. At least over here.

[agcatter]

1) polls now don't mean jack regardless who is ahead.

2) Bush is not trailing by double digits or anything like that if he's trailing at all.

3) I prefer Rasmussen simply because it is the only daily tracking poll I've seen.  It's easy to follow on a daily basis and, because it's daily, you can see slippage much quicker.  It has shown a statistical tie since Feb 8th with almost no movement.  My guess is that is probably close to right picture.  I base that on the various state polls I've seen.  They are mostly pretty much in line with the 2000 state outcomes and we all know that produced a tie in 2000.

[Reaganfan]

The real US election today would probably be:
Bush: 300
Kerry: 238

[Gustaf]

The real US election today would probably be:
Bush: 300
Kerry: 238

What would Bush be picking up/losing from the 2000 election?

[Fmr. Gov. NickG]

It's not clear from the graphic: is this a poll of registered voters or ADULTS?!

The results shown are among the REGISTERED voters surveyed

That's good to know.  Why would they mention the number of adults surveyed if they only include registered voters?  Was this a more general poll of the American public that just happened to include some political questions?  

[Gustaf]

Pollsters often sell out to their contractors.

[Fmr. Gov. NickG]

On the subject of weighting, one thing I wish was more clear on many of these polls is whether or not they are weighting by party registration.  Although vote preference and party ID can fluctuate a lot, party registration doesn't and the actual population numbers are verifiable (in most states).  For state polls, this can give you a good read on whether your sample is representative of the population politically, as well as demographically (which is accomplished by weigthing by sex, age, and geography).  I don't think most political polls weight by income, although this frequently done with corporate polling.

[agcatter]

Some states like mine and a lot of other Southern states don't have registration by party.

[Fmr. Gov. NickG]

On the subject of weighting, one thing I wish was more clear on many of these polls is whether or not they are weighting by party registration.  Although vote preference and party ID can fluctuate a lot, party registration doesn't and the actual population numbers are verifiable (in most states).  For state polls, this can give you a good read on whether your sample is representative of the population politically, as well as demographically (which is accomplished by weigthing by sex, age, and geography).  I don't think most political polls weight by income, although this frequently done with corporate polling.

We are getting into the "long" answer here but...

Some pollsters weight for self identified party Id, some do not...

Most who weight use something fairly close to 37/35/28 Dem/Rep/Ind for a national poll

For the record, the pollsters professional association recomends against this type of weighting...

Zogby weights his polls by party ID, as does Ed Goaes (Battleground), the Washington Post/NyTimes poll (sort of in a really weird way) weights, as does Teeter/Hart(WSJ), TIPP (kinda) weights

In THEORY all the polls weight for income, but this is very, very iffy in practice because when you survey people for their income they lie!

CNN/UsaToday does NOT, which is why it bounces around so much...

Where do you get the info on which poll companies weight on which variables?  

It's true that you can't weight on party registration in national polls, since not all states have party registration.  Party ID is OK, but moves around alot itself, so you can never be sure you're weighting to an accurate population estimate.

[Gustaf]

I don't trust weighting, it screwed up way too many times here in Sweden. Many pollsters have actually stopped doing it b/c it worked so badly. But I guess they were probably just incompetnent.

[The Vorlon]

You must be logged in to read this quote.

Usually just by folllowing the "footnotes" at the bottom of the polls - they usually link back somewhere.. just keep following...

Gallup is amazing BTW, if you e-mail them a question you don't get back a response,,, you get a %^%@!ing textbook...

More on weighting...

From a strictly mathematical point of view, weighting is nothing more that a boundry constraint you apply to a statistical model...to the degree that your boundry conditions are accurate the accuracy of your model improves.

If you assume that Dems/Reps/Inds break out 37/35/28 and that break down is correct, you get a more accurate poll, if your breakout is wrong, you get a less accurate result..

Even the very best screw up here now and then - Ed Goaes, who has been the Polling association's "Pollster of the Year" 4 times and was the most accurate pollster in both the 1992 and 1996 races got his turnout model wrong in 2000 and missed the final mark by 5%...

He made turnout sssumptios, they we wrong, his pol was wrong

Any time you weight a poll based on an assumption.. you are right.... or you are not...

Zogby, for good or evil, builds HUGE assumptions into many of his polling models, which is why Zogby is often spectatularly right when most are wrong, and spectatularly wrong when most are right...

[Gustaf]

You must be logged in to read this quote.

Usually just by folllowing the "footnotes" at the bottom of the polls - they usually link back somewhere.. just keep following...

Gallup is amazing BTW, if you e-mail them a question you don't get back a response,,, you get a %^%@!ing textbook...

Since you're the expert, how good id Gallup really? Their Swedish branch S-U-C-K-S, but they seem to have a pretty good reputation here?

[agcatter]

Gallup's swings give me a headache.

[Inmate Trump]

Not sure if this has been mentioned yet or not, but the newest Rasmussen poll has Bush back in the lead too.  There's perhaps a more reliable poll than the IBD/TIPP Poll.

[12th Doctor]

You must be logged in to read this quote.

Usually just by folllowing the "footnotes" at the bottom of the polls - they usually link back somewhere.. just keep following...

Gallup is amazing BTW, if you e-mail them a question you don't get back a response,,, you get a %^%@!ing textbook...

Since you're the expert, how good id Gallup really? Their Swedish branch S-U-C-K-S, but they seem to have a pretty good reputation here?

Gallup is... different.  It has a tendency to swing wildly before every election and settle down the week before.  It is typically relyible toward election day.  Usually it is within 2% of the acctual total near election day, but it has been know to be Very inaccurate in some cases.

[The Vorlon]

You must be logged in to read this quote.

Gallup is still very good.  They used to be all alone at the top, and now there are a bunch of other highly skilled firms, but yes.. Gallup is still excellent.

Where Gallup has really lost their lustre is that the tracking poll they ran in 2000 was a disaster.

Gallup signed a contact with CNN/USAToday to do a tracking poll and agreed to design the thing to be hair trigger responsive to even very slight changes.  Componding this fact was that the sample size Gallup used was very small (+/- 180 people a day or so) so on top of a wildly over sensitive turnout model, they also had huge amounts of plain old statistical noise that resulted from the small sample size.

The result was utterly predictable - the 2000 Gallup tracking poll had wild, crazy, absurd swings - Candidates would go from down 10 to up 12 in like 4 days...

The (highly) abreviated version of what is wrong with the CNN?Gallup tracking polls is as follows.

1) We know that turnout is usually a tad over 50% - if you simply ask if people are going to vote, about 80% or so say they will... This is where the mythical "likely voter" comes in.

2) There are two basic approaches to sorting the "likely" voter from the "unlikely" voter.  

One way is to base your judgement on past behavior - Did you vote in the last presidential election, Did you vote for Governor, Do you know where you polling place is, etc...  Based on these replies a score of some kind is assigned to each respondant, and then an assumption is made as to likely turnout.  For example, if you ASSUME turnout to be 50%, you take the 50% of the respondants who had the highest "scores" and you count them in your result. - This insulates your poll from the day to day events of the current election in terms of who islikely and who is not.

Gallup does things (or at least did in 2000) things a bit different - they asked, in addition to the questions above, a bunch of attitude questions.. How exited are you about the race, how closely have you been paying attentiion, etc.  Based on how "excited" a voter was they were then deemed to be likely, or not...

This method wildly skewed the pool of people deemed likely to vote, and produced just crazy numbers...

To use a sports anaology, lets say you wanted to do a poll on if more people were fans of the Green Bay Packers or the Miami Dolphins.  In reality, this ratio would be relaitively fixed, and is highly unlikely to change by 20% in two days.

Suppose now that the Packers just played the Dolphins and won 38-6.  If we did a poll among "likely" Football fans and you were a Dolphin fan and I asked you "How excited are you about the dolphins" - you wouldlikely answer that you were not all that excited...  Similarly, more Packer fans would likely be excited as their team had just won..  The net effect is that you would get a wildly inflated score amoung "likely" football fans in this poll.. Packer fans would be over represented in the "likely" foorball fan pool, while Dolphins would be underrepresented...

As a consequence, depending on the weeks events, the ratio of "likely" Gore to Bush voters within the "likely" same bounced around like a yo-yo and gave rather crazy results...

Now if you are just a few days or a week away from the election, this methodology works just fine, but 8 months, or even 2 months out... it produces gibberish...  Instead of asking "who are you going to vote for" You might as well just ask "Which candidate had a better week on the campiagn trail" because that is, in effect what you are asking..

I notice that the l CNN/UsaToday poll from last week arbitrarily padded their turnout to 50%, so the MAY have changed their methodology for 2004 (I hope so!)

This is the highly abbreviated answer about how "good" gallup is.