Can polls be trusted?

[Starbucks Union Thug HokeyPuck]

Wisconsin polls had Kerry up 30 pts. on Edwards the day before the primary.  Edwards came within 5.  Are these polls REALLY that accurate.  Most of Edwards voters said they made a last minute decision.  

I'm just holding on to one little thread of hope that Edwards somehow shocks the entire world tomorrow.
: (

Get your @sses out there and vote Deaniacs!!!!

[SayNoToNader.com]

Wisconsin is proof that they can't be trusted then. I didn't realize there was that big of a difference there. I hate polls !

[StevenNick]

These polls are about as useful as tits on a boar hog.

[angus]

I had a mathematics professor who used to say, "figures don't lie but liars figure."

That's an important question HockeyDude.  You may not be old enough to remember the chewing gum commercial line "4 out of 5 dentists recommend trident for their patients who chew gum."

What does that mean?  Well, it could mean they went out and asked the first five dentists they could find whether they recommend chewing gum to their patients, but that's unlikely because they'd all say don't chew gum.  Or maybe the asked several thousand dentists the question and 4 of them said "chew trident."  And they threw out 995 of them for other reasons.  It may mean they asked a thousand dentists, and around 800 of the dentists said, "Well I don't recommend chewing gum, mind you, but if they must then they should chew a sugarless gum."  The questioner would have no doubt prompted, "Like Trident gum?" and those 800 said, "sure, why not, if it has no sugar."

You should always question polls.  You should pay attention to who collected the data and who's presenting the data.   If the poll was conducted by the DNC, the RNC, the NOW, the Gun lobby, the Cato institute, FOX news channel, or the Sierra Club, you may reasonably question its legitimacy because those groups all have agendas.  But if the poll came from Gallup or Mason-Dixon it may be somewhat more objective.  Some polling groups round up people off the streets and pay them with cigarettes, others spend the money and do it right.  

You should pay attention to the group polled.  For example, note whether the poll was taken over a weekend (which favors Democrats because they have less money and may be found at home then), or over weekdays (which, some argue, favor Republicans as they are generally more traditionalist and would be spending time with their families).  Note whether the poll mentions that it was a sample of adults, registered voters, or likely voters.  Good representative samples are elusive sometimes, and much modelling goes into this inexact science.

Also, you should pay attention to the exact wording.  Good news sources will give you this information but you might have to dig a little deeper to find it.  For example, if you ask the question "Do you think innocent little babies who have never harmed anyone have the right to live?" you'll likely get a very different answer than if you ask "Do you think that jackboot-wearing gun-toting fascists should control your reproductive freedoms?"  Even though these questions are trying to get at the same essential data, their wording will produce starkly different results.  The wording of the questions is very important and reputable polling agencies spend a great deal of time and energy to assure neutral wording of questions.  (an excellent example of this appears in my post on  Re:Nader to announce on Meet the Press.)

You should also pay attention to the poll's margin of error.   Estimation of error can be difficult.  Even a good estimation may indicate slop.  49% plus or minus 3 percent, for example, is called a statistical tie with 47% plus or minus 3 percent.  

Having said all that, it's better than nothing.  If I tell you I tried my hardest and got the following result for the Iowa Primary:  Kerry -46%, Edwards-20%, Dean-18%, then you'd probably reasonably think Kerry will win it and Edwards and Dean will duke it out for 2nd.

[California Dreamer]

The big problem with the Wisconsin poll were that they all underestimated the Republican and Independent turnout which was around 42% of the total vote

if you look at just the democratic vote in Wisonsin, Kerry won by the predicted margins.

Most other polls in other states were pretty accurate.

Also Edwards has often been a good 'closer' and has done well with people who decide in the last two days...therefore they are unpollable.

All that being said I think the polls are accurate and Kerry will sweep all 10 with Edwards getting close in MD and GA.

[classical liberal]

The CATO Institute seems fiarly nonpartisan to me from their presentations on C-SPAN condemning Bush.

[angus]

The CATO Institute seems fiarly nonpartisan to me from their presentations on C-SPAN condemning Bush.

Dude, I'm totally down with CATO.  I mean, cato is right on with me.  Now, doesn't that tell you something??

[Gustaf]

I had a mathematics professor who used to say, "figures don't lie but liars figure."

That's an important question HockeyDude.  You may not be old enough to remember the chewing gum commercial line "4 out of 5 dentists recommend trident for their patients who chew gum."

What does that mean?  Well, it could mean they went out and asked the first five dentists they could find whether they recommend chewing gum to their patients, but that's unlikely because they'd all say don't chew gum.  Or maybe the asked several thousand dentists the question and 4 of them said "chew trident."  And they threw out 995 of them for other reasons.  It may mean they asked a thousand dentists, and around 800 of the dentists said, "Well I don't recommend chewing gum, mind you, but if they must then they should chew a sugarless gum."  The questioner would have no doubt prompted, "Like Trident gum?" and those 800 said, "sure, why not, if it has no sugar."

You should always question polls.  You should pay attention to who collected the data and who's presenting the data.   If the poll was conducted by the DNC, the RNC, the NOW, the Gun lobby, the Cato institute, FOX news channel, or the Sierra Club, you may reasonably question its legitimacy because those groups all have agendas.  But if the poll came from Gallup or Mason-Dixon it may be somewhat more objective.  Some polling groups round up people off the streets and pay them with cigarettes, others spend the money and do it right.  

You should pay attention to the group polled.  For example, note whether the poll was taken over a weekend (which favors Democrats because they have less money and may be found at home then), or over weekdays (which, some argue, favor Republicans as they are generally more traditionalist and would be spending time with their families).  Note whether the poll mentions that it was a sample of adults, registered voters, or likely voters.  Good representative samples are elusive sometimes, and much modelling goes into this inexact science.

Also, you should pay attention to the exact wording.  Good news sources will give you this information but you might have to dig a little deeper to find it.  For example, if you ask the question "Do you think innocent little babies who have never harmed anyone have the right to live?" you'll likely get a very different answer than if you ask "Do you think that jackboot-wearing gun-toting fascists should control your reproductive freedoms?"  Even though these questions are trying to get at the same essential data, their wording will produce starkly different results.  The wording of the questions is very important and reputable polling agencies spend a great deal of time and energy to assure neutral wording of questions.  (an excellent example of this appears in my post on  Re:Nader to announce on Meet the Press.)

You should also pay attention to the poll's margin of error.   Estimation of error can be difficult.  Even a good estimation may indicate slop.  49% plus or minus 3 percent, for example, is called a statistical tie with 47% plus or minus 3 percent.  

Having said all that, it's better than nothing.  If I tell you I tried my hardest and got the following result for the Iowa Primary:  Kerry -46%, Edwards-20%, Dean-18%, then you'd probably reasonably think Kerry will win it and Edwards and Dean will duke it out for 2nd.

Sums it up. I stated this before, but, as much as I have to admit it, I think you did it better. Good examples also.

[© tweed]

The polls this time around, on this Super tuesday, seem to be fairly accurate.  survey USA has done a good job.

[zachman]

Iowa and South Carolina were the only major polling flaws, but the caucus system makes for wild results, so that may explain the pre-caucus support.

[© tweed]

Iowa and South Carolina were the only major polling flaws, but the caucus system makes for wild results, so that may explain the pre-caucus support.

They weren't terrible though.  Wisconsin was the worst, the last SUSA poll before Wisconsin had it Kerry 53% Edwards 16%.