MSNBC Poll



  • I know, by now just about everyone on the internet has had just about enough of hearing about Ron Paul.

    But still... that's some nice counting there.

    MSNBC Poll

    Link



  • Near as I can tell from the comments, etc., the author assumed or decided that the Paul numbers weren't accurate - in spite of the fact that so much of Ron Paul's support comes from the Internet, which would make an Internet Poll a highly likely place to find Paul supporters...

    The funny thing is that the media are finally revealing just how ignorant they are to the fact that the American people are just as sick of them as they are of the political "powers that be" in Washington...



  • @GalacticCowboy said:

    The funny thing is that the media are finally revealing just how ignorant they are to the fact that the American people are just as sick of them as they are of the political "powers that be" in Washington...

    True. Sadly, though, the American people are still just as gullible as they always have been. As an exercise, compare the reasons a supporter of [i]any[/i] candidate (from [i]any[/i] party) will give for their support of the candidate with the voting record and/or history of public statements of that candidate, and watch the mismatches pile up. It isn't so much that the American people have wised up as that they have fully embraced Creationist "the facts are whatever you wish them to be so ignore records and investigations" evidentiary standards, so not even the media can sway them any more; only the lies with which they already agree get through.



  • <sarcasm> I must profess astonishment that polls are so perfectly done that they ALWAYS choose to reflect the views of the poll sponsor.</sarcasm>

    I know this is true because  I commissioned a poll this morning of males, 34.9-34.10 years of age, and wearing my pants, within a 1 foot radius of myself, and the results are accurate to within no decimal places, 100 times out 100.



  • I get pummeled with political "opinion poll" calls constantly. I answered a few but now I just hang up.

    So the results would be of those people who actually didn't hang up. I think there's a inherent fundamental bias there no matter what the survey is actually about.


  • @medialint said:

    I get pummeled with political "opinion poll" calls constantly. I answered a few but now I just hang up.

    So the results would be of those people who actually didn't hang up. I think there's a inherent fundamental bias there no matter what the survey is actually about.

    There's also a slowly growing bias in that polsters and hucksters aren't allowed to call cell phones, so the slowly growing number of wireless-only houses is throwing things off. In other words, taken to its logical extreme, eventually polls will be answered only by people who don't know they don't have to rent that bakelite handset from AT&T anymore.

    I'm sure the marketroids and polsters will eventually wiggle their fingers inside some politico's ass and have an amendment legalizing such calls stuffed inside a totally unrelated must-pass bill, as is their usual modus operandi.



  • @GalacticCowboy said:

    Near as I can tell from the comments, etc., the author assumed or decided that the Paul numbers weren't accurate - in spite of the fact that so much of Ron Paul's support comes from the Internet, which would make an Internet Poll a highly likely place to find Paul supporters...



    Well they have a point.  If you're running a poll like that and it gets posted on Digg or some similar place, you might as well cut the Ron Paul results in half (at least) and double Guliani's votes.



  • @MarcB said:

    I know this is true because  I commissioned
    a poll this morning of males, 34.9-34.10 years of age, and wearing my
    pants, within a 1 foot radius of myself, and the results are accurate
    to within no decimal places, 100 times out 100.

    Dude... WTF?  You have ninety-nine other men in your pants with you?  That's pretty damn kinky IMO :-O 

     



  • @Cap'n Steve said:

    @GalacticCowboy said:

    Near as I can tell from the comments, etc., the author assumed or decided that the Paul numbers weren't accurate - in spite of the fact that so much of Ron Paul's support comes from the Internet, which would make an Internet Poll a highly likely place to find Paul supporters...



    Well they have a point.  If you're running a poll like that and it gets posted on Digg or some similar place, you might as well cut the Ron Paul results in half (at least) and double Guliani's votes.

    Well this is a Facebook poll, which I presume means you have to be on Facebook to vote.  And they frequently put crap like this on the Facebook front page anyway...  But if you start with the premise that an Internet poll is fundamentally flawed anyway, then a) why are the results even important and b) why ignore *part* of the results?  If the Paul numbers are invalid, none of the others amount to a hill of beans either.



  • @GalacticCowboy said:

    Well this is a Facebook poll, which I presume means you have to be on Facebook to vote.  And they frequently put crap like this on the Facebook front page anyway...  But if you start with the premise that an Internet poll is fundamentally flawed anyway, then a) why are the results even important and b) why ignore part of the results?  If the Paul numbers are invalid, none of the others amount to a hill of beans either.

    Given that it's only a measly 5000 people, it only amounts of a hill of beans if it is valid. A group that small couldn't elect the Communist Party of China. 



  • @asuffield said:

    it only amounts of a hill of beans if it is valid.

    to a hill of beans. Bloody edit timeout is shorter than the time it takes for me to read the other threads and then check my post. 



  • @asuffield said:

    @GalacticCowboy said:

    Well this is a Facebook poll, which I presume means you have to be on Facebook to vote.  And they frequently put crap like this on the Facebook front page anyway...  But if you start with the premise that an Internet poll is fundamentally flawed anyway, then a) why are the results even important and b) why ignore part of the results?  If the Paul numbers are invalid, none of the others amount to a hill of beans either.

    Given that it's only a measly 5000 people, it only amounts of a hill of beans if it is valid. A group that small couldn't elect the Communist Party of China. 

     

    You fail at the concept of random sampling.

     That said, internet sites also fail at the concept of random sampling due to not being able to be 'random'.


     



  • @Jonathan Holland said:

    @asuffield said:
    @GalacticCowboy said:

    Well this is a Facebook poll, which I presume means you have to be on Facebook to vote.  And they frequently put crap like this on the Facebook front page anyway...  But if you start with the premise that an Internet poll is fundamentally flawed anyway, then a) why are the results even important and b) why ignore part of the results?  If the Paul numbers are invalid, none of the others amount to a hill of beans either.

    Given that it's only a measly 5000 people, it only amounts of a hill of beans if it is valid. A group that small couldn't elect the Communist Party of China. 

    You fail at the concept of random sampling.

    I do not think you know what those words mean. This is not random sampling, nor anything remotely like it. Even if it was, given that it's only a measly 5000 people, it only amounts of a hill of
    beans if it is valid. A random sample that small couldn't elect the Communist
    Party of China.



  • @asuffield said:

    @Jonathan Holland said:
    @asuffield said:
    @GalacticCowboy said:

    Well this is a Facebook poll, which I presume means you have to be on Facebook to vote.  And they frequently put crap like this on the Facebook front page anyway...  But if you start with the premise that an Internet poll is fundamentally flawed anyway, then a) why are the results even important and b) why ignore part of the results?  If the Paul numbers are invalid, none of the others amount to a hill of beans either.

    Given that it's only a measly 5000 people, it only amounts of a hill of beans if it is valid. A group that small couldn't elect the Communist Party of China. 

    You fail at the concept of random sampling.

    I do not think you know what those words mean. This is not random sampling, nor anything remotely like it. Even if it was, given that it's only a measly 5000 people, it only amounts of a hill of
    beans if it is valid. A random sample that small couldn't elect the Communist
    Party of China.

    It is not a random sampling. However, 5000 participants is quite adequate for a poll - most polls seen on the nightly news consult only 500 to 1500 people. 



  • @R.Flowers said:

    However, 5000 participants is quite adequate for a poll

    That depends on the size of the population.

     

    most polls seen on the nightly news consult only 500 to 1500 people. 

    Which is why their results are statistically meaningless when they are polling on a subject that affects the entire nation. 



  • @asuffield said:

    @R.Flowers said:
    However, 5000 participants is quite adequate for a poll
    That depends on the size of the population.
    most polls seen on the nightly news consult only 500 to 1500 people.
    Which is why their results are statistically meaningless when they are polling on a subject that affects the entire nation.
    Not true. Statistically, the margin of error depends only on the sample size and is independent of the population size. The margin of error is 1/sqrt(sample size), which is about 4.5% for a sample of 500 people.

    The difficulty with election polls is that it is nigh impossible to get a representative sample. For starters you're bound to people who have land lines, are patient enough to politely answer pollsters during dinner, are likely voters, etc. Internet polls are completely useless in that respect. Pollsters weigh the statistic results with an estimate on how bad the quality (not: quantity) of the sample is. That is based on post mortems on how well the polls where in past elections.

    It is therefore quite understandable that the pollsters are wrong early in an election, because the issues and motives are different now then in past elections. In Iowa, they over-weighed Clinton, found themselves wrong, adjusted, and then under-weighed her in New Hampshire. One may expect that in the coming weeks the polls get more accurate.



  • any poll that doesn't come from a respected poller (like Zogby or Rasmussen) should be immediately ignored.  Any poll that comes from the internet should be immediately ridiculed.



  • @JvdL said:

    Statistically, the margin of error depends only on the sample size and is independent of the population size. The margin of error is 1/sqrt(sample size), which is about 4.5% for a sample of 500 people.

    That applies only to scenarios where the statistic under test is independent and approximately evenly distributed, which only happens when you're drawing coloured balls from a bag. Also, that's the expression for the maximum margin of error, not the actual one, and it applies only when the sample size is small compared to the population, although those points are not particularly important here.

    In general, you can only concoct a sampling strategy where the margin of error is independent of the population size when you already know the nature of the distribution of the test statistic, and are merely sampling its parameters. Obviously you can't do that in an election poll.

    Statisticians are very good at making small samples of a large population of people seem significant, though.



  • @tster said:

    any poll that doesn't come from a respected poller (like Zogby or Rasmussen) should be immediately ignored.  Any poll that comes from the internet should be immediately ridiculed.

    And any poll that does come from a respected poller working in meatspace should still be treated with extreme suspicion. It is horrendously difficult to check the mathematics and they have a strong tendency to come up with the result that polls are good and pollsters should be paid more.



  • @asuffield said:

    @tster said:

    any poll that doesn't come from a respected poller (like Zogby or Rasmussen) should be immediately ignored.  Any poll that comes from the internet should be immediately ridiculed.

    And any poll that does come from a respected poller working in meatspace should still be treated with extreme suspicion. It is horrendously difficult to check the mathematics and they have a strong tendency to come up with the result that polls are good and pollsters should be paid more.

    lol. 



  • @asuffield said:

    @JvdL said:

    Statistically, the margin of error depends only on the sample size and is independent of the population size. The margin of error is 1/sqrt(sample size), which is about 4.5% for a sample of 500 people.

    That applies only to scenarios where the statistic under test is independent and approximately evenly distributed

    That's another way of saying that the sample has bad quality as elaborated in the remainder of my post. However, there is no reason to assume that quality of the sample improves when you increase its size. Election polls typically get less than 20% response, so even a poll that would query the entire US population would still capture only a 20%, non-representative, sample.



  • @JvdL said:

    @asuffield said:
    @JvdL said:

    Statistically, the margin of error depends only on the sample size and is independent of the population size. The margin of error is 1/sqrt(sample size), which is about 4.5% for a sample of 500 people.

    That applies only to scenarios where the statistic under test is independent and approximately evenly distributed

    That's another way of saying that the sample has bad quality as elaborated in the remainder of my post. However, there is no reason to assume that quality of the sample improves when you increase its size. Election polls typically get less than 20% response, so even a poll that would query the entire US population would still capture only a 20%, non-representative, sample.

    Yeah? Well you work for Microsoft! And you suck!



  • @MasterPlanSoftware said:

    Yeah? Well you work for Microsoft! And you suck!

    Damn, you blew my cover. 



  • @JvdL said:

    @MasterPlanSoftware said:

    Yeah? Well you work for Microsoft! And you suck!

    Damn, you blew my cover. 

    I know the feeling!



  • @tster said:

    any poll that doesn't come from a respected poller (like Zogby or Rasmussen) should be immediately ignored.  Any poll that comes from the internet should be immediately ridiculed.

    Agreed.

    After all, didn't know know that 67% of statistics are made up on the spot?
     



  • @JvdL said:

    @asuffield said:
    @JvdL said:

    Statistically, the margin of error depends only on the sample size and is independent of the population size. The margin of error is 1/sqrt(sample size), which is about 4.5% for a sample of 500 people.

    That applies only to scenarios where the statistic under test is independent and approximately evenly distributed

    That's another way of saying that the sample has bad quality as elaborated in the remainder of my post.

    It's a little stronger than that - you can't actually have a good quality random sample of a small fraction of the population in a plurality-voting-by-district system. Unlike the balls-from-bag case, the errors accumulate rather than cancelling out. Hence, small random samples are entirely useless. That's why real statisticians don't use them, they use other sampling methods. Which brings us back again to: you couldn't elect the CPC with that.


    However, there is no reason to assume that quality of the sample improves when you increase its size.

    I am certain that a sample of size equal to the population would be an improvement. I am reasonably confident that a sample of 75% of the population would still be a significant improvement. I would not expect to see any real change if they went up from a few thousands to a few tens of thousands.



  • @asuffield said:

    @JvdL said:

    However, there is no reason to assume that quality of the sample improves when you increase its size.

    I am certain that a sample of size equal to the population would be an improvement. I am reasonably confident that a sample of 75% of the population would still be a significant improvement. I would not expect to see any real change if they went up from a few thousands to a few tens of thousands.

    Such a large sample size improves the statistical quality but not the sample quality. The point is that 80% of the U.S. population refuses to participate in polls, and nobody has thus far been able to qualify them statistically; which is why the sample quality will never be good, why pollsters don't bother to better the 3-5% margin of error, and why a 500-1000 sample size is good enough regardless of population size

     

    Pop SizeSample SizeM.o.E
    100,000,000 750 3.7%
    10,000,000 750 3.7%
    1,000,000 750 3.7%
    100,000 750 3.6%
    10,000 750 3.5%
    1,000 750 1.8%

    (p=50%)


Log in to reply