cBS poll the left is trying to tout

[ShadesOfGrey]

Originally posted by: dullard

Originally posted by: ShadesOfGrey
This poll was about politics. When polling people about politics and political issues your sample should reflect the political breakdown of the country(or area you are polling). Otherwise, why not just ask a bunch of Democrats? The rest of the Demographics might be fine if you did so but if you were trying to present the data as representative of anything but what Democrats think, you are misrepresenting things.

Why use just a poll of likely voters (such as Zogby)? If you apply a poll of likely voters to the American public, you are misrepresenting things. Not all Americans vote. Many non-voters care about public policies.

Again, the results are irrelevant to the discussion of the sample. You derive results from the sample, not prove sample by results. If cBS would have adjusted their sample for political affiliation, this thread wouldn't exist even if the results were the same. This has always been about the bad sample done by cBS.

The results ARE relevant. If after adjusting you get the correct results, then your weighting was correct. If after adjusting you get incorrect results, then your weighting was incorrect. You cannot judge a poll's merrits without using the results.

HUH? How do you know if your results are correct or not? It's a poll. You can't verify a sample based on it's results because you don't know what the result should be.

Now as for Zogby, I agree some political issues should be more than just likely voters but I don't think you can say that he doesn't do that. The likely voters are for elections, not necessarily normal polls. If you can find contrary evidence from Zogby, please share.
The important piece of the puzzle though is the fact that he does use political affiliation to correct his sample so it better represents the US. Without it, it is meaningless and can't be extrapolated out to represent the US.

 

[arsbanned]

A majority think Bush is a liar. Put that in your crackpipe and smoke it.

 

[sandorski]

Dear God,

Why have you chose to smite us with this brain sucking demon?

Yours Truly
sandorski

 

[Bowfinger]

Originally posted by: ShadesOfGrey
[ ... ]
The important piece of the puzzle though is the fact that [ Zogby ] does use political affiliation to correct his sample so it better represents the US. Without it, it is meaningless and can't be extrapolated out to represent the US.

That's where you keep missing the boat. Zogby weights by political affiliation for election polls. That's appropriate since the demographics of people who vote in a typical election do not match those of the adult population as a whole. Your OP was not an election poll, however, so your incessant yammering about biased samples and Zogby is both ignorant and irrelevant.

That's the fact you refuse to accept.

 

[ShadesOfGrey]

Originally posted by: Bowfinger

Originally posted by: ShadesOfGrey
[ ... ]
The important piece of the puzzle though is the fact that [ Zogby ] does use political affiliation to correct his sample so it better represents the US. Without it, it is meaningless and can't be extrapolated out to represent the US.

That's where you keep missing the boat. Zogby weights by political affiliation for election polls. That's appropriate since the demographics of people who vote in a typical election do not match those of the adult population as a whole. Your OP was not an election poll, however, so your incessant yammering about biased samples and Zogby is both ignorant and irrelevant.

That's the fact you refuse to accept.

It doesn't change the fact that it's a political poll and should be treated and weighted accordingly. But ofcourse you are too ignorant or rather beligerant to accept that. Since no one seems to want to answer my question about whether or not they think a 24% vs 35% breakdown is acceptable, I'll mark it down as yet another dodge.
Have fun wallowing in your own ignorance and stupidity.:laugh:

 

[smack Down]

Originally posted by: ShadesOfGrey

Originally posted by: Bowfinger

Originally posted by: ShadesOfGrey
[ ... ]
The important piece of the puzzle though is the fact that [ Zogby ] does use political affiliation to correct his sample so it better represents the US. Without it, it is meaningless and can't be extrapolated out to represent the US.

That's where you keep missing the boat. Zogby weights by political affiliation for election polls. That's appropriate since the demographics of people who vote in a typical election do not match those of the adult population as a whole. Your OP was not an election poll, however, so your incessant yammering about biased samples and Zogby is both ignorant and irrelevant.

That's the fact you refuse to accept.

It doesn't change the fact that it's a political poll and should be treated and weighted accordingly. But ofcourse you are too ignorant or rather beligerant to accept that. Since no one seems to want to answer my question about whether or not they think a 24% vs 35% breakdown is acceptable, I'll mark it down as yet another dodge.
Have fun wallowing in your own ignorance and stupidity.:laugh:

There is a big difference between a poll trying to predict an election and a poll measuring the populations opionion. The first does depended alot on what precent of the voters for each party turns out and is key to predicting the outcome. If the poll results show a 24% to 35% break down then that is what the americian population is until proven otherwise.

 

[Bowfinger]

Originally posted by: ShadesOfGrey

Originally posted by: Bowfinger

Originally posted by: ShadesOfGrey
[ ... ]
The important piece of the puzzle though is the fact that [ Zogby ] does use political affiliation to correct his sample so it better represents the US. Without it, it is meaningless and can't be extrapolated out to represent the US.

That's where you keep missing the boat. Zogby weights by political affiliation for election polls. That's appropriate since the demographics of people who vote in a typical election do not match those of the adult population as a whole. Your OP was not an election poll, however, so your incessant yammering about biased samples and Zogby is both ignorant and irrelevant.

That's the fact you refuse to accept.

It doesn't change the fact that it's a political poll and should be treated and weighted accordingly. But ofcourse you are too ignorant or rather beligerant to accept that. Since no one seems to want to answer my question about whether or not they think a 24% vs 35% breakdown is acceptable, I'll mark it down as yet another dodge.
Have fun wallowing in your own ignorance and stupidity.:laugh:

Speaking of ignorance and stupidity, you remain unwilling to acknowledge that it is merely your own ignorant, partisan OPINION that political weighting was relevant to this poll. It is simply, inarguably NOT fact, no matter how many hundred times you bleat otherwise. You should give your over-inflated ego a rest before your head explodes. The Secret Service will undoubtedly frown on you rupturing Bush's lower GI tract.

 

[shira]

I've been avoiding this thread since my earlier posts because SOG's follow-ups have been so utterly inane as to prove that any rational discussion was useless. Against my better judgement, I make one final appeal to rationality.

To sum up what has come before: SOG seems to think that because the CBS poll was a "political poll" (whatever that means), the percentages of the political parties are the most important things to match actual the real-world percentages. And SOG seems to think that if the poll-percentages of political parties are off from their actual percentages in the general voting population, the poll results are questionable.

I think I've summed up SOG's essential argument in those two sentences. And refuting that argument is as simple as providing a simple counterexample:

Let's suppose the "true" voting percentages are 39, 35, and 24 % (Dems, Repubs, and Indies, respectively). So an intrepid pollster randomly calls 1000 registered-voter households and (joy of joys) gets 390 dems, 350 republicans, and 240 independents. He administers the poll and (since the political percentages are "perfect") has no reason to "weight" the responses by randomly eliminating respondants from any of the political groups. And he comes up with the following numbers:

Approve of Bush: 50%
Disapprove: 39%
No opinion: 11%

Since the results seem quite different from those obtained by other pollsters, he decides to analyze the underlying demographic data. And he finds that (horror or horrors) 100% of those polled are evangelical Christians.

Now, we all agree that there are evangelical Christians in every political group. And randomly, their numbers might be over- or under-represented in a poll. And if they are under- or over-represented, the poll results are not in general going to be as accurate as if their numbers are correctly represented (all other things being equal). So what are we to do about the data above?

According to SOG, nothing. The political percentages are perfect, so there's nothing to adjust. But we all realize that the poll results above are in fact almost meaningless, since a true cross-section of the voting population isn't even remotely represented in the demographic data.

And that, essentially, is why no single demographic statistic can ever be allowed to dominate during weighting. Of course, during a real-world poll, the underlying demographics aren't going to be off by the huge amount I've invented in my example. But there can be significant deviations between the polled population and the American voting population. And where those deviations are found, weighting occurs in an attempt to improve the estimates.

Weighting increases some groups' representation and decreases other groups' representation. The objective isn't to get any one group right, but rather to arrive at the overall optimal balance that produces the most accurate results. Since weighting is an extremely complex statistical process, it would be extremely difficult for anyone - without the benefit of seeing the full, pre-weighted numbers for ALL demographic groups (by age, by sex and sexual preference, by marital status, by income, by education, by religious views, and (yes) by political party) to determine what the "right" percentages are.

Thus, SOG has absolutely no basis for concluding that the final, weighted percentages used by CBS aren't the optimal ones.

 

[Tab]

Maybe, I should re-read the thread...

How does CBS figure out the "best" way to weight the polls exactly?

I am kind of confused...

 

[shira]

Originally posted by: Tab
Maybe, I should re-read the thread...

How does CBS figure out the "best" way to weight the polls exactly?

I am kind of confused...

I would give you a one paragraph answer. But really, taking a good course in factor analysis and population statistics would provide a much more complete answer.

 

[ShadesOfGrey]

Originally posted by: shira
I've been avoiding this thread since my earlier posts because SOG's follow-ups have been so utterly inane as to prove that any rational discussion was useless. Against my better judgement, I make one final appeal to rationality.

To sum up what has come before: SOG seems to think that because the CBS poll was a "political poll" (whatever that means), the percentages of the political parties are the most important things to match actual the real-world percentages. And SOG seems to think that if the poll-percentages of political parties are off from their actual percentages in the general voting population, the poll results are questionable.

I think I've summed up SOG's essential argument in those two sentences. And refuting that argument is as simple as providing a simple counterexample:

Let's suppose the "true" voting percentages are 39, 35, and 24 % (Dems, Repubs, and Indies, respectively). So an intrepid pollster randomly calls 1000 registered-voter households and (joy of joys) gets 390 dems, 350 republicans, and 240 independents. He administers the poll and (since the political percentages are "perfect") has no reason to "weight" the responses by randomly eliminating respondants from any of the political groups. And he comes up with the following numbers:

Approve of Bush: 50%
Disapprove: 39%
No opinion: 11%

Since the results seem quite different from those obtained by other pollsters, he decides to analyze the underlying demographic data. And he finds that (horror or horrors) 100% of those polled are evangelical Christians.

Now, we all agree that there are evangelical Christians in every political group. And randomly, their numbers might be over- or under-represented in a poll. And if they are under- or over-represented, the poll results are not in general going to be as accurate as if their numbers are correctly represented (all other things being equal). So what are we to do about the data above?

According to SOG, nothing. The political percentages are perfect, so there's nothing to adjust. But we all realize that the poll results above are in fact almost meaningless, since a true cross-section of the voting population isn't even remotely represented in the demographic data.

And that, essentially, is why no single demographic statistic can ever be allowed to dominate during weighting. Of course, during a real-world poll, the underlying demographics aren't going to be off by the huge amount I've invented in my example. But there can be significant deviations between the polled population and the American voting population. And where those deviations are found, weighting occurs in an attempt to improve the estimates.

Weighting increases some groups' representation and decreases other groups' representation. The objective isn't to get any one group right, but rather to arrive at the overall optimal balance that produces the most accurate results. Since weighting is an extremely complex statistical process, it would be extremely difficult for anyone - without the benefit of seeing the full, pre-weighted numbers for ALL demographic groups (by age, by sex and sexual preference, by marital status, by income, by education, by religious views, and (yes) by political party) to determine what the "right" percentages are.

Thus, SOG has absolutely no basis for concluding that the final, weighted percentages used by CBS aren't the optimal ones.

:roll: No where did I suggest that they should only use political affiliation. It should however be part of the demographics they use when weighting a POLITICAL POLL.

As you have again pointed out, not factoring key demographics can mess with your poll. Lets take a religious type poll for example. The question is about what people think of the Pope. The imaginary weighted results show that the pope is liked by 30% of the respondents. But if you looked at the weighted sample used in the poll you saw that the sample had a big percentage of athiests in it although it corrected for all the other normal demographics like race, age, etc. Would it not be correct to think that including a person's religion might have something to do with their response? Ofcourse it would and it would be insanely stupid to not to consider that part of the demographics correction process.

So while you continue to misrepresent what I have been saying and think I only want to use political affiliation, I do not think it is the end all or be all of a poll. But it is a very vital part if you want it to reflect society's political opinions which this poll purports to do.

I guess you'll just have to join bowfinger wallowing around in your own arrogance and ignorance.

 

[Tab]

Originally posted by: shira

Originally posted by: Tab
Maybe, I should re-read the thread...

How does CBS figure out the "best" way to weight the polls exactly?

I am kind of confused...

I would give you a one paragraph answer. But really, taking a good course in factor analysis and population statistics would provide a much more complete answer.

I actually may take STAT 330 next semster...

 

[totalcommand]

Originally posted by: Tab

Originally posted by: shira

Originally posted by: Tab
Maybe, I should re-read the thread...

How does CBS figure out the "best" way to weight the polls exactly?

I am kind of confused...

I would give you a one paragraph answer. But really, taking a good course in factor analysis and population statistics would provide a much more complete answer.

I actually may take STAT 330 next semster...

U.S. Census data is used, and the population of various groups, classified by race, age, sex, etc. is used to make weighted adjustments in the sample.

 

[totalcommand]

Originally posted by: Bowfinger

Originally posted by: ShadesOfGrey
[ ... ]
The important piece of the puzzle though is the fact that [ Zogby ] does use political affiliation to correct his sample so it better represents the US. Without it, it is meaningless and can't be extrapolated out to represent the US.

That's where you keep missing the boat. Zogby weights by political affiliation for election polls. That's appropriate since the demographics of people who vote in a typical election do not match those of the adult population as a whole. Your OP was not an election poll, however, so your incessant yammering about biased samples and Zogby is both ignorant and irrelevant.

That's the fact you refuse to accept.

Not to mention, a FoxNews poll, even with different political ID, gave similar results. Which proves that the CBS poll was correct in its weighting.

 

[shira]

Originally posted by: totalcommand

Originally posted by: Tab

Originally posted by: shira

Originally posted by: Tab
Maybe, I should re-read the thread...

How does CBS figure out the "best" way to weight the polls exactly?

I am kind of confused...

I would give you a one paragraph answer. But really, taking a good course in factor analysis and population statistics would provide a much more complete answer.

I actually may take STAT 330 next semster...

U.S. Census data is used, and the population of various groups, classified by race, age, sex, etc. is used to make weighted adjustments in the sample.

What makes this adjustment so complicated is that you have what is essentially a non-continuous function consisting of numerous independent or partially-independent variables (sex, income level, party affiliation, etc), the relationship of each of which to the question being asked is only incompletely understood.

If each variable represented a continuous function (for example, if there were infinite values of "party affiliation" along a continuum from "Democrat" to "Repulican"), it would be possible to compute partial derivitives for each variable and to create a correction matrix that could be applied to the raw data. That's still not a trivial task, but it would be MUCH easier than working with the discrete values of the independent variables that actual raw polling data has.

And as I wrote, each variable is not truly independent. For example, if the relationship between religious affiliation and Bush-approval were truly independent, then you'd see the same "curve" for religion-vs-approval for Democrats and for Repulicans (this assumes you could somehow hold all other variables constant, something that's impossible in practice). So the "independent" variables are only partially independent.

Add it all up and you have a real mess. People get PhD's figuring this stuff out. I kind of know how it works, based on my understanding of factor analysis in psychology research studies (which, for example, try to figure out cause-and-effect relationships between factors in childhood environment and later psychological pathologies), so I would be shocked if the mathematics used in polling isn't similar (and extremely complex).