Obama '08?

joefromchicago wrote:

Thomas wrote:
That wasn't the question though. The question was whether including cellphone-only households would make the poll more representative of the general population. Comparing the two polls, and finding no difference after correcting for age, is good reason to think that the answer is "no".

That's only true if one accepts that the second poll is representative of its population. But to test the representativeness of one poll by looking at another poll, the representativeness of which is merely assumed, is question-begging.

Huh?

The question at hand was whether the first poll - and by extension, polls in general - are representative since they do not include cellphone-only households.

A poll that surveys cellphone-only households, and finds that if they were included in an overall poll in proportional numbers, they would not make a significant difference, resolves that question.

Yes, one could still maintain that all polls are unrepresentative - including, then, by definition the second as well as the first one - for some other reason. Of course one would have to actually bring an argument forth about what that reason would be.

The argument that it is the exclusion of cellphone-households, specifically, that makes polls unrepresentative is void when it is found that a poll that includes them is as representative - or as unrepresentative if you will - as one that excludes them.

0 Replies

JPB wrote:

The compounding of the error rates need to be taken into consideration. Most survey results publish a margin of error based on the sample size of the survey. Large national polls usually publish a margin of error of +/- 3%. Smaller surveys typically have a 5% error band. These numbers indicate with 95% confidence that the result is within the margin of error. The 5% associated error rate is seldom discussed (they intend to be wrong 5% of the time). 99% confidence intervals are almost never calculated for political surveys because they require a much larger sample size to stay within the same margin or error. The cost of being wrong (within a given margin of error) is greater as the confidence of the estimate increases. Another way to look at survey results with a 3% margin of error is to say that the pollsters are 95% confident that the actual result is somewhere within 3% of a given number.

Absolutely. Thats why it is unwise to base any evaluation of current political standings on any one given poll. It is only when you find a trend taking hold across a series of polls, that you can say anything meaningful about what is going on. Which is why I like the pollster.com site so much, because it maps all poll results and then charts out a cautiously calculated trendline based on their running average.

0 Replies

nimh wrote:

JPB wrote:
The compounding of the error rates need to be taken into consideration. Most survey results publish a margin of error based on the sample size of the survey. Large national polls usually publish a margin of error of +/- 3%. Smaller surveys typically have a 5% error band. These numbers indicate with 95% confidence that the result is within the margin of error. The 5% associated error rate is seldom discussed (they intend to be wrong 5% of the time). 99% confidence intervals are almost never calculated for political surveys because they require a much larger sample size to stay within the same margin or error. The cost of being wrong (within a given margin of error) is greater as the confidence of the estimate increases. Another way to look at survey results with a 3% margin of error is to say that the pollsters are 95% confident that the actual result is somewhere within 3% of a given number.

Absolutely. Thats why it is unwise to base any evaluation of current political standings on any one given poll. It is only when you find a trend taking hold across a series of polls, that you can say anything meaningful about what is going on. Which is why I like the pollster.com site so much, because it maps all poll results and then charts out a cautiously calculated trendline based on their running average.

Wow, I had just written a post about how much I liked Pollster.com!

Politicalarithmetik.blogspot.com is another good one.

Cycloptichorn

0 Replies

JPB wrote:

The compounding of the error rates need to be taken into consideration. Most survey results publish a margin of error based on the sample size of the survey. Large national polls usually publish a margin of error of +/- 3%. Smaller surveys typically have a 5% error band. These numbers indicate with 95% confidence that the result is within the margin of error. The 5% associated error rate is seldom discussed (they intend to be wrong 5% of the time). 99% confidence intervals are almost never calculated for political surveys because they require a much larger sample size to stay within the same margin or error. The cost of being wrong (within a given margin of error) is greater as the confidence of the estimate increases. Another way to look at survey results with a 3% margin of error is to say that the pollsters are 95% confident that the actual result is somewhere within 3% of a given number.

And all of those margins of error are established by taking other polls.

Look, I'll make this simple: you can't check the accuracy of a poll by using another poll that relies on the same methodology. That's question-begging. Now, I'm not saying that polls are necessarily inaccurate (indeed, they usually are), I'm just saying that the method for confirming the accuracy of polls through the use of other polls is logically suspect.

0 Replies

nimh wrote:

Huh?

The question at hand was whether the first poll - and by extension, polls in general - are representative since they do not include cellphone-only households.

A poll that surveys cellphone-only households, and finds that if they were included in an overall poll in proportional numbers, they would not make a significant difference, resolves that question.

No it doesn't, it just adds another poll that might be equally flawed.

nimh wrote:

Yes, one could still maintain that all polls are unrepresentative - including, then, by definition the second as well as the first one - for some other reason. Of course one would have to actually bring an argument forth about what that reason would be.

The reason is that it begs the question.

nimh wrote:

The argument that it is the exclusion of cellphone-households, specifically, that makes polls unrepresentative is void when it is found that a poll that includes them is as representative - or as unrepresentative if you will - as one that excludes them.

You'd be satisfied if both polls were equally unrepresentative? That's a rather odd position to take.

0 Replies

I think the veracity of polls must wait until the real event takes place. Trends are good to know, but when something dramatic changes from previous polls, we must consider how those new changes may affect future polls.

Let's wait; it'll tell us what we want to know. All else is conjecture.

0 Replies

joefromchicago wrote:

JPB wrote:
The compounding of the error rates need to be taken into consideration. Most survey results publish a margin of error based on the sample size of the survey. Large national polls usually publish a margin of error of +/- 3%. Smaller surveys typically have a 5% error band. These numbers indicate with 95% confidence that the result is within the margin of error. The 5% associated error rate is seldom discussed (they intend to be wrong 5% of the time). 99% confidence intervals are almost never calculated for political surveys because they require a much larger sample size to stay within the same margin or error. The cost of being wrong (within a given margin of error) is greater as the confidence of the estimate increases. Another way to look at survey results with a 3% margin of error is to say that the pollsters are 95% confident that the actual result is somewhere within 3% of a given number.

And all of those margins of error are established by taking other polls.

Look, I'll make this simple: you can't check the accuracy of a poll by using another poll that relies on the same methodology. That's question-begging. Now, I'm not saying that polls are necessarily inaccurate (indeed, they usually are), I'm just saying that the method for confirming the accuracy of polls through the use of other polls is logically suspect.

Not exactly. It depends on if they are claiming significance or presenting confidence limits on their outcomes. In the straight representation of point estimates culminating in a difference of -2 to 2 on the 46 items, each point estimate will have it's own associated error and the errors are, in fact, compounded. There are statistical methods that calculate an overall confidence interval when comparing two proportions. You're correct in stating that that was not done here and we have no idea what the upper limit of the 95% CIs are for these data. I could calculate them, but I'm not that interested in doing the researcher's work for them. What they are presenting as -2 to 2 percent differences could easily be five times that, depending on the samples.

One thing I found interesting in the article was

Quote:

While they found statistically significant differences between landline and cell-phone-only respondents, once they weighted the landline sample to Census estimates for age and other demographics, they found that "none of the measures would change by more than 2 percentage points when the cell-only respondents were blended into the landline sample."

In fact, 41 of the 46 measures changed either not at all or by only a single percentage point once the cell-phone-only samples were "blended" in (these included party identification and generic vote preference for the U.S. House). The five measures that changed by 2 percentage points are noteworthy:

Notice that they indicate that significance testing was performed on the unblended samples and that significant differences were found. They then switch to discussing a flat percentage of 2% without any indication of confidence intervals or significance testing. I believe I've mentioned elsewhere that expressing data as point estimates without confidence intervals is misleading in that there is no representation of the upper limit of the estimate.

0 Replies

nimh wrote:

Huh?

The question at hand was whether the first poll - and by extension, polls in general - are representative since they do not include cellphone-only households.

In that case, Joe is right and you are begging the question after all.

nimh wrote:

A poll that surveys cellphone-only households, and finds that if they were included in an overall poll in proportional numbers, they would not make a significant difference, resolves that question.

Only if you can take for granted that your reference poll is representative.

nimh wrote:

Yes, one could still maintain that all polls are unrepresentative - including, then, by definition the second as well as the first one - for some other reason. Of course one would have to actually bring an argument forth about what that reason would be.

Only if the burden of proof is on the skeptics about the reference poll. If it isn't, one doesn't have to -- and the faithful in the reference poll are begging the question.

nimh wrote:

The argument that it is the exclusion of cellphone-households, specifically, that makes polls unrepresentative is void when it is found that a poll that includes them is as representative - or as unrepresentative if you will - as one that excludes them.

I agree; that's what I said in the first place. But now I'm confused: what was your "huh?" about?

0 Replies

joefromchicago wrote:

nimh wrote:
The argument that it is the exclusion of cellphone-households, specifically, that makes polls unrepresentative is void when it is found that a poll that includes them is as representative - or as unrepresentative if you will - as one that excludes them.

You'd be satisfied if both polls were equally unrepresentative? That's a rather odd position to take.

Not really. I think you sort of jumped into the conversation midway, but we were not actually discussing the overall merits of either poll, or polls in general - but whether it was missing out on the cellphone-only households that made polls unreliable.

If two polls, done according to identical methodology, with the only difference being that one reaches a large cross-section of cellphone-only households and the other doesnt, yield roughly the same results, one can take this as a piece of evidence that the cellphone-only household argument, in any case, does not make much difference.

You can still stake out some position that both polls were unrepresentative for some other reason - I've got no idea what exactly you're actually thinking of, here, but who knows - but that then is irrelevant to the cellphone-only household question we were discussing.

0 Replies

nimh, Do you think one poll that includes cellphone users is dependable to predict future outcomes no matter what the mix between line and cellphone users?

0 Replies

cicerone imposter wrote:

nimh, Do you think one poll that includes cellphone users is dependable to predict future outcomes no matter what the mix between line and cellphone users?

No poll ever is dependable to predict future outcomes. Polls arent in the prediction business.

Edit: I mean, if you would take a long and "rich" sequence of polls, you could chart out different future scenarios based on the trends that've been shown over the course of polls so far.. but any one poll itself only ever claims to reflect what people are thinking right now.

0 Replies

I understand that one poll is never a predictor of anything; but that the sequence of polls is more reliable.

The question was, how one poll that considers both line and cellphone will be somewhat consistent with future such polls. 1) Will the ratio between the 2 make any difference? 2) Will demographics make a difference?

0 Replies

cicerone imposter wrote:

The question was, how one poll that considers both line and cellphone will be somewhat consistent with future such polls. 1) Will the ratio between the 2 make any difference? 2) Will demographics make a difference?

Thats what the article that we were discussing is all about. It explains it much better - and more carefully worded - than I could do. http://www.pollster.com/blogs/cell_phones_and_political_surv.php

0 Replies

nimh, Thanks for the link; that answers all my questions.

0 Replies

Obama, Clinton Redistribute Campaign Cash to Poor Rivals

Senators Barack Obama, D-IL, and Hillary Clinton, D-NY, today turned presidential campaigning on its head when they announced that the combined $52 million in primary campaign cash they raised in the second quarter would be redistributed to less fortunate candidates like Delaware Sen. Joe Biden, Alaska Sen. Mike Gravel, and Ohio Rep. Dennis Kucinich.

In a joint news release, Senators Obama and Clinton said, "The fundamental principles of the Democrat party say that the rich and powerful have an obligation to help the poor and downtrodden."

Former North Carolina Sen. John Edwards, not slated to benefit from the redistribution, denounced the move as a "crass political ploy which lacks the weight of traditional Democrat moral leadership, because it was voluntary."

"If Hillary and Barry really believed in our Democrat principles," Mr. Edwards said, "they wouldn't voluntarily give their money to a few poor candidates...money which was voluntarily given to them. Instead, they would introduce legislation to mandate that all presidential candidates be given an equal amount of taxpayer dollars. This could be done without raising taxes on anyone but the filthy rich, and by shifting money from the Pentgon's quagmire budget to this new program."

0 Replies

Sounds like sour grapes from Edwards. <Barry?>

I have mixed feelings on the plan. On one hand people contribute to a candidate in order to enable that candidate to further his/her campaign. I'm not sure how those folks will feel about the wealth being spread. On the other hand, it's an interesting concept within the Democratic party.

I'll be curious to see how it plays with the party loyalists and with Clinton and Obama supporters.

0 Replies

JPB wrote:

Sounds like sour grapes from Edwards. <Barry>

I have mixed feelings on the plan. On one hand people contribute to a candidate in order to enable that candidate to further his/her campaign. I'm not sure how those folks will feel about the wealth being spread. On the other hand, it's an interesting concept within the Democratic party.

I'll be curious to see how it plays with the party loyalists and with Clinton and Obama supporters.

On the Gripping hand, it's a parody piece from Scott Ott at Scrappleface.

Cycloptichorn

0 Replies

Er, I think it's satire...

0 Replies

Well, that makes more sense! That's what I get for not checking links.

0 Replies

I'm certainly glad it's satire!

I'll go ahead and run with the sentiment of it though and point out that Obama has raised money from over 250,000 donors (as compared to Hillary's ~50,000 donors). Lots of different people making donations, not just the usual suspects.

0 Replies

Related Topics

Quick Links

My Account

able2know