[Tor]

REGISTER TO REMOVE THIS AD

I been watching CNN For the past two weeks. They have polls. Everytime they state the results they say "80% America hates Bush" But then at the end or at the very bottom of the t.v. they have something that says "500 people interviewed margin of error 2%".

I gotta argue, how the hell does 500 people equal 300million people in the U.S O_o It's less than .001% of the U.S Population. I hate it when they say "America thinks that 65% of blah blah is doing a bad job" and they have 100 people surveyed in this poll. WTF. Drives me nuts man.

[hazelmurphy]

There's some degree of accuracy in such polls, but only if they are taken using a fair representative of the voting base.

You can't take 500 people out of inner city Philadelphia and ask them who is going to win the presidential election, and present the answer as an accurate nationwide poll - as that wouldn't be accurate.

You can (usually) take 100 people out of Philadelphia, 100 people from Bumblefuck Montana, 100 from Crookedtooth Arkansas, 100 from Sisterbang West Virginia, and 100 from Liberalville California - ask them the same question and get a more representative answer. 500 people would be a bit small of ocurse . . .

Anyway - I rarely pay attention to political polls - there are just too many variables.

[Fipherion]

The pollers know what they're doing. Theres a lot of statistical theory around the idea of polling and I'm sure they don't just walk up to a group of people and say ask them questions. If you pick a fairly random sample is usually safe to say that their opinion is representative of the whole to within a margin of error that is calculated on various factors about the data.

The +-2% error they claim is some sort of confidence interval where you they can say with a certain amount of confidence that the REAL number is within that range. I don't know what level of confidence they usually use but its probably somewhere in the 80-95% confidence range. No matter what, there is always the chance that the actual value will be outside the range. Even with 95% confidence, 5% is not 0% so theres a fairly reasonable chance for your conclusions to be wrong. When that happens, it doesn't mean you should throw out all your faith in statistical data. It just means you didn't get a very representative sample.

In the end, its always good to question the pollers data but they're going to get it right more than they get it wrong.

[aaerdan]

Accurate as in 100% accurate or representative of the population as a whole? No.

[Avarion]

I was going to say exactly what Aaerdan said. The same goes for current UK polls; some of the statistics put the Tories 11% ahead of Labour, where others reckon that both parties are separated by only a few points. You can't survey an entire country; specific areas will lean one way where others will lean another, and some will be more mixed. Consensus is difficult to achieve where there are localised pockets of political opinion.

[bhamlaxy]

I been watching CNN For the past two weeks. They have polls. Everytime they state the results they say "80% America hates Bush" But then at the end or at the very bottom of the t.v. they have something that says "500 people interviewed margin of error 2%".

I gotta argue, how the hell does 500 people equal 300million people in the U.S O_o It's less than .001% of the U.S Population. I hate it when they say "America thinks that 65% of blah blah is doing a bad job" and they have 100 people surveyed in this poll. WTF. Drives me nuts man.

Like mentioned before, polling is a scientific and statistical process. They plug the numbers into an equation, figure out the confidence level and all that, and they calculate that they will need X number of people to get within a specific error of margin. They can get pretty accurate. I'd probably have more to contribute had I been in class a few weeks ago. In my "managing political campaigns" class we had Mark Blumenthal come in, the creator of Pollster.com and the unofficial king of political polling. I would have known a lot more about the topic had I been there.

[Moshineko]

Lucky bastard, Blumenthal is great. He's on NPR quite a bit, and always has some very interesting things to say.

Now, for polling, look at a few things:

What are the questions?
If they are worded correctly, the questions can be fairly representative of an actual view. Questions like "Is Obama a terrorist?" are not appropriate questions, nor are they usually asked by decent pollers.
It is more likely that they will ask a series of questions about the subject, none of which actually come out and ask the poll-taker to give a socially unacceptable answer, but which will tell the pollsters the true opinion of the poll-taker, anti-social or not.

For instance, rather than "Is Obama a terrorist?", one might ask "Do you associate the name 'Hussein' with negative, positive, or neutral events?", "Is a candidate's name important to their campaign?", "Is your view of the Democratic stance on Iraq favorable, unfavorable, or neutral?", and so on. Using the answers to these questions, one can come up with a set of data that supports many conclusions, most of which are likely accurate.

Which brings the second question. Who is giving the poll? What is their goal?
If you see a poll by Reuters, you're probably okay. They're known for an almost unbaised polling procedure, and have no (obvious) underlying motives for skewing data. If your poll was taken by a Christian organization, watch out for that bias. If Democratic, watch out for a bias against Republicans. In the West, a bias against Muslims, and so on. Know your source, as always.

What is the poll set?
Most polls now suffer from a poor polling set. The people who take these polls are usually elderly, as they're the only ones who will sit still long enough for a decent poll. Other than that, you have very motivated, and thus biased, subjects answering, or people who pop in on the crap, one-question, "Is Bush dumb?" online polls.
If you can find a decent number of people, and a fairly varied group, you might be okay. Don't do that, and your poll is crap. For instance, in the Spring semester at my college, we had a speaker talk to us about counselors' and psychologist's views on having an interpreter in a session, for deaf or deaf-blind patients. The conclusions were based on a polling of one hundred and eight psychologists and counselors. Immediately, it was apparent that the data had no relevance at all. Taking a percentage out of such a small group means that minorities and outliers are likely to skew your data by an unrealistic amount. If one of them had set an interpreter on fire, the data would show that 10% of counselors will set interpreters on fire. Probably not a good indicator of actual opinions.

So, the question isn't "are polls accurate", but rather "which polls are accurate". For that, you need to know a bit about the poll and pollsters in question. It's likely that most of the polls you see are accurate for their data set, it's just that data set that's in question. Human nature dictates that we'll usually hedge our answers, so as not to be caught out in the future, or told off for an unfavorable answer, so it's unlikely that any poll will ever be truly accurate. But then, that's where the margin of error comes in, of course, and the skill of the pollsters.

[bhamlaxy]

Lucky bastard, Blumenthal is great. He's on NPR quite a bit, and always has some very interesting things to say.

Dammit! Now I'm even more mad I missed him. I am FURIOUS I had to miss this weeks speaker, Joe Trippi, author of one of my favorite books (The Revolution Will Not Be Televised) and the campaign manager for Dean.

I agree with everything you said. I saw an interesting study about the wording of questions, where the same question asked in even minutely different ways would lead to dramatically different outcomes. Additionally, on the poll set stuff, its so hard to get a solid representation of America. Phone calls to people? Heavily skewed because certain types of people who tend to hold certain beliefs might just hang up. The only people who can forcibly poll you is the government. I was called in by the residence director of my school saying I was part of a randomly chosen group for a census-related poll thing, and if I didn't complete it we would both go to jail. I doubt the jail part, but we were still forced by law. The poll set is such a hard thing to get right, most polls are always going to be skewed to an extent.

[Houman]

no poll is ever correct

[meimei]

Originally Posted by Houman

no poll is ever correct

They won't ever be 100% correct, but they can be pretty close, and give an idea of what the public opinion about a certain topic is.

[zuesRenzo]

This really is not a debate, polls are never accurate. There is no way to understand how the entire population feels, off the opinion of 500 or 1000 people. Theoretically it is possible to get 500 people who are all bush supporter.

[bhamlaxy]

This really is not a debate, polls are never accurate. There is no way to understand how the entire population feels, off the opinion of 500 or 1000 people. Theoretically it is possible to get 500 people who are all bush supporter.

I think polls are very accurate because statistically they can poll the right amount of people, and then determine the margin of error using a formula. I don't know how I would calculate the probability of getting 500 Bush supporters, but with current polling methods it would be extremely improbable.

[zuesRenzo]

It is very unlikely but nevertheless possible. Polls do tend to show the data correctly when the
topic being contested is very clear cut. However polls are not only less accurate when the debates are close(which accounts for many of those "surprise" victories, but polls also tend to sway the opinions of voters. There are some people who will not vote for say a candidate people everyone will vote for him, and those people unknowingly might add up to sway the entire decision.

[Moshineko]

Actually, polls are much more accurate when the answers are shades of gray.

For instance, "Are you for abortion?" Is a very bad poll question. It has no wiggle room, and doesn't reflect the reality of the question. Thus, people who have varied opinions on the matter have only two choices, yes or no.

"Is abortion permissible in the event birth will kill the mother?", "Is third-term abortion ever justified?", "Should children of rape be born no matter what?" Are much more open questions, and are used to create a broad set of data from which conclusions are drawn.

Politically, they are accurate in the moment. For instance, a poll today has no bearing if, tomorrow, we find out Obama's family was secretly smuggling Jews' gold teeth out of Germany during WWII, or that McCain is an evil cyborg. We already know both those about Hillary, so she wouldn't be affected.

Polls themselves don't sway, it's the conclusions that are drawn from the polls. Statistics can be made to say anything.